You are on page 1of 70

( Experimental investigation,prediction and minimization of gd&t

error using taguchi analysis)

A PROJECT REPORT Submitted in


partial fulfilment of the requirement for
the
award of the degree of
BACHELOR OF TECHNOLOGY (B.Tech)
In
Mechanical Engineering
by

(Name of the Student)


(Reg. No.)

MECHANICAL ENGG. , MANIPAL


UNIVERSITY JAIPUR JAIPUR-303007
RAJASTHAN, INDIA

Month/Year
DEPARTMENT OF MECHANICAL ENGG
MANIPAL UNIVERSITY JAIPUR, JAIPUR – 303 007 (RAJASTHAN), INDIA

Date:

CERTIFICATE

This is to certify that the project titled Experimental investigation,prediction


and minimization of gd&t error using taguchi analysis is a record of the
bonafide work done by STUDENT NAME (Reg No.) submitted in partial
fulfilment of the requirements for the award of the Degree of Bachelor of
Technology (B.Tech) in Mechanical Engineering of Manipal University
Jaipur, during the academic year 2017-18.

Dept Guide
Name
Project Guide, Dept of (Name of the
Dept.) Manipal University Jaipur

HOD
Name
HOD, Dept of (Name of the
Dept.) Manipal University
Jaipur
Date:

CERTIFICATE

This is to certify that the project entitled Experimental


investigation,prediction and minimization of gd&t error using taguchi
analysis was carried out by STUDENT NAME (Reg. No.) at COMPANY
NAME, CITY NAME under my guidance during Starting Month, 2017 to
Ending Month 2018.

Supervisor Name
Designation,
Organisation Name, City
ACKNOWLEDGMENTS

It is indeed a great pleasure and privilege to present this report on Experimental investigation,
prediction and minimization of gd&t error using taguchi analysis.

I would like to express my gratitude to Er JITENDRA SINGH, SITE INCHARGE for his invaluable
suggestions,motivation,guidance and support through out the training.His methodology to start from
simple and then deepen through made me to bring out this project report without anxiety.

I am very thankful to friends, colleagues and all other persons who rendered their assistance directly or
indirectly to complete this project work successfully.
ABSTRACT

Tolerance design has become a very sensitive and important issue in product and
process development because of increasing demand for quality products and the
growing requirements for automation in manufacturing. This chapter presents
tolerance stack up analysis of dimensional and geometrical tolerances. The stack up of
tolerances is important for functionality of the mechanical assembly as well as
optimizing the cost of the system. Many industries are aware of the importance of
geometrical dimensioning & Tolerancing (GDT) of their product design. Conventional
methods of tolerance stack up analysis are tedious and time consuming. Stack up of
geometrical tolerances is usually difficult as it involves application of numerous rules
& conditions. This chapter introduces the various approaches viz. Generic Capsule,
Quickie and Catena methods, used towards tolerance stack up analysis for geometrical
tolerances. Automation of stack up of geometrical tolerances can be used for tolerance
allocation on the components as well as their assemblies considering the functionality
of the system. Stack of geometrical tolerances has been performed for individual
components as well as assembly of these components.

The technology has undergone major changes over the centuries to meet the changing
requirement of the society. During World War II, the United States manufactured and
shipped spare parts overseas for the war effort. Many of these parts were made to
specifications but would not assemble. The military recognized that producing parts
that do not properly fit or function is a serious problem since lives depend on
equipment that functions properly. After the war, a committee representing
government, industry, and education spent considerable time and effort investigating
this defective parts problem; this group needed to find a way to insure that parts would
properly fit and function every time. The result was the development of GDT.
Features toleranced with GDT reflect the actual relationship between mating parts.
Drawings with properly applied geometric tolerancing provide the best opportunity for
uniform interpretation and cost-effective assembly.
GDT was created to insure the proper assembly of mating parts, to improve quality,
and to reduce cost. Before designers can properly apply geometric tolerancing, they
must carefully consider the fit and function of each feature of every part. Properly
applied geometric tolerancing insures that every part will assemble every time.
Geometric tolerancing allows the designers to specify the maximum available
tolerance and consequently design the most economical parts.

LIST OF FIGURES
Figure

A detailed drawing of the bench vice assembly ..................................... ,29

Mean response graph ........................................................................... 44

S/N ratio response graph ....................................................................... 47


LIST OF TABLES

Table

2.1 . Table of cost minimization techniques...............................................................1

2.2 . Table of proposed cost tolerance relationships................................................19

4.1 . An L16 InnerlOuter Orthogonal Array Structure.........................................33


4.2 . Variables for the experiment...................................................................................35

4.3 . An L 16 Orthogonal Array.....................................................................................37


5.1 . Response table.............................................................................................................41

5.2 . Mean response table..................................................................................................42

5.3 . S/N response table.....................................................................................................46

5.4 . Mean response ANOVA table................................................................................45

5.5 . S/N ratio ANOVA table...........................................................................................48

5.6 . Confirmation experiment results..........................................................................53


Contents
ACKNOWLEDGEMENTS....................................................................................................................i

LIST OF FIGURES.................................................................................................................................iv

LIST OF TABLES.....................................................................................................................................v

LIST OF APPENDICES........................................................................................................................vi

CHAPTER

1. INTRODUCTION.........................................................................................................1

2. LITERATURE REVIEW............................................................................................6

Heuristic Approaches...................................................................................6

Minimum Cost Optimization methods...................................................9

Lagrange Multiplier.......................................................................11

Linear programming.....................................................................13

Dynamic programming................................................................14

Non-linear programming............................................................15

Design of Experiments (DOE)................................................................19

Taguchi's Parameter Design.....................................................22

Taguchi's Tolerance Design......................................................23

Tolerance Design using Taguchi's Parameter Design.....25

3. CASE STUDY - A BENCH VICE.....................................................................27

4. EXPERIMENTAL DESIGN SETUP AND PROCEDURE.......................3 1


Inner and outer arrays...............................................................................3 1

Designing an experiment using orthogonal arrays......................32

Experimental variable and level selection......................................34

Selection of orthogonal arrays and assignment of factors......36

Conducting the experiment....................................................................37

5. ANALYSIS AND RESULTS..............................................................................39

The mean response analysis.................................................................40

The S/N response analysis....................................................................43

Confirmation experiment.......................................................................49

. Geometrical interpretation 54

6. DISCUSSION AND CONCLUSIONS..........................................................57

BIBLIOGRAPHY..................................................................................................................................60
CHAPTER 1

INTRODUCTION
Manufacturing is a significant portion of the U.S. economy and manufacturing has

experienced a loss in its global competitive position ( Otto and Finnie, 1993). One of the key

factors contributing to the loss of market share to foreign competitors is poor product quality

(Juan, 1988; Deming, 1986). Therefore, much attention has been focused on the issue of

producing quality products, which we will define as a product produced with minimum If a

variation around a target dimensional value. product exhibits too much variation, it is a

non-conforming product or rejects. A reject can be defined as a part produced "out-of-

specification", i.e. not within the acceptable tolerance region. Tolerancing, the method by

which tolerances are assigned and their cumulative effects predicted, plays a key role in

reducing rejects.

A component tolerance is "the total amount by which a specific dimension in an

engineering drawing is permitted to vary [ANSI, 19821". It is the difference between the upper

and lower limits of the specification . Tolerances reflect a designer's intentions regarding product

functional and behavioral requirements with corresponding implications for manufacturing and

quality control. They are important not only because they spec@ product performance, but also

because they have a significant impact on the choice of manufacturing


processes, which in turn determines the final production cost. Tight tolerances can result in

excessive process cost, while loose tolerances may lead to increased rejects and assembly

problems. The engineer must design high quality products and processes at low cost, by

specifying the allowable variation that can be tolerated without loss of component

interchangeability and functionality.

The typical design problem is that the product designer has identified the allowable

product performance variation and must determine the allowable component tolerances,

that when combined, will result in an acceptable product.

Analysis of the effects of tolerances on product performance is difficult because it

requires determining the functional relationship between the component and assembly

tolerances, which is often unknown or very complicated. Thus, traditional tolerancing

methods are the use of past designs, handbooks, or "rules of thumb". These methods are,

often imprecise, not based on relevant data, or insufficient to guarantee a cost effective,

quality design.

Let Y, be the product or assembly response of interest which is a function of n

component features Xi :

Y, = f (XI, X2, ............. X)


9 [l]
Equation [I] is known as the stackup equation I function, which may be linear or non-linear.
Modern tolerancing is divided into two subareas: tolerance analysis and tolerance

allocation or synthesis (Lee and Woo, 1986). Tolerance analysis means to determine the resulting

assembly tolerance, Y,, when the individual component tolerances are, Xi , given;

tolerance allocation means to determine the required component tolerances, X,,


when the

assembly tolerance, Y,, is given.

Tolerance stackup or analysis methods are worst case analyses, statistical tolerancing

(Evans, 1974; Gerth, 1992), and Monte Carlo simulation (Evans, 1975; Araj and Ermer, 1989;

Bjorke, 1989). Tolerance allocation methods include standards, uniform and proportional

scaling (Chase and Greenwood, 1988), various minimum cost opGmization algorithms (Gerth, 94;

Sayed and Kheir, 1985; Spotts, 1973), and Taguchi methods (Liou, et. al., 1993).

These methods have various problems including the following:

← 1.All current tolerancing algorithms, with the exception of Taguchi methods,


require the funcctional relationship between component and assembly features

to be known/specified.

2.Uniform and proportional scaling are iterative and do not relate

tolerances to cost or quality, and the minimum cost methods require cost

information that is typically unavailable (Gerth, 94).


3 . The methods are complicated and cumbersome, and thus, typically

conducted only on "critical" dimension chains, leaving other chains

unanalyzed. This can result in excessively tight tolerance and a costly

product, or loose tolerance and future production problems.

When tolerance analysis/allocation is not conducted, it is possible to have

processes that are very capable, but are still the major sources of product

variation, because their tolerances were set too loosely (Gerth, 92).

The current thesis attempts to primarily address the first two criticisms, i.e.,

conducting a tolerance allocation study when the stacking function is not known, and cost

tolerance information is not available.

The focus of this research is two-fold:

 to develop a method for tolerance allocation based on design of experiments

to determine:

[1] the stacking function,

[2] the significant component tolerances that affect the stackup function, and

[3] the levels of the tolerances that result in the highest quality product;

to demonstrate the use of the method on the specific case study of a bench vice.
The main advantage of this methodology is the functional relationship need not to be

known.

The next chapter will review the literature on heuristic tolerance allocation approaches,

minimum cost optimization methods, and Taguchi's parameter and tolerance design. The case

study bench vice will be introduced in chapter 3. The experimental design method of

tolerance allocation as applied to the bench vice is described in chapter 4. The results are

presented in chapter 5.
CHAPTER 2

LITERATURE REVIEW

Details of past studies have been critically reviewed and are divided into three

sections. It is assumed the reader is familiar with the most common tolerance analysis

methods: worst case, statistical, and Monte Carlo tolerancing (Gerth, 1992; Bjorke, 1989).

The first section briefly explains heuristic approaches to tolerance allocation. The second

section describes minimum cost optimization methods and presents Lagrangian multiplier,

linear programming, dynamic programming, and non-linear programming techniques for

solving the tolerance allocation problem. The third section presents Taguchi's parameter

design, tolerance design, and a new method, tolerance design using Taguchi's parameter

design.

Heuristic Approaches

The important issue in tolerance allocation is to determine which combination of

component tolerances is best. One method is to use non-mathematical techniques such as rules of

thumb, past practice, or standards. Standards, such as IS0 Recommendation 286 and
ANSI B 4.2-1978 are a set of tables through which tolerances can be identified, and hence

manufacturing processes selected (Gutrnann and Caldwell, 1987).

Standards may be used to determine the initial component tolerances, from which the final

assembly tolerance is computed by any of the tolerance analysis methods, such as worst case

tolerancing, statistical tolerancing, or Monte Carlo simulation. If the resulting assembly tolerance

is too large the component tolerances may be adjusted iteratively until an acceptable

set of tolerances is found. Standards are popular because of their ease of use, and they provide a

good starting point. However, they provide no guidelines on how tolerances should

be adjusted, nor do they consider cost or quality tradeoffs between tolerances. Also, they

only spec@ sizes, no geometric tolerances.

Uniform scaling, proportional scaling, and allocation by precision factor are the

tolerance allocation methods which provide non-iterative heuristics on how to allocate the

allowable assembly tolerance among the individual components (Chase and Greenwood, 1988).

Uniform scaling assigns equal tolerances to all parts. The individual component

tolerances, Toli , for linear function can be calculated as:

Where, TolA= the given assembly tolerance,


Tolf, = component tolerance of supplied components (fixed), ni =

total number of components, and nf, = number of fixed

components.

Proportional scaling assumes a set of initial tolerances, Tolinitial , have been selected based on

experience, process, guidelines, or standards. The individual component tolerances are then stacked

according to a tolerance analysis method (worst case tolerancing, statistical tolerancing, Monte Carlo

simulation) and the resulting assembly variation is compared with the

specified assembly tolerance. If the calculated assembly variation is too large or too small, the

individual component tolerances are scaled by a constant proportionality factor, P.


The precision factor method assumes that s i i a r l y siied parts are machined to

similar precision levels. This is based on the theory that the part tolerance increases

approximately with the cube root of part size:

All of the above methods are simple and provide non-iterative solutions to compute

component tolerances. However, their underlying assumptions ignore tolerance tradeoffs

and are not based on cost or quality considerations.

Minimum cost optimization methods

In order to consider the cost tradeoff between tolerances, a number of different cost
minimization algorithms to allocate component tolerances have been developed In
this type
of approach, a manufacturing cost is associated with each tolerance level (cost-tolerance

relationship), and the problem becomes W i n g the component tolerance configuration that

minimizes total cost without exceeding a specified assembly tolerance.

Depending on whether the stackup function is linear or non linear, and whether the

cost tolerance model is discrete or continuous, Lagrange multiplier, dynamic

programming, linear programming and non-linear programming have been used to solve

for the optimal component tolerance set (see Table 2-1).


Linear Programming~:

Hoffman (1982) discussed the problem of tolerance allocation using linear

programming (LP) techniques. LP assumes a known assembly tolerance specification and


operation sequence plan.

Ostwald and Huang (1977) introduced a model of optimal tolerance selection for

functional dimensions assuming discrete production costs. This model was solved using linear
programming with 0-1 variables. In their model, all discrete cost values, CG, weighted by

binary coefficients, Xj, were summed in the system cost function . Only one cost value for each

component was permitted in a trial evaluation. A complex enumeration algorithm was used to solve the

problem. The result depended on the discretization of tolerances. Although it was claimed that only a

small part of the total number of combinations was required for evaluation, the 0- 1 discrete search

method was found insufficient for any practical application.

Dynamic Programming:

Dynamic programming allows one to evaluate several possible tolerance combinations

based on their cost (Enrick, 1985). The method, not only considers the statistical stackup of

the tolerances but also identifies the combination which results in the lowest total costs.

Dynamic programming assumes different discrete costs for each tolerance, Ti . Each tolerance and its

associated cost is considered one by one in each phase of the program. Beginning with

XI , Xz , and their associated costs, all possible combinations of their stack is evaluated. Thus if two

possible tolerances existed for each, there would be 4 possible combinations (branches). Any branch

that exceeds the required assembly tolerance is immediately pruned (discarded).

Also any combination that results in a looser tolerance at higher cost than another

combination is also pruned.


In this manner an efficient search of all combinations of tolerance cost fbnctions is

possible. The limiting factors are that this method can only be applied to discrete cost

tolerance relationships. It should be possible to apply it to non-linear tolerance chains,

although this has not been done to date and would be an area of h r e research.

Non- linear Programming~:

Sayed and Kheir (1985) describe a technique that assigns system element tolerances in a

manner that minimizes cost. They present a continuous tolerance solution which gives an absolute

minimum cost and serves as a basis for the selection of discrete tolerance values.

Similar methods have been developed by Lee and Woo (1986) for the continuous non-linear

solution. They propose probabilistic optimization by associating the tolerance, Toli, with a

standard deviation oi. Thus, all parameters are described by random variables and their

first and second moments:


.
This method works well for non-linear systems, and it allows rapid evaluation of

tolerance analysis embedded in tolerance synthesis. However, care should be taken in

setting up the problem formulation because no existing algorithm guarantees a globally

optimal solution unless the objective function and the constraints are of certain forms.

All of the above minimum cost optimization methods require knowing the following:

1.the function between component and assembly tolerance,

2. the assembly tolerance,

3. the cost-tolerance relationship,

4. tolerance analysis model.

Whether it is a linear or non-linear tolerance allocation method, it is important to have the

continuous function of the cost-tolerance relationship. This relationship is shown in Table 2-2.

Usually, it is difficult to have the functional stackup relationship (Liou, et. al., 1993), therefore, a

design of experiment approach is reviewed, and presented in the next section.


Design of Experiment(DOE)

Relatively few studies have been conducted in the past to allocate tolerances using

designed of experiments. Design of experiments is a discipline that enables one to

systematically vary a number of independent (input) variables to evaluate their effect on a


number of dependent (output) variables or responses. As the variables are changed in the

experimental process, their relationships, effects, and interactions are measured, analyzed

and mapped.

Taguchi introduced a method that provides a simple way to design efficient and cost

effective experiments (Taguchi and Wu,1979; Taguchi, 1986; Ross, 1988). Taguchi's method

has become very popular and is widely used to optimize industrial designs and processes. The

method identifies those factors (independent variables) that have a significant effect on the

performance (dependent) variable by using designed experiments.

Taguchi's philosophy is based on the loss function concept. "The quality of a product

is the (minimum) loss imparted by the product to society from the time the product is shipped"

(Elyrne and Taguchi, 1987). The idea is that loss occurs not only when the product falls

outside the specifications, but also that the loss continually increases as the part deviates

further from its nominal dimension (target value). Taguchi uses a simple quadratic function

to approximate the behavior of the loss.

His quadratic loss function can be represented as follows:

L=K(X-T)2

where, L = Loss in dollars

K= Cost coefficient
X= Value of quality characteristic

T= Target value (m)

Thus, the minimum cost product is the one that produces the minimum variation

around the nominal target. This concept is also applied to this research. The question

becomes, how to design a minimum variance product.

Taguchi views the design of a product or process as a three phase program (Byrne and

Taguchi, 1987):

1. System design

2. Parameter design

3. Tolerance design

System design is the phase when new concepts, ideas, and methods are generated to

provide new or improved products to customers.

The parameter design phase is crucial to improving the uniformity of a product and

can be done at no cost or even at a savings. This means certain parameters of a product or

process design are set to make the performance less sensitive to causes of variation.

The tolerance design phase improves quality. Quality is improved by tightening

tolerances on product or process parameters to reduce the performance variation. This is

done only after parameter design.


The research project assumes that the system and parameter design have already been

properly conducted. However, s i i the proposed tolerancing method involves parameter design

concepts, a brief description of traditional parameter design will be provided followed by a

discussion of Taguchi's tolerance design and why it is not a preferred method.

Taguchi's Parameter Design Paradigm:

Taguchi's Parameter design evaluates alternative nominal values for selected control

variables by statistical experimental design to determine the best combination of values that will

result in a product that is least sensitive to noise factors. This is called robust design and is the key

to achieving high quality without increasing product cost (Byrne and Taguchi, 1987).

The key to achieving robustness against noise is to discover and utilize interactions

between controllable factors and uncontrollable (noise) factors. Thus, the strategy in parameter

design is to separate controllable versus noise (uncontrollable) factors into an inner and outer array

and to study their effect in a statistical experiment. The most important. purpose of the

outer (noise) array is to deliberately create noise during the experiment so that the controllable

factor levels which are least sensitive to noise can be identified. Dr. Taguchi adopted orthogonal

arrays, or fdly saturated fractional factorial designs, to simplify the experimental

design procedure. The signal to noise ratio is used as a data transformation, and the equation
for calculating the S/N ratios are based on the characteristics of the response variables being

evaluated.

Taguchi's Tolerance Design Paradigm:

The objective of tolerance design is to improve quality by reducing variation in response. It is

only used when efforts of parameter design have not proved adequate in reducing variation. Tighter

tolerances are then required of those factors identified through parameter design to reduce performance

variation. Tolerance design typically means "spending money", buying better grade materials,

components, or machinery to control the range of design parameters (Byrne and Taguchi,

1987).

Taguchi's tolerance design uses the loss function concept and assumes that components are

statistically independent and normally distributed. The first step of tolerance design is to

determine the contribution of each noise factor to the quality loss (Phadke, 1989). In order

to improve the joint economics of product cost and quality loss, one should consider the

following two issues:

1. Ways of reducing the variation of the noise factors that contribute a large amount to

the quality loss

2. Ways of saving cost by allowing wider variation for the noise factors that contribute

only a small amount to the quality loss.


DErrico and Zaino ( 1988) presented a modification of Taguchi's tolerance design paradigm

where they assume a 3 point distribution that provides a closer approximation to the

normal.

Taguchi's tolerance design paradigm is better than traditional statistical

tolerancing methods (DErrico and Zaino, 1988) because:

1. The method is easy to use and can be easily described to scientists and engineers.

2. The method does not require the assembly function to be expressed in analytical form.

3. The method requires little computation as compared to Monte Carlo simulation, and

the Taylor-series expansion method.

4. This method is comparable in accuracy to Taylor Series expansion.

However, this method has the following disadvantages:

1 . When the number of components is large it requires an excessive number of experiments.

For example, to determine 12 component tolerances would require 256 experiments.

2. It is difficult to handle non normal distributions.

3. The loss function is difficult to quantify.


Tolerance Design using Taguchi's Parameter Design:

A new application of Taguchi's parameter design technique is utilizing the inner

and outer orthogonal array concept to the tolerance specification of robot kinematic

parameters (Liou et al., 1993). In order to illustrate the methodology, they use two different types

of robot manipulators, a two link planar manipulator and a five-degree-of-freedom Rhino robot.

Initially, they assigned a reasonable tolerance range for each joint using traditional past design

practices (heuristic methods). However, they believed that there were tolerance tradeoffs between

the manipulator joints, which if explored, would result in more consistent

placement of the robot's end effector. The advantage of larger tolerances at the joints would be

that less expensive servo controllers would be required. They used the inner orthogonal array to

study specific alternative tolerance ranges (loose, tight), and the outer orthogonal array to provide

the noise to each control factor, namely the direction that tolerance value deviates from a nominal

position (+, -). They then measured the deviation of the end effector position from nominal, and

analyzed the results using Taguchi's S/N ratio and analysis of variance (ANOVA) technique. They

compared the experimental results with results from a

Monte Carlo (MC) simulation and found that both methods reveal the same trends for

performance improvement. Their analysis indicated that tightening the tolerance of one factor

was the most cost effective way to improve the end effector's performance and to reduce the

overall cost. They concluded that parameter design was more computationally efficient than

MC methods and an effective method of tolerance allocation.


Application of Taguchi's parameter design to the tolerance allocation problem

appears promising. Its major advantage is that the functional stackup relationship between the

performance parameters and the component factors need not be known. Another advantage is

that it need not rely on a loss function, which cost coefficient is often difficult to quantify.

This thesis explores the application of this approach to a hypothetical design situation of a bench
vice to determine it's suitability for tolerance allocation compared tominimum
cost approaches.
CHAPTER 3

CASE STUDY - A BENCH VICE

The case study is introduced at this time to provide the reader with a concrete example

that will aid in the understanding of the parameter design approach to tolerance allocation. The case

study is a non fixing bench vice made from plastic as shown in Figure 3-1. The vice is made of six

components, namely, a front plate, a movable jaw, an end plate, two rods, and a screw rod. Each

component is manufactured separately and then assembled. Each component feature has a nominal

dimension and a tolerance specified by the designer.

One of the functional characteristics of the bench vice is that the end plate and the

movable jaw be parallel. This means that there should be no gap between the two plates when

the vice is filly closed. But due to the variation in the dimension of the parts and overall

stackup tolerance of the assembly, the two plates may not be perfectly parallel.

The goal of this thesis is to apply the parameter design concept to determine the set

of component feature tolerances that will result in the lowest cost product subject to quality

constraints on the product's function. Lowest cost means the largest tolerances and quality

constraints on the product function means a minimum gap between the jaw and the end

plate when the jaw is closed.


The methodology was evaluated by "building" and "assembling" the required bench

vices. Since actual construction was not feasible, a simulation program based on a 2-D

mathematical model of the manufacturing and assembly processes was written.

A parametric model of the bench vice was developed (see Figure 3 - 1). Points P 1 and

P3 are the end points of the end plate. Similarly, points P2 and P4 are the end points of the

movable jaw. Point P5 is the center of the rod hole in the end plate and point P6 is the center of

the rod hole in the front plate. Points P7, P8, P9, and PI0 are at the outer diameter of the rod

where it fits into the movable jaw. The complete Geometric Dimensioning & Tolerancing

(GD&T) component drawings are given in Appendix A

In the mathematical model the following assumptions are made:

1. The analysis is restricted to 2 dimensions, i.e., there is no variation along the x-

axis (into the page) for any feature (see Figure 3-1).

2. As shown in Figure 3-1, it is assumed that point P2 is very close to point P1 and the

difference (A) between point P2 and P 1 along the Y-axis is negligible. This assumption

is justified because the difference (A) is very small compared to the

length of the movable jaw.

3. The rod is straight, i.e. not curved or eccentric, although assembly variation can

cause it to be at an angle with respect to the z-axis.


BENCH VICE ASSEMBLY

Figure 3-1. A detailed drawing of the bench vice assembly


4. The center of the rod was assembled to the center of its mating holes at the front and

end plates, i.e. no clearance.

5. All part surfaces are flat, but not necessarily parallel.

Appendix B (Figure B1) shows a detailed analysis of the movable jaw and rod position.

Appendix B (Figure B2) shows the complete analysis for the case of Pl = P2. For illustration

purposes we consider the case of P1 = P2 and a complete mathematical formula is developed for

this case. Appendix C shows the list of formulae and calculations.


CHAPTER 4

EXPERIMENTAL DESIGN SETUP AND

PROCEDURE

The tolerance allocation methodology involving Taguchi's parameter design was

evaluated on the case study of bench vice. Details pertaining to the experiment are

presented below.

Inner and Outer Arravs :

In this experiment, the concept of inner and outer orthogonal may (OA) is used. The

inner OA is used to study the effect of the controllable factors and the outer OA provides noise

to each control factor. The objective of the inner OA is to determine the significance of the control

factors, and to select the levels of the significant factors that optimize the performance measure.

Optimal performance is one that exhibits minimum variation around a target value.

The variation is considered to be due to noise factors. The outer OA introduces noise

factors in the experiment in a systematic manner. Thus, it is possible to analyze and select

the proper levels of the control factors that are least sensitive to noise and will result in the

minimum variance response.


Each outer OA noise combination is treated as a replication of the inner OA. Thus, if an

outer OA requires Lo runs for each run of an inner OA that require L, runs, the size of the

experiment is L,* Lo runs. The inner array specifies the combination of tolerances ( loose = L1,

tight = L2 ) used for each control variable ( component feature ). The outer array specifies

the level of noise combinations, i.e., the direction ( + = N1, - = N2 ) that the tolerances
deviate from their nominal direction. Togethers, they define a unique product which

component features are at the extremes of their tolerances. The structure of the inner and

outer orthogonal array is shown in Table 4-1.

Designing an experiment using orthogonal arrays:

When designing an experiment, the factors, relevant interactions, and the factor levels

need to be determined. In the Taguchi method, two levels (minimum and maximum) are

usually recommended, but three levels (minimum, medium and maximum) may also be used.

Interaction effects between factors are usually assumed to be negligible compared to main

effects, but can be investigated. While using the Taguchi method, the experiment is designed by

following the column assignments specified by an orthogonal array (OA). The orthogonal

design employed is based on the number of factors, their levels and the number of selected

interactions. The most common O K s involve two level factors, and are the Ld ( 23), L g ( 2'1, L16 (

215),and L32( 231)orthogonal arrays. In an orthogonal array designated as La( bc), the
letters a, b, and c represent the number of runs, the number of levels for each factor, and

the number of columns in the array respectively. After an orthogonal array is selected,

designing an experiment becomes a "column assignment" task.

Experimental Variable and Level Selection:

Response Variable ( Dependent Variables ) :

The response variable is the maximum gap between the end plate and movable jaw

in the closed position measured at the corners of the plates.

Factors ( Independent Variables ) :

Twelve tolerances (control factors) can affect the gap between the end plate and the movable

jaw. Two tolerance levels were selected for each control factor: L1 (Loose tolerance)

and L2 (Tight tolerance). A linear model was assumed, and interactions were assumed to be negogible.

The specific factor levels were selected based on standards and past experience with machining

processes and their capabilities (see Table 4-2). The factor levels were generally selected to be far apart

so that the factor effect would large relative to the mean squared error.

However, certain tolerances, namely, B, D, H, and K had very little difference between

their loose and their tight tolerance. Thus, their perceived effect may appear to be

relatively small. (see Chapter 5)

Noise Variables:

The noise factor for each tolerance is the direction, + or - , the feature deviates from its

nominal.
Table 4-2 summarizes the variables and their levels used in the experiment.

Table 4-2. Variables for the experiment


Selection of Orthogonal Arrays and Assignment of Factors:

The selection of OA depends on the following :

1. The number of factors and interactions of interest

2. The number of levels for each factor of interest

These two items determine the total degrees of freedom required for

the entire experiment. The degrees of freedom for each factor is the number

of levels minus one. The degrees of freedom for the factors under

investigation, v, assuming no interactions, is given as v = 12 *(2-1) = 12

Thus, an OA is required that will accommodate v the total number of degrees of freedom. The

total degrees of freedom available in an 04 VLN , is equal to the number of trial minus

one: Vln = N-1

In order to select the particular orthogonal array for an experiment the following

inequality must be satisfied.

Vln>V

Since we have 12 control factors, an L 16 OA with 15 degrees of freedom is required.

The standard an L16 orthogonal array is shown in Table 4-3.


Table 4-3. An L 16 orthogonal array

After selecting the appropriate OA, the factors and interactions can be assigned to the

appropriate columns.

conducting the experiment :

To conduct the experimental investigation and simulate the various bench vices, a simulation

program was developed in Microsoft Excel version 4.0 (Microsoft, 1992) based on a geometrical

model of the vice to measure the gap between the end plate and movable jaw.
An L16 orthogonal array was used for both the inner and outer array resulting in 256

experiments.

Usually complete randomization should be used during an experiment to prevent the

influence of unknown and uncontrolled factors (Mead, 1988; Baker, 1990). However, s i i

the experiment was conducted as a computer simulation, all variables were known and

controlled, and thus, randomization was not necessary.

A macro program in Microsoft Excel Version 4.0 (Microsoft, 1992) was developed to

collect and analyze the data automatically. The detailed listing of the simulation program,

formulae, and macro program is in Appendix E, K, L, respectively.


CHAPTER 5

ANALYSIS AND RESULTS

In the Taguchi method, the signal to noise ratio ( S M ) is used as the data

transformation method that consolidates the data for each control array row over the various noise

levels into one value which computes both the mean and the variation present in the data. The

equations for calculating the signal to noise ratios are based on the characteristics of the response

variables being evaluated; nominal the better, smaller the better, and larger the better. In the bench

vice case, the response variable, i.e., the gap between the end plate and the

movable jaw is a smaller the better characteristic, and the S/N equation for this

characteristic is given as:

n = number of repetitions
where,
yi = individual data.

The S/N ratio for the smaller the better characteristic is essentially a measure of both the

mean value (signal) and the standard deviation (noise) of the response (Montgomery, 1991).

Because of the negative sign in [22], higher S/N ratios are indicative of experimental
conditions that will be more robust i.e., less sensitive to variation in the noise variable

(Clausing, 1988). Thus, it can be expected that statistically significant results found in the

analysis using S/N ratio will be the "best set" of tolerances to produce the minimum gap

between the end plate and movable jaw regardless of the other tolerance levels and their

direction.

The data is presented in Appendix F. The following analysis were performed: the

mean response analysis and the S/N response analysis. The mean response analysis shows

which factors have the greatest impact on the mean of the response distribution. The S / N

analysis show which factors have the greatest impact on the variance and mean of the

response distribution.

The mean response analysis :

The following steps are used to conduct the mean response analysis ( Peace, 1993):

1. Determine the mean for each row (experimental run). The mean for each row is

shown in Table 5-1.

2. Calculate the mean response for each control factor and develop a mean response table. This

is performed by grouping the mean responses by factor level for each column in the array,

taking the sum,and dividing by the number of responses. The absolute difference
or delta between the two results (two levels) is the effect of the factor. For factor A,

the calculations are as follows:

A, = ( 0.03068 + 0.03039 + ........................+ 0.02325 ) / 8 = 0.0214662 inches

A, = ( 0.01999 + 0.01999 + ........................+ 0.02539 ) / 8 = 0.0225062 inches

Average Gap
Run No. (inches) S/N
1 0.03068 28.78188
2 0.03039 28.86940
3 0.01 198 37.04638
4 0.01 138 37.39182
5 0.02037 32.43002
6 0.02012 32.54097
7 0.02356 30.45576
8 0.02325 30.57666
9 0.01999 32.60752
10 0.01999 32.72910
11 0.02353 30.57395
12 0.02323 30.70830
13 0.02141 33.00008
14 0.02083 33.22325
15 0.02568 30.12808
16 0.02539 30.19306

C
Total 0.35178 5 11.25623

Table 5- 1. Response Table


The results for the other columns, is presented in Table 5-2.

Table 5-2. Mean Response Table

3. Construct a mean response graph for each factor based on the mean response table

(see Figure 5-1).

4. Analyze the mean response table and mean response graphs. From Table 5-2 and Figure 5-1,

it can be seen that factor F (End plate P3 plate depth), Factor E (End plate P1 plate depth),

and Factor G ( movable jaw hole position 1) have the greatest effect on the gap
between the end plate and movable jaw, followed by factors D (End plate hole

diameter l), B (Front plate hole diameter I), and A (Front plate hole position 1). The

remaining factor effects is negligible by comparison.

5. Determine the statistical significance of the factors with an ANOVA (see Table 5-4). The

conclusions from the graphical analysis are verified by the mean response ANOVA.

Since our quality characteristic is smaller-the-better, we want to choose the tolerance

levels from the response table (or response graphs) that result in smaller average response values.

The recommended levels are E2 , F2, and G2. Since the effect of the other factors is

small, their levels can be selected based on other considerations, such as cost. In this case we

will assume that larger tolerances result in lower cost, and thus the other factors should all

be set at their loose tolerance level 1.

The S/N response analysis :

The same steps are used to conduct the S/N response analysis.

1. Determine the S/N ratio for smaller-the-better case for each row (experimental run).

2. Calculate the SM response for each control factor and develop a response table. For

factor A, the calculations are as follows:


Figure 5- 1. Mean Response Graph
Figure 5-2. S/N ratio Response Graph
3 . Construct S/N response computed for each factor, S/N ratio response graphs

are constructed for each factor as shown in Figure 5-2.

4. Analyze the S/N response table and response graph. From Table 5-3 and Figure 5-2, it

can be seen that the same factors that had the greatest effect on the mean response also

have the greatest effect on the variance of the gap, namely, E, F, G followed by B, C, A

with the others having a negligible effect.

5. Determine the statistical significance of the factors with an ANOVA (see Table 5-5). The

conclusions from the graphical analysis are strongly supported by the S/N ratio ANOVA.

Since our quality characteristics is smaller-the-better, we want to choose the level from the

response table (or response graphs) that shows smaller average response value. From the

above analysis, the recommended levels are selected as E2 ,F2, and Gz, which are the

same levels recommended by the mean analysis.

Confirmation Experiment:

To verify the experimental conclusion obtained above, a confirmation experiment was

conducted. During experiment it was assumed that the interaction effect will be negligible

because the response was hear over the narrow range of tolerances. The confirmation

experiment is the final step to verify the interaction assumption. Sixteen bench vices were
simulated on Microsoft Excel version 4.0 (Microsoft, 1992) spreadsheet under the

recommended conditions A1 ,B1,C1, D l , E2, F2, G2, HI , I1 , J1, Kl , and Li .

The predicted average gap between the end plate and movable jaw when the

factors are at their recommended levels is calculated as follows: (Ross, 1988)

= 0.0191 + 0.0195 + 0.01936 - 2 * 0.022


= 0.014 inches

The 99 % confidence interval for the confirmation experiment of 16 runs at the

recommended levels is given by


The confirmation experiment result are shown in Table 5-6.
Run No. Minimum gap at closure position
1 0.01243
2 0.01031
3 0.01243
4 0.01031
5 0.00104
6 0.00275
7 0.02596
8 0.02072
9 0.013 18
10 0.01152
11 0.01318
12 0.01152
13 0.02189
14 0.00188
15 0.02447
16 0.02114
Average 0.01342
S/N 36.0699

Table 5-6. Confirmation experiment results

The average and S/N ratio of the confirmation experiment results are 0.0134 inch

and 36.069 db. It can be seen that both are well within their predicted values of the mean

response and S/N ratio analysis.


The result of the experiment and analysis
can be summarized as follows:

1. The three significant factors that influence the gap between the end plate and

movable jaw at closure position are E (End plate P1 plate depth), F (End plate P3 plate

depth) and G (Movable jaw hole position 1). The recommended levels for these three

factors are Ez, Fz, and GZ. Therefore, the significant factors E, F, and G should

be set at the tighter tolerance levels and the insignificant factors should be set at

loose tolerance levels as their tolerance at the chosen level does not have any

significant effect on the gap at the closure position.

2. The interaction between the control factors are not significant.

3. The interaction between the noise factor (direction of tolerance) and control

factors are not significant.

4. The two analysis results are compatible.

5. The confirmation experimental results under the recommended conditions Ez, Fz, and

Gz are well within the intervals for the confirmation experiment.

Geometrical Interpretation :

The above analysis indicates that tightening the tolerance of factor E (End plate P1

depth), F (End plate P3 plate depth) and G (Movable jaw hole position 1) is the most effective

way to minimize the gap between the end plate and movable jaw at closure position. If

necessary and feasible, the tolerances of factor D (End plate hole diameter I), B (Front plate
hole diameter I), and A (Front plate hole position 1) can also be set at a tighter level to further

reduce the gap between the end plate and movable jaw. Of equal importance, tolerances of

insignificant factors such as C, H, I, J, K, and L can be widened to reduce manufacturing

cost without significantly affecting the gap.

Examination of Figure 3-1 shows that the above results makes physical sense and can be further

elaborated through the use of a gain or sensitivity matrix. The statistical tolerancing

equation for non-linear systems is given by

From Figure 3-1 it can be seen that the points PI, P3, P2, and P4 directly affect the gap, since

they define the parallelness of the mating surfaces. Points P1 and P3 have a greater effect than P2 and

P4 as the clearance between the rod and the hole in the movable jaw, reduce

the effective gain in the slope of the P2 - P4 surface. Specifically, the true position of the
movable jaw has a greater impact on the gap because it effectively moves the plate along

the Y-axis increasing and decreasing the gap size when the jaw is skewed. This effect is

magnified
by the fact that the hole is off-center creating a longer lever arm (P7 - P4 versus P9 - P2).
Because the movable jaw surface is shorter than the length of the rod, the jaw's skewness has
a greater impact than the slope of the rod (P5 - P6). Thus, the physical interpretation of the
problem confirm the results obtained from mean response analysis and S/N ratio response
analysis.
CHAPTER 6

DISCUSSIONS AND CONCLUSIONS

A new method has been demonstrated for tolerance allocation. The method is based on

Taguchi's parameter design. It is a new application of Taguchi's parameter design to tolerance

design where control factors are the tolerance levels (loose and tight tolerance) and the noise

factors are the direction of tolerances (positive and negative). The concept of employing inner and

outer orthogonal arrays to identify the significant tolerances and select the optimal levels was

successfully demonstrated on a bench vice case study.

From the experiment and the data analysis result, the following conclusions can be

drawn:

1. As with other systems, the Pareto principles applies: there are a few tolerances that

must be held tightly to minimize the variation in the performance parameters. The

other tolerances may be wide to reduce the manufacturing cost. This conclusion is

consistent with other tolerance allocation studies (Gerth, 94).

2. The linear model assuming negligible interactions was appropriate. This is not

surprising, given that the response varies over a tolerance range and not a choice

between nominal values which is typically much wider than the tolerances.
The major advantage of using Taguchi's parameter design method for tolerance

allocation is that the actual stackup function need not be known. This is very important because many

systems are so complex that the function can not be easily determined.

For example, in a direct current motor the exact relationship between the angular deviation of the

commutator and the current draw and magnetic field strength is not precisely known. All other

tolerancing methodologies assume that the relationship is known.

Furthermore, the method identifies the high impact tolerances. Thus, the engineer will

know where to focus engineering resources to either a) change the design to reduce the

gain, or b) develop more capable processing technology to reduce the variation.

The method is easy to use and can be easily described to scientists and engineers who

are familiar with Taguchi's parameter design methodology for selecting optimal nominal

values. The method is comparable in accuracy to statistical tolerancing which uses a first

order Taylor-Series expansion, since Taguchi's method also assumes a linear response

finction based on main effects.

The major disadvantage is that the method requires an inordinate number of experiments.

For example, to determine 15 tolerances requires 256 experiments, and to determine 3 1 tolerances

requires 1024 experiments. Gerth (1992) analyzed a complex system


using Monte Carlo analysis that contained 160 tolerances. Systems of such complexity could

not be reasonably handled by this method.

In conclusion, if only a few tolerance are to be studied, the logistics of conducting the

experiment are reasonable, and the tolerance stack is too complex to be described

deterministically, then this method can be a very powerful tolerance allocation procedure.
BIBLIOGRAPHY

American National Standard Institute. Dimensions and Tolerancing- ANSI Y 14.5 M


-1982. New York, New York: The American Society of Mechanical Engineers, 1983.

Araj, S. and D.S. Ermer. "Integrated Simultaneous Engineering Tolerances." In Quality


Improvement Techniques for Manufacturing Products and Services, editor A.H.,
Abdelmonem. 97- 114. Dearborn, Michigan, USA: American Society of Mechanical
Engineers, 1989.

Barker, T. B. Engineering Quality by Design, Marcel Dekker, Inc. (1990).

Bjorke, oyvind. Computer-Aided Tolerancing, 2nd edition. New York, New York:
ASME Press, 1989.

Byrne, D. M. and Taguchi, Shin. "The Taguchi approach to parameter design." 40th
Annual Quality Congress Transactions, 1987.

Chase, K. W. and W.H. Greenwood. " Design Issues in Mechanical Tolerance


Analysis." Manufacturing Review 1, 1 (March, 1988): 50-59.

Chase, K. W., et. al., "Least Cost Tolerance Allocation for Mechanical Assemblies
with Automated Process Selection". Manufacturing Review 3, 1 (March, 1990):
49-59.

Clausing, D. "Taguchi methods to improve the development process". IEE


International Conference on Communications, 12- 15, June, 1988.

You might also like