Professional Documents
Culture Documents
PREDICTIVE MAINTENANCE
EDWARD SZCZERBICKI
WARREN WHITE
Department of Mechanical Engineering, The University
of Newcastle, Callaghan, New South Wales, Australia
The authors are indebted to the members of Pacific Power International Technical
Service Group. Without their kind assistance this simulation project would not have
started.
Address correspondence to Edward Szczerbicki, Department of Me chanical Engi-
neering, The University of Newcastle, Newcastle, NSW 2308, Australia. E-mail: mees@
cc.newcastle.edu.au
No. Skill level No. Skill level No. Skill level No. Skill level No. Skill level
The simulation model looks at only the durations when a staff resource
is being utilized by performing duties of condition monitoring s i.e.,
disregarding meal breaks, staff meetings, other duties, etc...
Staff are available when requested s e.g., not on leave ..
Staff numbers are not changed during the simulation apart from the
alterations in the program.
Ratios of staff members at the various locations do not change during
the simulation.
Staff skill levels do not change during the simulation.
Staff are safety aware and have no accidents during the simulation.
For staff at the same skill level, no priority is given for any machine s s..
Any staff member capable of performing a given task will have no
priority regardless of skill level.
The percentage of workload per staff member for the tasks is constant
throughout the simulation.
All staff at a given skill level have obtained the necessary training and
qualification s s. required to perform their duties.
All machines are in good working order and safe from causing injury to
personnel.
All machines have designated and labeled targets for vibration and
some form of oil access s no setting up required..
All proximity switches on machines are in working order.
All machines consist of s on average . one motor, one gearbox, and a
drive of some sort.
Each machine s on average . has six reading points as shown in Figure 1
in accordance with the Standards Association of Australia AS
2625-198 3 s AS 2625-1983 , 1983..
Each point where applicable has horizontal, vertical, and axial direc-
tions for the vibration readings in accordance with AS 2625-198 3
s Figure 2..
The model uses only the duration times for tasks performed in pro-
cesses such as data collecting, oil sampling, analyses, filtergram
preparation, discussions with plant owners, and reporting s i.e., no
delays, no breaks, no miscellaneous functions that require time ..
There are no times for the chemical analysis; it is assumed that the
results have returned from the laboratory before a final analysis is
performed.
The chemical analysis is assumed to be correct and any question about
the validity means that a new sample must be sent to the laboratory
for a new result.
MODEL DESCRIPTION
The model was developed in the SLAM II simulation language and
implemented in the SLAMSYSTEM modeling environment s Pritsker,
1995.. It creates single entities at set intervals and then splits into five
parts, one for each of the locations. From there they are again split into
two parts, one for vibration analysis and the other for oil analysis
including wear debris. This enables the model to keep track of entity
locations and type of analysis being conducted.
From here the model tests various resources s staffing levels, data
collectors and oil sample pumps. for availability and utilizes them if
possible, or they wait in queues. Once the entity has accepted a
resource, it is put through various activities corresponding to the
flowchart diagrams, which have duration times and conditions applied.
At the end, various relevant forms of data are collected and the results
of the data displayed in a summary report, another aspect compiled
automatically by SLAM II.
Attributes
The attributes allow the individual entities to be traced during the
simulation model execution. The attributes used are specified in Table
2.
Files
These are the queues and await nodes that indicate where particular
entities are being stored while requesting some action or resource. The
files used in the model are listed in Table 3.
488 E. SZCZERBICKI AND W. WHITE
Attribute
number Definition Values Description
Duration Times
Duration times are measurements of how much time is allocated for
each task or delay, and some have included probability functions used
to determine the number of entities or the type of entity that moves
through that activity. Table 4 specifies duration times used in the
model. Global variables XX are listed in Table 7.
Resources
The resources are the staffing groups availabl e and the equipment
needed to simulate the system. Each group has separate tasks or
interactions with various other groups. Table 5 describes resources.
Resource allocation shows where the various staffing groups and
equipment are allocated to perform or be utilized in the system s Table
6.. It is easy to note some of the overlaps or common interactions.
SYSTEM SIMULATION FOR MAINTENANCE 489
Global Variables
These variable s are used to apply probability distribution duration times
to activities and also determine the probability of certain events hap-
pening, such as data audit failure. They keep the same values anywhere
in the program at any one instant in time s Table 7..
490 E. SZCZERBICKI AND W. WHITE
1 Entity creation 0 1
2 Number of machines for location A 0 }
3 Number of machines for location B 0 }
4 Number of machines for location C 0 }
5 Number of machines for location D 0 }
6 Number of machines for location E 0 }
7 Taking vibration reading XXs 3. 1
8 Downloading to computer s Vib. 2 1-XXs 2.
9, 11, 12, 13 Check data validity and audit s Vib. XXs 4. 1
10 Recheck of equipment and method RNORMs 3, 0.35. 1
s Vib.
14 Analyse vibration results XXs 5. 1-XXs 1.
15 Check results and compare data 4 0.99
history
16 Update database s machine `ok’ ., 2 0.08
s Vib.
17 Discussions with plant owner XXs 6. 1
s Vib.
18 Further testing and inspection XXs 7. 0.2
required
19 Blank r dummy 0 1
20 ª 29 Sample preparation XXs 8. 1
30, 32 Sample collection XXs 3. 1
31, 33, 34, 35 Filtergram preparation XXs 9. 1
36 Preview results s oil. XXs 10 . 1
37 Update database s oil. 2 1-XXs 1.
38 Check results s oil. 3 XXs 1.
39 Update database s oil. 2 XXs 1.
40 Compare history in database s oil. 3 1-XXs 1.
41 Discussions with plant owner s oil. XXs 9. 1
42 Inspection and r or additional XXs 9. 1
testing s oil.
43 Complete testing required s oil. XXs 7. 0.05
44 No further testing required s oil. 0 0.95
45 Non urgent-update database s oil. XXs 11 . 0.99
46 Specialist advice required s oil. XXs 12 . 0.01
47 Duration time for additional staff 480 1
s 11.
48 Duration time for additional staff 960 1
s 5.
SYSTEM SIMULATION FOR MAINTENANCE 491
1 STAFF1 10 1 11 STAFF11 1 9
2 STAFF2 2 2 12 STAFF12 1 18
3 STAFF3 1 17 13 STAFF13 2 6
4 STAFF4 1 3 14 STAFF14 4 8
5 STAFF5 1 10 15 DATALOG1 2 11
6 STAFF6 1 4 16 DATALOG2 3 12
7 STAFF7 1 15 17 DATALOG3 1 13
8 STAFF8 1 7 18 DATALOG4 2 14
9 STAFF9 1 21 19 OILPUMP1 1 20
10 STAFF10 1 5 20 OILPUMP2 1 19
Data Collection
These are collection nodes on the simulation model and are used to
collect information for the results printed out in the summary report,
automatically produced by the SLAMSYSTEM package s Pritsker, 1995..
The model uses them to collect various duration times for entities in the
system, times between entities leaving the system, count the number of
extra staff required for staff5, and count the number of repeat loops
caused by invalid or inadequate data s Table 8.. Another aspect of the
collection nodes is that they are able to produce histograms of the data
that are being collected, allowing the results to be displaye d graphically
as well as tabulated.
A 1 and 4 4 1, 2, 3, and 5 3
B 6 7 6, 5, and 8 5 and 9
C 10 11 5 and 12 5 and 9
D 8, 13, and 14 8 and 14 5, 8, and 14 5 and 9
E 11 and 5 11 5 5
492 E. SZCZERBICKI AND W. WHITE
XXs 1. TRIAGs 0, 0.08, 1.0. Probability that downloaded data are not
acceptable for vibration analysis
Probability problems s. in the oil wear debris
analysis
XXs 2. TRIAGs 0, 0.06, 1.0. Probability of extra tests; vibration data are
inadequate for the machine.
XXs 3. RNORMs 9.5, 1.5. Duration to take vibration reading
Duration to take an oil sample
XXs 4. UNFRMs 2.0, 5.0. Duration to download vibration data into
computer
XXs 5. RNORMs 4.50, 0.75. Duration to scan the results for potential
problems, especially those listed in the exception
report s Vib.
XXs 6. TRIAGs 10.0, 15.0, 16.5. Duration for discussions with plant owner s Vib.
XXs 7. RNORMs 20.0, 1.2. Duration of further testing s Vib.
Duration of complete testing s Oil .
XXs 8. TRIAGs 3.5, 5.0, 6.2. Duration to prepare oil samples
XXs 9. UNFRMs 8.0, 11.0. Duration for filtergram preparation
Duration for discussions with plant owner s Oil.
Duration for additional testing s Oil .
XXs 10. RNORMs 7.0, 0.75. Duration to preview oil results
XXs 11. RNORMs 8.0, 1.5. Duration to update database with nonurgent data
XXs 12. RNORMs 180.0, 40.0. Duration for specialist advise
XXs 13. ATRIBs 7. Number of times a machine is repeated on job
XXs 14. ATRIBs 8. Number of times data we re invalid
XXs 15. ATRIBs 9. Number of times extra tests required
a
All duration times for the oil process refer to the preanalysis or the wear debris
analysis.
without this aid the overall duration would have been much closer or
possibly over the limit. Extra staff members were used on a 16-hour
time period. This reflects real-life situations, as with this condition
monitoring service, there are other members of the staff that may not
necessarily work in these processes but possess the skills required and
help when needed.
The program is designed with the use of assign nodes at the start, to
make it easy to alter any of the duration times or percentage rejections.
On this basis the adaptability of the program is very useful, as it has
already proved itself in arriving at the values implemented and the
results produced. Table 9 gives the summary of all times for each
location.
Location A
The times in the system for location A are 37.9 minutes per machine for
the oil process and 39.4 minutes per machine for the vibration process,
giving a total time per machine of 38.0 minutes. The geographic
separation of the staff is one reason for the fast completion times, as
these staff do not share their workload with another location. An
exception is staff5, who performs a minimal amount and is based at
anothe r location. Another reason is that the collection stage predomi-
nantly takes longer to complete and with sufficient staff, not involve d in
the analysis, allows reduction of the machine completion times. Times
for the oil side also had the aid of greasers, who supplied the labor for
20% of the collection, thus contributing to the short time periods. Time
between completions was found to be 39.9 minutes, approximately the
same as the creation time, indicating that the flow through the system
was constant.
Location B
Times for the system resulted in 85.8 minutes per machine for the oil
process and 24.9 minutes per machine for the vibration process, giving
an overall time per machine of 40.0 minutes. Staff6 collects oil and
vibration data, and staff7 performs vibration analysis for this location
only, which, as in location A, assists in reducing the times. The oil
analysis staff members, staff5 and staff9, are requested for several
locations, hence the higher times for the oil side even though the
number of oil machines is one-fifth that of the vibration machines. Also,
it generally takes longer to analyze from the oil side of the system
because of the types of procedures involved. Hence, if these staff
members are in more demand, the times in the syste m increased
considerably. The flow through the system was 39.8 minutes between
completions, again showing good consistency, as with the system time
results.
Location C
Times for the system resulted in 88.2 minutes per machine for the oil
process and 43.0 minutes per machine for the vibration process, giving
an overall time per machine of 61.4 minutes. Here one staff member is
dedicated to vibration data collection and one to oil sample collection,
benefiting the system times. For the analysis stage the times reflect the
availability of the staff members, hence the longer times for the oil and
slight increase in times for the vibration. Oil analysis utilizes two shared
staff members, staff5 and staff9, whereas the vibration utilizes only
staff11, who is also shared. Another point to note that the number of oil
machines is 41% of the total number of vibration machines, again
affecting the times in the system. The time between completions is 40.1
minutes, again indicating flow consistency through the system.
Location D
For this location the times in the system results were 81.6 minutes per
machine for the oil process and 23.5 minutes per machine for vibration,
giving a total time in the system of 31.1 minutes per machine. The times
for the vibration were reasonably short because of the number of staff
members available for both collection and analysis. These staff mem-
496 E. SZCZERBICKI AND W. WHITE
bers were also availabl e for the oil sample collection, which helped, but
only two shared staff members s staff5 and staff9 . were available for the
analysis, hence the longer times. If staff14 had fewer members, longer
times would be observed because the collection times have a great
bearing on the total times. As in location C, there were 25% fewer
machines, but less staff members to collect the samples. Thus the
results having similar outcomes is an indication of the effect of staffing
levels at the analysis stage. This means that the number of collectors
had little bearing on the result; the driving force behind the longer
times stems from the analysis staffing levels. The time between comple-
tions was 40.0 minutes, again indicating the consistency of the flow
through the system.
Location E
The location times in the system resulted in 106 minutes per machine
for the oil process and 44.4 minutes per machine for vibration, giving a
total time in the system of 61.5 minutes per machine. These were the
longest times in the system for any of the five locations, and looking at
the staffing numbers for the whole processes provides the justification.
The connection between the staffing numbers and the times in the
system, shown here to have the full effect, is something that the other
locations had to a smaller degree. Although there are fewer machines
than in most locations, the numbers are comparable with those for
location C, but the times were influenced by the staffing numbers and
their shared workloads. Again, this is more evident with the oil sampling
and analysis taking longer to perform than those of the vibration,
especially in the analysis stage. Longer times had no effect on the flow
through the system, with a time between completions of 39.6 minutes.
CONCLUSIONS
Several points arose from the program results, ranging from the quality
of the program itself to the possible implementation of system alter-
ations. For the system modeled, simulation demanded some intrinsic
logical loops that proved successful. Although the staffing distribution
had complications, SLAM II handled the logic of the system with simple
execution. In the early stage s of programming some problems evolved
related to the logic required. This logic was simplified and the program
SYSTEM SIMULATION FOR MAINTENANCE 497
REFERENCES
AS 2625-1983. 1983. Rotating and reciprocating m achinery} mechanical v ibra-
tions, Parts 1, 2, 3, and 4. North Sydney: Standards Association of Australia.
Beebe, R. 1988. Machine condition m onitoring. Victoria: Engineering Publica-
tions.
Chen, Y., and X. Li. 1996. Integrated diagnosis using information-gain-weighted
radial basis function neural networks. Comput. Ind. Eng. 30:243 ] 255.
Huang, H.-H., and B. Wang. 1996. Machine fault diagnostic using a transputer
network. Compu t. Ind. Eng. 30:269 ] 281.
Lin, D., and B. Wang. 1996. Performance analysis of rotating machinery using
enhanced cerebellar model articulation controller s E-CMAC. neural net-
works. Com put. Ind. Eng. 30:227 ] 242.
498 E. SZCZERBICKI AND W. WHITE