CHAPTER-I

OBJECTIVE AND SCOPE OF THE PROJECT

1.1 INTRODUCTION

In the early days, equipment maintenance was conducted only when

equipment actually failed. The work was more “fix it” than maintenance. Shortly
thereafter, came the recognition that performing regular maintenance and
refurbishment tasks on equipment could keep equipment operating longer between
failures. This became known, variously, as Periodic Maintenance, Calendar Based
Maintenance or Preventive Maintenance (PM). The goal was to have most of the
equipment be able to operate most of the time until the next scheduled maintenance
outage. This approach is also outdated. Now Condition monitoring has made good
progress in recent years maintenance is being carried out based on condition of
machine which reduces the cost of unnecessarily opening of equipment. Most of
the defects encountered in the rotating machinery give rise to a distinct vibration
pattern (vibration signature analysis techniques)Vibration Monitoring is the ability to
record and identify vibration “Signatures” which makes the technique so powerful
for monitoring rotating machinery. Vibration analysis is normally applied by using
transducers to measure acceleration, velocity or displacement. The choice largely
depends on the frequencies being analyzed.
Condition monitoring has made good progress in recent years in identifying
any types of deterioration in plant machinery, so that pro-active maintenance can
be performed, improving overall plant productivity. Vibrations are found almost
Everywhere in power plants. Rotating machinery vibrates due to unbalances,
misalignments and imperfect bearings; Vibration, in general, reduces equipment
life and, in extreme cases, can result in equipment damage or even catastrophic
failures. On the other hand, existence of vibration can also be used to diagnose
equipment problems and provide.

1
1.2 OBJECTIVE OF THE PROJECT
The new generation of condition monitoring and diagnostics systems differs by
the detailed solution of diagnostic problems that allows making a step from machine
vibration state monitoring to the monitoring of the machine technical condition. Most
rotating machine defects can be detected by such a system much before dangerous
situations occur.
The aim of vibration monitoring is the detection of changes in the vibration
condition of the object under investigation during its operation. The cause of such
changes is mainly the appearance of a defect. The number of such points can be reduced
to one or two for each object to be monitored if there is a common casing.
The main objective of this project is to identify the causes of significant vibrations
developed in the main pump driving end and main pump non driving end and to rectify
those vibrations by proper action and to develop an simple ANN for the fault diagnosis
boiler feed pump.
1.3 ORGANIZATION OF THE WORK
The project is organized into following activities:
Chapter I highlights the importance, objective of the study and methodology of the work.
Chapter II deals with the information from the research papers.
Chapter III gives the introduction of maintenance strategies, predictive maintenance
procedure, and condition monitoring techniques and briefly about the vibration
monitoring.
Chapter IV describes the theory regarding basics of vibration, vibration instrumentation,
and DATAPAC 1500 the instrument used for the vibration analysis.
Chapter V deals with the vibration analysis procedure and fault diagnosis of the
machinery.
Chapter VI gives the overview of the artificial neural networks, back propagation
algorithm and application ANN to vibration analysis.
Chapter VII deals with the case study of the BOILER FEED PUMP. The vibration
spectrum analysis and the experimentation and the fault diagnosis of the machine.
Chapter VIII presents the application of ANN for the fault recognition on BFP.
Chapter IX lists out the results, conclusions and future scope of the work.

2
 CHAPTER-II
LITERATURE REVIEW

2.1 INTRODUCTION
A literature survey was taken up to review present status of research in the field of
condition monitoring. Machine condition monitoring is gaining importance in industry
because of the need to increase reliability and to decrease the possibility of production
loss due to the machine breakdown. The use of vibration and acoustic emission signals is
quite common in the field of condition monitoring of rotating machinery. By comparing
the signals of a machine running in the normal and faulty conditions. Detection of faults
like unbalance, rotor rub, shaft misalignment, gear failure and bearing defects is possible.
These signals can also be used to detect the incipient failures of the machine components,
through the online monitoring system, reducing the possibility of catastrophic damage
and the down time. Some of the recent works in this area are.

R.K BISWAS, [1] Scientist and head, condition monitoring group, CMERI,
DURGAPUR. Presented paper on “Vibration based condition monitoring of rotating
machines” states that Condition Monitoring is defined as the collection, comparison and
storage of measurements defining machine condition. Almost everyone will recognize the
existence of a machine problem sooner or later. One of the objectives of Condition
Monitoring is to recognize damage that has occurred so that ample time is available to
schedule repairs with minimum disruption to operation and production. In this aspect
vibration is probably the best operating parameter to judge dynamic condition of
machines. Condition monitoring is essentially a screening process in which
measurements and other data are compared to pre-established norms for the purpose of
recognizing abnormal variations.

A machine seldom breaks down without warning. The signs of impending

breakdown are almost present long before the catastrophic failure. Vibration signals
define the dynamic property of the machine including various faults of machine like
bearing instability, unbalance, coupling misalignment, looseness, rubs, etc. Vibration

3
characteristics also define early indication of defects on components such as rolling
element bearing and gears.

M.Todd, S.D.J.McArthur, G.M.West, J.R.McDonald, S.J.Shaw. J.A.Hart [2]

paper on “the design of a decision support system for the vibration monitoring of
turbine generators” they discussed about the Condition Monitoring (CM) systems
monitor the health of expensive plant items such as turbine generators. They interpret
turbine parameters by signaling an alarm when pre-defined limits are breached. This is a
time consuming and laborious process due to the volume of data interpreted for each
alarm. In order to reduce the burden of alarm assessment, a Decision Support System
(DSS) is proposed. The DSS will feature a Routine Alarm Assessment (RAA) module
which provides an initial analysis of the alarms, highlighting those with no further
operational consequence and enabling the expert to focus on those which indicate a
genuine problem with the turbine. The implementation of an RAA prototype is discussed
along with how this will act as a foundation for a full alarm interpretation and fault
diagnostic system.

David Clifton,[3] St. Cross College, December, 2005 made research on

“Condition Monitoring of Gas-Turbine Engines” This report describes preliminary
research into condition monitoring approaches for modern gas-turbine aircraft engines,
and outlines plans for novel research to contribute to machine learning techniques in the
condition monitoring of such systems. A framework for condition monitoring of aircraft
engines is introduced, using signatures of engine vibration across a range of engine
speeds to assess engine health. Inter- and intra-engine monitoring approaches are
presented, in which a model of engine normality is constructed using vibration data from
other engines of its class, or from the test engine itself, respectively.

T.W. Verbruggen[4] a book on “Wind Turbine Operation & Maintenance

based on Condition Monitoring”. This report is part of the project entitled WT_Ω
(WT_OMEGA = Wind Turbine Operation and Maintenance based on Condition
Monitoring) which has been carried out in co-operation withLagerwey the Wind Master,
Siemens Nederland, and SKF.

4
 A Ramachandra, S B Kandagal, [5] worked on “Prediction of Defects in
Antifriction Bearings using Vibration Signal Analysis” Condition monitoring of
antifriction bearings in rotating machinery using vibration analysis is a very well
established method. It offers the advantages of reducing down time and improving
maintenance efficiency. The machine need not be stopped for diagnosis. In order to
prevent any catastrophic consequences caused by a bearing failure, bearing condition
monitoring techniques, such as, temperature monitoring, wear debris analysis, oil
analysis, vibration analysis and acoustic emission analysis have been developed to
identify existence of flaws in running bearings. Among them vibration analysis is most
commonly accepted technique due to its ease of application.

Sadettin Orhan, Nizami Aktu¨rk, Veli C¸ elik,[6] worked on “Vibration

monitoring for defect diagnosis of rolling element bearings as a predictive
maintenance tool: Comprehensive case studies” Vibration monitoring and analysis in
rotating machineries offer very important information about anamolies formed internal
structure of the machinery. In this study, the vibration monitoring and analysis case
studies were presented and examined in machineries that were running in real operating
conditions. Failures formed on the machineries in the course of time were determined in
its early stage by the spectral analysis. It was shown that the vibration analysis gets much
advantage in factories as a predictive maintenance technique.

Peter W. Hills, Mechanalysis (India) Limited, India, “A more intelligent

approach to rotating equipment monitoring”,[7]. Proactive condition management of
rotating machinery is not new to the power sector and is applied widely but with varying
degrees of success. The financial benefits have long been recognized and widely
reported, but the cost of implementation, required expertise and continuity of the systems
remain as constraints to its broader use. To date, the focus of condition monitoring of
rotating equipment has been on detecting the mechanical aspects of a machine, such as
imbalance, alignment, etc, with little attention being paid to the on-line detection of its
electrical system.

Cornelius [8], Scheffer, describes the paper on “Pump Condition Monitoring

through Vibration Analysis” It is well-known that vibration analysis is a powerful tool
5
for the condition monitoring of machinery. This especially applies to rotating equipment
such as pumps. Through the years a variety of vibration-based techniques have been
developed and refined to cost-effectively monitor pump operation and the onset of
failures. This paper is an overview of a variety of vibration-based condition monitoring
techniques for pumps. In some instances these techniques are also applicable to improve
the operation and efficiency of pumps. Specific aspects to consider when taking vibration
measurements on pumps are for instance where to take readings, which type of probe to
use, what frequency range should be used, what the settings on the analyzer should be,
etc.

Sheng Zhang, Joseph Mathew, Lin Ma, Yong Sun and Avin Mathew,[9]
presented a paper on “Statistical condition monitoring based on vibration signals”.
Designing control limits for condition monitoring is an important aspect of setting
maintenance schedules and has been virtually ignored by researchers to date. This paper
proposes a novel statistical process control tool, the Weighted Loss function CUSUM
(WLC) chart, for the detection of condition variation. The control limit was designed
using baseline condition data, where the process was fitted by an autoregressive model
and the residuals were used as the chart statistic. The condition variation is reflected by
the changes of mean and variance of the statistic’s distribution against baseline condition,
which can be detected by a single WLC chart. The approach was evaluated using a case
study which showed that the chart can detect faulty conditions as well as their severity.
The proposed approach has the advantage of requiring healthy baseline data only for the
design of condition classifiers. It is applicable in numerous practical situations where data
from faulty conditions are unavailable.

P. Caselitz, J. Giebhardt,[10], presented a paper on “Condition Monitoring

and Fault Prediction for Marine Current Turbines”. This paper introduces the
concept of condition monitoring and fault prediction for marine current turbines. It will
describe the required hardware to perform condition monitoring measurements and some
appropriate fault prediction algorithms specific for marine current turbines. Furthermore,
concepts for communication and data base handling will be introduced. For the above
mentioned items, some relevant standards and technical guidelines will be addressed.

6
 Steven M. Schultheis,[11], Charles A. Lickteig, presented a paper on
“RECIPROCATING COMPRESSOR CONDITION MONITORING”. This paper
will discuss risk-based decision making in regard to measurements and protective
functions, online versus periodic monitoring, proven and effective measurement
techniques, along with a review of both mechanical- and performance-based
measurements for assessing machine condition. Case histories will also be presented to
demonstrate some of the concepts.

Peter W. Hills,[12] Mechanalysis (India) Limited, India presented a paper on

A more intelligent approach to rotating equipment monitoring in the journal The
article is based on the paper ‘Intelligent Condition Management On-line’ The majority of
condition monitoring regimes for power plants’ rotating equipment is focused the
detection of mechanical faults, with little attention paid to electrical faults in equipment.
This could be about to change with the introduction of an on-line monitoring system that
‘learns’ to detect both types of fault.

L. B. Jack, A. K. Nandi,[13] presented a paper on “Feature Selection for ANNs

using Genetic Algorithms in Condition Monitoring”. The work presented in this paper
the work presented in this paper is based around experimental results per- formed on
vibration data taken from a small test rig which was tested with a number of
interchangeable faulty roller bearings. This is used to simulate the type of problems that
can commonly occur in rotating machinery. Rolling elements, or ball bearings, are one of
the most common components in modern rotating machinery; being able to detect
accurately the existence of a fault in a machine can be of prime importance in certain
areas of industry.

Ms S Wadhwani, Dr S P Gupta, Dr V Kumar, [14] “Wavelet Based

Vibration Monitoring for Detection of Faults in Ball Bearings of Rotating
Machines” this paper describes the application of wavelet transform (WT) for detection
of bearing damage from the vibration signal of the bearing. The wavelet transform
approach enables instant to instant observation of the contribution of different
frequency components over the full spectrum from. Actually, wavelet transform acts as a
mathematical microscope in which one can observe different parts of the signal by

7
adjusting the focus. A new technique combining the WT with neural network for
detection and classification of ball bearing fault in a three phase, 3.75 kW inductions
motor is presented in this paper. The method is tested successfully for three faulty
bearing conditions: crack in inner race, crack in outer race and defect in balls.

Zhigang TIAN, [15] “An Artificial Neural Network Approach for Remaining
Useful Life Prediction of Equipments Subject to Condition Monitoring”. Accurate
equipment remaining useful life prediction is critical to effective condition based
maintenance for improving reliability and reducing overall maintenance cost. An
artificial neural network (ANN) based method is developed for achieving more accurate
remaining useful life prediction of equipment subject to condition monitoring. The ANN
model takes the age and multiple condition monitoring measurement values at the present
and previous inspection points as the inputs, and the life percentage as the output.
Techniques are introduced to reduce the effects of the noise factors that are irrelevant to
equipment degradation. The proposed method is validated using real-world vibration
monitoring data.

N.M. ROEHL C.E. PEDREIRA" H.R. TELES DE AZEVEDO [16] presented

a paper on “Fuzzy art neural network approach for incipient Fault detection and
isolation in rotating machines”. A neural network approach for on-lie detection and
isolation of faults in rotating machines is proposed. The methodology is based on
clustering of shaft vibration monitoring data by using fuzzy art neural networks. Fault
isolation is obtained by retrieving stored associations among known physical faults and
clusters. The proposed scheme is implemented to detect and isolate different operation
modes in a hydro generator.

P.A.L. Ham, B.Sc.C.Eng..F.I.E.E. [17] “Trends and future scope in the

monitoring of large steam turbine generators” Current practices in the monitoring of
large steam turbine generators are briefly discussed, consideration being given to the
traditional range of turbine supervisory equipment, and the more extended facilities
which are sometimes now associated with rotating machinery, such as vibration
monitoring, together with the more generalized data logging systems now specified by
some Utilities. Consideration is given to the possible range of parameters and equipment

8
areas which may now be incorporated into a monitoring scheme, and attention is drawn
to the advances in display technology and operator interfaces which are now possible at
moderate cost. In a concluding section, a range of monitoring functions which could be of
wide general application in the field of steam turbine generators is discussed.

2.2 SUMMURY
This literature review presents an overview of the vibration based condition
monitoring of the rotating equipments in the thermal power plants. This literature review
also contains the review methodologies of the predictive maintenance technology.

9
 CHAPTER-III

MAINTENANCE STRATEGIES

3.1 INTRODUCTION
General industrial profitability is effected both by on stream functions and
maintenance costs. Any system must account for the optimum maintenance that can be
performed by an organizational setup. Maintenance besides trying to better its own
efficiency and mechanical performance must solve the problem of failure.

3.2 CLASSIFICATION OF MAINTENANCE STRATEGIES

Maintenance strategies are classified by three developmental stages:

1. Break down maintenance
2. Preventive maintenance

3. Predictive maintenance

3.2.1 Break Down Maintenance

This provides the replacement of defective part or machine after the machine
becomes incapable of further operation. Break down maintenance is the easiest method to
follow and it avoids the initial costs on training personnel and other related upfront costs.
Draw backs of the break down maintenance are
1. Failures are untimely.
2. Since machine is allowed to run till to failure repair is more expensive, sometimes total
replacement is required.
3. Failures may be catastrophic. Hence loss will be more.
4. Production loss will be more, as it requires more time to restore normalcy.
5. It reduces the life span of the equipment.

10
3.2.2 Preventive Maintenance

In preventive maintenance, maintenance is scheduled on calendar or hours to run

and is performed irrespective of machine conditions.
Advantages:
1. Damage to machine is less.
2. Down time of machine is reduced by 50-80%.
3. Lower expenses of overpay may same as much as 30%.
4. Increases the equipment life expectancy.
5. Reduces maintenance cost by reducing the

I. Capital spending by 10-20%.

II. Labor cost by 10%.
III. Material cost by 30%
6. Improve the employee’s safety.
7. Preventive maintenance results in a catastrophic failure and down time is required
to complete all schedule maintenance costs.
Disadvantages:
1. periodically dismantling of each and every critical machine is expensive and time
consuming.
2. It may lead to unnecessary inspections even on healthy machine also which may
further lead to more complications.
3. It is difficult to predict time interval between inspections, which ultimately may
lead to break down maintenance.

Fig: 3.1 Failure rate or bath tub curve

11
 Preventive maintenance alone cannot eliminate break down. The causes of
equipment failure change with the passage of time fig: shows the failure rate curve which
is also called as life span characteristic curve or bath tub curve. Failure rate is taken on
ordinate and time is taken on abscissa. When the equipment is new there is a high failure
rate due to design and manufacturing errors. Failure rate is increases once again since the
equipment approaches the end of its failure.
3.2.3 Predictive Maintenance
Predictive maintenance (PdM) techniques help determines the condition of in-
service equipment in order to predict when maintenance should be performed. This
approach offers cost savings over routine or time-based preventive maintenance, because
tasks are performed only when warranted.

Predictive maintenance or condition-based maintenance, attempts to evaluate the

condition of equipment by performing periodic or continuous (online) equipment
condition monitoring. The ultimate goal of PdM is to perform maintenance at a scheduled
point in time when the maintenance activity is most cost-effective and before the
equipment loses optimum performance. This is in contrast to time- and/or operation
count-based maintenance, where a piece of equipment gets maintained whether it needs it
or not.
Most PdM inspections are performed while equipment is in service, thereby
minimizing disruption of normal system operations. Adoption of PdM can result in
substantial cost savings and higher system reliability. Trending and analyzing machinery
parameters we can detect the developing problems in early stages. Hence repair works
can be carried out before failure of a machine
Advantages:
• Shut down can be done at convenient times.
• Work schedule can be prepared for mobilizing men, tools and replacement parts
before shut down reducing machinery down time.
• Identifying problem, costly trial and error procedures to solve a problem can be
avoided.
• Machine in good running condition can run continuously as long as problem
develops.

12
Disadvantages:
• Requires skilled labor.
• It is costly affair.
Predictive Maintenance Program:
For all machine common characteristic is vibrations and hence vibrations become
a powerful tool in implementing predictive maintenance program. The vibration
predictive maintenance program has four steps:
1. detection
2. analysis
3. correction
4. confirmation
Detection
First select all available critical machines in the plant. prepare a schedule for all
these machines for data collection identify bearing locations of the machine train motor
non drive end, MNDE, FNDE, FDE, PNDE, PDE, etc. identify the directions where
vibration data is collected like h, v, a etc. define which vibration parameters are to be
collected via displacement, velocity, acceleration etc. after doing all these, start collecting
vibrating data and related data and record them. Collect the data for every fortnight or
monthly or so .by trending and interpreting the data identify source of vibrations.
Analysis
After identifying the source of vibrations analyze to pin point the root cause for
vibrations. This can be achieved by eliminating process. Follow confirmative procedures
in support of analysis
Correction
Open and inspect the machine at a convenient time and make necessary corrections.
Confirmation
After corrections put the machine in service and again collect vibration data and
look for elimination of the source.

13
3.3 CONDITION MONITORING

3.3.1 Introduction

Condition monitoring pre-supposes knowledge of machines condition and its rate

of change, which can be ascertained by selecting a suitable parameter for measuring
deterioration and recording its value at intervals either on a routine or continuous basis.
This is done while the machine is running. The data obtained may then be analyzed to
give a warning on failure. This activity is called as condition monitoring.
Condition monitoring essentially involves regular inspection of equipment
using human sensory facilities and a mixture of simple aids and sophisticated instruments
The central emphasis is however on the fact that most inspections should be preferably
done while the machine is running.
Condition monitoring is concerned with the analysis and interpretation of
signals from sensors and transducers installed on operational machinery, employing
sensors positioned outside the machine, often remove from the machine components
being monitored, normally does the monitoring of a machine condition and health, using
established techniques, the analysis of information provided by the sensor output and
interpretation of the evaluated output is the needed to establish what actions to be taken.
Condition monitoring can also be a test and quality assurance, system for
continuous processes as well as discrete component manufacture. It maximizes the
performance of the company’s assets by monitoring their condition and ensuring that they
are installed and maintained correctly, it aims of detecting condition leading to
catastrophic breakdowns and loss of service, reducing maintenance overhauls, fine
turning of operating equipment increasing production and operating efficiency and
minimizing the replacement parts inventory. This is because a readily monitor able
parameter of deterioration can be found in every plant, Machinery and probabilistic
element in future prediction is highly reduced or almost eliminated thus maximizing the
items life by minimizing the effect of failure

14
3.3.2 Condition Monitoring Techniques

There are only seven main techniques of condition monitoring.

They are:
a) Visual monitoring
b) Contaminant or debris monitoring
c) Performance and behavior monitoring
d) Corrosion monitoring thermograph
e) Sound monitoring.
f) Shock pulse monitoring.
g) Vibration monitoring.

a. Visual Monitoring
Visual monitoring involves the inspections and recording of surfaces to detect
Such as surface cracks and their orientation. Oxide films, weld defects and the presence
of potential sources such as sharp notches or misalignment.

b. Contaminant Monitoring

Debris analysis is well proving in all types of industrial and works on the
principle of taking or known quantity. Sample example: a gear box, then for analyzing
the amount and type of foreign particles present in the sample. This will be show such
problems, as gear wear, to the sample detects particles of gear material .oil analysis
differs from debris analysis so for as this technique allows an assessment of the actual
condition of the oil in use. That is whether the oil quality is good enough for the
application after period of use or it is burnt or exceeded its useful use.

c. Performance and Behavior Monitoring

Performance and behavior monitoring involves checking the performance of

machine or component to see whether it is behaving correctly. Monitoring the
performance of the bearing by measuring its temperature to see whether it is carrying out
its function.

15
d. Corrosion Monitoring
Corrosion monitoring has actually applied to the fixed plane containing
aggressive material to monitor the rates of internal corrosion of walls of the equipment. It
is the system systematic measurement of corrosion or degradation of an a item of
equipment, with the aim of assisting and understanding the correct corrosion process of
obtaining information for the use of controlling corrosion

e. Thermography

Thermograph is a rapidly developing; it provides color cameras and videos,

clean indicator of heat loss, hot spot, cold spot, such as switchgear or any piece of plant
or production where temperature or its effects is important, it can be used both as
maintenance tools or a quality assurance tool. Shock pulse method is unique technique
for monitoring the true operation of the bearing by measuring the pressure wave
generated by the instantaneous mechanical impact.

f. Sound Monitoring:
Human operators are normally highly sensitive to the detection of defects as a
result of sudden change of sound due to the looseness of component results of wear or
slackening of fastening are particularly susceptible to such forms monitoring The most
widely available micro phones for sound or piezoelectric moving coils and condensers.

16
g. Vibration Monitoring:

Vibration monitoring measures the frequency and amplitude of vibrations. It is

Known that readings will change as machinery wear sets in. such readings can be
interpreted as indicators of the equipments condition, and timely maintenance actions can
be scheduled accordingly. Electrical machines and mechanical reciprocating or rotating
machines generate their own vibration signatures (patterns) during operation. However
such raw signals contain a lot of background noise, which makes it difficult or even
impossible to extract useful, precise information by simply measuring the overall signal.
It is thus necessary to develop an appropriate filter to remove the operationally and
environmentally contaminated components of signals (the background noise) so as to
reveal the clear signals generated by the events under study. To capture useful condition
monitoring data, vibration should be measured at carefully chosen points and directions.
Vibration monitoring is a well established method for determining the physical
Movements of the machine or structure due to imbalance mounting an alignment this
method can be obtained as simple. Easy to use and understand or sophisticated real time
analysis, vibration monitoring usually involves the attachment of a transducer to a
machine to record its vibration level special equipments is also available for using the
output from sensor to indicate nature vibration problem and even its precise cause.
Transducers for the measurement of vibrations employ electromagnetic
electrodynamics, capacitive, piezoelectric, or strain gauge principles out of these
piezoelectric accelerometers is most widely used since the recent past, Among the
monitoring techniques vibration monitoring as gained considerable importance because
of following fundamental factors
1) All rotation and reciprocating machines vibrate either to a smaller or greater
extent machines vibrate because of defects or incurrence in system
2) When inaccuracies or more it results in increased vibration each kind of defect
provides a vibration characterized in the unique way.
Therefore vibration characteristics reveal the health condition of machine.

17
 CHAPTER-IV

BASICS OF VIBRATION

Definition
Vibration can be defined as simply the cyclic or oscillating motion of a machine
or machine component from its position of rest. Vibration refers to mechanical
oscillations about an equilibrium point. The oscillations may be periodic such as the
motion of a pendulum or random such as the movement of a tire on a gravel road.

Fig 4.1 basic vibration representation

Vibration is occasionally "desirable". For example the motion of a tuning fork, the
reed in a woodwind instrument or harmonica, or the cone of a loudspeaker is desirable
vibration, necessary for the correct functioning of the various devices.

More often, vibration is undesirable, wasting energy and creating unwanted sound
– noise. For example, the vibration motions of engines, electric motors, or any

18
mechanical device in operation are typically unwanted. Such vibrations can be caused by
imbalances in the rotating parts, uneven friction, the meshing of gear teeth, etc. Careful
designs usually minimize unwanted vibrations.

4.1 What Causes Vibration?

Forces generated within the machine cause vibration. These forces may be one that
 Change in direction with time, such as the force generated by a rotating
unbalance.
 Change in amplitude or intensity with time, such as the unbalanced magnetic
forces generated in an induction motor due to un equal air gap between the
motor armature and stator (field).
 Result in friction between rotating and stationary machine components in
much the same way that friction from a rosined bow causes a violin string to
vibrate.
 Cause impacts, such as gear tooth contacts or the impacts generated by the
rolling elements of a bearing over flaws in the bearing raceways.
 Cause randomly generated forces such as flow turbulence in fluid handling
devices such as fans, blowers and pumps, or combustion turbulence in gas
turbines or boilers.
4.2 What is Machine Vibration?
Most of us are familiar with vibration; a vibrating object moves to and fro, back
and forth. A vibrating object oscillates. We experience many examples of vibration in our
daily lives. A pendulum set in motion vibrates. A plucked guitar string vibrates. Vehicles
driven on rough terrain vibrate, and geological activity can cause massive vibrations in
the form of earthquakes.

19
 Fig 4.2 Examples of vibration

There are various ways we can tell that something is vibrating. We can touch a
vibrating object and feel the vibration. We may also see the back-and-forth movement of
a vibrating object. Sometimes vibration creates sounds that we can hear or heat that we
can sense.
What is machine vibration? Machine vibration is simply the back and-forth
movement of machines or machine components. Any component that moves back and
forth or oscillates is vibrating. Machine vibration can take various forms. A machine
component may vibrate over large or small distances, quickly or slowly, and with or
without perceptible sound or heat. Machine vibration can often be intentionally designed
and so have a functional purpose.
At other times machine vibration can be unintended and lead to machine damage.
Most times machine vibration is unintended and undesirable. This book is about the
monitoring of undesirable machine vibration. Shown below are some examples of
undesirable machine vibration.

20
 Fig 4.3 vibrating parts
4.3 Vibration and Machine Life
Your first question may be: "Why worry about a machine's vibration?" obviously,
once a machine is started and brought into service, ft will not run indefinitely. In time, the
machine will fail due to the wear and ultimate failure of one or more of its critical
components. And, the most common component failure leading to total machine failure is
that of the machine bearings, since it is through the bearings that all machine forces are
transmitted. Of course, the next question is: "How long will be bearings last?" Although
an exact answer to this question is impossible, the manufacturers of rolling element
bearings attempt to estimate bearing life using the following calculation:

L 10 LIFE (HOURS) = 16.666/ RPM X (RATE / LOAD)3

Where: RPM = Machine rotating speed in Revolutions per Minute

RATE = the rated load capacity of the bearing (lbs.)
Load = the actual load to which the bearing is subjected.
This includes not only the static load due to the weight of the rotor, but the
dynamic load due to forces of unbalance, misalignment, etc., FORCES THAT CAUSE
VIBRATION.
According to this calculation to estimate bearing life, doubling the rotating speed
from, say 1800 RPM to 3600 RPM, would cut bearing life in half. However, by cutting
the load on the bearing by one-half would increase its service life by eight times (2-cubed

21
or 2 x 2 x 2 = 8). Of course, this estimate of bearing life does not take into consideration
other factors such as inadequate lubrication, lubricant contamination or damage from
improper storage or installation techniques.

From the above calculation, it can be seen that bearing load, including dynamic
load from vibratory sources such as unbalance and misalignment, has a significant effect
on bearing life and, ultimately, machine life. Further, the amount of vibration exhibited
by a machine is directly proportional-to the amount of force generated. In other words, if
the unbalance force is doubled, the resultant vibration amplitude will be doubled also. Or,
if the unbalance force is cut in half the unbalance -generated vibration will be cut in half
also. Therefore, the answer to the question: "Why worry about a machine's vibration?" is
simple:

1. Increased dynamic forces (loads) reduce machine life.

2. Amplitudes of machinery vibration are directly proportional to the amount of dynamic
forces (loads) generated. If you double the force, you double the Vibration.
3. Logically then, the lower the amount of generated dynamic forces, the lower the levels
of machinery vibration and the longer the machine will perform before failure
It's that simple. Low levels of vibration indicate low vibratory forces which, in turn,
results in improved machine life.
With few exceptions, when the condition of a machine deteriorates, one of two possibly
both things will generally happen:
 The dynamic forces generated by the machine will increase in intensity, causing
an increase in machine vibration. Wear, corrosion or a build-up of deposits on the
rotor may increase unbalance forces. Settling of the foundation may increase
misalignment forces or cause distortion, piping strains, etc.
 The physical integrity (stiffness) of the machine will be reduced, causing an
increase in machine vibration.
 Loosening or stretching of mounting bolts, a broken weld, a crack in the
foundation, deterioration of the grouting, increased bearing clearance through
wear or a rotor loose on its shaft will result in reduced stiffness to control even
normal dynamic forces, Thus, it should be obvious that an increase in machinery

22
 vibration is a positive indicator of developing problems. In addition, each
mechanical or operational problem generates vibration in its own unique way. As
a result, it is also possible to identify the specific nature of the problem by simply
measuring and noting its vibration characteristics. The techniques of identifying
specific defects and problems are presented in the section on VIBRATION
ANALYSIS.

4.4 Characteristics of Vibration

Whenever vibration occurs, there are actually four forces involved that
determine the characteristics of the vibration. These forces are:
• The exciting force, such as unbalance or misalignment.
• The mass of the vibrating system, denoted by M.
• The stiffness of the vibrating system, denoted by the symbol K.
• The damping characteristics of the vibrating system, denoted by the symbol C.
The exiting force is trying to cause vibration, where as the stiffness, mass and damping
forces are trying to oppose the exiting force and control or minimize the vibration.

The characteristics needed to define the vibration include:

• Frequency
• Displacement
• Velocity
• Acceleration
• Spike energy
• Phase

4.4.1 Vibration Frequency

The amount of time required to complete one full cycle of the vibration is called
the period of the vibration. If, for example, the machine completes one full cycle of
vibration in 1/60th of a second, the period of vibration is said to be 1/60th of a second.

Although the period of the vibration is a simple and meaningful characteristic, a

characteristic of equal simplicity but more meaningful is the vibration frequency.

23
Vibration frequency is simply a measure of the number of complete cycles that occur in a
specified period of time such as "cycles-per-second" (CPS) or "cycles-per-minute"
(CPM). Frequency is related to the period of vibration by this simple formula:

Frequency = 1/Period
In other words, the frequency of a vibration is simply the "inverse" of the period of the
vibration. Thus, at the period or time required to complete once cycle is 1 / 60th of a
second, then the frequency of the vibration would be 60 cycles-per-second or 60 CPS.
Given a frequency expressed in Hz, you can convert it to CPM:
CPM = Hertz x 60 Seconds/Minute
Given a frequency expressed in CPM, you can convert it to Hz:
Hertz = CPM/60 Seconds/Minute
Significance of Vibration Frequency
There are literally hundreds of specific mechanical and operational problems that
can cause a machine to exhibit excessive vibration. Obviously, when a vibration problem
exists, a detailed analysis of the vibration should be performed to identify or pinpoint the
specific cause. This is where knowing the frequency of vibration is most important.
Vibration frequency is an analysis or diagnostic tool.
The forces that cause vibration are usually generated through the rotating motion
of the machine’s parts. Because these forces change in direction or amplitude according
to the rotational speed (RPM) of the machine components, it follows that most vibration
problems will have frequencies that are directly related to the rotational speeds.
To illustrate the importance of vibration frequency, assume that a machine,
consisting of a fan operating at 2400 RPM and belt driven by a motor operating at 3600
RPM, is vibrating excessively at a measured frequency of 2400 CPM (1 x fan RPM), this
clearly indicates that the fan is the source of the vibration and not the motor or belts.
Knowing this simple fact has eliminated literally hundreds of other possible causes of
vibration.
Predominant Frequency: Predominant frequency is the frequency of vibration having
the highest amplitude or magnitude.
Synchronous Frequency: Synchronous frequency is the vibration frequency that occurs
at 1 x RPM.

24
Sub synchronous Frequency: Sub synchronous frequency is vibration occurring at a
frequency below 1 x RPM. A vibration that occurs at 1/2 x RPM would be called a Sub
synchronous frequency.
Fundamental Frequency: Fundamental frequency is the lowest or first frequency
normally associated with a particular problem or cause. For example, the product of the
number of teeth on a gear times the RPM of the gear would be the fundamental gear-
mesh frequency. On the other hand, coupling misalignment can generate vibration at
frequencies of 1 x, 2x and sometimes 3 x RPM. In this case, 1 x RPM would be called the
fundamental frequency.
Harmonic Frequency: A harmonic is a frequency that is an exact, whole number
multiple of a fundamental frequency. For example, a vibration that occurs at a frequency
of two times the fundamental gear mesh frequency would be called the second harmonic
of gear mesh frequency. A vibration at 2 x RPM due to, say, misalignment, would be
referred to as the second harmonic of the running speed frequency (1 x RPM).
Order Frequency: An order frequency is the same as a harmonic frequency.
Sub harmonic Frequency: A sub harmonic frequency is an exact submultiples (1/ 2, 1/3,
1/4, etc.) of a fundamental frequency. For example, a vibration with a frequency of
exactly 1/2 the fundamental gear-mesh frequency would be called a sub harmonic of the
gear mesh frequency. Vibration at frequencies of exactly 1/2, 1/3 or 1/4 of the rotating
speed (1 x RPM} frequency would also be called . Sub harmonic frequencies; and these
can also be called Sub synchronous frequencies. However, not all Sub synchronous
frequencies are sub harmonics. For example, a vibration with a frequency of 43% of the
running speed (1 x RPM) frequency is a Sub synchronous frequency but it is not a sub
harmonic.

4.4.2 Vibration Amplitude

As mentioned earlier, vibration frequency is a diagnostic tool, needed to help

identify or pinpoint specific mechanical or operational problems. Whether or not a
vibration frequency analysis is necessary, depends on how "rough" the machine is
shaking. If the machine is operating smoothly, knowing the frequency or frequencies of
vibration present is not important. The magnitude of vibration or how rough or smooth
the machine vibration is, is expressed by its vibration amplitude. Vibration amplitude can
25
be measured and expressed as:
Displacement
Velocity
Acceleration
SPIKE ENERGY
Vibration Displacement
The vibration displacement is simply the total distance traveled by the vibrating
part from one extreme limit of travel to the other extreme limit of travel. This distance is
also called the "peak-to-peak displacement". Peak-to-peak vibration displacement is
normally measured in units called mils, where one mil equals one-thousandth of an inch
(1 mil = 0.001 inch). Measured vibration amplitude of 10 mils simply-means that the
machine is vibrating a total distance of 0.010 inches peak-to-peak.
In Metric units, the peak-to-peak vibration displacement is expressed in micrometers
(sometimes called microns), where one micrometer equals one-thousandth of a millimeter
(1 micrometer = 0.001 millimeter).

Vibration Velocity

The vast majority of machine failures caused by vibration problems are fatigue
failures, & the time required to fatigue failure is determined by both how far an object is
deflected.(displacement) and the rate at which the object is deflected (frequency), of
course, displacement is simply a measure of distance traveled and frequency is a measure
of the number of times that “trip” is taken in a given period of time such as a minute or
second, if it is known how far one must travel in a given period of time, it is a simple
matter to calculate the speed or velocity required. Thus, a measure of vibration velocity is
direct measure of fatigue in short
Fatigue=displacement * frequency
Velocity=displacement *frequency
Thus: velocity=fatigue
Vibration velocity is measurement of the speed at which a machine or machine
component is moving as it undergoes oscillating motion.
Vibration velocity is expressed in inches-per-second peak (in/sec-pk) for English
units in metric units, vibration velocity is expressed in millimeters-per-second peak.
26
Vibration Acceleration

VIBRATION ACCELERATION is another important characteristic of vibration

that can be used to express the amplitude or magnitude of vibration. Technically,
acceleration is simply the rate of change of velocity. The acceleration of the weight is
maximum or at its peak value at the upper limit of travel where the velocity is zero (0).
As the velocity of the weight increases, the rate of change of velocity or acceleration
decreases. At the neutral position, the weight has reached its maximum or peak velocity
and at this point, the acceleration is zero (0). After the weight passes through the neutral
position, it must begin to slow down or "decelerate" as it approaches the lower limit of
travel. At the lower limit of travel the rate of change of velocity (acceleration) is, again, at
its peak value.
Expressed in in/sec/sec-peak or mm/sec/sec-peak.
This can also be written as;
in/sec/sec = in/sec2
Or
mm/sec/sec = mm/sec2

4.4.3 Spike Energy

When flaws or defects appear in a bearing, the resulting vibration will appear as a
series of short duration spikes or pulses such .The duration or "period" of each pulse
generated by an impact depends on the physical size of the flaw; the smaller the flaw, the
shorter the pulse period will be. As the size of the defect increases, the period of the pulse
becomes longer. A short-term (40 millisecond sec) time waveform that was taken on a
ball bearing with a small nick purposefully ground on the bearing inner race way. It can
be seen that the pulse period lasts only a few microseconds (1 microsecond = 1 millionth
of a second). Of course, if the period of a vibration signals is-known, the frequency of the
vibration can be found by simply taking the inverse of the period. For example, if it takes
1/3600 minute to complete one cycle of a vibration, then the vibration frequency is 3600
cycles per minute (CPM) or the inverse of the period.
In the case of the pulses generated by the bearing defects, since the pulse periods
are so short, the period inverses (frequencies) are typically very high. To illustrate, a

27
MICRO-FLAW is generally defined as a defect that is so small that it is essentially
invisible to the naked eye. The pulses generated by a micro-flaw are typically less than 10
micro-seconds (i.e. 10 millionths of a second). By taking the inverse of a 10 micro-
second pulse, the fundamental frequency becomes 100,000 Hz (TOOK Hz) or 6,000,000
CPM. As bearing deterioration progresses, the flaw gets larger. The next stage is a
MACRO-FLAW or one that is detectable with the naked eye. Since the macro-flaw is
larger, the duration or period of the pulse generated is longer and, thus, the fundamental
pulse frequency is lower. Typically, a macro-flaw will generate a pulse with a period
exceeding 20 microseconds, resulting in a fundamental pulse frequency of 50K Hz
(3,000,000 CPM) or less. Of course, as the bearing defects continue to increase in size,
the resultant pulse periods become even longer resulting in a decrease in fundamental
pulse frequency. Experimentation has revealed that by the time the fundamental pulse
frequency has reduced to approximately 5k Hz (300,000 CPM), bearing deterioration has
generally reached severe levels.
With the above facts in mind, the following outlines the basic features of the
SPIKE ENERGY (abbreviated gSE) approach developed by 1RD Mechanalysis
1. Since the frequencies of bearing vibration are very high, utilize a vibration acceleration
signal from an accelerometer transducer. Vibration acceleration tends to emphasize higher
frequencies as shown by the comparison in Figure.
2. Incorporate a "band-pass" frequency filter that will electronically filter out frequencies
above 50K Hz (3,000,000-GPM) -and below 5K Hz (300,000 CPM). By eliminating
frequencies above 50K Hz/, micro-flaws, defects that are undetectable with the naked
eye, will not affect the measurement. In other words, when the SPIKE ENERGY (gSE)
measurements reveal a significant increase, a visual inspection of the bearing should
provide confirmation with a visible flaw. For most predictive maintenance programs,
detecting micro-flaws is of little concern since deterioration to the macro-flaw stage may
take several months.
The lower cut-off frequency of 5K Hz (300,000 CPM) filters out or ignores most
other inherent sources of vibration including unbalance, misalignment, aerodynamic and
hydraulic pulsations, electrical frequencies, etc., that tend to dominate or "hide" the
vibration from bearing defects.

28
3. Since the spike-pulse signals generated by bearing defects have very low RMS values,
incorporate a true peak-to-peak detecting circuit instead of an RMS detecting circuit.
4.4.4 Phase
Phase, with regards to machinery vibration, is often defined as "the position of a
vibrating part at a given instant with reference to a fixed point or another vibrating part.
Another definition of phase is: "that part of a vibration cycle through which one part or
object has moved relative to another part".
The concept of "phase" is often the most confusing to newcomers to the field of vibration
detection and analysis; however, from a practical standpoint, phase is simply a convenient
mean of determining the "relative motion" of two or parts of a machine or vibrating
system. The units of phase are degrees, where one complete cycle of vibration equals
360 degrees.

4.5. INSTRUMENTS FOR VIBRATION DETECTION AND

ANALYSIS
4.5.1 Introduction
Instruments for measuring and analyzing machinery vibration are available in a
wide array of features and capabilities, but are generally categorized as:
1. Vibration meters.
2. Vibration frequency analyzers
4.5.2 The Vibration Transducer
Regardless of the vibration instrument being used, the "heart" of every instrument
is the vibration transducer. This is the device that is held or attached to the machine to
convert the machine's mechanical vibration into an electrical signal that can be processed
by the associated instrument into measurable characteristics of vibration amplitude,
frequency and phase. Many different varieties of vibration transducers have been used
over the years. However, with few exceptions, the transducer provided as standard with
nearly all present-day vibration meters, analyzers and data collectors is the vibration
accelerometer.
An accelerometer is a self-generating device that produces a voltage output
proportional to vibration acceleration (G's). The amount of voltage generated per unit of

29
vibration acceleration (G) is called the sensitivity of the accelerometer and is normally
expressed in milli volts-per-G (mv/G), where 1 milli volt equals one-thousandth of a volt
(1 mv = 0.001 volt). Accelerometers are available with sensitivities ranging from less
than 1 mv/G to 10,000 mv/G; however, most accelerometers for general purpose
vibration detection and analysis applications will have sensitivities ranging from 10 to
100 mv/G.

Fig: 4.3 Basic construction of an accelerometer

Theory of Operation
Figure 4.1 shows a simplified diagram of typical accelerometer construction. The
component of the accelerometer that generates the electrical signal is called a
"piezoelectric" element. A piezoelectric material is a non-conducting crystal that
generates an electrical charge when mechanically stressed or "squeezed". The greater the
applied stress or force, the greater the generated electrical charge.
Many natural and man-made crystals have piezoelectric properties. There are also
a number of ceramic (polycrystalline) materials which can be given piezoelectric
properties by the addition of certain impurities and by suitable processing. These are
called "Ferro-electric" materials. Most commercially available accelerometers used today
incorporate Ferro-electric materials because they can be fabricated in a variety of shapes
and their piezoelectric properties can be controlled more easily than crystals to suit many
applications.
Referring to the diagram in Figure 4-1 the accelerometer consists of a mass
(usually a stainless steel disk) compressed against a "stack" of piezoelectric disks. The
size and number of piezoelectric disks used in an accelerometer determines not only its

30
sensitivity (mv/G), but its usable frequency range as well. When the accelerometer is held
or attached lo a vibrating object, the piezoelectric elements will be subjected to resultant
"inertia" forces of the mass. Thus, a forces proportional to the vibration acceleration is
applied to the piezoelectric elements, resulting in an electrical charge signal proportional
to vibration acceleration.
The operation of an accelerometer used for measuring and analyzing machinery
vibration is exactly the same as that of a ceramic cartridge used on phonographs and
record players, where the vibration of a phonograph needle riding in the grooves of a
record is converted to an equivalent electrical signal.
The amount of electrical signal generated by the piezoelectric element is
relatively small and many times must be transmitted by an interconnecting cable to the
vibration instrument or analyzer which may be some distance away. For this reason, a
common practice is to incorporate an electronic amplifier directly inside the
accelerometer to amplify the signal so it can be transmitted through long cables without
worrying about signal loss or interference Horn radio frequencies (RF interference) or
high voltage electro-static interference or high voltage transformers, electrical fields
around motors, etc. Accelerometers built-in amplifiers can normally be used with
interconnecting cables up to 1000 feet (330 meters) in length without appreciable signal
loss or interference.

Where to Take the Readings

Fig: 4.4 Direction for placing the sensor

Since vibratory forces generated by the rotating components of a machine are
passed through the bearings, vibration readings for both detection and analysis should be

31
taken directly on the bearings whenever possible.
FFT means:

The term "FFT" stands for "Fast Fourier Transform". Nearly 200 years ago,
French mathematician, Baron Jean Baptist Joseph Fourier established that any periodic
function (which includes machinery vibration signals) can be represented mathematically
as a series of sines and cosines. In other words, it is possible to take a vibration time
waveform, whether simple or complex, and mathematically calculate the vibration
frequencies present along with their amplitudes. The process is called a "Fourier
Transform". Although a Fourier Transform can be done manually, the process is
extremely time consuming. However, with the introduction of digital technology, the
process can be carried out very fast. Hence the term: Fast Fourier Transform or FFT.
Digital vibration analyzers and data collectors actually include a computer chip
programmed to perform the FFT function.

Analog Signal

The FFT process begins with an analog signal from a vibration transducer.
Normally, the transducer will be a vibration accelerometer; however, signals from other
types of transducers can be processed as well such as microphones, pressure transducers,
current transformers, etc.

Input

Since a vibration accelerometer is normally used for vibration detection and

Analysis, it may be necessary to convert the acceleration signal to velocity by "single
integration" or to displacement by "double integration". These functions are carried out at
the input section. Calibration of the analog signal, based on transducer sensitivity, is also
performed at the input.

32
4.5.3 DATA PAC 1500

Instrument Details:

ENTEK IRD (company)

Data Pac 1500
Portable data collector \ analyzer.

Data Pac 1500 is part of Entek IRD is complete range of monitoring products
and services to all industry segments world wide .the data pac 1500 is a fully featured
portable data collector \ analyzer designed in a small lightweight package that monitors
the conduction of the equipment found in many process industries such as power
generators petrochemical pulp and paper and primary metals. This easy-to use instrument
features high frequency range and true zoom capabilities normally only found in high
priced, bulky real-time analyzers .The data Pac 1500 collects field data, including
vibration information and process variables with a frequency range of 10cpm
-4518000cpm (.18hz)-75.3khz .it also includes true zoom capability ,screen capture and
print utilities.

Fig: 4.5 DATA PAC 1500

The data Pac 1500 utilizes the latest advances in analog and digital electronics
including digital signal processing (DSP) and industry highest resolution A\D converter
to provide both speed and accuracy in the data collection process. The instrument
incorporates a large VGA resolution screen for easy reading and comprehensive data
presentation online context sensitive help is building to all applications so they are easy

33
to use and require minimum training . The data Pac 1500 accepts industry standard
type 1 or type 2 PC memory cards to provide both unlimited and reliable data storage and
is powered by long life, rechargeable, easily removable Ni-cad battery cells.

Features:

Inputs/outputs:

Single data channel, constant current interface standard +/- 10volts engineering
units (EU), providing for vibration inputs and process inputs (temperature prop is
optional). Reference input channel, supports a variety of externally powered TTL
compatible inputs. Including photocells, electromagnetic transducers or ENTEK IRD lace
tach.
Frequency:

Frequency response: 10CPM to 4518000CPM (0.18Hz to 75.3 kHz) non integrated

21CPM to 4518000CPM (0.36Hz to 75.3 kHz) integrated
Frequency ranges: 42 ranges between 600CPM and4518000CPM (10Hz and 75300Hz)
Frequency resolution: upto12800 lines
GSE corner frequencies: 100, 200, 500, 1000, 2000, 5000Hz

Amplitude range /resolution:

18 bit A\D converter is incorporated for a solid 96db dynamic range

Auto ranging capability sets full scale in 1,2and5 increments
Last hardware range is stored for each measurement to improve measurement speed

Supported measurement:

Acceleration
Velocity
Displacement
Spike energy
Temperature
Thrust or axial position
DC voltage
AC voltage

34
Adb, Vdb
Phase (1x-99x)
Speed
Time synchronous FFT’s
Time synchronous wave form
Amplitude vs. RPM (optional)
Start up (coast down FFT waterfall plots (optional)), NY quist plots
(Optional), speed
Profiling plots (optional).

Signal processing:
A wide range of options are available and controllable by the host software including
RMS, peak, peak to peak, and DC meter types
Linear, exponential, RMS and peak hold averaging
FFT processing (hamming, hanning, Kaiser-bessel flat top and rectangular window)
12.5 KHz real time data collection and processing rate
Time domain data collection
Automatic amplitude ranging

35
 CHAPTER-V
VIBRATION ANALYSIS

5.1 INTRODUCTION

There are literally hundreds of specific mechanical and operational problems that
can result in excessive machinery vibration. However, since each type of problem
generates vibration in a unique way, a thorough study of the resultant vibration
characteristics can go a long way in reducing the number of possibilities—hopefully to a
single cause. A simple, logical and systematic approach that has been proven successful
in pinpointing the vast majority of the most common day-to-day machinery problems.

5.2 DEFINE THE PROBLEM

The following lists some of the reasons for performing a vibration analysis:

1. Establish "baseline data" for future analysis needs. At the beginning of a

predictive maintenance program, even machines in good operating condition should be
thoroughly analyzed to establish their normal vibration characteristics. Later, when
problems do develop, this baseline information can be - extremely useful in performing
a follow-up analysis to show precisely the vibration characteristics that have changed.
2. Identify the cause of excessive vibration. Referring to the vibration severity
guidelines machines in service that have vibration levels in the "rough" regions or greater
should be thoroughly. Analyzed to identify existing problems for immediate correction.
Once corrections have been made, a follow-up analysis should be performed to insure
that problems have been solved and the machine returned to satisfactory condition. If all
significant problems have been solved, the follow-up analysis data will serve as the
baseline data for future analysis as outlined in (1) above.
3. Identify the cause of a significant vibration increase. Once a developing problem
has been detected by routine, periodic checks, the obvious next step is to perform a
detailed vibration analysis to identify the problem for correction. Here also, a follow-up
analysis will verify that the problems have been corrected and provide a baseline for

36
future comparison
4. Identify the cause of frequent component failures such as bearings, couplings,
seals, shafts, etc.
5. Identify the cause of structural failures such as the structure or foundation, piping
etc.
6. Identify the source of a noise problem.

5.3 DETERMINE MACHINE DETAILS

Some of the important detailed features of the machine that need to be known for
Accurate analyses include:
1. The rotating speed (RPM) of each machine component: Of course, direct-coupled
machines have only one rotating speed (RPM) that needs to be known. However,
machines that include gear drives will have more than one.' For single gear increasers or
reducers, both the input and output speeds are needed. For multiple gear increasers or
decreases, the rotating speeds of the various intermediate gears must be known along
with the input and output speeds.
2. Types of bearings: Of course worn or defective sleeve or plain bearings will have
different vibration characteristics than defective rolling-element bearings. Therefore, it is
most important to know whether the machine has plain or rolling element bearings. If the
machine has rolling-element bearings, it is also beneficial to know the number of rolling
elements and other details of bearing geometry; with this information, the vibration
analyst can actually calculate the frequencies of vibration caused by specific bearing
defects such as flaws on the outer and inner raceways, rolling elements, etc. Details on
determining specific bearing defect frequencies are presented in the ANALYSIS OF
ROLLING ELEMENT BEARINGS section of this chapter.
3. Number of fan blades: Knowing the machine RPM and number of blades on a fan
will enable the analyst to easily calculate the "blade-passing" frequency. This is simply
the product of the number of fan blades times fan RPM. This frequency of vibration is
also called the "aerodynamic pulsation frequency.
4. Number of impeller vanes: Similar to fans and blowers, knowing the number of
vanes on a pump impeller allows the analyst to calculate the vane-passing frequency, also

37
called the "hydraulic-pulsation" frequency.
5. Number of gear teeth: The rotating speed and number of teeth on each gear must be
known in order to determine the possible "gear-mesh" frequencies.
6. Type of coupling: Gear and other lubricated types of couplings can generate some
unique vibration characteristics whenever their lubrication breaks down or if lubrication
is inadequate.
7. Machine critical speeds: Some machines such as high speed, multi-stage centrifugal
pumps, compressors and turbines are designed to operate at speeds above the natural or
"resonant" frequency of the shaft. The resonant frequency of the shaft or rotor is called its
"critical" speed, and operating a near this speed can result in extremely high vibration
amplitudes. Therefore, knowing the rotor critical speed relative to machine RPM and
other potential exciting force frequencies are very important.
8. Background vibration sources: Many times the vibration being measured on a
machine is actually coming from another machine in the immediate area. This is
particularly true for machines mounted on the same foundation or that are interconnected
by piping or other structural means. Therefore, it is important to be aware of potential
"background" contributions. This is especially true with machine tools, due to the low
levels of vibration required! If possible, the machine under analysis should be shut down
and readings taken to directly determine the amount and significance of background
vibration.

5.4 VISUAL INSPECTION

Before collecting data, the vibration analyst should first make a visual check of
the machine to determine if there are any obvious faults or defects that could contribute
to the machines condition. Some obvious things to look for include;
1. Loose or missing mounting bolts
2. Cracks in the base, foundation or structural welds
3. Leaking seals
4. Worn or broken parts
5. Wear, corrosion or build-up of deposits on rotating elements such as fans.

38
Slow Motion Studies

Another test that is helpful in a visual inspection of the machine is slow motion
observation of the various rotating elements of the machine with a high-intensity
stroboscopic light. The strobe light must be one that has an adjustable flash rate. Simply
adjust the strobe to flash at a rate which is slightly faster or slightly slower than the
rotating speed (1 x RPM) of the machine. This will make the rotating components appear
to rotate slowly.
Slowing down the rotating motion of the machine makes it possible to visually
detect problems that may be difficult to detect any other way. Visual run out of a shaft
may pinpoint or verity a bent shaft condition. Eccentricity of "V" belt sheaves and
pulleys can be easily detected in slow motion. Slow motion studies are especially useful
in evaluating problems with belt drives. Worn grooves in pulleys or belts with variations
in thickness can easily be seen by observing the action of the belt riding up and down in
the pulley grooves. On multiple-belt drives, belt slippage can be determined by observing
the belts in slow motion.

5.5 PROBING STUDIES

The tendency in vibration analysis is to concentrate on analyzing vibration data

taken at the bearings of the machine. While this data is definitely an important part of any
vibration analysis, in many cases the vibration that is occurring at the machine's bearings
is actually the result of problems elsewhere in the "system. For example, in one case a
vertical pump had a vibration of 0.7 in/sec measured at the top bearing of the pump
motor. However, overall vibration readings taken on the pump base, foundation and
piping revealed that the discharge piping was vibrating at a level of 3.0 in/sec or over four
times higher than the pump motor itself. The problem turned out to be resonance of the
discharge piping and not a problem with the pump itself. The pump and drive motor in
this case were simply responding to the piping problem.

The only way other problems in the system can be detected, such as the piping
resonance described above, is to go looking for them. Depending on the anticipated
vibration frequencies, select overall displacement, velocity or acceleration for
measurement. Some of the areas that should be checked include:

39
1. Suction and discharge piping on pumps: Take overall measurements in three directions.
On long piping runs, take readings at several locations along the piping.
2. Externally mounted components such as exciters, lube-oil pumps, surge bottles, etc.
Here also, take overall measurements in three directions
3. Take overall measurements on nearby machines that may contribute background
vibration. If a nearby machine has higher vibration amplitudes than the one being
analyzed, it is very likely some of the vibration is coming from the background source.
4. Compare overall vibration readings across all mounting interfaces to detect obvious
signs of looseness or weakness.
5. In addition to taking comparative overall readings across the mounting interfaces
To detect Obvious looseness problems, the vibration, amplitudes taken vertically
at The mounting points of a machine, such as the four feet of a motor, can be compared to
reveal the possibility distortion or "soft-foot or distortion conditions caused by uneven
mounting or foot will usually be indicated if one or more of the feet reveals a
significantly higher amplitude than the other feet. If this is defected, the condition should
be verified and Corrected before further analysis is carried out. Soft-foot conditions can
be checked by placing a dial indicator directly on the foot and carefully loosening the
mounting bolt while observing the indicator reading. Any movement or "spring" in
excess of 0.002 - 0.003 inch is generally considered excessive and should be corrected.

5.6 OBTAIN HORIZONTAL, VERTICAL AND AXIAL SPECTRUMS (FFTS) AT EACH BEARING

OF THE MACHINE TRAIN

In many cases, the analysis steps carried out thus far may be sufficient to pinpoint
the specific problem causing excessive vibration. If not, the next step is to obtain a
complete set of amplitude-versus-frequency spectrums or FFTs at each bearing of the
machine train. For a proper analysis, the machine should be operating under normal
conditions of load, speed, temperature, etc.
In order to insure that the analysis data taken includes all the problem-related vibration
characteristics and, yet, is easy to evaluate and interpret, the following recommendations
are offered;

40
Interpreting the Data
Once horizontal, vertical and axial FFTs have been obtained for each bearing of
the machine train, the obvious next question is: "What is this data telling me?"
Essentially, amplitude-versus-frequency spectrums or FFTs serve two very important
purposes in vibration analysis:
1. Identify the machine component (motor, pump, gear box, etc.) of the machine train
that has the problem And
2. Reduce the number of possible problems from several hundred to only a limited few.

Identifying the Problem Component Based On Frequency

Figure 5-1 shows a fan operating at 2200 RPM, belt driven by an 1800 RPM
motor. The rotating speed of the belts is 500 RPM. Assume that a vibration analysis was
performed on this machine and the only significant vibration detected had a frequency of
2200 CPM or 1 x RPM of the fan. Since the vibration frequency is exactly related to fan
speed, this clearly indicates that the fan is the component with the problem. This simple
fact eliminates the drive motor, belts and possible background sources as possible causes.
Most problems generate vibration with frequencies that are exactly related to the
rotating speed of trip in trouble. These frequencies may be exactly 1 x RPM or multiples
(harmonics) of 1 x RPM such as 2x, 3x, 4x, etc. In addition, some problem's may cause
vibration frequencies that are exact sub harmonics of 1 x RPM such as 1/2x, l/3x or 1/4 x
RPM. In any event, the FFT analysis data can identify the machine component with the
problem based on the direct relationship between the measured vibration frequency and
the rotating speed of the various machine elements.

Identifying the Problem Component Based On Amplitude

Identifying the fan as the source of vibration based on vibration frequency was
quite easy in the above example because of the notable differences in the rotating speeds
of the various machine components. The obvious question, of course is: What about
direct-coupled machines that is operating at exactly the same speed?" In this case, the
component with the problem is normally identified as the one with the highest amplitude.
For example, consider a motor direct coupled to a pump. Examining the analysis data, it
is noted that the highest vibration amplitude on the motor is 1.0 in/sec compared to 0.12

41
in/sec on the pump. In this case, the motor is clearly the problem component since its
vibration amplitude is nearly 8 times higher than that measured on the pump.
In general, the machine component that has the problem is usually the one with
the highest amplitude of vibration. The forces that cause vibration tend to dissipate in
strength at increased distances from the source. However, there are exceptions to this rule
such as the example given earlier where a vertical pump was vibrating excessively due to
a resonance problem with the discharge piping. In this case, the exciting force was
actually generated by the motor/pump but was being amplified by the resonant condition
of the piping.
Another exception to this rule involves misalignment of direct coupled machines.
Sir Isaac Newton’s third law of physics slates that "whenever one body exerts a force on
another, the second always exerts on the first a force which is equal in magnitude but
oppositely directed." In other words, "for every action, there is an equal but opposite
reaction." In the case of coupling misalignment, the vibratory force (action) is generated
at the coupling between the driver a driven components. As a result, the "reaction" forces
on the driver and driven unit; will be essentially equal, resulting in reasonably
comparable vibration amplitudes. The only reason one component may have a slightly
higher or lower amplitude than the other is because of differences in the mass and
stiffness characteristics of the two components. But, in most cases with the coupling
misalignment, the vibration is fairly uniformly "shared" by the driver and driven units.

Fig 5.1 Different components generate different vibration frequencies

Reducing the List of Possible Problems Based On Frequency
In addition to identifying the problem machine component based on frequency

42
and/or amplitude characteristics, the second purpose of FFT analysis data is to limit or
reduce the list of possible problems based on the measured vibration frequencies.
As stated earlier, each mechanical and operational problem generates its own
unique vibration frequency characteristics. Therefore, by knowing the vibration
frequency, a list of the problems that cause or generate that particular frequency can be
made, which greatly reduces the long list of possibilities.
The chart lists the most common vibration frequencies is they relate to machine
rotating speed (RPM), along with the common causes for each frequency. To illustrate
how to use the chart, assume that the belt-driven fan pictured in Figure 4-1 has excessive
vibration at 2200 CPM which is 1 x RPM of the fan. Of course, this clearly indicates that
the fan is the component with the problem and not the drive motor or belts. In addition,
since the vibration frequency is 1 x RPM of the fan, the possible causes listed on the chart
are:
1. Unbalance
2. Eccentric pulley
3. Misalignment—this could be misalignment of the fan bearings or misalignment of the
fan and motor pulleys.
4. Bent shaft
5. Looseness
6. Distortion—from soft foot or piping strain conditions
7. Bad belts—if belt RPM
8. Resonance
9. Reciprocating forces
10. Electrical problems

Using this simple chart, along with the fact that the vibration frequency is 1 x
RPM of the fan has reduced the number of possible causes from literally hundreds to only
ten (10) likely causes, A little common sense can reduce this list even further. First, since
the vibration frequency is not related to the rotating speed (RPM) of the drive belts,
possible belt problems can be eliminated as a possible cause. Secondly, since this is a
reciprocating machine such as a reciprocating compressor or engine, the possibility of
reciprocating forces can be eliminated from the remaining list. Finally, since the

43
 frequency is not related to the drive motor or AC line frequency. In any way, the
possibility of electrical problems can be eliminated. Now, the number of possible causes
of excessive vibration has been reduced to only seven (7) by simply knowing that the
vibration frequency in this case is 1 x RF of the fan.

Table 5.1: VIBRATION FREQUENCIES AND THE LIKELY CAUSES

Frequency in Most Likely causes Other possible causes & Remarks

Terms Of RPM

1x RPM Unbalance 1)Eccentric journals, gears or pulleys

2)Misalignment or bent shaft if high axial
vibration
3) bad belts if RPM of belt
4} Resonance
5) Reciprocating forces
6) Electrical problems
2 x RPM Mechanical looseness 1) Misalignment if high axial vibration
2) Reciprocating forces
3) Resonance
4) bad belt if 2 x RPM of belt
3 x RPM Misalignment Usually a combination of misalignment and
excessive axial clearance (looseness).
Less than Oil Whirl {Less than 1) Bad drive belts
1x RPM 1/2 x RPM 2) Background vibration
3) Sub-harmonic resonance
4) "Seat" Vibration
Synchronous Electrical Problems Common electrical problems include broken rotor
(A.C line bars, eccentric rotor, and unbalanced phases in
frequency) poly-phase systems, unequal air gap.
2xSynch. Torque Pulses Rare as a problem unless resonance is excited
Frequency
Many Times Bad Gears Gear teeth times RPM of bad gear
RPM Aerodynamic Forces Number of fan blade times RPM
(Harmonically Hydraulic forces Number of impeller vane times RPM
Related Freq.) Mechanical May occur at 2, 3, 4 and sometimes higher
Looseness harmonics if severe looseness
Reciprocating Forces
High Frequency Bad Anti-Friction 1) Bearing vibration may be unsteady amplitude
(Not Harmoni- bearing and frequency
cally Related) 2) Capitation, recirculation and flow turbulence
causes random high frequency vibration
3)Improper lubrication of journal bearings
4)rubbing

44
Comparing Tri-Axial (Horizontal, Vertical and Axial) Data

Not only can specific vibration problems be recognized by their specific frequency
characteristics, but in many cases by the direction in which the vibration occurs. This is
why it is necessary to take analysis data in the horizontal, vertical and axial directions - to
further the process of elimination.
Table 5.1 shows a typical "set" of tri-axial data taken on one bearing of a belt
driven fan operating at 2200 RPM. Of course similar data would be taken on the, other
fan bearing as well as the motor bearings. "Stacking" the horizontal, vertical and axial
data for a particular bearing on the same sheet as shown, greatly simplifies the
comparison. Note
That the same full-scale amplitude range (0 to 0.3 in/sec) was used for all the data
to further simplify the comparison.
There are basically two comparisons that need to be made from the data in Figure 5-2.
First, how do the horizontal and vertical readings part; and secondly, how do the radial
readings (horizontal and vertical) compare1 to the axial readings.

Fig 5.2: Typical tri-axial data taken on a belt-driven fan

45
Comparing Horizontal and Vertical Readings

When comparing the horizontal and vertical data, it is important to take note of how
and where the machine is mounted and also, how the bearings are mounted to the
machine. Basically, the vibration analyst needs to develop a "feel" for the relative
stiffness between the horizontal and vertical directions in order to see whether the
comparative horizontal and vertical readings indicate a normal or abnormal situation.
Machines mounted on a solid or rigid base may be evaluated differently than machines
mounted on elevated structures or resilient vibration isolators such as rubber pads or
springs.
To explain the significance of machine stiffness, assume that the fan in Figure 5-1
is mounted on a rigid, solid concrete base which, in turn, is mounted on a solid
foundation located at ground level. This would be regarded as a "rigid" installation and
under normal conditions the vertical stiffness would be greater than the horizontal
stiffness. If such is the case/one would expect that normal problems, such as unbalance,
would cause higher amplitude of vibration in th2 horizontal direction than the vertical
direction, if a rigidly mounted machine has higher vibration in the vertical direction than
the horizontal direction, this would generally be considered as 'abnormal', and may
indicate a looseness or weakness condition. On the other hand, if this same machine was
mounted on springs or rubber pads, a higher amplitude in. the vertical direction may not
be considered unusual or an indication of structural problems.
Another factor that needs to be considered is the "ratio" between the horizontal and
vertical Amplitudes. As explained, it is not unusual for rigidly mounted machines to have
higher amplitudes of vibration in the horizontal direction, compared to the vertical
direction. However, the ratio between the horizontal and vertical amplitudes should be
checked to see if it is normal or indicative of some unusual problem. As a normal
unbalance response, it is not unusual for machines to exhibit ratios between the horizontal
and vertical amplitudes of 1:1, 2:1, 3:1 or 4:1, depending on the particular installation. In
other words, it would not be unusual for a rigidly mounted fan, motor or pump to have a
vibration amplitude at 1 x RPM as much as 4 times higher in the horizontal direction than
the vertical direction due to unbalance. Ratios beyond 4:1 somewhat unusual and
typically indicate an abnormal condition such as looseness or resonance.

46
Comparing Radial (Horizontal & Vertical) Data to Axial Data

The second important comparison that needs to be made to tri-axial analysis data is
how the radial (horizontal and vertical) readings compare to the axial readings. Relatively
high amplitudes of axial vibration are normally the result of:
1. Misalignment of couplings
2. Misalignment of bearings
3. Misalignment of pulleys or sheaves on belt drives
4. Bent shafts
5. Unbalance of "overhung" rotors such as the fan in Figure 5.1
A general rule, any time the amplitude of axial vibration exceeds 50% of the
highest radial (horizontal or vertical) amplitude, the possibility of a misalignment or bent
shaft condition should be considered.) CM course, extremely high amplitudes of axial
vibration may also be due to resonance or unbalance of an overhung rotor. Verifying the
cause of a high axial vibration using "phase analysis" techniques will be covered in the
sections to follow.
Examining the axial vibration in the examples given in Figures, it can be seen that in
neither instance is the amplitude of axial vibration greater than 50% of the highest radial
amplitude. As a result, misalignment or bent shaft Conditions are not indicated examples.

'Where Do Multiple Harmonic Vibration Frequencies Come From?

Something that often worries or confuses the beginning vibration analyst is the
appearance of numerous "harmonic" frequencies that sometimes appear in their FFT
analysis data. A good example is the frequency analysis data presented in Figure -5-3.
Although the predominant vibration is clearly 2200 CPM (1 x RPM of the fan), vibration
frequencies can also be seen at 4400 CPM (2 x RPM), 6600 CPM (3 x RPM) and 8800
CPM (4"x RPM). Although their amplitudes are considerably lower than that at 1 x RPM,
these "harmonic" frequencies are very important and should not be ignored, as will be
explained in the following Paragraphs.
The presence of multiple or "harmonically" related vibration frequencies is not
uncommon, and their presence in the FFT data can be easily explained by examining the
frequency characteristics of various vibration waveforms. Figure 5.3 illustrates four (4)

47
different types of vibration waveforms — a sinusoidal is a sine wave, a square wave, s
triangular or "saw-tooth" wave and a spike pulse. These waveforms can be readily
generated by various machinery problems, depending on the nature of the problem and
the extent of the exciting forces. The '1 Linda -mental" frequency of each of the
waveforms in Figure 5-3 is the same; however, the frequencies presented in the FFTs will
be considerably different.

Fig 5.3: Different wave forms result in different frequency characteristics

A sinusoidal or "sine" wave could be the result, of a simple unbalance or

misalignment problem. If a frequency analysis (FFT) is performed on a true sinusoidal
waveform, the result will be a single frequency of vibration with certain amplitude and
NO multiple frequencies.
By comparison, a frequency analysts (FFT) of & square waveform will not only
display the fundamental frequency (1x), but the odd multiple or harmonic frequencies as
well (i.e. 3x, 5x, 7x, etc.). The number of odd multiple frequencies present in the FFT
data will depend on how close the waveform is to a true square wave, the intensity or
amplitude of the vibration and the response characteristics (peak or RMS) of the
instrument as well as its dynamic range. Figure 4-8 shows a 6000 CPM (1 00 Hz) square
waveform signal obtained from an electronic signal generator along with the FFT

48
frequency analysis. Note that the frequency analysis not only includes the fundamental
frequency of 6000 CPM, but the odd multiples as well (i.e. 18,000 CPM, 30,000 CPM,
42,000 CPM, etc.). One possible explanation (or a square wave vibration would be an
unbalance condition combined with system looseness. If the unbalance force was great
enough, the machine could literally be lifted off the foundation and held to the limit of
looseness until the unbalance force has rotated to a position where the upward force is
reduced, allowing the machine to drop. Another possibility is a mild rubbing condition
that might "flatten" the unbalance sine wave whenever the rub occurs.
The fundamental (1x) frequency accompanied by the odd multiple or harmonic
frequencies, similar to a square wave. However, the amplitudes of the odd harmonics of a
triangular waveform decrease more quickly at higher frequency than do those of a square
waveform as shown in Figure 5-3. Triangular or saw tooth waveforms can also be
generated by conditions such as looseness or excessive bearing clearance that result in
"distortion" of an unbalance sine we Here also, the number of odd multiple frequencies
that accompany the fundamental frequency will depend not only on the amplitude of the
fundamental frequency, but the dynamic range and circuit response characteristics (peak
RMS) of the analysis instrument.
Some problems such as a cracked or broken tooth on a gear, or a flaw on a
bearing raceway or rolling element, will generate vibration in the form of impact or
spike-pulses. A frequency analysis or FFT of a spike-pulse signal will reveal the
fundamental impact frequency, followed by the entire multiple or harmonic frequencies
(i.e. 2x, 3x, 4x, 5x, 6x, etc.) as shown in Figure 4-3. As before, the number of harmonic
frequencies evident in the FFT will depend on the amplitude of the fundamental
component and the dynamic range and circuit response characteristics (peak or RMS) of
the analysis instrument.
The presence of multiple, harmonic frequencies in an FFT are definitely important
and should not be ignored, even though their amplitudes may be considerably less than
that of the fundamental frequency. Their mere existence indicates that the vibration is not
a true sine wave, and may provide clues to other significant problems such as looseness
conditions, gear tooth problems, bearing problems, etc. In the case of the belt driven fan
in Figure 5-1, the harmonic frequencies only appeared at the drive-end bearing (bearing

49
C). Ultimately, the problem was found to be a loose pulley on the fan shaft, which was
allowing the pulley to "rattle" on the shaft during rotation. This caused a spike-pulse
distortion of the unbalance sine wave, resulting in the harmonic vibration frequencies.
Once the set-screws were tightened securing the pulley to the shaft, the multiple
harmonic frequencies totally disappeared, leaving only the 1 x RPM unbalance vibration
frequency.
Distortion of a sinusoidal vibration waveform, resulting in multiple vibration
frequencies, may not only be the result of mechanical problems such as looseness,
bearing defects, rubbing or gear defects as described above. (Waveform distortion can
also result from the setup and operation of the vibration analysis equipment. For
example, if a magnetic holder is being used to mount the vibration accelerometer to the
machine, any looseness or rocking of the magnet on the surface of the machine can result
in the appearance of multiple frequencies in the analysis data. In addition, if the
amplitude of machine vibration exceeds the full-scale amplitude range selected on the
analyzer instrument, the true vibration signal may be "chopped", resulting in multiple
frequency components in the FFT data that do not physically exist. For example, if the
actual level of machine vibration was 1.0 in/sec but the analyzer was set for a 0.3 in/sec
full-scale range, the vibration signal would be chopped off, creating an approximate
square waveform. The result, of course, would be odd multiple frequencies in the analysis
data that do not actually exist.

Side-Band Frequencies
"Side-band" frequencies are an additional vibration frequency that often appears
FFT data that can be confusing to the beginning vibration analyst. Side band vibration
frequencies are the result of a variation in the amplitude of given vibration frequency
signal as a function of time. This variation in amplitude with time is also called
"amplitude modulation". For example, consider a Rolling element bearing with a
significant flaw or defect on the rotating inner raceway.)

50
 Fig: 5.4 spike pulses due to flaw on the inner race of the bearing
As the inner raceway rotates, spike pulses will be generated each time a rolling
element impacts the flaw. However, the amplitude or intensity of the pulses generated
will vary as the defect rotates into and out of the load zone of the bearing. This is shown
in Figure.
Impacts that occur when the defect is within the load zone will obviously be mw
intense than those that occur out of the load zone. The result is a modulation the
fundamental bearing defect frequency. The fundamental bearing defect frequency in this
case is the frequency at which rolling elements impact the inner raceway flaw and is
called the "ball passing frequency of the inner raceway" or simply BPFI. When
discussing side-band frequencies, the fundamental bearing frequency in this case would
be called the "carrier" frequency. The frequency at which the amplitude of the carrier
frequency varies is called the "modulating" frequency. The modulating frequency in the
case of a defect on the inner raceway will be 1 x RPM, since the defect is rotating into
and out of the bearing load zone at the rotating speed of the shaft.

5.7 DETERMINE IF THE VIBRATION IS DIRECTIONAL OR NON-

DIRECTIONAL

In addition to a comparison of tri-axial (horizontal, vertical and axial data) other analysis
techniques such as simple probing studies has been discussed to show how the list of
possible problems can be reduced. A vibration frequency of 1 x RPM is probably the
most common "predominant" vibration encountered during analysis because so many
different yet common day-to-day problems can cause it. These problems include.
1. Unbalance
2. Bent shafts

51
3. Misalignment— of couplings, bearings and pulleys.
4. Looseness
5. Resonance
6. Distortion—from soft foot or piping strain conditions
7. Eccentricity---of pulleys and gears
8. Reciprocating forces
Of all the problems listed above, the only ones that generate uniform radial forces
and resultant vibration are unbalance and bent shafts. All of the remaining problems
typically generate forces and resultant vibration which is very highly directional in
nature. Therefore, determining whether or not the radial vibration directional or non-
directional can be an extremely valuable analysis tool in reducing the list of possible
problems.
To explain the difference between directional and uniform or non-directional vibration,
consider the response of a machine to a simple unbalance problem. An unbalance
condition generates a certain amount of radial force which is governed by the amount of
unbalance weight (ounces, grams, etc.), the radius of the weight or its distance from the
shaft centerline and the rotating speed (RPM) of the machine.
In any case, an unbalance generates a fixed amount of force that is simply
changing in direction with shaft rotation. If the stiffness of the machine was the same in
the horizontal and vertical directions, the machine would literally move in a circular path,
and the radial vibration amplitudes would be the same in all radial directions. Of course,
the horizontal and vertical stiff nesses will probably not be exactly the same, so the radial
motion will probably be somewhat elliptical, resulting in slightly different amplitudes
measured in various radial directions. In any case, a simple unbalance, uncomplicated by
other problems, generates a fairly uniform, non-directional radial vibration. In terms of
radial vibration, a bent shaft reacts in much the same way as simple unbalance. However,
remember ' from our earlier discussion that a bent shaft will also be characterized by rela-
tively high axial vibration amplitudes as well.
Compared to unbalance and bent shafts, the other listed causes of 1 X RPM
Vibration DOES NOT generate uniform radial vibration. Instead, they create radial
vibration which is very highly directional. For example, consider the radial vibration

52
generated by coupling misalignment. When a coupling is misaligned, obviously it is
misaligned in a certain direction. As a result, the radial forces and, hence, the radial
vibration will be most pronounced in the direction of misalignment.
Similar to coupling misalignment, a distortion problem from a soft foot or
piping strain problem creates misalignment. Of the machine's bearings in a certain
direction with the result being highly directional radial vibration.
An eccentric pulley used in a "V" belt drive produces variations in belt tension with each
revolution, causing a highly directional vibration measured in the direction of belt
tension. An eccentric gear creates directional forces due to the "cam-like" action with the
mating gear, similar to an eccentric pulley. Unbalanced reciprocating forces due to
misfiring, valve leakage and other operational problems with engines and reciprocating
compressors generate vibration in the direction of reciprocation which are, of course,
highly directional. Structural looseness or weakness problems such as loose mounting
bolts or deterioration in grouting, simply allow the machine more freedom to move in the
direction of the looseness. Although some other exciting force must be present, such as
unbalance, structural looseness usually results in highly directional vibration.

Detecting Directional versus Non-Directional Vibration

There are basically three ways to determine whether the vibration of a machine
reasonably uniform or highly directional in nature. These include:
1. A comparison or horizontal, vertical and axial FFT data
2. Comparing the horizontal and vertical phase measurements
3. Multiple radial amplitude measurements
Comparing Horizontal and Vertical Phase Readings
Figure 5-5 illustrates two types of radial vibration—uniform and highly directional. Of
course, uniform or reasonably circular radial vibration is typically the result of a simple
unbalance problem. Phase measurements, which were discussed, are obtained by either
triggering a stroboscopic (strobe) light with the vibration signal or by comparing the
vibration signal with a refer pulse such as that obtained with a photo-electric pickup
(photocell) or laser In any case, phase was described as the "relative" motion of two or
more parts of a machine.

53
 Fig: 5.5 Comparative horizontal and vertical phase readings are used to distinguish
between uniform and highly directional radial vibration
However, phase can also describe how a single element of a machine is vibrating, such as
a bearing. In other words, comparative phase measurements taken in the horizontal and
vertical directions of a particular bearing can reveal whether the bearing is vibrating in a
highly directional or a reason uniform manner.
In the case of a normal unbalance, it is the "heavy spot" of unbalance that; dually
causes the vibration. And, since it is the vibration signal that triggers the strobe light or is
compared to the reference pulse, it can be said that, indirectly, the heavy spot is
responsible for the ultimate phase angle measurement. With this in mind then, it should
be apparent that if the machine bearing is vibrating in a reasonably uniform or circular
pattern, from the time that the heavy spot rotated to a position to cause the machine to
move horizontally, the heavy spot had to rotate an additional 1 /4 revolution or 90 degrees
to cause it to move vertically. Therefore, if the machine is, intact, vibrating radially in a
reasonably uniform or circular manner, a comparison of phase readings taken in the
horizontal & vertical directions should show a difference of approximately 90 degrees.
Of course, even with simple unbalance, the stiffness in the horizontal and vertical
directions will probably not be exactly the same and, as a result, the radial motion will
probably not be a perfect circle. Therefore, when comparing phase readings taken in the
horizontal and vertical directions, the phase difference may not be exactly 90 degrees.
Because the radial motion may not be a perfect circle, a tolerance or plus or minus 30
degrees is normally allowed. In other words, comparative horizontal and vertical phase
readings between 60 degrees and 120 degrees indicate a fairly uniform radial vibration or
one that is NOT highly directional.

54
 In the case of highly directional radial vibration, comparative horizontal and
vertical phase readings will either be nearly the same (0 degrees difference) or 180
degrees out-of-phase, depending on which side of the machine bearing the readings are
taken. Referring to Figure 4-5, if the vibration was occurring virtually along the straight
line designated "A", the machine would reach its maximum motion toward the horizontal
measurement location at exactly the same instant it reached its maximum motion toward
the vertical measurement location. In this case, comparative horizontal and vertical phase
measurements would be the same (i.e. differ by 0 degrees). However, if the directional
vibrate was occurring along the straight line designated as "B", then at the exact instant
the machine reached its maximum motion towards the horizontal measurement location,
it would have reached its maximum location away from the vertical measurement
location. In this case, comparative horizontal and vertical phase readings would differ by
180 degrees.
In summary, a simple comparison of horizontal and vertical phase readings can
quickly tell whether the radial vibration is fairly uniform or highly directional. In
addition, it should be pointed out that it does not make any difference how the machine is
vibrating radially relative to the directions of vibration measurement. Any time the ratio
between the maximum and minimum radial vibration amplitude exceeds much beyond
5:1, comparative horizontal and vertical phases readings will indicate a highly directional
vibration by virtue of a 0 degree or 180 degrees phase comparison.
5.8 IDENTIFYING THE MOST COMMON MACHINERY
PROBLEMS
5.8.1 Vibration due to Unbalance

Unbalance of rotating machine components is, perhaps, the easiest problem lo

pinpoint with confidence. Simple unbalance, uncomplicated by other problems, can be
readily identified by the following characteristics:

1. The vibration occurs at a frequency of 1 x RPM of the unbalanced component. The

presence of multiple, harmonic frequencies (i.e. 2x, 3x, 4x, times RPM) usually indicates
additional problems such as looseness, rubbing, etc

55
2. The radial vibration is reasonably uniform and not highly directional. A comparison of
horizontal and vertical phase readings will normally show a difference between 60
degrees and 120 degrees. If comparative horizontal and vertical phase readings cannot be
taken, multiple radial amplitude readings should not show a discrepancy in excess of 5:1.
3. If a specific machine component such as a motor or fan is the source of unbalance, that
component will! Have significantly higher amplitudes of vibration at the 1 x RPM
frequency. Unbalance of couplings will likely reveal comparable amplitudes on both the
driver and driven machine components.
Unbalance conditions can often be affected by other operating conditions such as
load or temperature. For example, machines operating at elevated temperatures can
physically distort or change shape due to thermal changes, resulting!! a change rotor
balance. Large, fabricated boiler draft fans must often be balanced at operating
temperature due to thermal distortion. They may run smoothly when cold but vibrate
excessively when hot.
In addition, due to minor variations in the track and pitch-angle of the fan blades.
Large fabricated fans may show significant changes in the unbalance vibration
characteristics with changes in flow conditions. In other words, a change in the damper
setting may result in a significant change in the unbalance amplitude & phase
characteristics. Such affects are referred to as "aerodynamic unbalance', and [joint out
the importance of balancing a rotor under its normal operating conditions of temperature
and flow conditions.
5.8.2 Bent Shaft Problems

Bent shafts are a common problem encountered on machinery, and are often the
result of manufacturing errors or mishandling and damage during transportation or
machine installation. In addition, a rotor may "bow" as the result of thermal distortion at
elevated temperatures or due to excessive unbalance forces.

56
 Fig: 5.6 uneven rotor and bent shaft problems
Regardless of the cause, bent shafts will usually generate a predominant vibration at 1 x
RPM, very similar to simple unbalance. And, like unbalance, the radial vibration caused
by a, bent shaft will be fairly uniform and not highly directional. However, unlike
unbalance, bent shaft conditions will normally cause a relatively significant vibration in
the axial direction as well. As staled earlier, any time the amplitude of vibration measured
in the axial direction exceeds 1/2 (50%) o! the highest measured radial vibration, a bent
shaft is a very possible cause.
Because bent shafts cause significant vibration in the axial direction, a bent shaft
problem can normally be verified using a phase analysis of the axial vibration. However,
there are actually two different types of bent shaft conditions:
1. Rotors that have a simple "bow and:
2. Shafts that have a bend or "kink", but only near a particular bearing.
Each type of bend will result in significant axial vibration, but each type will
cause the various bearings of the machine to vibrate in the axial direction in a notice-ably
different manner. Therefore, an axial phase analysis cans not only verity a bent shaft
condition, but can also help in identifying the nature and location of the bend as well.

57
5.8.3 Shaft with a Kink or Bend Close To the Bearing

The axial vibration caused by a bent shaft can actually occur in two different
ways. Normally, if the bend is fairly close to a particular bearing, such as a "kink" in the
stub shaft of a motor or pump caused by bumping the shaft during transportation or
installation, the bearing will tend to vibrate axially in a "twisting", motion This twisting
motion can be easily recognized by taking comparative axial phase measurements at
multiple axial positions as shown in Figure 4.7. Four axial phase readings at each bearing
of the machine are recommended; however, physical constraints may make it impossible
to take all the readings desired. In any case, more than one axial phase reading is needed,
so try to take &s many as possible.
If the bearing is, in fact, "twisting" due to a kink in the shaft that is very close or
actually through the bearing itself, the result will be a drastic difference in the phase
readings obtained at the four axial positions, as shown in Figure 5-7. In Figure 5-8, it can
be seen that the upper and lower measurement points {1 and 3) are actually 180 degrees
out-of-phase, as are the measurement points on opposite sides of the shaft (2 and 4). This
clearly indicates that the bearing is vibrating axially in a twisting fashion. Further
verification of a bend or "kink" in the shaft can be carried out using a dial indicator.

Fig 5.7: four axial phase readings at each bearing are needed to see how each bearing is
vibrating axially

Fig 5.8: These axial phase readings show a twisting

58
 axial motion.
Bearings that are "cocked" in the machine housing may also cause significant
vibration amplitudes in the axial direction, and may reveal the same "twisting" action as
that caused by a kinked shaft. However, a cocked or misaligned bearing can usually be
distinguished from a kinked shaft by comparing the amplitudes of vibration measured at
the four axial positions. Normally, if a bearing is cocked in the housing, it will be cocked
in a specific direction and show a significant difference in the amplitudes measured at the
four axial positions. On the other hand, a shaft that has a simple bend or "kink" will
reveal fairly uniform amplitudes in the four axial positions.

5.8.4 Identifying a Simple Shaft Bow

A shaft that has a simple bow may not cause the supporting bearings to vibrate
axially in a "twisting" type of motion. Instead, a simple bow may cause the supporting
bearings to vibrate axially in a "planer" fashion
In order to identify a simple bow as the cause of high axial vibration, it will be
necessary to compare the "relative" axial motion of the support bearings. If the shaft is
simply bowed, the supporting bearings of the rotor will reveal a substantial "out-of-
phase" condition. Although a pronounced bow may reveal as much as a 180 degrees
difference in the axial phase of the rotor supporting bearings, an out-of-phase condition
of only 90 degrees or more is significant enough to indicate a possible bow in the shaft.
Run out checks with a dial indicator should be performed to verify the bent shaft
condition—especially if the amplitudes of axial vibration far exceed 50% of the highest
radial amplitudes.

When comparing the axial phase readings at the supporting bearings of a rotor, it is
most important to keep in mind the direction of the transducer. To illustrate, when taking
axial phase- readings on the left side bearing of the rotor, the vibration transducer may
have been pointing to the right. However, when axial phase readings were taken on the
right side bearing, it may have been necessary to point the vibration transducer to the left.
If this was the case, it will be necessary to correct the phase readings for one of the
bearings by 180 degrees to compensate for the necessary 180 degree change in transducer
direction. Of course, if the direction of the pickup axis can be kept the same at all bearing

59
locations, then no correction factor is necessary.

5.8.5 Vibration Due to Misalignment

Surveys have shown that at the beginning stages of most predictive maintenance
programs, misalignment of direct coupled machines is by far the most common cause of
machinery vibration. In spite of self-aligning bearings and flexible couplings, it is
difficult to align two shafts and their bearings so that no forces exist which will cause
vibration. Although machines may be well aligned initially, several factors can affect
alignment, including:

Fig: 5.9 parallel misalignment and angular misalignment

1. Operating temperature: Machines aligned when cold may "grow" out of alignment due
to variations in thermal conditions,
2. Settling of the base or foundation
3. Deterioration or shrinkage of grouting
The following are general characteristics to look for:
1. The radial vibration caused by coupling misalignment is typically highly directional on
both the driver and driven units. Misalignment occurs in a certain direction and, as a
result, the radial forces are not uniformly applied in all radial directions like that from
unbalance.
2. The vibration frequencies due to misalignment are usually 1 x, 2x and 3x RPM, and
may appear in any combination depending on the type and extent of misalignment.
Angular misalignment normally causes vibration at 1 x RPM, whereas offset or parallel
misalignment causes vibration predominantly at 2 x RPM. In fact, offset misalignment is

60
probably the most likely cause of a predominate 2 x RPM vibration. Combinations of
angular and offset misalignment may show combinations of 1x and 2 x RPM and in some
cases even 1x, 2x and 3 x RPM.
3. Whenever misalignment is suspected, an axial phase analysis comparing the relative
axial motion of the driver and driven units can be most helpful. As stated earlier, "for
every action there is an equal but opposite reaction". As a. result, misalignment
problems will normally reveal a significant phase difference \ up to 180 degrees.
However, phase differences as little as 60 degrees Tin relative axial motion is
sufficient to suggest misalignment.
5.8.6 Vibration Due To Looseness
Excessive vibration may exist due to looseness conditions; however, looseness is
not the actual cause of the vibration. Some other exciting force such as unbalance or
misalignment must be present to actually cause the vibration.

Fig: 5.10 vibrations due to looseness (excessive clearance in bearings and loose bolts)
Looseness is simply a loss or reduction in the normal stiffness of the machine or
system, perhaps due to loose mounting bolts, cracks in the base or foundation,
deterioration in grouting, cracked welds, loose lags or anchors or rotors loose on the
shaft. Looseness conditions simply allow whatever exciting forces that exist in the
machine to exhibit or generate higher amplitudes of vibration than they would if no
looseness problems existed. If the predominate exciting force is an unbalance at 1 x
RPM, then the predominate vibration due to looseness would be 1 x RPM in this case.
However, if the predominate exciting force is occurring at 2 x RPM due to an offset
misalignment, then the looseness would occur at a frequency of 2 x RPM. Looseness
does not have to occur at a frequency of 2 x RPM as many published vibration diagnostic

61
charts would lead one to believe.
Two general types of looseness are:
1. Looseness associated with the rotating system, including rotors loose on the shaft,
bearings loose on the shaft or in the machine housing and excessive sleeve bearing
clearance.
2. Looseness of the support system of a machine such as loose mounting bolls, grouting
deterioration or cracks in the structure.

Mechanical Looseness and "Sub harmonic" Frequencies

On occasion, certain looseness conditions can also result in "sub harmonic"

Frequencies of vibration (i.e. 1/2, 1/3, or 1/4 x RPM) with frequencies at 1/2 x RPM
being the most common. For example, there have been many reported cases of excessive
wear and clearance in sleeve bearings of large motors that have resulted in vibration
frequencies of 1/2x, 1x, 1-1/2x, 2x, 2-1/2x, 3x, 3-1/2x and Higher orders of 1/2 x RPM.
Why and how vibration frequencies at multiples of ' half-order (1/2 x RPM) are generated
has never been fully explained. However, when detected, possible looseness conditions
including bearing clearance Problems should be suspected.

5.8.7 Vibration Due To Eccentricity

Of course, no rotor or shaft can be made perfectly round. Some eccentricity or

"out-of-roundness" will be present on nearly every rotating assembly. Eccentricity is a
common cause of unbalance, and for common machines such as fans, blowers, pumps,
etc., normal balancing procedures can be carried out to minimize the effects of
eccentricity. However, in certain situations, eccentricity can result in "reaction" forces
that cannot be totally compensated by simply balancing the rotor. Probably the most
common examples are eccentric belt pulleys and chain sprockets, eccentric gears and
eccentric motor armatures.

In the case of an eccentric belt pulley or chain sprocket, each revolution of the
eccentric pulley or sprocket will cause a variation in belt or chain tension. The result will
be a vibration frequency at 1 x RPM of the eccentric element, with a directional force on
a line between the centers of the driver and driven pulleys or sprockets. Although this
could be easily mistaken as an unbalance problem, a simple test for- the directionality of

62
the radial vibration by taking comparative horizontal and vertical phase readings or by
taking multiple radial amplitude readings will quickly reveal the highly directional nature
of the vibration. Slow motion studies with a stroboscopic light or run out checks with a
dial indicator will confirm the eccentricity problem.

Fig: 5.11 Vibrations Due To Eccentricity

Eccentric gears will cause highly directional vibration at 1 x RPM of the eccentric
gear in a manner similar to that of eccentric belt pulleys and chain sprockets, and can be
identified by taking comparative horizontal and vertical phase readings or by taking
multiple radial amplitude readings as described earlier.

5.8.8 Vibration Due To Resonance

Resonance is a very common cause of excessive vibration on machines because

1. Machines consist of many individual elements or components such as suction and

discharge piping, bearing pedestals, bases, and accessory items such as exciters and lube
oil pumps, etc... Of course, each component has its own mass and stiffness characteristics
and, hence, its own unique natural frequency.

2. The stiffness of each machine component is different in different directions. As a

result, each machine component will likely have several different natural frequencies.
For example, consider a fan bearing. Most likely, the stiffness of the bearing will be
different in the horizontal, vertical and axial directions. As a result, the natural
frequencies of this particular machine component will also be different in the horizontal,
vertical and axial directions.

63
 When one considers all of the various machine components, along with the
multiple natural frequencies possible for each component, the reason that resonance is
such a common problem is quite understandable. All that is required is that the natural
frequency of one machine component, in one of its directions of vibration, be the same as
one exciting force frequency inherent to the machine, when this happens, resonance and
high levels of vibration will result.
Although machines that are installed and brought into service may not exhibit
resonance problems initially, resonance may become a problem in the future if changes in
machine stiffness occur as the result of bearing wear, grouting deterioration, loosening of
mounting bolts or other problems.

5.8.9 Vibration due to defective rolling element bearings

When a rolling element bearing develops flaws on the raceways and/or rolling
elements, there are actually a number of vibration frequency characteristics that can result
depending on the extent of deterioration.

5.12 vibrations due to worn rollers, worn gear teeth and worn belts
Thus, identifying these characteristic frequencies cannot only help to verify that a
bearing is definitely failing, but can also give some indication of the extent of
deterioration. The following is a discussion of the four stages that a bearing will typically
go through from the earliest stage of deterioration to that approaching catastrophic
failure. Catastrophic failure id defined here as simply the total in ability of the bearing to
perform its intended functions of minimizing the friction generated through rotating
motion and keeping rotating and non-rotating parts from coming into contact with one
another. In other words, failure will occur when either the bearing literally comes apart or
seizes due to excessive heat buildup.

64
5.9 VIBRATION ANALYSIS

5.9.1 Introduction

Interpreting vibration characteristics carries out diagnostics studies based on

vibration analysis. The following vibration characteristics are utilized to interpret the
machinery health condition.
1. Vibration displacement
2. Vibration velocity
3. Vibration acceleration will include the health condition.
4. Vibration frequency will indicate the defect
5. Vibration base will help to indicate the defective location

5.9.2 Dominant Frequency

It is sometimes useful to know the dominant frequency of vibration of a machine

that has several rotating parts. Consider for example a belt driven blower. The motor and
fan run at different speeds. The dominant frequency will reveal which part of the machine
is causing the most vibration. The measurements taken on the motor bearings would
include vibration of the motor plus vibrations from the fan transmitted through the drive
belt and mounting structure. Conversely, measurement taken on the fan bearings would
include vibrations of the fan plus vibrations transmitted from the motors as shown in the
figure. In many cases the point with the most vibration would pin point which part has
the trouble, but not always. Measurements taken on the fan contains vibrations of the fan
plus transmitted vibrations from the motor.

The following sequence is used to find the dominant frequency:

1. Measure and record the displacement (D) at a given point.

2. In same manner and at the same point, measure and record the velocity (V).

3. The dominant frequency can be found by dividing the velocity measurement (V) by the
displacement measurement (D) and multiplying by the constant 19100. The answer will

65
be the dominant frequency of vibration in cycles per minute.

Dominant frequency (CPM) =

velocity (V) millimeters/second x 19100 (metric units)
Displacement (D) micrometers

For example, if the motor runs at 1750 RPM and the fan at 2600 RPM.
Measurements taken on the motor are displacement (D) = 47 micrometers and velocity
(V) = 6.5 millimeters / seconds.

The dominant frequency then becomes:

(6.5/47) XI 9, 120 =2,644 cpm

2644 CPM is nearest the fan speed. The fan is the dominant part and is causing
vibration.

Generally the dominant frequency will be equal to the rotating speed of the part
cause the vibration, assuming that the trouble is unbalance. In any event the dominant
vibration frequency will normally be some multiple of RPM of the part. After
determining the dominant frequency, the type of the trouble present may be assumed
from the following table.

66
 CHAPTER-VI

ARTIFICIAL NEURAL NETWORKS

6.1 INTRODUCTION
An artificial neural network (ANN), usually called "neural network" (NN), is a
mathematical model or computational model that tries to simulate the structure and/or
functional aspects of biological neural networks. It consists of an interconnected group of
artificial neurons and processes information using a connectionist approach to
computation. In most cases an ANN is an adaptive system that changes its structure based
on external or internal information that flows through the network during the learning
phase. Neural networks are non-linear statistical data modeling tools. They can be used to
model complex relationships between inputs and outputs or to find patterns in data.
A neural network is an interconnected group of nodes, akin to the vast network of
neurons in the human brain.

Fig: 6.1 a simple neural network

6.1.1 Background

Component based representation of a neural network. This kind of more general

67
representation is used by some neural network software.

There is no precise agreed-upon definition among researchers as to what a neural network

is, but most would agree that it involves a network of simple processing elements
(neurons), which can exhibit complex global behavior, determined by the connections
between the processing elements and element parameters. The original inspiration for the
technique came from examination of the central nervous system and the neurons (and
their axons, dendrites and synapses) which constitute one of its most significant
information processing elements (see Neuroscience). In a neural network model, simple
nodes (called variously "neurons", "neurodes", "PEs" ("processing elements") or "units")
are connected together to form a network of nodes — hence the term "neural network."
While a neural network does not have to be adaptive per se, its practical use comes with
algorithms designed to alter the strength (weights) of the connections in the network to
produce a desired signal flow.

These networks are also similar to the biological neural networks in the sense that
functions are performed collectively and in parallel by the units, rather than there being a
clear delineation of subtasks to which various units are assigned (see also
connectionism). Currently, the term Artificial Neural Network (ANN) tends to refer
mostly to neural network models employed in statistics, cognitive psychology and
artificial intelligence. Neural network models designed with emulation of the central
nervous system (CNS) in mind are a subject of theoretical neuroscience (computational
neuroscience).

In modern software implementations of artificial neural networks the approach

inspired by biology has for the most part been abandoned for a more practical approach
based on statistics and signal processing. In some of these systems, neural networks or
parts of neural networks (such as artificial neurons) are used as components in larger
systems that combine both adaptive and non-adaptive elements. While the more general
approach of such adaptive systems is more suitable for real-world problem solving, it has
far less to do with the traditional artificial intelligence connectionist models. What they
do have in common, however, is the principle of non-linear, distributed, parallel and local

68
processing and adaptation.

6.1.2 Models

Neural network models in artificial intelligence are usually referred to as artificial

neural networks (ANNs); these are essentially simple mathematical models defining a
function. Each type of ANN model corresponds to a class of such functions.

6.1.3 The Network in Artificial Neural Network

The word network in the term 'artificial neural network' arises because the function f(x)
is defined as a composition of other functions gi(x), which can further be defined as a
composition of other functions. This can be conveniently represented as a network
structure, with arrows depicting the dependencies between variables. A widely used type

of composition is the nonlinear weighted sum, where, where K (commonly referred to as

the activation function) is some predefined function, such as the hyperbolic tangent. It

will be convenient for the following to refer to a collection of functions gi as simply a

vector.

Fig:6.2 ANN dependency graph

This figure depicts such a decomposition of f, with dependencies between variables

indicated by arrows. These can be interpreted in two ways.

69
The first view is the functional view: the input x is transformed into a 3-dimensional

vector h, which is then transformed into a 2-dimensional vector g, which is finally

transformed into f. This view is most commonly encountered in the context of

optimization.

The second view is the probabilistic view: the random variable F = f(G) depends upon
the random variable G = g(H), which depends upon H = h(X), which depends upon
the random variable X. This view is most commonly encountered in the context of
graphical models.

The two views are largely equivalent. In either case, for this particular network
architecture, the components of individual layers are independent of each other (e.g., the

components of g are independent of each other given their input h). This naturally
enables a degree of parallelism in the implementation.

6.1.4 Learning

What has attracted the most interest in neural networks is the possibility of

learning. Given a specific task to solve, and a class of functions F, learning means using
a set of observations to find which solves the task in some optimal sense. This entails

defining a cost function such that, for the optimal solution f *, (i.e., no solution has a cost
less than the cost of the optimal solution).

The cost function C is an important concept in learning, as it is a measure of how

far away a particular solution is from an optimal solution to the problem to be solved.
Learning algorithms search through the solution space to find a function that has the
smallest possible cost.

For applications where the solution is dependent on some data, the cost must
necessarily be a function of the observations; otherwise we would not be modeling
anything related to the data. It is frequently defined as a statistic to which only

70
approximations can be made. As a simple example consider the problem of finding the

model f which minimizes, for data pairs (x,y) drawn from some distribution . In practical

situations we would only have N samples from and thus, for the above example, we
would only minimize. Thus, the cost is minimized over a sample of the data rather than
the entire data set.

When some form of online learning must be used, where the cost is partially
minimized as each new example is seen. While online learning is often used when is
fixed, it is most useful in the case where the distribution changes slowly over time. In
neural network methods, some form of online learning is frequently used for finite
datasets.

6.1.5 Employing artificial neural networks

Perhaps the greatest advantage of ANNs is their ability to be used as an arbitrary

function approximation mechanism which 'learns' from observed data. However, using
them is not so straightforward and a relatively good understanding of the underlying
theory is essential.

Choice of model: This will depend on the data representation and the application. Overly
complex models tend to lead to problems with learning.

Learning algorithm: There are numerous tradeoffs between learning algorithms. Almost
any algorithm will work well with the correct hyper parameters for training on a
particular fixed dataset. However selecting and tuning an algorithm for training on
unseen data requires a significant amount of experimentation.

Robustness: If the model, cost function and learning algorithm are selected appropriately
the resulting ANN can be extremely robust.

With the correct implementation ANNs can be used naturally in online learning
and large dataset applications. Their simple implementation and the existence of mostly
local dependencies exhibited in the structure allows for fast, parallel implementations in
hardware.

71
6.1.6 Applications

The utility of artificial neural network models lies in the fact that they can be used
to infer a function from observations. This is particularly useful in applications where the
complexity of the data or task makes the design of such a function by hand impractical.

6.1.7 Real life applications

 The tasks to which artificial neural networks are applied tend to fall within the
following broad categories:
 Function approximation, or regression analysis, including time series prediction,
fitness approximation and modeling.

 Classification, including pattern and sequence recognition, novelty detection and

sequential decision making.

 Data processing, including filtering, clustering, blind source separation and

compression.

 Robotics, including directing manipulators, Computer numerical control.

Application areas include system identification and control (vehicle control, process
control), quantum chemistry, game-playing and decision making (backgammon, chess,
racing), pattern recognition (radar systems, face identification, object recognition and
more), sequence recognition (gesture, speech, handwritten text recognition), medical
diagnosis, financial applications (automated trading systems), data mining (or knowledge
discovery in databases, "KDD"), visualization and e-mail spam filtering.

6.2 PATTERN CLASSIFICATION NETWORK:

A two layer feed-forward network with non linear output functions for the units in
the output layer can be used for the task o f pattern classification. The number of units in
the input layer corresponds to the dimensionality of input pattern vectors. The units in the
input layer are all linear, as the input layer merely contributes to fan out the input to each
of the output units. The number of output units depends on the number of distinct classes
in the pattern classification task. We assume for this discussion that the output units are
binary. Each output unit is concerned to all the input units, and a weight is associated

72
with each connection. Since the output function of a unit is a hard limiting threshold
function, for a given set of input-output patterns, the weighted sum of the input values
compared with the threshold for the unit to determine whether the sum is greater or less
than the threshold. Thus, in this case a set of in equalities are generated with the given
data. Thus there is no unique solution for the weights in this case. Typically, if the
weighted sum of the input values to the input values to the output values exceeds the
threshold, the output signal is labeled as 1, otherwise as 0. Multiple binary output units
are needed of the number of pattern classes exceeds 2.

6.3 BACK PROPOGATION ALGORITHM (generalized delta rule)

Given a set of input-output patterns (al,bl),l=1,2,…….L,

Where the lth input vector and the al = (al1,al2…..all)T and the lth output vector
bl=(al1,al2……alk)T.

Assume only one hidden layer and initial setting of weights to be arbitrary.

Assume input layer with only linear units.

Then the output signal is equal to the input activation value for each of these units. Let η
be the learning rate parameter.

Let a =a (m) =a and b=b (m) =bl.

Activation of unit I in the input layer, xi= ai(m)

Activation unit j in the hidden layer,

Output signal from the jth unit in the hidden layer,

Activation of unit k in the output layer,

Output signal from unit k in the output layer,

73
Error term for the kth output layer,

Calculate the error for the lth pattern,

Total error for all patterns,

Apply the given patterns one by one, may be several limits, in some random order and
update the weights until the total error reduces to an acceptable value.

6.4 APPLICATION OF ANN TO VIBRATION ANALYSIS

6.4.1 Reasons for Applying Ann

 Process upsets and abnormal situations often entail a certain degree of

unpredictability

 Unknowns and unpredictable behavior of processes operating in combination with

one another call for artificial intelligence

 Artificial neural networks are widely preferred due to their pattern recognition
capabilities

6.4.2 Developing an ANN system

In order to overcome the ever present restriction in the development of expert

systems based on neural networks a four phased data production analysis procedure for
training has been developed. The four interleaved approaches for data collection are
simple fault simulation, finite element method(FEM) simulation, test rig simulation and
real data gathering using this technique the amount of data can be extended to fulfill the
strictest requirements of a neural network capable of handling almost all rotating
machinery at the user sites. The adopted approach makes it possible to vary such
parameters as type and structure of the machinery in question, together with the
operational parameters such as running speed, load and environmental factors. The data
flow scheme during the development process is shown in figure. In this approach data
from the simple fault simulator FEM model test rig and real world can be combined and
74
analyzed in one module and then treated in the same way when training the neural
network. Basically this approach overcomes the problem of gathering sufficient amount
of data from faulty machinery in the field which for the defined fault and machinery
scenarios would take hundreds of years to be adequate for training.

Simulated
Simulated
fault
data
simulation Vibration
parameters
statistical
FEM Simulated FFT spectrum
simulation data ceptrum auto
correlation

Simulated
Data test,
Test
conversio Test vibration
measureme Test data
n data paramete
nt
rs

Plant Data
Test
measureme Test data conversio
data
nt n

Data Optimized
optimization data
ANN

Fig 6.3 the data flow chart

75
6.4.3 Test rig simulation

A test rig has been built to enable the generation of fault data for training the
neural networks and neuro fuzzy system in addition; the test data is used to refine the
simulation model. The test rig consists of a motor, brake and a shaft with two bearings.
The test rig has been used for data collection for different fault types such as unbalance,
misalignment and different types of bearing faults. All these fault types have been studied
at different stages of fault severity. The data from the test rig gives the basic information
for the development of the hybrid expert system. As work on the project developed, the
test rig was modified to represent a more complicated item of plant.

6.4.4 Simple fault simulator

A simple PC based fault simulator (SFS) has been developed. The SFS can be
used to predict the characteristic features of vibration acceleration signals for different
fault types such as unbalance, misalignment and different types of bearing faults. The
SFS can also introduce noise to the signals due to various sources, i.e the effects of other
machinery and amplifier/transducer performance. Linked to the SFS there is a module for
the calculation of different measuring parameters, which have been programmed the
background for this is simply that there are a great no of different measurement
parameters used in practice.

6.4.5 FEM simulation

A finite element model simulation FEM was developed to provide a more

sophisticated degree of control over the synthesized data. The complexity of the FEM
model is much greater than the SFS and allows a variety of changes to be made to the
model for generating different fault situations. The great advantage of the fem simulation
is the possibility of making structural changes to the machine with minimum effort,
especially if the model has been built for this purpose. In this case the model has been
parameterized using super element techniques in order to keep the no of degrees freedom
of the model at a reasonable level for calculation purposes. The model basically
represents the test rig but offers the capability of making dimensional changes such as
shaft diameter variation and bearing distance variation thus introducing to the response
the effect of transfer functions due to the variation of natural frequencies

76
6.4.6 Problems developing diagnostic expert systems

Rules are a common form of knowledge representation and are present in most AI
applications such as expert systems and decision support systems. Such rules are
normally obtained from human experts as a result of a lengthy knowledge engineering
process other systems may obtain their knowledge from machine learning programs. Rule
based systems are widely used because they allow the incorporation of multiple clauses,
enable the use of confidence measures, can be subjected to mathematical rigor i.e formal
logic enable modular systems to be built and most important of all a lot of human
reasoning can be expressed as rules. Rule based systems have some serious disadvantages

1. The process of acquiring knowledge from human experts is very time consuming,
prone to errors and is quite an expensive process to carry out.

2. Rules are brittle when presented with nosy, incomplete or non linear data.

3. Rules do not scale up very well, inconsistencies and errors start to creep in as the
knowledge based grows.

The main cause of brittleness with in rule based systems is a requirement that every
possible combination of antecedents and their values must be explicitly provided for.
Thus in order to cope with the complexities of real world data a practical rule based
system must have a large number of rules to account for the most common cases that
should occur. Except for smaller, restricted applications it is most unlikely that any given
rule based system will be robust or complete enough (i.e have sufficient rules) to provide
reasonably correct answers when presented with deviant input. Most expert systems
experience a sharp drop of in-operational capability when confronted with novel
situations for which no specific rules exist. However, vibration data has a great deal of
variability in the amplitude, frequency and phase of the measured data points. This
variability of data input is the main cause of diagnostic failure within rule based system.
Vibration data often stresses the pattern matching ability of rule based to breaking point.

77
6.4.7 Advantages of applying ANN’s for fault detection

One of the tools used for fault detection is the application of artificial neural
networks for the analysis of vibrations generated by a piece of equipment.

Although there are numerous efficient methods for modeling of mechanical

systems, they all suffer the disadvantage that they are only valid for a particular machine.
Changes within the design or the operational mode of the machine normally require a
manual adaptation. Using neural networks to model technical systems eliminates this
major disadvantage. The basis for a successful model is an adequate so called knowledge
base on which the network is “trained”. Without prior knowledge of the machines
systematic behavior or its history, training of a neural network is possible.

A second advantage of neural network is that the trained network is not only valid
for a small section of the complete operational range of the machine. A neural network
that has been trained and there by possessing a tolerable residual is valid for all
conditions included in the knowledge base.

78
 CHAPTER VII

CASE STUDY

BOILER FEED PUMP 4B

7.1 Description of the Equipment

Boiler Feed Pump (BFP) is used to pump the feed water (chemically treated water)
in to the boiler. The FK6D30 type BFP consists of FAiB56 Booster Pump (BP) directly
driven from one end of the shaft of an electric motor. BFP is driven from the opposite end
of Motor shaft through a spacer type flexible coupling.

Fig: 7.1 Boiler feed pump train

The BP is a single stage, horizontal, axial split casing type, having the suction and
discharge branches on the casing bottom half, thus allowing the pump internals to be
removed without disturbing the suction and discharge pipe work or the alignment
between the pump and the driving motor. The rotating assembly consists of the shaft,

Impeller, nuts, keys, seal sleeves, thrust collar, the rotating parts of the mechanical seals,

79
pump coupling. The rotating assembly is supported at each end of the shaft by a white
metal lined journal bearing. The residual axial thrust is taken up by a tilting double thrust
bearing mounted at the non drive end of the pump. The present work deals with the main
pump which is connected to the motor by flexible hydraulic coupling. The boiler feed
pump is supported by the journal bearings. The line diagram of the entire unit is shown in
figure 7.1.

7.2 Specifications:

No of stages……………………………...6

Design flow rate…………………………448.5 m3/sec.

Speed ……………………………………5178 rpm.

No of impeller vanes of main pump…….7no’s

No of impeller vanes of booster pump….5no’s

Booster pump speed…………………….1485 rpm

Variable speed geared coupling

Motor speed……………………………..Ne=1482 rpm

Gear ratio………………………………..u1=Ze/Zi=143/41

Primary speed N1………………………..N1 = 5289 rpm

Full load slip…………………….………..S = 2.1%

Max output speed………………………..Ne =5178 rpm.

Regulating range………………………...4:1 down loads.

80
7.3 Experimentation

The measurements are recorded using “Data PAC 1500”, dual channel seismic
pick-up, with a frequency range of 10cpm to 4518000cpm(0.18hz to 75.3 Hz), with A/D
converter, VGA resolution screen data collector of Entek IRD,USA make, over a period
of 6 months at regular monthly intervals. The instrument is mounted on the bearing
supports along horizontal (H), vertical (V), and axial (A) directions, the axial direction
being in line with the axis of the shaft. The measurements are made in displacement and
velocity modes. Accelerations have been computed. Regular logging of the data has
provided on the basis for performance trend monitoring of the rotating structure and
prediction of faults to apply reasoning to trace the root cause.

7.4 Data collected before rectification

Vibration readings before rectification:

DATE SUPPORT DISPLACEMENT(um) VELOCITY(mm/sec)

POINT H V A H V A

11th NOVEMBER BPNDE 12.0 2.5 4.5 1.6 1.5 1.3

2008 BPDE 8.2 3.2 5.0 1.5 0.6 1.5
MMDE 7.2 4.3 5.8 0.9 0.8 1.0
MMNDE 5.3 2.8 1.8 0.8 0.4 0.4
MPDE 18.4 11.2 14.5 4.0 3.3 3.2
MPNDE 28 38 25 9.2 5.0 8.3
25th November BPNDE 17.2 12.3 13.5 2.5 1.8 1.4
2008 BPDE 14.5 5.3 6.8 1.8 0.8 1.6
MMDE 9.8 3.6 6.4 0.9 1.0 1.2
MMNDE 0.8 0.4 0.4 0.9 0.6 0.6
MPDE 4.0 3.3 3.2 7.8 3.8 5.2
MPNDE 9.2 5.0 8.3 12 12.2 9.2
10th December BPNDE 13 5.2 14 1.8 1.5 1.4
2008 BPDE 9.5 2.2 5.2 1.2 0.6 1.4
MMDE 9.8 3.6 6.4 0.9 0.8 1.2
MMNDE 5.8 2.6 2.2 0.8 0.5 0.5
MPDE 13 14 12.5 3.6 4 3.2
MPNDE 26 25 18 8.1 5.5 7.0
23rd December BPNDE 12 16.5 5.8 2.3 1.4 1.0
2008 BPDE 8.5 3.4 4.6 1.5 0.7 0.9
MMDE 6.6 4.5 5.8 0.8 0.7 1.3
MMNDE 5.6 2.2 2 0.8 0.5 0.6
MPDE 17 16.5 13 4.8 5.1 4.0
MPNDE 28 45 42 10 7.2 6.6
81
06th January 2009 BPNDE 29 11 13.4 6.8 4.7 3.4
BPDE 21 6 19.5 4.7 2.2 2.9
MMDE 10.5 9 13.5 2.1 2.1 3.1
MMNDE 13.5 4.5 5.5 1.8 1.6 1.6
MPDE 27.5 21.5 21 10.8 10.5 11.5
MPNDE 54 46 22 16.5 16.3 14.5
20th January 2009 BPNDE 10.5 12 8.5 1.8 1.4 1.1
BPDE 7.5 2.8 4.5 1.2 0.6 1.4
MMDE 6.2 4.5 5.3 0.8 0.6 1.0
MMNDE 5.8 2.1 2.2 0.7 0.6 0.5
MPDE 15 45 13.5 4.8 4.5 4.2
MPNDE 28 29 25 9.2 6.2 10
O3rd January 2009 BPNDE 10 4.8 8.5 1.6 1.3 1.2
BPDE 8 8.2 5 1.3 0.7 0.8
MMDE 6.5 4 5.5 0.8 0.6 1.0
MMNDE 5.8 3 2.1 0.8 0.5 0.6
MPDE 13.5 17 6 2.8 3.0 2.4
MPNDE 21 40 25 6.5 3.0 7.0
17th January 2009 BPNDE 22 10 12 5.1 4.6 3.1
BPDE 17 5.0 10 3.9 2.4 2.9
MMDE 12 10 11 2.4 1.7 2.8
MMNDE 10 6.0 5.0 2 1.3 1.5
MPDE 170 14 15 6.2 8.4 6.4
MPNDE 36 22 48 6.4 8.5 13.7
17th FEBRUARY BPNDE 24.0 12.0 13.8 4.70 3.90 3.10
2009 BPDE 19.0 7.00 16.0 4.30 2.50 3.70
MMDE 13.0 9.00 13.0 2.70 2.10 3.10
MMNDE 10.0 7.00 6.00 1.90 1.20 1.40
MPDE 24.0 170 4.00 9.10 9.40 7.80
MPNDE 38.0 28.0 12.0 9.80 9.70 16.2
30th MARCH 2009 BPNDE 17.0 6.42 8.88 2.95 2.34 1.49
BPDE 12.9 3.56 9.56 1.85 1.21 1.60
MMDE 10.3 7.35 8.55 1.47 1.13 1.18
MMNDE 8.95 5.51 3.63 1.48 1.04 1.15
MPDE 11.3 11.0 9.05 4.97 5.74 5.40
MPNDE 31.8 18.1 9.19 11.3 4.40 20.7
14th APRIL 2009 BPNDE 26.8 8.60 10.0 5.10 3.90 3.00
BPDE 16.6 5.20 10.5 3.70 2.10 3.20
MMDE 12.4 8.20 11.4 2.30 2.20 3.00
MMNDE 12.6 6.40 5.70 2.50 1.80 2.30
MPDE 68.6 29.6 26.4 15.2 17.1 18.1
MPNDE 68.4 98.0 27.2 23.6 27.7 18.1
17th APRIL 2009 BPNDE 26.8 8.60 10.0 5.10 3.90 3.00
BPDE 16.6 5.20 10.5 3.70 2.10 3.20
MMDE 12.4 8.20 11.4 2.30 2.20 3.00
MMNDE 12.6 6.40 5.70 2.50 1.80 2.30

82
 MPDE 35.8 21.1 20.2 9.50 6.60 11.7
MPNDE 62.89 87.9 20.7 16.6 21.5 10.0

7.5 ANALYSIS OF DATA

Observing the readings from 11-11-2008 to 17-04-2009 the vibration velocity on

14-04-2009 at the main pump non driving end was tending towards abnormality. Also On
17-04-2009 the vibration velocity was intolerable. Added to these the reading at the main
pump drive end on 17-04-2009 was high.

Dominant frequency = V/D * 19100

= 21.5/87.9*19100=4671.786cpm

Thus, X=4671.786/5178=0.98=1 approximately,

Where X=the ratio of the dominant frequency to the speed

Since X is more than one times the speed of the main pump, the defect was found
to be unbalance or looseness.

The main pump vane passing frequency

= number of impeller vanes X main pump running speed.

= 7 X 5178 =36,248 cpm.

In the frequency vs. velocity spectrums the vane passing frequency harmonics of
main pump were present and 2x is having higher value. So, due to the 1x of main pump
running speed and 2x of the main pump vane passing frequency are the dominant
frequencies the cause for the frequent vibration increase was the looseness of the main
pump.

83
7.6 Spectrum Analysis before Rectification

Fig 7.1 Velocity spectrum of MPDE bearing in horizontal direction

Fig 7.2 Velocity spectrum of MPDE bearing in vertical direction

84
 Fig 7.3 Velocity spectrum of MPDE bearing in axial direction

Fig 7.4 Velocity spectrum of MPNDE bearing in horizontal direction

85
Fig 7.5 Velocity spectrum of MPNDE bearing in vertical direction

Fig 7.6 Velocity spectrum of MPNDE bearing in axial direction

86
7.7 Observations from the Spectrums:

 In all the spectrums of Main Pump Drive End and Main Pump Non Drive End 1x
running speed harmonic are present.

 In the vertical spectrum of Main Pump Non Drive End 1x running speed
frequency peak is predominant and having higher value 19.9 mm /see

 In all the spectrums of Main Pump Drive End & Main Pump Non Drive End main
pump vane passing frequency and its harmonics are found and 2x of vane passing
frequency peak is predominant having value 7.9 mm /see

 In all the spectrums of Main Pump Non Drive End side bands to vane passing
frequency peak harmonics are found.

 In all the spectrums of Main Pump Drive End both the booster pump and main
pump vane passing frequency peaks are found.

7.8 Conclusions:

 All the vibration values of booster pump non drive end, booster pump drive end,
main motor booster pump end, and main motor non drive end are in the good
zone. But Main Pump Drive End & Main Pump Non Drive End is in alarm level.

 Here 1x is predominant frequency peak in the Main Pump Drive End & Main
Pump Non Drive End spectrums Misalignment/ Looseness of the main pump
were suspected.

 A harmonic with vane passing frequency peak of the main pump are found and is
having predominant peak value. So the looseness/misalignment of the pump
impeller was suspected.

 The main pump foundation bed bolts are checked for looseness and must be
tightened.

87
7.9 Action taken

Due to high vibration on pump side (at NDE vertical) pump stopped on 18-04-
2009 and attend for the rectification. Pump NDE side bed bolts were tightened and thrust
bearing top cover is lifted by 0.01mm. After attending the work pump taken into the
service on 21-04-2009 and vibration readings were taken and observed that vibration
readings were reduced to normal value i.e. in the good zone.

7.10 Data collection after rectification

DATE SUPPORT DISPLACEMENT(um) VELOCITY(mm/sec)

POINT H V A H V A

21st APRIL BPNDE 16.3 6.65 9.05 3.21 2.53 2.02

2009 BPDE 13.4 3.84 9.40 2.26 1.10 1.55
MMDE 12.4 7.99 9.17 1.63 1.61 1.59
MMNDE 10.9 5.94 4.71 1.50 1.14 1.52
MPDE 16.4 19.7 12.6 5.27 5.61 6.19
MPNDE 33.1 23.2 8.35 8.44 6.20 6.34
28th APRIL BPNDE 19.0 9.08 11.0 3.19 3.19 2.30
2009 BPDE 15.5 3.60 9.68 2.26 1.26 1.51
MMDE 11.4 7.08 9.18 1.48 1.27 1.98
MMNDE 9.24 5.29 4.29 1.92 1.34 1.11
MPDE 20.4 15.2 6.25 5.52 6.54 5.82
MPNDE 30.0 19.3 6.38 8.10 5.29 14.2
12th MAY 2009 BPNDE 25.0 10.8 12.0 5.40 4.10 3.10
BPDE 18.4 4.60 10.8 3.20 2.50 3.00
MMDE 12.6 9.00 10.8 2.50 2.10 2.90
MMNDE 12.2 6.00 5.20 3.40 1.80 2.40
MPDE 22.0 20.8 15.0 9.30 11.6 11.4
MPNDE 36.4 28.0 18.2 13.9 9.70 10.1

ISO standards iso-10816

Good : 0 to 5.4mm / sec

Satisfactory : 5.4 to 10.6mm / sec

Alarm : 10.6 to 16.mm / sec

Not permitted : >16. mm / sec

88
7.11 Spectrums after rectification

Fig 7.7 Velocity spectrum of BPNDE bearing in horizontal direction

Fig7.8 Velocity spectrum of BPNDE bearing in horizontal direction (Booster pump vane
passing frequency peak)

89
Fig 7.9 Velocity spectrum of BPNDE bearing in horizontal direction (Booster pump vpf
& harmonics)

90
Fig 7.10 Velocity spectrum of MPDE bearing in horizontal direction (1x main pump
harmonics &vane passing frequency and its harmonics)

Fig 7.11 Velocity spectrum of MPDE bearing in vertical direction (1x harmonics)

91
 Fig 7.12 Velocity spectrum of MPDE bearing in axial direction

Fig 7.13 Velocity spectrum of MPNDE bearing in horizontal direction

92
 Fig 7.14 Velocity spectrum of MPNDE bearing in vertical direction

Fig 7.15 Velocity spectrum of MPNDE bearing in horizontal direction (side bands of

93
 main pump vpf)

Fig 7.16 Velocity spectrum of MPDE bearing in vertical direction (main pump vpf &
harmonics)

94
 Fig 7.17 Velocity spectrum of MPNDE bearing in vertical direction (1x harmonics)

Fig 7.18 Velocity spectrum of MPDE bearing in axial direction (main pump vpf &
harmonics)

CHAPTER VIII
APPLICATION OF ANN TO FAULT RECOGNITION
ON BFP

8.1 NETWORK SPECIFICATIONS:

For the purpose of recognizing the faults present in the given boiler feed pump
from the vibration patterns, the ANN was developed and trained using actual data
pertaining to the boiler feed pump and applying the back propagation algorithm. The final
network architecture is shown below:
The parameters pertaining to the ANN and the training process are
Network architecture:

95
Input layer: 3 units
Hidden layers: 2
Number of neurons in each hidden layer: 2
Model of neuron: perceptron model.
Output function: hyperbolic tangent

8.2 Training of the network:

A ‘c’ program was used for back propagation training of the ANN based on the
back propagation algorithm documented above. The data collected was used for training
the network. In addition to this, past data pertaining to the same boiler feed pump was
also used.
Once the network was trained, 10 new sets of data were used for testing the network in
addition to the 58 training sets.

8.3 Results From ANN Program:

The back propagation learning law depends upon random initialization of
weights, and hence yields different results with each execution of the program. The
results of one such execution are documented below:
Inputs to the program are:
 Learning Rate =0.1
 Momentum Rate=0.1
 Number of Iterations: 500

Results after training ANN are:

Row Target Output Match
TRN 1 unbalance unbalance Ok
TRN 2 unbalance unbalance Ok
TRN 3 misalignment misalignment Ok

96
TRN 4 unbalance unbalance Ok
TRN 5 misalignment misalignment Ok
TRN 6 misalignment misalignment Ok
TRN 7 misalignment misalignment Ok
TRN 8 misalignment misalignment Ok
TRN 9 misalignment misalignment Ok
TRN 10 misalignment misalignment Ok
TRN 11 looseness looseness Ok
TRN 12 looseness looseness Ok
TRN 13 looseness looseness Ok
TRN 14 looseness looseness Ok
TRN 15 looseness looseness Ok
TRN 16 looseness looseness Ok
TRN 17 misalignment misalignment Ok
TRN 18 misalignment misalignment Ok
TRN 19 misalignment misalignment Ok
TRN 20 unbalance unbalance Ok
TRN 21 looseness looseness Ok
TRN 22 looseness looseness Ok
TRN 23 misalignment misalignment Ok
TRN 24 looseness looseness Ok
TRN 25 unbalance unbalance Ok
TRN 26 unbalance unbalance Ok
TRN 27 unbalance unbalance Ok
TRN 28 unbalance unbalance Ok
TRN 29 unbalance unbalance Ok
TRN 30 unbalance unbalance Ok
TRN 31 unbalance unbalance Ok
TRN 32 unbalance unbalance Ok
TRN 33 unbalance unbalance Ok
TRN 34 unbalance unbalance Ok
TRN 35 unbalance unbalance Ok
TRN 36 unbalance unbalance Ok
TRN 37 unbalance unbalance Ok
TRN 38 unbalance unbalance Ok
TRN 39 unbalance unbalance Ok
TRN 40 unbalance unbalance Ok
TRN 41 unbalance unbalance Ok
TRN 42 unbalance unbalance Ok
TRN 43 unbalance unbalance Ok
TRN 44 unbalance unbalance Ok

97
TRN 45 unbalance unbalance Ok
TRN 46 unbalance unbalance Ok
TRN 47 unbalance unbalance Ok
TRN 48 unbalance unbalance Ok
TRN 49 unbalance unbalance Ok
TRN 50 unbalance unbalance Ok
TRN 51 unbalance unbalance Ok
TRN 52 unbalance unbalance Ok
TRN 53 looseness looseness Ok
TRN 54 looseness unbalance Wrong
TRN 55 unbalance unbalance Ok
TRN 56 unbalance unbalance Ok
TRN 57 unbalance unbalance Ok
TRN 58 unbalance unbalance Ok
TST 59 misalignment misalignment Ok
TST 60 misalignment misalignment Ok
TST 61 misalignment misalignment Ok
TST 62 misalignment misalignment Ok
TST 63 misalignment misalignment Ok
TST 64 unbalance unbalance Ok
TST 65 looseness looseness Ok
TST 66 unbalance unbalance Ok
TST 67 misalignment misalignment Ok
TST 68 looseness looseness Ok

98
 CHAPTER IX
RESULTS AND DISCUSSIONS

9.1 VIBRATION SPECTRUM ANALYSIS

The reason for the frequent increase in vibration was found to be due to looseness

problem in the main pump bed bolts and it is also the fact that, the unbalance is caused

due to the flow related unbalance in the main pump. This problem was conformed using

the spectrum analysis. The spectrums were collected using DATAPAC 1500.

In order to rectify this problem various operations were performed. It was found

that the main pump bed bolts are not properly tightened and having some looseness. To

rectify this, the main pump bed bolts were tightened and the thrust bearing top cover is

lifted by 0.01mm to overcome the misalignment. So that the vibration must be reduced.

After the operations were performed the vibration readings and spectrums were

taken using the analyzer. The readings were found to be feasible to satisfy ISO standards.

99
9.2 SUMMARY OF ANN RESULTS:

The results obtained from ANN application may be summarized as below: For
Testing Data Sets:

Network Output

Target Unbalance Misalignment Looseness

Unbalance 2 0 0

Misalignment 0 6 0

Looseness 0 0 2

Table 9.1

For All Data Sets:

Network Output

Target Unbalance Misalignment Looseness

Unbalance 38 0 1

Misalignment 0 16 0

Looseness 0 0 13

Table 9.2

Matrix represents the number of times a fault has been diagnosed correctly or
incorrectly.

100
For example: In Table 9.2.,

Looseness has been diagnosed correctly 13 times and has been incorrectly
diagnosed 1 time as unbalance.

9.3 CORRELATION OF THE MATHEMATICAL AND ANN

ANALYSIS;

SET1 SET 2 SET 3

SPEED (N) 5178 5178 5178

VELOCITY (V) 16.6 21.5 10

DISPLACEMENT (D) 62.89 87.9 20.75

MATHEMATICAL LOOSENESS LOOSENESS UNBALANCE

ANALYSIS

ANN ANALYSIS LOOSENESS LOOSENESS LOOSENESS

Table: 9.3

Results obtained by conventional methods and by application of the Artificial

Neural Network show a high degree of correlation. Thus, it may be concluded that the
pattern recognition and pattern classification properties of Artificial Neural Networks
make it highly feasible to apply them on a regular basis in industries.

101
REFERENCES

1. R.K.Biswas “vibration based condition monitoring of rotating machines” national

conference on condition monitoring [NCCM-2006] December 2006 pg no 34-40.
2. M.Todd, S.D.J.McArthur, G.M.West, J.R.McDonald, S.J.Shaw. J.A.Hart paper
on “the design of a decision support system for the vibration monitoring of
turbine generators”
3. David Clifton, research article on “Condition Monitoring of Gas-Turbine
Engines”
4. T.W. Verbruggen a book on “Wind Turbine Operation & Maintenance based on
Condition Monitoring”
5. A Ramachandra, S B Kandagal, paper on “Prediction of Defects in Antifriction
Bearings using Vibration Signal Analysis”
6. Sadettin Orhan, Nizami Aktu¨rk, Veli C¸ elik, “Vibration monitoring for defect
diagnosis of rolling element bearings as a predictive maintenance tool:
Comprehensive case studies”
7. Peter W. Hills, Mechanalysis (India) Limited, India, “A more intelligent approach
to rotating equipment monitoring”
8. Cornelius, Scheffer, the paper on “Pump Condition Monitoring through Vibration
Analysis”.
9. Sheng Zhang, Joseph Mathew, Lin Ma, Yong Sun and Avin Mathew, a paper on
“Statistical condition monitoring based on vibration signals”.
10. P. Caselitz, J. Giebhardt,[10], presented a paper on “Condition Monitoring and
Fault Prediction for Marine Current Turbines”.
11. Steven M. Schultheis,[11], Charles A. Lickteig, presented a paper on
“Reciprocating Compressor Condition Monitoring”.
12. Peter W. Hills, Mechanalysis (India) Limited, India presented a paper on “A more
intelligent approach to rotating equipment monitoring”
13. L. B. Jack, A. K. Nandi, presented a paper on “Feature Selection for ANNs using
Genetic Algorithms in Condition Monitoring”.

102
14. Ms S Wadhwani, Dr S P Gupta, Dr V Kumar, “Wavelet Based Vibration
Monitoring for Detection of Faults in Ball Bearings of Rotating Machines”
15. Zhigang TIAN, “An Artificial Neural Network Approach for Remaining Useful
Life Prediction of Equipments Subject to Condition Monitoring”.
16. N.M. ROEHL C.E. PEDREIRA" H.R. TELES DE AZEVEDO presented a paper
on “Fuzzy art neural network approach for incipient Fault detection and isolation
in rotating machines”.
17. P.A.L. Ham, B.Sc.C.Eng..F.I.E.E. “Trends and future scope in the monitoring of
large steam turbine generators”
18. Dukkipati.S.Rao vibration technology published by Narosa publishers, 2004
19. G.K.Groover “mechanical vibrations” published by Nem Chand and pros 1996
20. R.A Collacat condition monitoring published by M c Graw Hill
21. G.D Rai an introduction to power plant engineering khanna publishers
22. R.S Khurmi theory of machines
23. Dr.NTTPS Vibration Analysis Material
24. Dr.NTTPS Vibration Basics Material
25. NBC bearing journals
26. http://www.skf.com
27. http://www.nbc.com
28. A.V.Barkov, N.A.Barkova, and A.Yu. Azovtsev, "Condition Monitoring and
Diagnostics of Rotating Machines Using Vibration", VAST, Inc., St. Petersburg,
Russia, 1997.
29. Eshleman R I“Some recent advances in roto dynamics” 3rd International
conference on vibrations of rotating machinery .University of York 1984.
30. A collection of condition diagnostics papers on the Internet site:
http://www.vibrotek.com/ref.htm

103
APPENDIX
THE CODE USED TO DETECT THE FAULTS:
#include<stdio .h>
#include<math. h>
#include<ctype.h>
#ifndef VAX
#include<string.h>
#include<process.h>
#include<conio.h>
#include<stdlib.h>
/*for declaration of calloc() on PC or compatible*/
#include<malloc.h>
#endif

/*define constants used throughout functions*/

#define NMXUNIT 10 /*MAXIMUM NUMBER OF UNITS IN A LAYER*/
#define NMXHLR 5 /*MAXIMUM NUMBER OF HIDDEN LAYERS*/ #define
NMXOATTR 5 /*MAXIMUM NUMBER OF OUTPUT FEATURES*/ #define
NMXINP 30 /*MAXIMUM NUMBER OF INPUT SAMPLES*/
#define NMXIATTR 7 /*MAXIMUM NUMBER OF INPUT FEATURES*/
#define SEXIT 3 /*EXIT SUCCESSFULLY*/
#define RESTRT 2 /*RESTART*/
#define FEXIT 1 /*EXIT IN FAILURE*/
#define CONTNE 0 /*CONTINUE CALCULATION*/
/*Data base: declaration of variables*/
float eta; /**learning rate**/
float alpha; /**momentum rate**/
float err_curr; /**normalized system error**/
float maxe; /* *maximum allowed system error* */

104
float maxep; /* *maximum allowed pattern error* */
float *wtptr[NMXHLR+l];
float *outptr[NMXHLR+2];
float *errptr[NMXHLR+2];
float *delw[NMXHLR+l];
float target[NMXINP][NMXOATTR];
float input[NMXINP][NMXIATTR], ep[NMXINP];
float outpt[NMXINP][NMXOATTR];
int nunit[NMXHLR+2], nhlayer, ninput, ninattr, noutattr;
int result, cnt, cnt_num;
int nsnew, nsold;
char task_name[20];
FILE *fpl, *fp2, *fp3,*fl,*f2,*f3,*f4,*f5,*fopen();
int fplotl10;
double x[15][5],y[15][5],P[15][5];
float a,al,a2;
int i, j,number=10;
*random number generator(computer independent*/

long randseed = 568731L;

int random()
{
randseed = 156251 * randseed + 22221L;
return((randseed>> 16) & 0x7FFF);
}

/*allocate dynamic storage for the net*/

void initQ
{

105
int lenl, len2, i, k;
float *pl,*p2,*p3,*p4;
lenl=len2 = 0;
nunit[nhlayer+2] = 0;
for (i=0;i<(nhlayer + 2);i++) {
lenl += (nunit[i] + 1) * nunit[i+l];
len2 += nunit[i] + 1;
}
pl=(float *) calloc(lenl+l,sizeof(float)); /*weights*/
p2=(float *) calloc(len2+l,sizeof(float)); /*outputs*/
p3=(float *) calloc(len2+l,sizeof(float)); /*error*/
p4=(float *) calloc(lenl+l,sizeof(float)); /*delw*/
/*set up initial pointers*/
wtptr[0] = pl;
outptr[0] = p2;
errptr[0] = p3;
delw[0] = p4;
/*set up the rest of pointers*/
for(r=l;i<(nhlayer+l);i++) {
wtptr[i]=wtptr[i- 1 ]+nunit[i] *(nunit[i- 1 ]+ 1 );
delw[i]=delw[i-1]+nunit[i]*(nunit[i-1]+1);
}
for(i=1;i<(nhlayer+2);i++)
{
outptr[i]=outptr[i-1]+nunit[i-1]+1;
errptr[i]=errptr[i- 1 ]+nunit[i- 1 ]+ 1 ;
}
/*set up threshold outputs*/
for(i=0;i<nhlayer+l ;i++)
{
*outptr[i]+nunit[i])=1.0;

106
}
// getch();
}
/* Initialize weights with random numbers between -0.5and+0.5*/
void initwt()
{
int i , j;
for(j=0;j<nhlayer+l ;j++)
for(i=0;i<(nunit[j]+l)*nunit|j+l];i++)
{
*(wtptr[j]+i)=random()/pow(2.0,15.0)-0.5;
*(delw[j]+i)=0;
}
// getch();
}
/*specify architecture of net and values of learning
parameters*/
set_up()
{ int i;
eta=0.9;
printf(“\nmomentum rate eta (default=0.9)?: ");
scanf("%f”,&eta);
alpha=0.7;
printf("\n learning rate alpha (default=0.7)?:");
scanf("%f”,&alpha);
maxe=0.01;
maxep=0.001;
printf("\n maximum total error (default=0.0l)?: ");
scanf("%f”,&maxe);
printf("\n Max individual error (default=0.001)?:");
scanf("%f”,&maxep);

107
cnt_num=1000;
printf("\n Max number of iterations(default=1000`)?: ");
scanf("%d",&cnt_num);
printf("\nNo. of hidden layers?: ");
scanf("%d",&nhlayer);
for(i=0;i<nhlayer;i++){
printf("\n\t no.of units for hidden layer %d?: ",i+l);
scanf("%d",&nunit[i+l]);
printf("\ncreate error file? If so typel, or type 0: " );
printf("\nExecution starts");
nunit[nhlayer+1 ]=noutattr;
nunit[0]=ninattr;
}
*read file for net architechture and learning parameters. File name has suffix_v.dat*/
dread(taskname)
char *taskname;
{
int i , j,c;
char var_file_name[20];
strcpy(var_file_name,taskname);
strcat(var_file_name,"_v.dat");
if((fp 1 =fopen(var_file_name,"r"))==NULL)
{
perror("\nCannot open datafile");
exit(0);
}
fscanf(fpl,
“%d%d%d%f%f%d%d”,&ninput,&noutattr,&ninattr,&eta,&alpha,&nhlayer,&cnt_num);
for(i=0;i<nhlayer+2;i++)
fscanf(fpl,''%d",&nunit[i]);
if((c=fclose(fpl))!=0)

108
printf("\nfile cannot be closed %d", c);
}
/*Read file containing weights and thresholds, filename has suffix _w.dat*/
wtread(taskname)
char *taskname;
{
int i , j,c;
char wt_file_name[20];
strcpy(wt_file_name,taskname);
strcat(wt_file_name,"_w.dat");
if(( fp2 = fopen(wt_file_name, "r")) == NULL)
{
perror("\nCannot open data file");
exit (0);
}
for (i=0;i<nhlayer +1 ;i++)
{
for(=0;j<(nunit[i]+l)*nunit[i+l];j++)
{
fscanf(fp2,"%f “, (wtptr[i]+j));
}
}
if((c=fclose(fp2)) !=0)
printf("\nFile cannot be closed %d”, c);
}
/*Create file for net architecture and learning parameters. File name has suffix _v.dat*/
dwrite(taskname)
char *taskname;
{
int i, j, c;
char var_file_name[20];

109
strcpy(var_file_name,taskname);
strcat(var_file_name,"_v.dat");
if (( fpl=fopen(var_file_name,"w+")) == NULL)
{
perror("Cannot open data file");
exit(0);
}
fprintf(fpl,”%u%u%u%f%f%u%u\n”ninput,noutattr,ninattr,eta,alpha,nhlayer,cnt_num);
for(i=0;i<nhlayer+2;i++)
{
fprintf(fpl, "%d",nunit[i]);
}
fprintf(fpl,"\n%d%f”,cnt,err__curr);
fprintf(fpl,"\n");
for(i=0;i<ninput;i++)
{
for(j =0 ;j <noutattr ;j++)
fprintf(fpl,"%f “,outpt[i][j]);
fprintf(fpl,"\n");
}
if((c=fclose(fpl))!=0)
printf("\nFile cannot be closed %d",c);
}
/*Create for file for saving weights and threshold values learned for training. File name
has suffix __w.dat* /
wtwritte(taskname)
char *taskname;
{
int i, j,c,k;
char wt_file_name[20];
strcpy(wt_file_name,taskname);

110
strcat(wt_file_name, "_w.dat");
if((fp2=fopen(wt_file_name,”w+”)) = NULL)
{
perror("Cannot open data file");
exit(0);
}
k=0;
for(i=0;i<nhlayer+l ;i++)
for(=0;j<(nunit[i]+l) * nunit[i+l]; j++)
{
if(k==8)
{
k=0;
fprintf(fp2,"\n”);
}
fprintf(fp2,"%f” , *(wtptr[i]=j));
k++;
}
if((c=fclose(fp2))!=0)
printf(“\nFile cannot be closed %d", c);
}
/*Bottom_up calculation of net for input pattern i*/
void forward(i)
{
int m,n,p,offset;
float net;
/*input level output calculation*/
for(m=0;m<ninattr;m++)
*(outptr[0]+m) = input[i][m];
/*hidden and output layer output calculation*/
for(m=l ;m<nhlayer+2;m++)

111
{
for (n=0;n<nunit[m];n++)
{
net=0.0;
for(p=0;p<nunit[m-l]+lp++)
{
offset = (nunit[m-l]+l)*n+p;
net += *(wtptr[m-l]+offset)
*(*(outptr[m-l]+p));
}
*(outptr[m]+n)=l/(l+exp(-net));
}
}
for (n=0;n<nunh[nhlayer+l];n++)
outpt[i][n] = *(outptr[nhlayer+l]+n);
}
/* Several conditions are checked to see whether learning should terminate*/
int introspective (nfrom,nto)
int nfrom;
int nto;
{
int i,flag;
/*reached maximum iteration?*/
if (cnt>=cnt_num) return(FEXIT);
/*error for each pattern small enough?*/
nsnew=0;
flag-1;
for (i=nfrom; (i<nto) && (flag = 1); i++)
{
if (ep[i] <=maxep) nsnew++;
else flag =0;

112
}
if (flag = 1) return (SEXIT);
/*System total error small enough?*/
if(err_curr<=maxe) return (SEXIT);
return(CONTNE);
}
/*Threshold is treated as weight of link from
a virtual node whose output value is unity*/
int rumelhart(rrom_snum,to_snum)
int from_snum;
int to_snum;
{
int i,j,k,m,n,p,offset, index;
float out;
char *err_file = "criter.dat";
nsold =0;
cnt = 0;
result=CONTNE;
if(fplotl10==l)
if((fp3=fopen(err_file,"w")) == NULL)
{
I
perror("Cannot open error file");
exit(0);
}
do{
err_curr =0.0;
/*for each pattern*/
for(i=from_snum; i<to_snum;i++)
{
/*bottom_up calculation*/

113
forward(i);
/*top_down error propagation*/
/*output_level error process*/
for(m=0;m<nunit[nhlayer+1];m++) {
out= *(outptr[nhlayer+1] +m);
*(errptr[nhlayer+1]+m) = (target[i][m]-out)*(l-out)*out;
/*Hidden and output layer errors*/
for(m=nhlayer+1;m>=1;m--) {
for(n=0);n<nunit[m-1]+l;n++) {
*(errptr[m-1]+n)=0.0;
for(p=0;p<nunit[m];p++) {
offset = (nunit[m-1]+l) * p+n;
*(delw[m-1]+offset)=eta * (*(errptr[m]+p))*(*(outptr[m-1]+n)) +alpha *(*(delw[m-
1]+offset));
*(errptr[m-1]+n)+-=*(errptr[m]+p)* (*(wtptr[m-1]+offset));
}
*(errptr[m-1]+n)= *(errptr[m-1]+n)*(l-*(outptr[m-1]+n)) * ( *(outptr[m-1]+n));
}
}
/*Weight changes*/
for(m=l;m<nhlayer+2;rn++) {
for(n=0;n<nunit[m];n++) {
for(p=0;p<nunit[m-l]+l;p++) {
offset = (nunit[m-l]+l) * n+p;
*(wtptr[m-l]+offset) += *(delw[m-l]+offset);
}
}
}
ep[i] = 0.0;
for (m=0;m<nunit[nhlayer+l];m++){
ep[i] += fabs((target[i][m]-*(outptr[nhlayer+l]+m)));

114
}
err_curr+=ep[i]*ep[i];
}

/*Normalised system error*/

err_curr = 0.5*err_curr/ninput;
/**Save errors in file to draw the system error with plot 10**/
if(fplotl10==l)
fprintf(fp3,"%ld,%2.9f\n",cnt,err_curr);
cnt++;
/*Check condition for terminating learing*/
result = introspective(from_snum,to_snum);
} while (result = CONTNE);
/*Update output with changed weights*/
FILE *r;
r=fopen("c:\\output\\reporttxt","w");
for (i=from_snum; i<to_snum;i++) forward(i);
for(i=0;i<nhlayer+l ;i++)
{
index = 0;
for(j=0;j<nunit[i+l];j++)
{
fprintf(r," \n\nweights between unit %d of layer %d", j,i+l);
fprintf(r,"and units of layer %d\n", i);
for(k=0;k<nunit[i] ;k++)
fprintf(r,"%f\n",*(wtptr[i]+index++));
fprintf(r,"\n Threshold of unit %d of layer %d is %f “, j, i+1, *(wtptr[i] + index++));
}
}
fprintf(r,"\n\nTotal number of iteration is %d", cnt);
fprintf(r,"\nNormalized system error is %f\n\n\n", err_curr);

115
return(result);
}
/*Read in the input data file specified by user during the interactive session*/
user_session()
{
inti,j,showdata;
char fnam[20],dtype[20];
FILE *fp;
printf("\n Start of learning session");
/*For task with name task_name, input data file of the task
is automatically said to be task_jiame.dat by the program*/
//printf("\n\t Enter the task name: ");
//scanf("%s", task_name);
printf("\nHow many features in input pattern?: ")
scanf("%d",&ninattr);
printf("\nhow many output units?: ");
scanf("%d",&noutattr);
printf("\nTotal number of input samples?:");
scanf("%d",&ninput);
strcpy (fnam,task_name);
strcat(fnam,".dat");
printf("\nlnput file name is %s ", fnam);
if(( fp=fopen(fnam, "r")) == NULL)
{
printf("\nFile %s does not exist", fnam);
exit(0);
}
printf("\nDo you want to look at data just read?");
printf("\n Answer yes or no: ");
scanf("%s",dtype);
showdata = ((dtype[0] == ‘y’) || (dtype[0] =='y'));

116
for (i=0;i<ninput;i++) {
for(j=0;j<ninattr;j++) {
fscanf(fp, "%f ',&input[i][j]);
if (showdata) printf("%f ',input[i][j]);
}
for(j=0;j<noutattr;j++) {
fscanf(fp,"%f”,&target[i][j]);
if(showdata)printf("%f\n",target[i][j]);
}
}
if((i=fclose(fp)) !=0)
printf("\nFile cannot be closed %d",i);
exit(0);
}
}
/*Main body of learning*/
Leaming()
{
int result;
user_session();
set_up();
init();
do{
initwt();
result = rumelhart(0,ninput); }
while (result = RESTRT);
if (result = FEXIT)
printf("\nMax number of iterations reached,");
printf("\n but failed to decrease system");
printf("\n error sufficiently");
}

117
dwrite(task_name);
dwrite(task_name);
}
/*Main body of output generation*/
output_generation()
{
int i,m,nsample;
charans[10];
chardfile[50];
FILE*fl;
/* If task is already in the memory, data files for task donot
need to be readin. But, if it is a new task, data files
should be read into reconstruct the net*/
printf("\nGeneration of outputs for a new pattern");
printf("\n\t Present task name is %s",task_name);
printf("\n\t Work on a different task? ");
printf("\n\t Answer yes or no: ");
scanf("%s", ans);
if ((ans[0]=’y’) || (ans[0]='Y'))
{
printf("\n\t Type the task name: ");
scanf("%s", task_name);
dread(task_name);
init();
wtread(task_name) ;
/*Input data for output generation are created*/
printf("\nEnter file name for patterns to”);
printf(" be processed;");
scanf("%s",dfile);
if ((fpl=fopen(dfile,"r")) = NULL)
{

118
perror("Cannot open dfile");
exit(0);
}
printf("\nEnter number of patterns for processing: ")
scanf("%d",&nsample);
for(i=0;i<nsample;i—)
for(m=0;m<ninattr;m—)
fscanf(fp1,”%f”&input[i][m]);
/*0utput generation calculation starts*
fl=fopen("c:\\output\\output.txt"."w");
for(i=0;i<nsample;i++)
{
forward(i);
for(m=0;m<noutattr;m++)
{
fprintf(fl,"%f\n",outpt[i][m]*8.43472);
}
printf("\n");
}
printf("\nOutputs have been generated");
if((i=fclose(fpl))!=0)
printf("\nFile cannot be closed %d",i);
}
void main()
{
char select[20],cont[10];
//FILE *f5;
clrscr();
strcpy(task_name, "train" );
do
{

119
printf("\n** Select L(earning) or 0(utput generation)**\n");
do
{
scanf("%s",select);
switch(select[0])
{
case 'o':
case 'O’:
output_generation();break;
case T:
case 'L':
learning();
break;
default:
printf("\nanswer learning or output generation");
break;
}
}while((select[0]!=’o’)&&(select[0]!=’O’&&(select[0]!=T)&&(select[0]!='L1));
printf("\nDo you want to continue? ");
scanf("%s",cont);
}while ((cont[0]=’y’) 11 (cont[0]='Y'));
}

120