An Expert Sys tem for the Design and Se lect ion of Ball Bearing Parameters

M A Pathak & R S Ahluwalia, Industrial Engineering Department, West Virginia University, Morgantown, WV-26506-6101, USA

Abstract Artificial Intelligence (AI) is an emerging technology. Research in AI is focused on developing computational approaches to intelligent behaviour. The computer programs with which AI could be associated are primarily symbolic processes associated with complexity, ambiguity, indecisiveness, and uncertainty. One of these computer programs is referred to as Knowledge. based Expert System (KBES) as it represents knowledge acquired from various experts in a particular field of interest to the user. The expert system emulates human behaviour in solving problems thought to require experts for their solution by utilizing computer programs that incorporate experts' heuristic reasoning. In this paper, the application of KBES to aid the design of ball and roller bearings is discussed. The precision rolling-element bearing of the twentieth century is a product of exacting technology and sophisticated science. A bearing supports radial and axial loads, at the same time allowing relative motion between two elements of a machine. Various requirements and steps in the design of baU and roller bearings are discussed. Equations are developed for the relevant design parameters and input into the expert system shell called VP-Expert. The expert system rules are also provided.

I n t r o d u c t i o n Artificial Intelligence (AI) has been gaining popularity since the 1960s. Expert Systems (ES) within artificial intelligence are now beginning to be applied in industrial and manufacturing areas as a means of solving problems and as decision making aids. Expert systems are also called Knowledge Based Expert Systems (KBES). F e i g e n b a u m (1~ has de f ined KBES to be intelligent compute r programs which use knowledge and inference procedures to solve problems that are difficult enough to require significant human expertise for their solutions. This human expertise is obtained from an expert. An expert is a person who through training and exper ience can perform a task with a high degree of skill.

People acquire knowledge through exper ience and study of various subjects. In different fields of technology, this knowledge and expertise is obtained by few experts. When these people retire or leave their job, their expert ise goes with them. Expert systems provide a way to capture this expert knowledge so that it is not permanent ly lost.

Expert systems are needed where:-

• the expertise to solve a problem is either expensive or scarce

• to speed up the time an expert analysis takes • to upgrade the performance of less experienced and

less skilled personnel • to apply expert analysis to situations for which

specialists do not have time • to minimize or eliminate the clerical and other low-

level errors made.

An expert system can perform several possible expert tasks eg.,

• as an au tonomous system - the expert system analyses a situation, makes a decision, and then acts

on it directly or co m m an d s a human action - all without human participation in the decision process.

• As an au tonomous system with human override - the expert system acts as an autonomous system but presents its decisions to a human expert, who can override or modify them.

• as an expert consultant - the expert system provides consulting at an expert level for a task practitioner in the manner similar to that of a human expert consultant.

• as a colleague - the expert system gives suggestions to a human peer. Although it may not have a higher skill level than the human, it may be able to apply itself to analysis that a human does not have the time to do.

• as an intelligent assistant - the expert system acts as an intelligent assistant to the practitioner. It might make a set of plausible suggest ions or determine a list of items to consider when a certain condition o c c u r s .

The number of problem types in which expert systems have been and are being applied is continually growing and includes such fields as diagnosis, scheduling, planning, monitoring, process control, design, forecasting, signal interpretation, configuration and training. Although an expert system can perform better than the best experts in some cases, it can never contain the complete expertise and exper ience of a leading expert. And since expert systems are based on the knowledge of humans, it is unlikely that they can solve problems that no human experts could solve. An expert system has three main componen t s (Figure 1), namely the knowledge base, the inference engine, and the working memory.

Knowledge base The knowledge base stores the facts and heuristics of domain experts. This knowledge is acquired from

MATERIALS & DESIGN Vol. 11 No. 6 DECEMBER 1 9 9 0 0261-3069/90/060317-06 © 1991 Butterworth-Heinemann Ltd 317

J User data CllO) l ~ i n f e r e n c e l ~ J K n o w l e d g e l engine I T M ~I base I

Fig 1 Expert system architecture

documents and consultation with experts. Knowledge acquisition is accomplished by meetings between people developing the expert system and domain experts. The knowledge may be represented as semantic networks, frames, or rule based knowledge. Each section of the knowledge base may be independent in terms of conclusions and input data.

The knowledge acquired is represented by production rules, also called IF-THEN rules. Such rules have the basis form:-

IF load = 10000 lb THEN bearing = roller.

Therefore the production rule formalisation is a flexible mechanism, providing modulari ty and standardization of representation. It is used to represent different types of knowledge eg situation action/, premise/conclusion, sufficiency, definition. A set of production rules embody a large collection of expert knowledge.

The inference engine It provides the system control. It applies the expert domain knowledge to what is known about the present situation to determine new information about the domain. This process will lead to the solution of the problem. The inference engine also enables the expert system's interfaces to data sources and to the user. The method used to reproduce part of an expert's reasoning process is called chaining. There are two methods presently available, namely forward and backward chaining. Forward chaining involves reasoning from data to the goal while backward chaining finds data to end on a goal. In forward chaining, the applicable set of known facts are used to trigger applicable rules. These facts are stored in working memory. From the total set of facts collected, additional rules are triggered till a goal condition is reached. In backward chaining, the expert system starts out with a known goal searching for rules or subgoals that satisfy this or part of this goal. Each of the subgoals is checked against already existing data or data input by the user. This process continues until the goal is verified as true or false, based on data.

The working memory The working memory contains the information that the

system has received about the problem at hand. in addition, any information the expert system derives about the problem is also stored in the working memory. ES have been applied in fields where the need for diagnosis predominates; applications where heuristic knowledge and human expertise is costly and scarce. The area of diagnosis is not limited to medical fields but also to applications of the manner of equipment failure, program debugging etc. If there is an established theory for the subject, then that knowledge can be docu- mented, but if there is no such theory available and if there are only heuristic rules then ES is one of the ways to capture that knowledge and implement an effective computer based system.

Literature survey Expert systems Edward Feigenbaum~ 1~ developed a mass-spectrogram interpreter ES called DENDRAL. Its immediate successor was MYCIN, an ES for diagnosing bacterial infections of the blood, developed by Shortliffe~2L MYCIN proved to be extremely successful and thus opened up the doors for ES to be applied for various applications.

In engineering field, ES have been applied to

• selection of components • selection of analytical program's • trouble-shooting and diagnosis • process planning • welding (selection of welding electrode) • stress corrosion cracking etc.

Freed and Wright (3) developed FAX, an expert system for the analysis of mechanical failure. It was designed to train non-professionals in failure analysis and prevention. Chester, Lamb and Dhurjati c4~ devel. oped FALCOM, an on-line alarm analysis ES for power plants and process plants. The paper by Basden and Kelly ~5~ deals with the development of a prototype ES to predict the risk of stress corrosion cracking. The method of developing this into an operational tool for regular use by materials engineers is described.

Vaghul et a~ 6~ addressed the deyelopment of an on- line ES to operate in computer aided design environments for evaluating designs regarding their manufacturability. The work by Calkins ~7~ demonstrates an application of the high level computer graphics ES for an automated sculpture surface CAD procedure. The work includes the automatic recognition and development of 3D structures from 2D representations. In the paper by Brown ~8~ a description of the Air-Cyl ES for designing air cylinders from given requirements is given.

Ball & roller bearings History The precision rolling.element bearing of the twentieth century is a product of exacting technology and sophis- ticated science~gL The Greek, Roman and Chinese civilizations all played an important role in the overall development of ball bearing technology. The modern forms of roiling-element bearings finally emerged during the period 1850-1925. At that time, general industrial

318 MATERIALS & DESIGN Vol. 11 No. 6 DECEMBER 1990

T Outs ide d iameter

Bearings The purpose of a bearing is to support a load while permitting relative motion between two elements of a machine. The term rolling contact bearing refers to the wide variety of bearings that use spherical balls or some type of roller between the stationary and the moving elements. The most common type of bearing supports a rotating shaft, resisting purely radial loads or a combination of radial and axial loads. The following characteristics of ball bearings make them desirable over other types of bearings:

• low starting and good operating friction. • the ability to support combined radial and thrust

loads. • less sensitivity to interruptions in lubrication. • no self-excited instabilities. • good low-temperature starting.

With these advantages, there are also some disad- vantages, eg some bearings are designed to carry only axial loads.

• finite fatigue-life subject to wide fluctuations. • larger space required in the radial direction. • low damping capacity. • higher noise level. • more severe alignment requirements. • higher cost.

The essential parts of a ball bearing are shown in Figure 2. They are inner and outer ring, the balls, and the separator. The inner ring is mounted on a shaft and has a groove in which the balls ride. The outer ring is usually the stationary part of the bearing and also contains a groove to guide and support the balls. The separator prevents the contact between the balls and thus reduces friction, wear and noise from the regions where sliding conditions would occur. There are three types of ball bearings based on supporting only radial loads, only thrust loads, and a combination of both loads. Here only those bearings that support radial loads are considered. In these radial bearings, there are two main types, deep groove and filled notch bearings.

In deep groove bearings, the inner and outer races are approximately one fourth as deep as the bali diameter. A filled notch bearing has a notch in the races thereby permitting the insertion of balls. Although there are more varieties of the radial bearings, the design principles remain the same.

The design of a bearing requires consideration of:-

Width i TM ~ [

Outer r ing

Separa tor

Inner r ing

Corner" radius

applications included capstans, turntables, axles and railway axle.boxes. However, it was the humble bicycle that provided the real spur to ball bearing development. Eschmann ~1°~ has given a complete treatise on the design of ball and roller bearings. Mott (11) discussed the general concepts in the design and applications of roller bearings. Matsumori ~12~ discussed the application of CBN abrasives in advanced ball bearing manufacture. Ashburn ~13~ provided a rare look at the manufacture of roller bearings at a bearing plant. In this paper, only the application of an expert system to the design of ball and roller bearings is discussed. The expert system rules are also provided.

Fig 2 Ball bearing nonemclature

• characteristics of the bearing load. • relative motion between the bearing elements. • geometry of the bearing surfaces. • physical and chemical properties of the lubricant and

bearing metals.

The requirements that a bearing must specify are:-

• factor of safety - which depends on the bearing application.

• bearing life, reliability and ambient conditions. • bearing precision. • power consumed in the bearing. • bearing installation and maintenance cost.

Before a bearing design can be performed, it is important to know what inputs are available. For ball bearings, radial load on the bearing, rotational speed on the shaft, nominal shaft diameter at the journal must be known beforehand.

The following are the initial steps in the design of a ball-roller type bearing:-

• determine axial force on the bearing from working condition.

• take radial and axial factors from the handbook; these factors are given in Table I.

It should be noted that radial loads are forces at right angles to the axis of the shaft, such as the loads imposed by straight spur gears, drive chain or V-belts ~141. Thrust loads are parallel to the shaft axis eg turntable on vertical shafts, loads on crane hook.

• calculate bearing life (L) in million revolution of the bearing: the formula applied is as follows:-

L = (Li fe in hours x rpm x 6 0 ) / ( 1 0 6 )

A list of bearing-life recommendations for various classes of machinery ~15~ is given in Table I!.

MATERIALS & DESIGN Vol. 11 No. 6 DECEMBER 1990 319

Table ! Radial and ax/al factors

Bear ing Load Radial load Axial load

Radial ball 1 0.85 Angular contact ball 0.5 0.4 Thrust ball - 0.25 Cylindrical roller 0.5 - Tapered and spherical roller 0.2 0.15

(According to bearing type) Deep groove 1 1 Filling notch 1.2 low Double row 2.2 2.6 Self aligning 0.76 0.5

Table !i Bearing life recommendations

M a c h i n e r y c l a s s Life, kh

Instruments & apparatus for infrequent use < = 0.5 Aircraft engine 0.5 - 2 Machines for short operation 4 - 8 Machines for intermittent service 8 - 14 Machines for 8 hour service which are not fully utilised 14 - 20 Machines for 8 hour service with full utilization 20 - 30 Machines for 24 hour service 50 - 60 Machines for 24 hour service with high reliability 100 - 200

• take the service factor depending upon the nature of the load. The calculation of bearing life and the dynamic capacity is based upon the knowledge of the load to be imposed on the bearing during its life and on the knowledge of the fatigue-life character- istics of the bearing. In some applications, the load may not be constant but may vary in a known manner, in still other applications, the bearing may be subjected to loads which can not be calculated readily. Some allowance must be made for these indeterminate loads and this is usually denoted by means of a service factor. These service factors obviously depend on the type of machinery where these bearings are applied. For some basic types of machinery the service factors are given in Table III.

Table I!! Service factors

Type of app l ica t ion Service factor

Precision gearing 1 - 1.1 Commercial gearing 1.1 - 1.3 Application with poor bearing seals 1.2 Machinery with no impact 1 - 1.2 Machinery with light impact 1.2 - 1.5 Machinery with moderate impact 1.5 - 3

• Determine the dynamic loading capacity (C). C = [L/L~o] 1~K x (x.F, + Y.Fa)Ks

Where,

L L~o

m

F r m

F~ -

K s -

x, Y - K -

required life of bearing in million revolution. life of bearing for 90% survival at o ne million revolution. radial load on bearing. axial load on bearing. service factor. radial and axial factors. exponent - 10 /3 for roller bearing and 3 for ball bearing.

3 2 0

Oe+O

l e + 1

2e+1

R=0.99 I , I j ~ 0J 100 200 300 400 50

Bore or OD

Fig 3 Diameter tolerance on inner ring bore. Mini- mum lmm)

6e+0

5e+O

4e+O

3e+O

2e+O

l e + 0

0e+0

~ S x 2+4.375e-8x^3

R = 0.99

i I l I , I , I L I , I00 200 300 400 500 600

Bore or OD

Fig 4 Diameter tolerance on inner ring bore. (Maxi- mum 0.0001 in)

• Select bearing that satisfies these requirements using manufacturer's catalogue for the required dynamic capacity and bore size.

• Determine fits and tolerances for the shaft and housing. As there are tables available for the tolerances, statistical relations are developed for diametral and width tolerances from the relevant graphs shown in figures 3 and 4. These linear or polynomial equations are then input into the expert system so that tolerances can be computed directly.

• If bearings are operated at high temperatures, employ correction factor to dynamic load bearing capacity.

• Determine the type of lubrication as shown in TablelV.

Table IV Type of lubrications

U s e g r e a s e U s e oil

Temperature Below 200 - 250 ° F Above 200 - 250 ° F Speed factor Below 200k - 300k Above 200k - 300k Load Low to moderate High Housing design Simple Complex Long period, no attention yes no

MATERIALS & DESIGN Vol. 11 No. 6 DECEMBER 1990

These are considerations that must be satisfied in order to design and select a bearing. When these considerations are put into the expert system in the form of IF-THEN rules, the expert system can suggest bearing parameters that may satisfy most of the bearing requirements and then the bearing can be chosen from the manufacturer's catalogue. In real-world applications, this catalogue will be directly interfaced with the expert system so that expert system can select the bearing and present it to the user without any need for searching through these catalogues.

Manufacture of bearings Rings The operations needed in the sequential order are, turning, boring, facing, forming in annealed condition, and grinding the tracks. Tracks or races are carefully polished to reduce friction and thereby increasing the bearing life. Then fine grinding is applied.

Steel balls • smaller size balls are formed in automatic cold-

heading presses from steel rod. Larger sizes are hot- pressed into shape from billets.

• formed blanks are ground to remove fins and give them true spherical shape.

• balls are rough-ground between two grinding-discs that rotate at different speeds and whose axes of rotation do not coincide.

• tumbling in an abrasive compound to correct size. • due to all this cold-working, internal strains are

developed in the balls which are relieved by heat treatment and tempering.

• precision grinding between cast iron disc and grinding wheel, both provided with concentric grooves.

• rotary tumblers containing a fine abrasive mixed with oil imparts superior finish to these balls.

• barrels containing sawdust clean these balls. • barrels containing strips of soft-kid leather produce

polished and glossy balls.

Bearing materials A good bearing material must have satisfactory compressive and fatigue strength to resist externally applied loads. Also it should exhibit low modulus of elasticity. Following is a list of the few requirements that these materials must satisfy,

• greater hardness • must not weld to the shaft. • exhibit conformability for misalignment. • permit abrasive particles to embed and sink into the

bearing material. • higher corrosion and oxidation resistance. • low cost and high reliability.

The two most common materials used for bearings are SAE 52100 and M.series steels. SAE 52100 material is preferred due to its high heat resistance property and the ability to sustain original hardness at high temperatures. M-series steels eg case hardening steels, chrome steel, rustless steel etc are superior at high temperatures and exhibit excellent corrosion and oxidation resistance over 350°E Other materials used are Babbits (lead or tin bearing alloys), copper alloys,

gray cast iron, graphite etc.

Assembly Installation of the balls between rings. The free space created by the eccentric positioning of the rings is fitted with the balls, whose size and number are so calculated that by utilizing the elasticity of the rings, the inner ring, located between the first and the last ball, can be displaced to a position concentric with the outer ring. The balls are then distributed uniformly over the circumference.

Conclusion All relevant aspects of the ball bearing design, manufacture and assembly are considered, although design is considered in more detail than other factors. The design steps and the design guidelines are converted into expert system rules (Appendix A) which if run by the user will provide him with the relevant bearing parameters and design guidelines.

R e f e r e n c e s

1 Barr, A, and Feigenbaum, E A, The Handbook of Artificial Intelligence, William Kaufman, Menlo Park, Californa, Vol 1, 1981.

2 Shortliffe, E H, Computer-based Medical Consultation: MYCIN, American Elsevier, New York, 1976.

3 Freed, D and Wright, D, "FAXS: An Expert System for the Analysis of Mechanical Failures", Proc. ASME Intl. Computers in Engineering Conf., Las Vegas, NV, August 1984.

4 Chester, D L, Lamb, D E, and Dhurjati, P, "An Expert Systems Approach to On-line Alarm Analysis in Power and Process Plants", Computers in Engineering, pp 345-351, 1984.

5 Basden, A, "On the Application of Expert Systems", lnt J Man- Mach Stud, Vol 19, No 5, pp 461-477, 1983.

6 Vaghul, M, Zinsmeister, G E, Dixon, J R and Simmons, M K, "Expert Systems in a CAD environment: Injection Moulding Part Design as an Example", Proc ASME lntl Computers in Engineering Conf, Boston, August 1985.

7 Calkins, D E, "An Automated Sculpted Surface CAD Procedure Based on High Level Computer Graphics and Expert Systems", Proc ASME lntl Computers in Engineering Conf, Boston, August 1985.

8 Brown, D C, "Failure Handling in a Design Expert System", CAD Journal, June 1985.

9 Hamrock and Dowson, Ball Bearing Lubrication, John Wiley Publications, 1981.

10 Eschmann, Hasbargen, and Weigend, Ball and Roller Bearings - Theory, Design and Application, Wiley Publishers, 1985.

11 R L Mott, Machine Elements in Mechanical Design, Merrill Publishers, 1985.

12 Matsumori, Noburu, "CBN Abrasives Advance Ball Bearing Manufacture", Manufacturing Engineering, V 101, pp 70-2, Nov 88.

13 Ashburn, Anderson, "A Rare Look at a Bearing Plant", American Machinist, V 128, pp 102-103, July 1984.

14 Wilcock and Booser, Bearing Design and Application, McGraw Hill, 1957.

15 Shigley, J E, Mechanical Engineering Design, McGraw Hill, 1977.

Appendix A Note: These rules are to serve as an example of how a bearing design expert system might look. From the AUTOQUERY command, VP-Expert will auto- matically ask the user for the values of the variables that it will need during the decision process.

VP-Expert Rules AUTOQUERY ACTIONS Find Bearing__type

MATERIALS & DESIGN Vol. 11 No. 6 DECEMBER 1990 321

Find Radial._.load__cap Find d ia__to l_out__max

Find load__char;

rule 1 IF bearing = deep groove then radial~load__cap = 1.0

thrust__load__cap = 1.0;

rule 7 if ax ia l~ load = (load) then d y n _ l o a d ~ c a p = ( (1 /110)*(x* f r+Y*fa)*ks) ;

rule 8 if ax ia l~ load = (load) then bearing__life = ((req__life) * (n) * 60) /10000000;

rule 18 if appl = c o m m e r c i a l ~ g e a r i n g or appl = poor bearing__seals then service factor = 1.2;

rule 24 if speed__factor > 2 0 0 0 0 0 and speed~factor < 3 0 0 0 0 0 then lubrication = grease;

rule 30 if bore ._dia = (outer._dia) then diametral__tol -- ( - 4 . 5 2 7 5 - 0 .0321 * (outer dia)) display "This is the min imum tolerance for inner ring bore d iameter in tenths of an inch with correlation coefficient = 0 .99 for the given equation: ' ;

rule 36 if bore__dia = (outer ._dia) then width__tol = ( - 5 5 . 0 1 7 6 + 0 . 2 6 9 2 ) * o u t e r _ _ d i a -

0 . 0018" (outer dia)**2) display "This is the min imum width tolerance for single width bearings in tenths of an inch, with correlation coefficient = 0.98";

rule 38 if Ioad_._char = unidirectional or l o a d _ c h a r = cyclic or I o a d ~ c h a r = starting or I o a d ~ c h a r = unbalance or l o a d ~ c h a r = shock then bearing type = rol l ing_contact ;

rule 39 if b e a r i n g ~ t y p e = ro l l ing_contac t then s p e e d ~ l i m i t = cen t r i fuga l~ load

speec l~l imi t = dynamic_e f f ec t ;

322 MATERIALS & DESIGN Vol. 11 No. 6 DECEMBER 1990