SPC Made Easier, Simpler, More Statistically Powerful

Keki R. Bhote

n the search for world class quali­ when they realized that SPC in pro­ come fashionable. They are good Ity, there must be synergy between duction was too little and too late. for employee morale, and they are management, customers, suppliers, Paradoxically, the U.S. rediscovered good for small-increment problem and employees. Organization, sys­ SPC at the same time that the Japa­ solving. However, their hunches tems, and training of the right kind nese were discarding it. (SPC was often lead to blind alleys, consuming are essential catalysts. A blueprint widely used in the U.S. during World needless cost and time. of improvement in these general War II, but then the baby was distributions and areas is beyond the scope of this thrown out with the bathwater, in a studies indicate the existence of var­ article. Instead, it concentrates on revolt against the mysteries of the iation, but like control , pro­ easy, simple, cost effective, and sta­ world of .) vide few clues as to the origin of the tistically powerful tools that can im­ problem. Engineering judgment and "If Japan Can, Why Can't We" prove product , not by an varying one cause at a time can What re-ushered the SPC age inconsequential 10 percent, 50 sometimes help, but the longevity of in America was the airing of the percent, or even 100 percent, but by chronic problems bears witness to NBC White Paper-"If Japan Can, factors of 10:1 and 100:1 in a short their futility. Lastly, there is some­ Why Can't We." It gave the Ameri­ period of time. If quality is the cen­ times a child-like faith that comput­ can public at-large its first glimpse of terpiece for a company's resur­ ers can solve quality problems, but the reasons behind Japan's suc­ gence, these statistical tools are the used with ineffective methods they cess: Quality, in general, and SPC piece de resistance, within quality. only process meaningless fast­ in particular. Deming was rescued er! The Traditional, Ineffective Quality from the U.S. quality wilderness and SPC Tools for the Japsnese Line Tools elevated to a prophet within his own Workers It is common knowledge that country. Major companies scurried The Japanese, as indicated ear­ Japan drew even with the U.S. in to jump on the SPC bandwagon. lier, have abandoned SPC tools for quality progress by the mid 1960s Unfortunately, SPC has become more powerful techniques, with two and has gone on to command an synonymous with control charts. exceptions. First, they still trot out ever-widening lead. What is less Control charts are complex, costly, control charts as show-and-tell for well-known are the quality tools and almost useless in their ability to visiting American firemen, so that used by the two countries. The U.S. their advanced methods would not persevered in the traditional tools of ... Control charts are receive much scrutiny! Secondly, inspection and sorting, detection complex, costly, and almost they have trained their entire direct and correction, fire fighting, exhorta­ labor work force in elementary SPC tion, and the crutch of useless in their ability to solve tools, so that they can tackle small­ plans, at least until 1980. These quality problems. step quality problems through quali­ tools consume costs but add little ty circles, (improvement) value. solve quality problems. The most groups, and employee suggestions.' Conventional SPC Tools- Too charitable rationale for their exis­ These elementary tools consist Little, Too Late tence is as a maintenance tool, to of: The Japanese, on the other be used only after a process is • P.D.C.A. (The Plan- Do- Check­ hand, abandoned these ineffective brought under control with more Action Circle) allegedly taught by tools and adopted statistical process powerful diagnostic techniques. Deming, but recently claimed as a control (SPC) in the 1950s, under Problem Solving-U.S. Style. Japanese innovation the coaching of Dr. W. Edwards Along with control charts, the • and analysis Deming. They rode the crest of the U.S. began to use other less-than­ SPC wave until the mid 1970s, effective problem-solving tools. Brainstorming, quality circles, and Kepner-Tragoe methods have be-

12 Target

..~ ~...... •...... •...•.•.•...... •~~.~~~~~-"'""""""'."""'"'"~".""'."'."".""'.","""""""" Design of

Characteristic Classical Taguchi Shainin

Technique Fractional factorials, Orthogonal arrays Multi-vari, vari~ble search, full EVOP, etc. factorials

Effectiveness Moderate Low to moderate Extremely powerful (20 percent to 200 percent (20 percent to 100 percent (100 percent to 500 percent improvement) improvement) improvement) Retrogression possible Retrogression likely No retrogression

Cost High High Low Many experiments Many Few experiments experiments

Complexity Moderate High Low Full ANOVA required Inner and outer array multipli­ Experiments can be under­ cation, SIN, Anova stood by line operators

Statistical Validity Low Poor High Higher order No Every variable tested with all effects confounded with main Even second order interaction levels of every other variable effects effects confounded with main Excellent separation and To a lesser extent, even effects qualification of main interac­ second order interaction SIN concept good tion effects effects confounded

Applicability Requires hardware Primary use as a substitute for Requires hardware Monte Carlo analysis Main use in production Can be used as early as prototype and engineering run stage

Ease of Moderate Difficult Easy Implementation Engineering and statistical Engineers not likely to use Even line workers can knowledge required technique conduct experiments

Fig. 1. Three approaches to the are compared.

• Check-sheets, , and These and allied techniques are jective is to disoover key variables in frequency distributions to measure often grouped together and called product and process design, drasti­ variation the "seven tools of QC" that every cally reduce the variations they • Pareto charts to separate the vital Japanese worker leams and uses. cause, and opan up the tolerancas few causes of defects from the The aggregate benefit is very pow­ on the lesser variables to reduce trivial many erful, but unfortunately it is these oosts. These experiments are oon­ ducted before the start of produc­ • Cause-and-effect diagrams to list elementary SPC tools that U.S. pro­ tion. Hundreds of Japanese oompa­ all possible causes of a problem fessionals are embracing in their un­ nies oonduct thousands of these and group them by families equal drive to catch up with Japan. Meanwhile, Japanese professionals designed experiments each year to • CEDAC (Cause-and-Effect Dia­ - in engineering, manufacturing, make designs "more robust." The grams with the Addition of Cards) and qual~y-have graduated to discipline is a variant of the classical - a refinement of cause and ef­ more sophisticated and powerful design of experiments fathered by fect diagrams to keep them cur­ tools. Sir of Britain 70 years rent and more meaningful to the ago. The principal Japanese expo­ Japan's Secret Weapon worker nent is Dr. Genichi Taguchi, who The central thrust and secret • Control charts. adapted the classical methods into a weapon of Japanese quality is de­ sign of expariments (DOE). The ob- t>

Fall 1987 13 system called the "Orthogonal portantly, the only way to reduce CPK = (1 - K) Cp Array." and eventually eliminate all scrap, (Where K is an absolute value, with­ In the U.S., by contrast, DOE rework, analyzing, etc., is to get a out regard to its sign) was practically unheard of, except product parameter at, or very close (Design Center, D, need not be at for a small band of academics and to, its design center. This is the only the mid-point of the specification missionary . A belated way to get to zero defects and 100 width.) movement, started in the early percent yields. Once this is done, When the process average, 1980s, imported Taguchi's methods production becomes a "breeze" x, and the design center, D, coincide, wholesale. without the necessity of inspections p that add no value! K is reduced to zero, so that C and Classical vs. Taguchl vs. Shainin CPK are equal. Fig. 3 shows two ex­ DOE Measures of Variation: Cp snd CPK amples of non-centered processes. Fig. 1 lists three approaches to Before variation can be re­ Process A has a narrow spread, the design of experiments - the duced, it must be measured. Two with a respectable Cp of 2.5. But classical, Taguchi, and those taught yardsticks-Cp and CPK-have be­ because it is skewed toward one by Dorian Shainin, less famous than come standard terminologies in re­ end of the specification Iim~s, ~s CPK Deming or Juran, but a giant of an cent years. Process capability, Cp, is is reduced to 1.0. By contrl;lSl, Proc­ authority and a consummate prob­ defined as specification width divid­ ess B has a wider spread and a lem solver. The classical tools start ed by process width. It is a measure poorer Cp of 1.67. Yet, it is more with fraction factorials and end with of spread. Fig. 2 depicts six frequen­ toward the design center and, there­ evolutionary optimization (EVOP). cy distributions comparing the speci­ fore, has a higher CPK of 1.33. The Taguchi's methods use orthogonal fication width (40 - 20 = 20) to the objective, again, is to aim for high arrays (inner and outer), analysis of process width. Process A in Fig. 2 Cp.'s of 5, 10, 15, and 20 without , and signal-to-noise. The has a Cp of 0.8. This is a process adding recurring costs. Shainin experiments start with multi­ out of control, with reject tails at Measuring Variation - The Easy vari charts, followed by variables both ends. This used to be the norm Way search and/or full factorials and end for U.S. processes before the 1980 Traditional process capability with B vs. C validation. Fig. 1 com­ SPC age. Process B, with a Cp of studies require 50 units to be mea­ pares each of these DOE approach­ 1.0 is barely in control. Any slight sured on the process, from which es in terms of effectiveness, cost, change will cause rejects. At least the average, x, and the standard de­ complexity, statistical validity, appli­ 50 percent of U.S. processes have viation(s) are calculated. The proc­ cability, and ease of implementation. not advanced beyond a Cp of 1.0. ess limits are then conventionally While all three approaches are far Process C has a Cp of 1.33 and has defined as ± 3s. (Sometimes, superior to conventional SPC tech­ a margin of safety between the x where production is limited or meas­ niques, the Shainin tools run circles tighter process limits and the specifi­ around the other two in almost urements are difficult, 30 units can cation limits. The Japanese used suffice as a statistical minimum.) every characteristic. They can help this as a standard in the early From these measurements, Cp and America leapfrog the Japanese in p 1980s. Process D, with a C of 1.66 C can be calculated. their own game of concentration on is better, while the more progressive PK However, there is a statistically design quality over production quali­ companies in the U.S. have estab­ sound short-cut to process capabili­ ty. The tragedy is that most U.S. lished Process E, with a Cp of 2.0 ty studies. The rule is to measure companies are unaware of their ex­ (that is, the process width is only five units in a row taken from the istence, much less users of them! half the specification width) as a process. If all five units fall within near-term goal. Process F, with a Cp Variation Is Evil the middle half of the specification of 8.0, is not only much better, it is In Just-In-Time (JIT) practices, limits, it can be proved, using the attainable. In fact, there is no limit to inventory is evil. In quality, variation binomial theorem, that a C of 2.00 a higher and higher Cp-even 10, PK is evil. There are two reasons. First, and over, exists. If even one of the 15, 20-as long as no recurring getting a product within broad speci­ five units falls outside the middle costs are added to the product or fication limits does not result in 100 half of the specification limits, the process and only the cost of design percent customer satisfaction. Cus­ process has excess variation, which of experiments-an investment, tomers want uniformity, consistency. must be reduced using engineering Any departure from a desired design rather than a cost - is incurred. judgment, or preferably, the design center or target value results in cus­ A better measure of variability of experiments. tomer dissatisfaction. The cost of than Cp is CPK. It reduces Cp values Control Charts-Putting the Csrt such dissatisfaction goes up expo­ if the process is not centered. The Before the Horse nentially as a product parameter penalty is assessed by a K (for cor­ The systematic reduction of var­ moves away from design center to rection) factor. The formulas are: iance should begin at the prototype, one end or the other of the specifi­ C- Specification width (S) or engineering run, stage of product cation limit. Secondly, and more im- p - Process width (P) K = [Design Center (D) ­ Process Average (x)] S/2

14 Target Capability Index Examples

C SPEC. WIDTH (S) P PROCESS WIDTH (P)

20 40 20 40 A B

20 Cp~25 ~0.8 Cp~-~1.020 20 C D 20 40 20 40

20 Cp~- ~1.33 Cp~-~1.6620 15 12

E F 20 40 20 40

Cp~-20 ~2.00 Cp~-20 ~8.00 10 2.5 FIg. 2. Six frequency distributions comparing the specification width to the process width.

and process design. This is classical below 2.0, the techniques of varia­ ophy: "Talk to the parts, they are prevention of defects over correction tion reduction apply equally well to smarter than the engineers'" What later on at much greater expense. current production. he is that parts contain much The knee-jerk U.S. reaction to on the causes of varia­ variation reduction and problem tion that can be unlocked with ap­ The systematic reduction of solving is to launch a program of propriate statistically-deslgned ex­ variance should begin at the control charts. At best, control periments. The analogy of a prototype, or engineering run, charts will indicate that variation ex­ detective story is appropriate in this stage of product and process ists. But they cannot accurately diagnostic journey. Clues can be design. This is classical point to the cause of variation. Con­ gathered, each progressively more prevention of defects over trol trend analyses provide pos~ive, until the culpr~ cause or correction later on at much clues comparatively as faint as the variable-the Red X in the Shainin greater expense. light from stars beyond the Milky lexicon-is captured, reduced, and Way! controlled. The second most impor­ Variation Reduction- A Proven tant cause is called the Pink X, the Nevertheless, since over 95 percent Roadmap third the Pale Pink X. of important U.S. product and proc­ Fig. 4 is a time-tested roadmap ess parameters have Cp,'s well to variation reduction. It is largely an accumulation of the DOE tools, taught by Dorian Shainin. His philos-

Fall 1987 15

"""""".e".'."'.""''">'''''''''"'''".' Variation Reduction Process Capability (CPK' Design of Experiments/Statistical Process Control: Process A A Roadmap

Direct Customer Inputs

Prioritization of Market Research Customer Requirements .-t L Multi-Attribute Competitive Analysis

(1) (3) (2) 10 12 14 15 20 LSL X D USL Monte Carlo Simulation I-- Multi-Vari Charts I-- Components Search Cp=2.5; C =1.0 PK Despite narrow distribution, poor CR because - K X far from design center. t

Process B Paired Comparisons Variables Search

Full Factorials

10 11 14 15 17 20 B vs. C LSL XD USL Cp=1.67; Cp .1.33 K PATH 1 Wider distribution than Process A, but closer If mathematical relationships! to design center, so acceptable Cp . formulas between input variables K are known Positrol Fig. 3. Examples of non-centered processes. PATH 2 Prototype stage: If only two to six units are available

PATH 3 Process Certification At the eng. run, pilot run or full production, with 30 or more units available

Control Charts! Pre-Control

FIll. 4. DOE tools taught by Dorian Shainin are reflected in this illustration.

.-

16 Target Speclflcatlons-Arbitrary, Multi-vari Chart Overbellrlng, Wrong The diagnostic journey begins with a challenge to engineering Bam 9am lOam 11 am 12am specifications that non-engineers as­ 0.2510 f-----.---__+-----+----+-+-~-~--+----- sume are God-given. There are sev­ I- \ 1\ eral reasons for poor specifications: ~ ~ • Customers are not consulted v ...... -the difference between market­ ing and selling 0.2500 1-\-l-----J.----\H---+l~~F+_~~~+_-J-V__\l_---l----:\~l__+\...... +~ • The importance of a product fea­ Xl ~ ture to a customer versus the cost "" of that feature is not value re­ .~~ searched I: • The engineer's ego to create a i5 0.2490 [--;:;:::=:::;::::::±:;----+-t--4----1~----+_---- "state of the art" design • The engineer's reliance on prevI­ ~Max. ous designs, boilerplate require­ ttr-!Min. ments, or suppliers' published Right specifications 0.24S0 p===t~~~___l_~~~--+~~~__+~~- • Reliability is seldom a specifica­ tion • In translating product specifica­ tions into component specifica­ Fig. 5. Cyclical variations in a multi-vari chart, tions, a reliance on formulas or guesses, rather than design of ex­ which confound main and interaction The purpose of a multi-vari periments. effects. chart is to reduce the very large It is this author's experience that a Multi-Veri Cherts-A Powerful number of variables (up to 100) to a good percentage of both quality Firat Clue to the Red X few prime suspects that can then be problems and design cycle time can There are two prerequisites to a investigated using variables search, be reduced if quality techniques are multi-vari chart, the starting point of if five or more variables remain, or used to determine specifications. the Shainin tools, (1) A , full factorials if four or fewer remain. Some of these include: Value Re­ to assure that the most frequent de­ A multi-vari chart is a stratified ex­ search, Quality Function Deploy­ fect is being investigated, with the periment, where the objective is to ment (a Japanese tool to translate highest pay-off and, (2) A cause­ determine whether the major varia­ the "" into and-effect diagram to identify all the tion pattern is positional, cyclical, or hard specifications), Multi-Attribute suspect causes of the problem. temporal. If the greatest variation is Analysis and, of course, DOE tools. However, a cause-and-effect dia­ temporal, there is less need to look at the other two types of variation. Computer Techniques gram will only indicate the numerous Examples in each pattern of varia­ Route 1 in Fig. 4 uses Monte variables or causes, without any tion are: Carlo simulation and other computer methods to determine the effect of ... A cause-and-effect Positional independent variables upon a given • Variation within a single unit (such diagram will only indicate the as porosity in a metal casting) or output. However, this is only useful numerous variables or causes, if the formula governing the respec­ across a single unit, with many tive variables is known and can be without any due of where to parts (such as a P.C, board with programmed. More often than not, start or what interactions exist many components) these mathematical relationships are between those variables. It • Variations by location in a batch not known, especially when second may also miss the boat loading process (such as cavny­ order and higher order relationships altogether. The real cause may to-cavny variations in a mold between the independent variables not even be on the list of press) can cause totally unexpected re­ causes drawn on the diagram. • Machine-ta-machine; operator-to­ sults. This is also one of the main operator; or plant-to-plant varia­ statistical weaknesses of the tion. , since the orthogo­ clue of where to start or what inter­ Cyclical nal array uses fractional factorials, actions may exist between those variables. It may also miss the boat • Variation among consecutive units altogether. The real cause may not drawn from a process even be on the list of causes drawn on the diagram.

Fall 1987 17 • Variation among groups of units noon, 110 percent of the allowed Red X, Pink X, and Pale Pink X variation and 88 percent of the actu­ respectively). • Batch-to-batch variations al variation had been accounted for. • Lot-to-Iot variations. Other DOE Tools In analyzing these three stratifi­ The multi-vari chart reduces a Temporal cations of variation, the time-to-time large number (even up to 100) of • Hour-to-hour; shift-to-shift; day-to­ variation is the largest, with the suspect variables to a narrow family day; week-to-week or month-to­ greatest change occurring between of variables - 20 or less. Other month variations. 10 am and 11 am. This provided a DOE tools such as Components strong possible clue-a coffee Multl-Varl Case Study: Rotor Search and Paired Comparisons break, with a cooling down of the Shaft can also be used to reduce hun­ A manufacturer, producing cy­ lathe. The next reading at 11 am dreds of suspect causes down to a almost repeated the am reading at lindrical rotor shafts, with a diameter 8 few. Variables Search is the Rolls requirement of 0.250 in. ±0.001 in., the start of the day. This led to Royce of DOE techniques that can was experiencing excessive scrap. temperature as a potential Red X. slug it out with the Taguchi method­ The foreman discovered, to his em­ A process capability study indicated ology and win every time! It is used barrassment, that the amount of a spread of 0.0025 in., against the when there are five or more suspect coolant in the lathe tank was low. requirement of 0.002 in., that is, a variables, following a multi-vari C of 0.8. The foreman was ready When the coolant was added to the "homing in" investigation. The full PK prescribed level, the time-to-time to junk the old turret lathe and buy a factorial is used when there are four new one, for $70,000, that could variation was reduced to an incon­ or fewer suspect variables left to in­ hold a tolerance of ±0.0008 in., that sequential figure. vestigate. Both tools will determine The unit-to-unit variation, at 5 is, a CPK of 1.25. The plant manager the contribution of each factor, or directed that a multi-vari study be percent of the total allowed, was not cause, to the total variation, as well conducted before the purchase of worth investigating. However, the as clearly separate the main effects the new lathe, even though the pay­ within-unit positional variation from the interaction effects - a key back period for the new lathe would showed a total variation of 45 per­ requirement that fraction factorials or be only nine months. cent (15 percent + 30 percent) of Taguchi's orthogonal arrays, with Fig. 5 shows the resu It of a the allowed tolerance. The variations their "saturated designs," cannot multi-vari chart. The positional (with­ in taper indicated a non-random pat­ achieve. tern, with the left side always higher in each shaft) variations describe SPC Tools: The Tall thaI's been taper changes, from the left side of than the right side. This led to the conclusion that the cutting tool was Wagging the DOE Dog the shaft to the right, and out-of­ It is only when the DOE diag­ round conditions (maximum diame­ not parallel to the axis of the rotor shaft. A slight adjustment in the set­ nostic tools have pinpointed the ter and minimum diameter) on each major variables and drastically re­ side of the shaft. The cyclical varia­ ting reduced taper to almost zero. Finally, the out-of-round condi­ duced their variability through re­ tions, from one shaft to the next, are design, manufacturing process con­ shown by the thin connecting lines tion was traced to a worn bearing guiding the chuck axis. New bear­ trol, or supplier process control that between each shaft. The temporal SPC can begin its mission - to as­ variations, from one time period to ings were installed for a total cost of $200, including labor. sume that variation, once reduced the next are also shown. Analysis of through DOE, is maintained at that each variation indicated: In summary, the total variation was reduced from 0.0025 in. down reduced level. Time-to-time variation: 60 percent of to 0.0004 in. or a CPK of 0.0021 Two Pre-SPC Tools: Posltrol and the allowed tolerances 0.0004 = 5.0! The benefits: zero Process certification Unit-to-unit variation: 5 percent of scrap and a cost avoidance, in re­ Even within this limited scope of the allowed tolerance taining the old machine, of almost SPC, control charts-or the pre­ Within unit variation: $70,000. ferred technique of precontrol-are (a) taper: 15 percent of the The multi-vari chart is an exam­ not the starting point. The process allowed tolerance ple of the power of the Shainin producing the product must be di­ (b) out-of-round: 30 percent of tools. They are simple (no complex rectly controlled. One of the weak­ the allowed tolerance. mathematical or statistical formulas). nesses of American industry is that Total variation: 110 percent of the They are quick (in this example, a we attempt to control a process by allowed tolerance. snapshot of the process in just four examining the product it produces. hours). They are cost effective (in That is too late. Key process param­ One of the rules of a multi-vari chart this case, a cost of $200 for a sav­ eters must first be separated from is to continue the plot until at least ings of $70,000 along with zero less important process parameters 80 percent of the out-of-control vari­ scrap). Finally, they are powerful in using the same DOE tools. Next, ation is captured. In this case, by 12 getting at the root-cause of the these key process parameters must problem (here, the coolant, bear­ ings, and cutting-tool setting are the

18 Target The Simple Mechanics of Pre.control

Simple Pre-control Rules P-C Line p-c Line 1. Draw two pre-control (P-C) lines in middle half of spec. width.

2. :0 determine process capability, five units In a row must be within P-C lines (green zone). If not, use diagnostic tools to reduce variation.

3. In ~roduction, periodically sample two units consecutively.

12 Condition Action 14 1. Two units in Green Zone Continue 86% 2. One unit in Green and one Continue 1 unit in Yellow 14 3. Two units in Yellow Slop 7% 4. One unit in Red Slop

Red l<-"7'qYe~lI!Qlowti..l._ Green Zone --of""Y~.I~'O~~..I Red 4. Frequency of sampling: _ __z_o--,n.;crt'=--~z~on~.~=:::::====±~zosn~e_~~z~o~n.=---__ Divide the time interval between two 30" 1.50" 1.S

Fig. 6. Application of pre-control with four rules.

be monitored through a discipline To guard against retrogression the this arti~le. On the other hand, pre­ called Positrol. It determines, for work stations should be periodically control IS only now coming into the each key process parameter, who, re-certified. consciousness of quality practition­ where, how, and when it is mea­ ers and needs an explanation of at sured. A Positrol log is then kept of The l'(rannlcal Use of Control Charts least its mechanics. Fig. 6 shows such measurements, which can be the application of pre-control with readily monitored for conformance It has now been shown that contr~l four simple rules to: (1) establish by an auditing activity. charts are the last step in pre-control lines; (2) determine proc­ Another important pre-SPC tool vanatlon reduction and control not the first. But even here, controi ess capability at the start of produc­ is Process Certification. Its founda­ tion; (3) monitor production on an tion is Murphy's Law: "If anything charts can be substituted with an easier, simpler, and more cost· on-going basis and (4) determine cango wrong, it will!" At any work the frequency of sampling. station, a number of peripheral is­ effective technique called pre­ control. The simplicity of pre-eontrol is sues can make or break quality, be· now obvious. Its mechanics can be Sides the major areas of design, When developed by Dr. Walter Shewhart 60 years ago, control taught to managers and blue-collar process, and materials. These in­ workers alike in five minutes. Even clude: operator goals, instruction charts were useful. So was the s~ch Model T. But both have outlived the least sophisticated line or ma­ training; environmental factors chine operator can use it and more as temperature, humidity, dust, gas their glamour and value today. Un­ fortunately, a number of OEM cus­ important, make quick adjustment to control, water cleanliness electro­ assure defect-free' production on static discharge; equipme'nt and tomers increasingly demand the use of control charts as a passport to thousands upon thousands of units. gage calibration; work layout, etc. ~hile there is no need for charting These issues have to be listed at their business. They force control charts down the throats of unknow­ In pre-control, plots of the two unit each work station, generally by the ing and unwilling suppliers, and they samples versus time can be main­ process engineer. An interdisciplin­ bludgeon into submission those tained as a record. From such plots, ary team then examines each item knowledgeable suppliers who dare frequency distributions or Cp 's can on the checklist to assure that the be derived easily. Pre-contr01 can necessary disciplines are in place to to point out that the emperor wears no clothes. also be used for tightened specifica­ prevent even the random occur­ tions (relative to broader customer Pre-Control-The Elegance of rence of poor quality. At that point, ~pecification for one-sided specifica­ Simplicity the work station is process-certified. tions); or for attributes (by weighing Since control charts are widely known, they will not be explained in I>

Fall 1987 19 .,,------The Advantages of Pre-Control Over Control Charts

Characteristic Control Charts Pre-Control

1. Simplicity Complex-calculations of control Simple~pre-controlare middle half of spec. width

2. Use by operators Difficult- charting mandatory, interpretation Easy-green and yellow zones, a practical unclear approach for all workers

3. Mathematical Involved~X, R, and process Elementary-must only know how to divide by limits must be calculated four

4. Small production Useless for production runs below 500 units­ Can be used for production runs above 20 runs sampling of 80-150 units before even trial limits units; pre-control lines pre-determined by can be established specs (which can be narrowed)

5. Re-calibration of Frequent-no such thing in industry as a None needed, unless specs "goal posts" are control limits constant cause system moved inward

6. Machine adjust­ Time consuming-any adjustment requires Instant-based on two units ments another trial run of 80-150 units

7. Frequency of Vague, arbitrary Simple rule: Six samplings between twc 'sampling stoppages/adjustments

8. Discriminating Weak- a risk of rejection by chart, when there Excellent-a risk of rejection by pre-control is power are no rejects, is high. ~ risk of acceptance by low, <2 percent under worst conditions; 0 wit~ chart (in control), when there are rejects, is CPK of 1.66. ~ risk <1.36 percent under worst high conttitions; 0 percent with C PKof 1.66 Little relationship to specs

9. Attribute charts P, C charts do not distinguish between defect Attribute charts can be converted to pre­ types or importance control charts by weighting defect modes and an arbitrany rating scale

10. Economy Expensive-calculations, papenwork, larger Inexpensive-calculations simple, minimal samples, more freauent samoling long trial papenwork, small sampies, infrequent sampling runs if quality is good, process capability deter­ mined by just five units

Fig. 7. Control charts and pre-control compared.

attribute defects in terms of impor­ control a far more effective tool than ence transactions and in trade journals. It tance) and by converting them to a control charts. The prescription for can be shown that the a (alpha) variables scale, say from 1 to 10. restoring American industry to world - producer's - risk cannot exceed 2 percent and the IJ (beta)-consumer's risk generally Fig. 7 compares the many class is straight forward. Without it, will not exceed 1.36 percent. However, if the weaknesses of control charts and we cannot succeed. With it, we will process width is only half of the specification width and the process is centered-a C of the strengths of pre-control. not fail! PK 2.0-the a and IJ risks are reduced to zero! The March to World Class Even with a C PK of 1.33, these risks are less Competitiveness 'The number of suggestions turned in by Jap­ than 10 parts per million. anese workers is legendary. Whereas the In conclusion, a company that average number of suggestions per employ­ seeks world class status must place ee per year in the U.S. is 0.1, the figure of Author: quality at center-stage as a superor­ Japan is 10. More important, over 80 per­ cent of these suggestions are approved by Keki A. Bhote is senior corporate con­ sultant, quality and productivity im­ dinate, sacred value. Within quality, Japanese management. The quality circles provement, , Inc., Schaumburg, the most important task is to reduce with their own ideas, try pilot variation drastically. This is best runs, and submit the suggestions to man­ IL. He is the author of Supply Manage­ ment, a management briefing published done with the design of experi­ agement when they are sure of success. Management approval then becomes almost by American Management Association ments. SPC can then be used for automatic. Membership Publication Division, 1987. "maintenance" purposes, with pre- 2For those interested in the theory of pre­ control, there is a considerable body of liter­ ature available in A.S.Q.C. annual confer-

20 Target

'''''''''''''''''''''''''''''''',"'J_}~~.",!,'