Vous êtes sur la page 1sur 6

Allocation of Geometric Tolerances: New Criterion and Methodology

Ashraf 0. Nassef, Hoda A. ElPJlaraghy (1) Industrial and Manufacturing Systems Engineering Dept., University of Windsor, Ont., Canada Received on January 8,1997

Abstract This paper presents a new approach to the synthesis of geometric tolerances in the design process. In the published literature minimum cost has been the main criterion for tolerance allocation. Although this is satisfactory for dimensional tolerances, extending it to geometric tolerances presents a major drawback. The reason is that for each feature the number of combinations of tolerance types is large This paper presents a novel approach to the tolerance allocation problem that consists of 2 steps: (i) the manufacturing processes that minimize the production cost are chosen, then (ii) the types and magnitudes of geometric tolerances are optimized using genetic algorithms to reduce the probability of rejecting good parts or accepting bad parts during inspection, where a bad part is a one which will cause a violation of the assembly functional requirements. Keywords: Design, Tolerancing, Optimization

1. INTRODUCTION

Tolerance allocation is an important step in the design phase of any product. It refers to a class of problems where the values of tolerances imposed on dimensions and geometries are selected to satisfy a certain criterion. For a long time tolerance allocation was confined to dimensional tolerances (5,8), where the minimum cost was used as the allocation criterion. The same criterion was used by researchers who explored the allocation of geometric tolerances (4,6).This paper shows that this criterion has drawbacks especially when allocating tolerance types and magnitudes. Farmer and Gladman (2) state that geometric tolerances should be allocated to assembly features to satisfy the design functional requirements. Hence, those parts which are not rejected during inspection should not cause a violation of the assembly functional requirements once they are assembled to the other parts. A new tolerance allocation criterion is presented in this paper along with an algorithm for the selection of types and magnitudes of geometric tolerances.
2. MINIMUM COST CRITERION

Figure 1 (a) shows a discrete cost vs tolerance curve that has been used for allocating dimensional tolerances (5). It related tolerance values to the production cost, where each process Si was associated with a tolerance value ti equal to the difference between the limits between which each dimension was produced. Since a tradeoff exists between decreasing the production cost and the decreasing the rejection rate, tolerances were allocated to minimize the production cost, while the rejection rate was constrained to a predetermined value. However, this approach poses a problem for the selection of types and magnitudes of geometric tolerances. Figure 1 (b) shows two cylindrical offset surfaces, between which a manufactured cylinder is produced according to a manufacturing process accuracy i and its associated probability distribution. Unlike the case of dimensional tolerances, there is no single tolerance value to be associated with the manufacturing process which can in turn be used in the optimization process. Tolerancing standards (1) define four major geometric controls for features of size. Each of these controls can be fulfilled by a number of tolerance types. These controls and types can be summarized as shown in Table 1 Control Tolerance Types
1. Size Tolerance

I I

Size Form

I ~ t j T P e r p e n d i c u l a r i t y 2. , Parallelism 3. Angularity Orientation 5,6,7, ... Any of the above tolerances with respect to different datums 1,2,3,4 ,... Position Tolerance with Position respect to different datums Table 1 Tolerance Types

I I

1. Straightness, Z.Flatness, 3. Circularity 4. Cvlindricitv

Annals of the ClRP Vol. 46/1/1997

101

.M
4

Raw Parts
224.0
0
w1

Manufactured Parts

171.0
0

Accepted Parts
I

Accepted Assemblies

V Rejected Parts

Relected Assemblies

Fig3 Flow of Parts through Industrial Processes

v
Fig.2 Speed Reducer's Casing

Consider the selection of an orientation tolerance (perpendicularity or parallelism) for the hole marked C5 of the speed reducer casing shown in Fig.2 . There are more than 6 possible orientation tolerances that can be used to control the hole's orientation as shown in Table 2 .

1. - w.r.t. datum 6 2. w.r.t. datum C 3. Nw.r.t. datum D 4. Nw.r.t. datum E 5. 1lw.r.t. datum H 6. //w.r.t. datum G 7,8,9 .... //w.r.t. datums established by more than one feature

__

Legend:

//= Parallelism -= Perpendicularity w.r.t. = with respect to

Table 2 Candidate Orientation Tolerances for Hole C5

If the minimum cost criterion is used to select tolerancetypes and magnitudes for the casing holes in Fig.2 , then at least six cosf vs orientation tolerance magnitude relationshipsfor orientation control of a single hole, are needed as a priori to formulate the problem. Similarly, there are at least five candidate datum reference frames to control the position of each hole and at least three form tolerance types to control each hole's form (circularity, cylindricity and straightness). Therefore, for the five holes, at least 30 cost vs orienfation tolerance magnitude relationships, 25 cost vs position tolerance magnitude relationships, 15 costvs form tolerance magnitude relationships and 5 cost vs size tolerance magnilude relationships, are needed as priori to select tolerance types and magnitudes using the minimum cost criteria. The input data become more cumbersome as the number of features in the problem becomes large, as in the case of a multi-step gear box casing.

Fig.3 shows the industrial processes which are used to manufacture assemblies from raw parts. The following processes have different costs associated with them: 1) manufacturing processes, which include the cost of machines, tools, operators and scrap, 2) manufacturing sequences, which dictate the fixturing set-up of parts, 3) inspection processes which include the cost of the inspection machines and operators, and 4) assembly processes, which include the assembly tools and operators. At the same time manufacturing processes are the ones which cause the geometric variability in parts' surfaces that need to be controlled by tolerances. Hence, a manufacturing process which produces parts' surfaces with high variability will produce parts that have a higher probability of violating the assembly functional requirements when assembled to other parts. The minimum cost criterion (while constraining the percentage of rejected assemblies to a predetermined value) can be used to select the manufacturing processes and sequences. So, where does the geometric tolerances stand? The answer is in the the inspection process. Since it is difficult to check whether the functional requirements are within the allowable ranges after the assembly process, tolerances are specified during the design phase on surfaces of individual parts such that a "bad" part (one which will cause a violation of the functional requirements when assembled to other parts) is rejected during inspection. Hence, if geometric tolerances are properly specified in the design phase, then all bad parts will be rejected during inspection, and the assemblies obtained after the assembly process will not violate the specified functional requirements.
4. PROBABILITY OF MISMATCH

3. FROM MANUFACTURING TO ASSEMBLY

Fig.4 (a) shows a top view of a cylindrical feature represented by a set of points generated from a manufacturing process distribution. The feature is included in some assembly having a set of functional requirements. Fig.4 (b) shows the effect of the functional requirements on two points with radii r and r' on the feature's surface. Since process S2 has less variance than S1, the probability that r and r' will satisfy the functional requirements by using but S1 process S2 is higher than that evaluated by using S,, has less cost than S2.The tradeoff between maximizing the probability of satisfying the functional requirements and minimizing the cost applies. What is selected here is the process generating each feature.

102

is equal to the probability that the assembly will belong to the region R,, where u(r1,r2,...,rn) is the joint probability distribution over all radii and n is the number of radii (random variables). P , will be denoted as the probability of mismatch. Hence, if the geometric tolerances are allocated such that all parts, are accepted during inspection, will produce assemblies which satisfy the design functional requirements, f'm would be equal to zero.
4.1 Evaluation of the Mismatch Probability

E &=Set of feasible values for r and r' satisfying


the functional requirements

629 R1=Set of values of rand r' produced by the


manufacturing process S1

CZZ Rp=Set of values of r and r' produced by the


manufacturing process S2

~7RGT=Set of feasible values of r and r' satisfying


the geometric tolerance specification

Fig.4 Feasibility Space for a Cylindrical Feature


Since it would be difficult to verify the functional requirements after completing the assembly, the geometry of the individual manufactured parts is controlled with a set of geometric tolerances in order to reject bad parts during inspection. The understanding of the relationship between geometric tolerances, assembly functional requirements and manufactured parts is facilitated by drawing the feasible areas over two radii (two points are used for demonstration since it is impossible to view a diagram of the whole feature in hyperspace) of a manufactured feature as shown in Fig.4. The area enclosed by the dashed lines in Fig.4 (RGT) represent the set of feasible values for r and r' satisfying the imposed geometric tolerances. If the imposed geometric tolerances are allocated to capture the design intent, then the set of feasible values of r and r' satisfying the functional requirements should satisfy the geometric tolerances and the set of values r and r' which violate the functional requirements should be rejected when checked for the geometric tolerances. In other words, geometric tolerances should be selected to minimize the hyper-volume R, shown in F i g 5 where:

The integral of equation 2 can be evaluated using Monte Cario simulation. where instances of the coordinates of feature points can be generated, using some probability distributions. These instances are used to simulate both the inspection and assembly processes and a counter is incremented whenever a mismatch is encountered. There are two Monte Carlo methods that can be usad to evaluate P , . These are the hit-and-miss method and the sample-mean method. The sample-mean method requires the prior knowledge of a joint probability distribution having its highest density over the region Rm. Since the region R, is not known, the hit-and-miss Monte Carlo method, which assumes a uniform distribution associated with each variable, is more appropriate. However, using the nit-and-miss method, a large error in the evaluated value of P , can develop especially with a large number of variables. This problem can be remedied by incorporating variance reduction techniques in the simulation such as Latin hypercube sampling and antithetic variates (7). The evaluation process is described in the following algorithm: Given: An assembly with the number of featureslsurfaces
NSUrf.

Each featurekurface is represented by a number of points Np, whose locations in space follow the probability distribution of the manufacturing process which generate the featurelsurface. The total number of random variables N is therefore equal to:
Nsud

N=

1",

(3)

J'1

iii. A number of functional requirements Nreq. iv. A Monte Carlo simulation sample size N,. 1. repeat Ns times, a. Latin hypercube sampling step Assume a uniform distribution over the range of possible values of the random variables. Divide the cumulative probability function of each random variable into m equal probability intervals. Generate a random number Ui for each interval and use the inverse uniform probability function to deliver an instance of the random variable creating the matrix r. where:
r:.1

Fig.5 Mismatch Region Assume that the range of values of each radius r;. is covered by a probability distribution q , then the integral:

r1.2 r13 ... r1.m

r =

P , =

...

... ...

u(rl, r2, ...,r,,) dr,drz...dr,,

(2)

( r l . r *....rn)E R,

For each random variable r, rearrange the m generated values in a random permutation

103

b. Antithetic variates step For every random variable 9. use the random number 1-Ui to deliver another set of m instances forming another matrix r. c. Since each column of the r and F matrices forms an instance of a set of manufactured parts, then for each column in the r matrix do: i. Calculate the geometric deviations on all features check them for the imposed tolerances. ii. Simulate the assembly sequence then verify the assembly functional requirements. iii.if the assembly requirements are in spec and the tolerances are not in spec or if the assembly requirements are not in spec and the tolerances are in spec then Increment Na = (Counter for belonging to Rm) else Increment Nf (Counter for not belonging to Rm). end if e. end-for f. Repeat the loop of step "d" for the f matrix obtaining E and 5. end-repeat 6. Calculate the probability that the assembly instances will belong to Rm.

~~

Control

1 Tolerance Tvpes

-~~

lMaqnitude
~~

I f,1

Orientation! i w.r.t. A for bl and bz] I- - ----- 2 1 i w.r.t. A for bz and 2 I I // w.r.t. D for bl

p.032
10.023 I

Form of bl

.~ Circularity - _ _ _ - - - - - - - - - -' - -1Cvlindricitv 1

0.059

Table 3 Probability of mismatch for different tolerances The above results indicate that higher mismatch probabilities were associated with the poor choices of tolerance types.

5. TOLERANCE ALLOCATION PROCEDURE


Tolerance allocation consists of two optimization procedures. The first is the selection of manufacturing processes and sequences which minimize the production cost and constrain the rejection rate. The second, allocates the types and magnitudes of the size, form and orientation tolerances in addition to the magnitudes of position tolerances (The datum reference frames for position tolerances are directly deduced from the manufacturing sequences). Since both procedures are of a combinatorial nature, genetic algorithms (3) are used as an optimization method for both procedures. 5.1 Selection of Processes and Sequences Given an assembly of Np parts and a set of Nf features. Each feature 4 has one discrete independent variable X , (The manufacturing process by which it is produced) and each part p; has another discrete independent variable Y; (The manufacturing sequence by which it is produced). Hence the total number of independent variables is equal to Np+Nf. Since the datum reference frame of the position tolerancing scheme of each part must follow the manufacturing sequence of the part, another formulation can be devised where each feature 5 can have a variable X , which stands for the manufacturing process and another variable Z , which is the datum reference frame used in its position tolerance. Since some reference frames cannot be used with others, each candidate reference frame can have a set of constraints on its usage as an attribute. The objective function to be minimized is the total production cost. The probability that the manufactured parts and features will violate the assembly functional requirements (Prej) is constrained to a specified value v. Therefore, the objective function is given by the following equation. min(Cost(X,Z)) , subject to Prei(X,Z) 5 v
x.z

m.

f , =0.5x

a_.=_) Na (Na + Nf Na + Nf

4.2 Example

1500.0

I ]
Fig.6 Simplified punching machine In order to show how the mismatch probability can be used to identify the effect of poor tolerance choices, consider the simplified diagram of a punching machine shown in Fig.6 which is composed of four main parts (base, stand, head and punch). The base has two holes bl and &,whose nominal diameters are 210 and 325 mm respectively. The functional requirement is to keep the clearance between the punch and the hole bl between 8.6 and 12 mm. Assume that all dimensions are produced at their nominal values except for the two base holes. and that only the form and orientation of hole bl and the orientation of hole b2 are to be controlled. For form control, cylindricity would constrain the hole's form than circularity assuming that both types have the same tolerance magnitude. The orientation of the punch in space is determined by the orientation of hole &, and hence the control of orientation of hole bl should relate to a datum established by hole (D). The probability of mismatch was evaluated for form control and then for orientation control. The results are shown in Table 3 .

(6)

5.2 Minimization of the Mismatch Probability

Given an assembly with a set of Nf features, each feature 6 in the assembly has two discrete-valued independent variables (Oi and Fi) for the types of tolerance controlling its orientation and form deviations respectively and four real-valued independent variables ( X i , Y,, Zi and Ui) for the magnitudes of the tolerances of size, position, orientation and form respectively. The datum reference frame for the position tolerance is already determined by the manufacturing sequence and controlling the size of a feature

104

needs only one type of tolerance (size tolerance). The objective function to be minimized is the mismatch probability P,.
0.F.X.Y.Z.U

($1

4.0 -

s1

(Features C1, C4 and C5)


3 2
I

min ( P J O , F, X, Y, 2 , U))

(7)

2.0
c.5
1.o

. S 3

s4

6. EXAMPLE

Dimensions in mm

300
3 0 e

c 2

,011

.012

.013 ,014 Process Limits (mm)

150 f.075

Cd

F
\ o
0

ca
0 1

2.0 1 .o
.012 ,013 .014 .015 Process Limits (mm)

Fig.9 Cost vs. Process Limits


150 100 Fig.7 Speed Reducer

The speed reducer shown in Fig.7 has a functional requirement, requiring the center-distance between the input and output shafts to bel50f 0.075 mm, otherwise, backlash between the gear teeth will increase, leading in turn to an unacceptable level of gear noise. Tolerances are to be allocated to the casing holes: C1, C2, C3, C4 and C5, (Fig.2 ) and to the bushings holes 81 and 8 2 (Fig.8 ).

&
Feature Candidate 1
C2

Candidate 2 CONSTRAINT C2 means that


B-D*Cl

A-B-F

C-E

frame can be used only if feature C2 is manufactured

+
82

D-A

4 D-A

D-C B1

Table 4 Candidate Datum Reference Frames

Fig.8 Bushing
C1 -LB

T y p e ? -

Figure 9 shows the cost functions used for the selection of manufacturing processes. All processes were assumed to generate manufactured surfaces between concentric off set cylinders, whose offset distance is equal to the process limit, and according to a normal distribution. Table 4 shows the candidate datum reference frames for the selection of manufacturing sequences. Table 5 shows the candidate tolerance types for controlling orientation. The procedures described in sections 5.1 and 5.2 are used to select the manufacturing processes and the geometric tolerances. The value v (equation 6) constraining the rejection rate is set to 0.0063.

IE

Range of Tolerance Magnitudes for Position and Orientation Control

Table 5 Candidate Orientation Tolerances

105

Reference Frame Probability of rejection due to the violation of functional requirements = 0.0057

Table 6 Optimum Processes and Sequences


Feature

I C1 1

C2

C3

C4

C5

I B1

j 82

P, = 0.0067
Table 7 Optimum Tolerance Types and Magnitudes Results of the optimization processes are shown in Tables

.6and 7, and the convergence curves of the cost vs. genetic


iterations are shown in figures 10 and 11. The rejection probability obtained by using the manufacturing processes and datum reference frames, shown in table 6, equals 0.0057, and the probability of mismatch achieved by the genetic algorithms is equal to 0.0067. The reason that the rejection probability is not exactly equal to the desired value of 0.0063 and that the probability of mismatch is not equal to zero is that the problem is of a discrete nature, where the search can only find solutions near to the desired ones.

10.00

9.00

r i
..............................

both types and magnitudes of geometric tolerances. The input data to the allocation process will be cumbersome, unreliable and difficult to obtain. A new criterion, mismatch probability,was developed to allocate geometric tolerances. The criterion was to illustrate the benefit of choosing some tolerance types over others. An algorithm was developed, based on genetic algorithms and using the proposed criterion. to allocate tolerance types in addition to tolerance magnitudes. The developed algorithms demonstrate a methodology which can be used to convert the geometric tolerance allocation problem from a human-based trial and error process into a mathematically formulated problem to search for the best solution. The proposed algorithms do not completely exclude the use of the expertise of tolerance designers. The choice of the set of possible tolerance types for each feature is still made by the designer. The algorithm took a relatively long time (six hours on an SGI workstation), because of the complexity of the objective function, and the use of genetic algorithms which use a large population of instances of the independent variables in its search. However, the tolerance allocation problem is solved only once at the detailed design stage. The benefit of selecting tolerance types and magnitudes outweighs the initial computational effort. The computational time can be reduced using a hybrid optimization model that incorporates genetic algorithms and direct search, where the search space is coursely discretized and genetic algorithms can be used to roughly estimate the location of the global minimum, and the extra search needed can be conducted using direct search.
8. REFERENCES

(1) ASME. 1995, Mathematical Definition of Dimensioning a n d a s , ANSI Y14.5.1M-1994.


(2) Farmer, L.E.. Gladman, C.A., 1986, Tolerance Technology Computer-Based Analysis, Annals of CIRP. 35:7-10.

.............................

. . . . . . ........................

(3) Goldberg, D, 1989. GeLetS Alaorithms in Search. Qpthization and Machine Learning, Addison Wesley. (4) Kanai, S., Onozuka, M. and Takahashi, H., 1995, Optimal Tolerance Synthesis by Genetic Algorithm under the Machining and Assembling Constraints, Proceedings of the 4th CIRP Seminars on Computer Aided Tolerancing, Tokyo, Japan. ( 5 ) Lee, W.4. and Woo, T.C., 1989, Optimum Selection of Discrete Tolerances, Trans. of ASME, Journal of Mechanisms, Transmissions and Automation in Design, 111:243-251. (6) Nassef, A.O. and EIMaraghy, H.A., 1993, Allocation of Tolerance Types and Values Using Genetic Algorithms, Proceedings of the 3rd CIRP Seminars on Computer Aided Tolerancing, CACHAN, France. (7) Rubinstein, R., 1981, Simulation and the Monte CarlQ Method, John Wiley and Sons, New York.

8.00

I 1

10 20 30 40 Fig.10 Cost vs Iterations

0.025 . . . . ..:... . . . . .:.


0.020 0.015

. . . . .:.. . . . .

......

Fig.11 Probability of mismatch vs Iterations

7. CONCLUSION
The cost vs. tolerance functions which are traditionally used to allocate dimensional tolerance cannot be used to allocate

(8) Zhang. C., Wang, H.P., Li, J.K., 1992, Simultaneous Optimization of Design and Manufacturing. Tolerances with Process (Machine) Selection, Annals of the CIRP, 41569-572.

106

Vous aimerez peut-être aussi