Justifying Six Sigma Projects in Manufacturing Management
Transcription
Justifying Six Sigma Projects in Manufacturing Management
Justifying Six Sigma Projects in Manufacturing Management Tapan P. Bagchi1 Narsee Monjee Institute of Management Studies, Shirpur Abstract This paper develops and illustrates a case in to examine and quantify such shortfalls—many manufacturing management, using the instance of being preventable by reduction of quality variance justifying quality improvement of ball bearings—a and/or part variety. Statistical and numerical common precision product whose correct models have been used. Thus, targeting beyond manufacture and assembly greatly affects their scrap and rework, this paper invokes modeling efficiency, utility and life. Mass-produced at high methods to quantify such not-so-visible constraints speed, bearings extend a fertile domain for that limit productivity and profits of high-volume benefiting from QA apparatus including Gage R&R, high-speed processes. ISO standards, sampling, and SPC to Six Sigma DMAIC (Pyzdek 2000). However, when large Keywords: Precision Manufacturing, Variance investments are involved, it becomes imperative Reduction, Hidden Costs of Poor Quality, Numerical that besides the obvious, the hidden costs of quality Modeling, Monte Carlo Simulation. be located and sized. This paper provides methods ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 1 Dr. Tapan Bagchi is the corresponding author who can be reached at tapan.bagchi@nmims.edu 73 exceptional domain in which quality assurance Introduction Managing precision manufacturing of specialized products at their highest achievable performance level is anything but trivial, but management frequently finds itself unable to justify the large investment entailed in superior technologies required to do so. We illustrate a procedure for this by using a real case—a firm's pursuit to upgrade the quality of automotive ball bearings (Figure 1) that it produced. Mounted on skateboards, passenger vehicles, machine tools and even a space shuttle's engine, bearings have been a major mechanical innovation that reduces surface to surface contact between moving surfaces, thereby reducing friction and saving motive energy requirement and its wasteful loss. Traced to drawings by Leonardo da Vinci around 1500, bearings today help the “bearing” of load typically between a shaft and a rotating surface. Bearings are mass-produced by manual to fully automated machining and assembly. Their precise manufacture greatly affects their efficiency, utility and life. Bearings, as contrasted with appliances, toys, furniture, etc., also are an methods from Gage R&R, ISO standards, SPC (Montgomery 2005) and sampling to Six Sigma DMAIC (Pyzdek 2000; Evans 2005) can impact business. A mid-size bearings manufacturer gave this writer an extraordinary opportunity to observe first hand the bearing production process, freely interact with the expert staff manning the machines and work stations and vary process parameters in experiments to observe their effect on product quality. This company had already trained its staff in TQM tools and TPM methods. However, no measurable impact from these on either the bottom line or top line could be discerned by management, as is often the case. Therefore, a rigorous and advanced method that could elevate profits and customer satisfaction was sought. Six Sigma appeared to promise such breakthrough—but, the gains from it could not be projected beforehand. This paper describes the modeling methodologies that led to successfully justifying state-of-the-art technology interventions in this company. Figure 1 The Components of a Ball Bearing 74 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects To scope the quality improvement project Cpk was Motivation of the current study was to help the assessed at quality-bottlenecked process steps. The manufacturer find economic justification for plant ran Orthogonal array experiments (Taguchi possible major technology intervention that could and Clausing 1990) to locate process factors cut tangible and intangible COQ (cost of poor speculated to affect quality by the plant. Thus, key quality) (Gitlow et. al. 2005, Gryna et. al. 2008) and quality deviations in need of attention could be delays and raise profits by reducing production of identified, but no firm basis could be cited to marginal quality bearings. With improved quality, motivate impacting them. However, such studies led the company could possibly sell to premium bearing to re-statement of the project's charter, which markets. became “predict variability of the final bearing assembly based on information available on part This paper is organized as follows. The next section variability.” Key parts in question here were the of this paper outlines the relevant aspects of bearing inner and outer rings of the bearing and the rolling parts manufacture and assembly, and then states balls (Figure 1). the problem of immediate focus—low yield (proportion of acceptable production) of quality Deductive variance prediction from parts to whole bearings, resulting from parts with high proved too complex as it led to queuing or inventory dimensional variability (σ ). The manufacturer type models (Bhat 2008) involving random variables discretised (rounded down) from real numbers. General forms of such models (see (1) and (2) later in this paper) have not yet been solved theoretically. Consequently, the process—the assembly of complete bearings from parts separately manufactured by grinding/honing machines with significant variability in them—was first numerically modeled and then studied by Monte Carlo simulation. The objective was to quantify the relationship of high variability (σ) in manufactured ring sizes (outer and inner) and the variety of bearing balls needed to complete the 2 wanted to be competitive in both quality and profitability. Subsequently, we portray a key operational bottleneck that the plant faced—the challenge of selecting balls of correct size to match a random pair of outer and inner rings produced by track grinding. Next, we provide a statistical perspective of bearing assembly since all machining operations are subject to random variation yielding rings with considerable variance in their dimensions. Then, we show the steps to numerically determine the dependence of distinct ball size requirements on ring grinding variance, and then relate this to yield. assembly. Till this point, “experience” had guided the creation of the large assortment of ball sizes that the plant used. Producing a wide assortment of ball sizes with frequent machine set up changes (a hidden cost) was a burden for the plant. But management could find no sound method to answer why this practice should be changed. They were “committed to deliver high performance bearings to customers”, so the issue remained stuck there. ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Subsequently, a simulation procedure is developed to predict process yield within stated precision given specified randomness of outer and inner ring sizes. Typical questions that management will confront that could be successfully tackled by such simulation are presented next. Results of a number of designed simulation experiments indicate that a rising variety is required in distinct ball sizes as Justifying Six Sigma Projects 75 grinding variance (σ ) goes up (Figure 4). Next we (CTQ) dimensions remain within 0 and 1.2 micron illustrate one such use of simulation to determine and surface finish is acceptable. 2 the distinct categories of “standard” ball sizes required in high yield assembly given Cpk ratings of The Production of Balls track grinding. Subsequently, we use assembly costs Balls are the most critical engineering component of and visible COQ (scrap and rework) to help project bearings as they directly “bear” the load while the justifiable capital expenditure in technology providing minimal resistance to movement. Each that could improve grinding precision (i.e., reduce ball must be precision-machined and polished, and σ ). The paper ends with a summary of conclusions together balls cost typically about 40% of a that management should expect to see in such a bearing's manufacturing cost. Ball manufacturing involves the following steps (The Manufacture of study. Ball Bearings 2009): Ball Bearing Manufacture Ball bearing production is now generic—used by industry worldwide. Some steps may be automated while others are kept manual. Many steps are augmented by automated inspection and SPC. All ball bearings comprise the outer ring, the inner ring, and the rolling balls along with some support parts (Figure 1). Each of these parts is a precision product made from special steel and it must be produced, tested and then assembled correctly in order to enable the completed bearing to perform at location as expected. Ring grinding also called track grinding comprises a sequence as follows (Ball Bearing 2009; The Manufacture of a Ball bearing 2009): • Cold or hot forming operation using steel wire or rods by a heading machine. This leaves a ring of metal (called flash) around the ball. • Removal of flash by rolling between grooved rill plates, giving each ball a very hard surface greatly needed for its load bearing capacity. Process settings include pressure and spinning speed while squeezing by rilling hardens the ball. • Heat treatment. • Setting ball grinder and grinding the ball to its specified dimension. • Lapping to render a perfectly smooth shiny surface, without removing any more material. • Turning of raw material—steel tubes and bars—into raw rings (a step that is often Bearing Assembly outsourced). Outer and inner ring pairs and the corresponding • Heat treatment of raw rings. correct size balls are then selected. Rings are • Precision face grinding. manually or automatically deformed lightly to • Precision outer diameter (OD) and inner insert the balls into the tracks between the rings. diameter (ID) grinding to produce tracks in rings Retainer rings and lubricants may be added. Each on which the balls will roll. final bearing assembly is 100% tested for clearance • Final honing to create surface finish. and noise. Ring width and track dia are the control targets in Critical in final bearing assembly is the selection of track grinding. At each step, sampled inspection is balls that will result in the specified radial clearance done to ensure that the final critical-to-quality between the balls and rings (see Section 2). Note 76 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects that the wider is the variation in dimensions of the Process Variability and its Hidden bearing's inner and outer rings, the larger will be the Impact on Productivity and Costs number of and variety in the size (diameter) of the high-precision hardened steel balls required to complete the ball selection step. To adapt to ring dia variation (imprecise grinding or high σ), industry produces balls of several different “standard” sizes—incrementing in 1 or 2 micron steps in diameter. Such availability of balls of different sizes helps the plant reach the desired radial clearance in the maximum proportion (measured as p) of bearings assembled, even with ring size variability. p measures the yield (= fraction of on-spec bearings automatically assembled from the total outer/inner ring pairs produced) of the assembly line. Ring pairs for which a matching ball size cannot be automatically found reduce p. Such rings are separated and assembled manually by using presorted matching rings. Rings that cannot be manually matched are scraped. A typical precision bearing costs about USD 10 to make. The subject plant made 30 million bearings/year and was incurring an internal failure financial loss of about 1% annually in scrap and rework, very significant and substantial, and visible to management. Additionally, poor quality caused intangible losses. For instance, the outsourced vendor put “extra” steel on the raw turned rings (and charged for it) that were machined off to produce the feed to the precision grinding process. Such non-value added machining reduced the plant's productive capacity. Besides, management remained curious whether improved machining precision (high Cpk at grinding or low σ) could reduce the variety in the size of components (here balls) that must be produced and stocked to provide the desired clearance and “custom” matching during assembly (explained in Section 2). This “variety” of required balls—each kind custom- Manual ring-ball matching is slow (<1/10th of the made—was a significant, but an unknown hourly yield of automated assembly) and a costly component of a bearing's production cost. operation. Note, also that each “standard” ball size Management speculated that the existing poor must be separately produced, requiring extra setups. grinding precision (high σ) caused this variety and Thus, COQ considerations will urge one to lower the lost production capacity due to frequent set up 2 variation σ (or σ) of the track grinding operation. changes on ball machines. This could perhaps be Intuitively, one feels that lower the σ, smaller will be optimized by a study of ball production which the needed number of (“standard”) balls of distinct currently accounts fo 40% of total production cost. diameters for final assembly. Hence, lower overall On the other hand, to find technology benchmarks, cost of bearing production. These considerations the plant had checked the output of outer rings on led us to develop the quantitative relationship two different track grinders, one 25 years old and between ring (track) grinding variability (σ) and the the other new. Cpk differed by 0.69 to 2.02 between variety in “standard” ball sizes needed to keep yield the two, showing a realizable possibility for (p) high. This methodology is described Section 4 reducing σ provided the monetary incentive for such onward. technology upgradation could be quantified. However, as noted, installation of all new machines was to be a large investment that implored quantification (monetizing) of the incentives. This ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 77 task the plant found difficult. Difference = (ID of OR) – (OD of IR) = RC + 2 Ball dia when RC > 0 (2) Thus, one could intuit that wide dimensional variation (σ) in bearing parts—inner rings, balls, The notations used here are ID for inner track dia of and outer rings—led to production of many rings the outer ring (OR) and OD for the outer track dia of and balls that could not be automatically assembled the inner ring (IR). Ball dias are stepped from a to make finished bearings while maintaining the smallest practical size (s) in discrete units of 1 or 2 desired RC (radial clearance, see Section 2) and micron. Hence, for a feasible assembly. other quality characteristics. One approach to raise hourly yield (the proportion of correct assemblies Ball Dia = [Difference - RC]/2 (3) from all parts produced) would be to reduce all variances. A partial solution to this quandary would Industry has found it expedient to manufacture be the use of “standard” sized large assortment balls balls used in bearings to “standard” dimensions in as most bearing manufacturers currently do. The specified steps, like shoes, and not in continuous convenient though expensive way would be to sort dimensions (SKF Bearings Handbook 2009). So the all parts produced and then find matches that meet selection is made by rounding down to the nearest CTQs including RC. Yet another approach to reduce size standard ball to the calculated “Ball Dia” the cost of poor quality (visible and hidden) would determined by (3) for each IR/OR pair being be to seek optimal variance reduction considering at assembled. In this plant, “standard balls” are made the minimum all measureable costs in ring grinding, in one micron steps. ring matching (pairing) and then the selection of balls from the resulting smaller assortment of Note that ID and OD are subject to grinding bearing parts. variation. Hence, as RC increases due to bearing design requirements, the random “Difference” (ID of Ball Selection Process for Bearing Assembly OR) – (OD of IR) in (2) for many IR / OR pairs will Sorting randomly produced rings and balls to find increase, and hence the variety required in standard matches that will successfully fit is slow and effort- ball sizes for the different randomly picked outer / (cost-) intensive, even if the task is automated. inner ring pairs. Conversely, if the “Difference” in (2) Nevertheless, ball selection is a critical practice in were small, a smaller number of standard ball bearing assembly worldwide due to the choices will be required. Of course, if extra effort considerable dimensional variation of machined was made to sort all IRs and ORs before assembly so parts—inner and outer rings. Such selection aims at that the pairs would result in final radial clearances achieving the CTQ target radial clearance (RC) that close to the target RC, perhaps only one or two must meet the engineering spec of each assembled standard-sized balls will be required. bearing. RC (using X+ to represent the maximum of X or 0) is given by the formula: RC (Radial Clearance) = [(Inner Dia of OR - Outer Dia of IR)+ – 2 Ball Dia]+ (1) 78 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects Such sorting of rings before assembly, unless done (using equation (3)) held in stock. The ball completely and cheaply, is not justifiable as it will selected should be such that the final assembly likely require such matching to be automated, with a should result in the target RC between the balls great deal of rejections and recycling of rings since and the ring tracks. IR/OR dimensions vary randomly, before a matched pair is passed to the ball insertion workstation. • Assemble the bearing by pushing the balls into the tracks. Selecting a “standard sized” ball from an assortment • Conduct visual and dimensional checks and of balls, therefore, is the preferred option performance tests (e.g., noise at full speed) on worldwide in bearings assembly and there are each final assembly. specialists who manufacture such automated bearing assembly machines. Swiss bearing makers • Reject bearings that are unacceptable. Accept others for further processing. use this procedure routinely. With wide variation within (ring-to-ring) the inner Ball selection logic is as follows. Generally, it is ring and also within the outer ring dimensions desired that the output produced by a produced, a relatively large number of trials are manufacturing process should fall within the required in Step 3 above to find the best matching of specified range, fixed by tolerance or “spec”. balls for the inner and outer ring. But, as such ring Furthermore, the larger the spec range the greater grinding variation (σ) decreases, within a few trials will be the permitted variation in the output that is the best fitting balls—due to lower dimensional acceptable. Since a bearing comprises the assembly variability of rings—may be found. This is why of the inner ring, outer ring and balls, each produced leading bearing producers are moving towards with some variation, to deliver a “quality” bearing, raising Cp/Cpk of ring manufacture. Such action the manufacturer has to find the best match of an reduces output variations and hence the average inner ring, an outer ring and a ball size such that the “Difference” in (2). The result is that then fewer balls fit correctly (with a clearance) within those “standard” ball sizes will be required to assemble inner and outer rings. The bearing will then possess the bearings while one would still deliver the the desired target RC to allow the balls to roll targeted finished bearing performance. between the rings and have good life. Therefore, ball selection is implemented in the following steps: So, as Cp/Cpk or the “Sigma” metric (Pyzdek 2000) of the ring grinding process goes up, it reduces not • Measure and dimensionally sort all finished inner and outer rings. only the process cycle time that includes matching, but also production of defective bearings (possible • Produce and sort an estimated required marginal misfits) and rejection of rings in bearing assortment (distinct sizes) of standard balls and assembly. With low Cp/Cpk many inner/outer ring keep them in stock. This step is guided by the pairs randomly picked will not match at all, creating shop's experience with the quantities of scrap and raising the cost of poor quality. unmatched rings generated at assembly. • Find matching ring pairs that will lead to on-spec This condition raises a question for the bearing assembly using the rings chosen and a ball size manufacturer: What should be the relative ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 79 precision with which the rings should be A Statistical Perspective of Variations in manufactured? In this paper, we outline procedures Bearing Assembly to help relate process variability (σ) in inner and outer ring machining to the variety required in ball sizes to make high performance bearings. Hence, rather than bear only on inductive or intuitive reasoning as alluded to in Section 2, we sought a stronger case for variance reduction based on analytical reasoning. It was clear that with incentives thus made visible, one would adopt a data-driven and fact-based rather than intuitive quality improvement stance. Assume that inner rings are produced with a nominal outer track mean dimension μI and standard deviation σI. Similarly, assume that outer rings are produced with inner track mean μO and standard deviation σO. Let the size of a randomly picked inner ring be I and that of an outer ring be O. Since rings are independently produced, there is no relationship between random dimensions I and O. However, during assembly, a “feasible” bearing can be assembled using I and O only when: Prima facie, as noted above, grinding Cpk estimates O - I – 2 B – RC ≥ 0 with RC > 0 obtained at the start of the project hinted at a as specified by engineering (4) significant opportunity to reduce cost of set up changes as well as rework and the production of Here, RC is the designed (targeted) radial clearance unacceptable rings. Management already intuited and B is the diameter of the balls to be placed that if ring size variation could be reduced, fewer between the outer and inner rings. In (4), RC is an standard ball sizes would do the job, a lot of set up engineering constant (> 0), dictated by bearing life hours could be reduced, and the manual assembly considerations. B is the size of the (identical) balls done with inner and outer rings rejected by the selected to be placed in the bearing to make the automatic assembly machine could perhaps even be assembly possible. As noted in Section 2, B has to be eliminated. However, quantitative estimates of such carefully chosen for each I and O pair so as not to let incentives were unavailable to them. Due to the the final radial clearance of the assembled bearing processes being random, the tools to help tackle this deviate too much from the design RC value. situation could either be the exact theoretical Otherwise, the fit will be too tight or loose, affecting modeling of the assembly process incorporating the the bearing's installed performance and life. rounding-to-the-lower-dimension practice to pick balls or a numerical approach or Monte Carlo As described in Section 2, in order to make the simulation. In this study, each of these methods was process workable, industry produces “standard” explored. precision balls of various sizes and keeps those in stock. But, these sizes (like shoe sizes in a shoe store) do not vary continuously. One produces balls only at certain “stepped” sizes, usually in steps of 1 (or 2) micron starting with the smallest ball. It is not difficult to see from (4) that the higher are track dimension variances σI2 and σO2+, the wider will be 80 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects the probable difference between the OR/IR rings intuitively as said in Section 1 and elsewhere. A pairs randomly picked for assembly and the larger small point to be noted is that the starting ball size s will be the number of differently sized balls required will be smaller when track grinding required to complete the assembly. In fact, if the variability (σ) is high. (This is shown later in Figure distributions of the outer and inner ring sizes 4 with RC = 20 and σ = 1 vs. cases when σ is higher.) overlap due to high variance, many candidate pairs will be rejected during the automatic assembly step Recall next that O and I are dimensions of two that picks one outer and one inner ring and checks rings—one outer and one inner—that we randomly their dimensions for a feasible assembly. picked with the hope of matching them successfully by fitting them with the appropriately sized balls. For illustration, let the balls be made precisely in Frequently, σ for the grinding process—inner or steps of 2 micron, starting with the smallest ball of outer—is nearly the same. However, in the general size s. If balls are sized successively, they will have case, let the variance of outer ring track dia O be σ02 diameters s, s + 2, s + 4, s + 6, …, s + 2(k – 1), … Then, if and that for the inner ring track dia I be σI2. Then, the the ball with size s + 2(k – 1) gets matched, given I, O larger are process variances σ02 and σI2, the wider the and RC (> 0), we shall have possible random difference or gap (O – I) is likely to be. And, if the distributions of O and I overlap, the O – I – RC = 2(s + 2(k – 1)) (5) Equation (5) leads to cannot lead to successful bearing assembly when O k - 1 = [(O – I – RC)/2 – s]/2 and I due to their high variability would not leave This leads to k = 1 + [(O – I – RC)/2 – s]/2 much clearance between them. In fact, random (6) This expression simplifies to k = 1 + (O – I)/4 – RC/4 – s/2 higher will be the proportion of ring production that variables O and I being independent (the rings are separately produced), the variance of the random (7) variable (O – I) is the sum of the variances of the two random variables O and I. Hence, Equation (7) indicates some important things. First, since RC (radial clearance) and s (the smallest size standard ball available for assembly based on engineering considerations) are constants, the number k is directly dependent on the difference (O – I), the dimensional difference between the outer and inner ring track diameters that we are trying to assemble into an acceptable and properly functioning ball bearing. Equation (7) suggests that the wider is the difference between O and I, the larger will k be, indicating the number of differently sized balls that we must have on hand must then also be large in order to complete bearing assembly with outer and inner rings produced with wide variability. This deduction confirms what is held ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 2 2 Variance (O – I) = σO-I = σO + σI 2 (8) Sometimes—since the outer and inner rings are independently produced and picked for assembly randomly—we may even have picked two rings when O < I! Those two rings must then be put aside for manual matching from a bin of assorted rings in stock. What, therefore, is the message? The first is that in order to reduce the fraction of outer and inner rings rejected by the automatic assembly machine because (O – I) does not leave enough room for two standard balls and RC, one should reduce both σO2 Justifying Six Sigma Projects 81 and σI . Next, when σO and σI are large, many more satisfactorily with another part. If part size outer and inner rings will have a wider gap between variability is too high, however, this process slows them, leading to a larger k, or a wider variety of down the generally high speed production rate and differently-sized balls to be stocked for the introduces hidden costs of poor quality. When outer automatic (or even manual) assembly. and inner rings are mass produced, to find a 2 2 2 matched (Ix, Ox) pair, a relationship is used to guide To summarize, this section has found why reduction ball selection that guarantees the required radial of variability (as urged for instance by the advocates clearance RC. This relationship, shown below, of Six Sigma) may raise yield, reduce the variety of determines the correct ball size for the (Ix, Ox) ring balls and reduce reprocessing rejected rings pair. manually. This will reduce COQ. In the next section, the impact of high machining variance is Ox – Ix = RC + 2 Bx = RC + 2(s + i)v numerically assessed. We see that as σI - O rises, so The correct ball size Bx is given by does the variety required of “standard” balls. Numerical Assessment of Impact of Variance Let the largest size standard ball available be of Reduction on Ball Variety k diameter (s + k). Our attempt here will be to link (Ox – Is the cry for variance reduction (Evans and Lindsay 2005) only hard sell? What if track grinding variability (σ) in ring production was reduced by half? How would that impact yield p at automatic assembly? In this section, we numerically analyze this issue to quantify this impact. We first define some variables as: Ix = diameter of a randomly produced inner ring. Ox = diameter of a randomly produced outer ring. RC = desired radial clearance after assembly. Bx = ball size used to assemble the ball bearing. Ix) to k. We expect that when the tracks of the outer and inner rings are ground, there will be some variability. Intuitively, we feel that larger the resulting track diameter variances σOx2 and/or σIx2, larger will be k or the variety in the sizes of “standard” balls required for correct assembly. This relationship can be numerically established as follows: Given k, the limits on the range or gap (Ox – Ix) to lead to a correct bearing assembly may be determined. Such an assembly will result in the desired radial Also, let s micron be the smallest size standard ball available, balls being made in sizes stepped by one unit (1 micron), starting from size s. Thus, the (i + 1)th ball in this sequence of “standard” balls will be of clearance RC to assure good life and other performance criteria (SKF Bearings Handbook 2009). These limits are given by dia (s + i) micron. This gives In bearing manufacturing almost nothing is thrown away since scrap is visible. It is collected and reprocessed, but at additional cost. In bearing and assembly, attempt is made to find a part that will fit 82 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects This produces the two “not a feasible assembly” When X and Y are independent and normally lower and upper limit conditions as distributed such that X ~ N[μx, σx2] and Y ~ N[μy, σy2], (9) then U is distributed as N[μx -μ y, σx2 + σy2]. Therefore, if FU(u) represents the cdf of U, then the probability and that a randomly picked inner and outer ring will (10) lead to a feasible bearing assembly is P[RC + 2s U RC+2(s + k)] = FU[RC+2(s + k)]-FU[RC + 2s] We will now attempt to find the distribution of (Ox – Ix), the random gap between the outer and inner ring track diameters in which the balls will sit. Let the probability density functions of two independent In the current application, samples of over 300 random variables X and Y be f(x) and g(y). Then the inner ring sizes were found to be normally distribution of U (= X – Y) will be convoluted (Rice distributed; hence, Ix ~ N[μI, σI2] (Sharma 2009). 2007, p. 97). Similarly, the machined outer ring track dia were also normal; thus Ox ~ N[μO, σO2]. Such observations Determining P[U u] for arbitrary distributions f(x) are commonplace in mechanical metal removal by and g(y) is difficult. However, we can determine P[U grinding (Gigo 2005). u] for some common distributions assumed for X and Y as follows. Figure 2 Effect of ring dia variability on the variety of ball sizes required for high yield bearing assembly Expression (11) is highly informative. Note first that will be the value of k, the variety in size of balls Φ(u)-the cdf of the standard normal distribution-is required to produce a feasible bearing assembly. a monotonic function of u. Next, note that in (11), engineering considerations fix the quantities RC, s, Note further that σO2 and σI2, respectively, represent μO and μI. Therefore, if we wish to have 99% of the the variances of the track diameters of the machined machined inner and outer rings correctly outer and inner rings. Hence, higher the variability assembled by picking two equal-sized balls from the in track grinding, larger will be k, the variety in size range [s, s + k], the larger is different sized balls required-the assortment ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 , the larger Justifying Six Sigma Projects 83 starting with the smallest dia s to the largest size s + Table 1 displays the numerically estimated fraction k. Figure 2 shows the numerically determined of correctly assembled ball bearings with assumed dependency between grinding sigma (standard track dia grinding variances σO2 and σI2 . We used deviation σ) and ball variety shown in ball size here numeric values of Φ(.), the standard normal increments in micron. For typical grinding cdf. Observe the impact of increasing minimum k on operations performed on outer and inner rings, yield (p). Desirable yields exceeding 99% with σ= 3 machining variances σO2 and σI2 on machines of were obtained with k values near 10. For lower σO2 c o m p a ra b l e c o n d i t i o n we re s t a t i s t i c a l ly and σI2 one would require a smaller k or fewer sizes indistinguishable, represented here by σ. of balls to complete the assembly without rejecting rings. When k is set at 0, i.e., when only balls of nominal size are available, very few ring pairs are expected to match; hence, almost no bearings could be assembled. Table 1 Impact of rising track dia variability (σ) on k (minimum variety of balls required for 100% yield in bearing assembly) Process scenario # σOuter ring σInner ring k = minimum distinct sizes of balls required p = % yield of correctly assembled bearings 1 1μ 1μ 3 100% 2 2μ 2μ 7 100% 3 3μ 3μ 10 100% 4 4μ 4μ 14 100% 5 5μ 5μ 17 100% 6 6μ 6μ 21 100% 7 1μ 5μ 12 100% 8 5μ 1μ 12 100% This analysis, done numerically, reasserts the also reduce scale (larger lot size) advantages. These fundamental credence—larger the ring machining considerations will prompt one to seek production 2 2 variability (here σO and σI ), larger will be the methods or strategies requiring fewer assortments variety balls needed to ensure correct assembly. of balls. All this is consistent with the spirit of Six From cost of quality standpoint, each additional Sigma®, which champions variability reduction “standard” sized ball requires its own independent (Pyzdek 2000; Evans and Lindsay 2005). set up and production run with its own settings on rilling and other machines. Each time such a set up is On the other extreme, if production machinery is set changed, it reduces the ball plant's throughput, up arbitrarily, a really large assortment of balls will hence, its productivity. Variety or frequent changes be needed by the assembler. The “rejects” from such 84 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects an automated assembly will require a wider effort, in probing the effect of operational variations support to hand-match rings that the assembly and different scenarios for which a numerical model machine cannot accept within its normal involving unusual distributions without probability setting—again raising COQ, hence, adding to tables may not be easy to build. production cost. Thus, a trade-off appears possible between high Cpk ring grinders (with small variance) Monte Carlo Simulation of Ball Bearing on one hand and producing and stocking a larger Assembly variety of “standard” balls with sizes varying from s Figure 3 is a screen shot of the Monte Carlo model 2 2 to (s + k) on the other. As shown, if σO and σI are built in Excel® for bearing assembly simulation. The reduced, k (the number of distinct ball sizes) will be top portion of the worksheet shows where first the smaller. Figure 2 displays this relationship. When process parameters (nominal dimensions μO and μI, the relevant costs are available, an optimization can and standard deviations σO and σI) are entered. be attempted—best done before capital is invested Parameters s and k indicate ball sizes. In this model, in plant machinery and technologies. We note that s is specified by the analyst while incremental ball the automatic selection of “fitting” parts is a sizes are automatically determined by the model by common technique implemented in thousands of finding the correct k using (7) above. At the production systems worldwide. As reconfirmed beginning, distribution of dia variations was here for bearing production, grinding variance established as normal—common in grinding (Gijo reduction will lead to the handling of fewer sizes of 2005). balls getting assembled into finished bearings, lowering the cost of poor quality. Such cost reduction explored during planning could possibly achieve optimum technology configuration in the plant. In the following section, we describe a Monte Carlo method to project the fraction of ring mismatches occurring during assembly, if the analyst specifies the ball sizes available, radial clearance desired, the nominal ring sizes μO and μI, and grinding standard deviations σO and σI. Prevailing dimensions and standard deviations may be obtained from the relevant X Bar-R control charts maintained on the shop floor. We envisage two reasons for attempting to study the bearing assembly process by Monte Carlo simulation. The first is the flexibility that it affords in respect to the distributions of the part dimensions. The second is the flexibility simulation extends, albeit at the cost of higher computational ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 85 Figure 3 The automated bearing assembly simulator In this simulation of automated bearing assembly, among the variables (e.g., relationships of the ring sizes O and I are randomly found, presently type (7), (9), (10), etc.). The variables presently normally distributed dimensions based on μO and μI, involved were outer and inner ring track dia, ball and σO and σI, respectively, as entered in the green size, and the ball variety k. zone of the worksheet. Random samples are drawn by Excel® using the NORMINV() function. Each row • Set up the experimental framework—the from Row 26 downward uses one simulated outer variables to be manipulated (here k), the ring (O value) and one simulated inner ring constants (the smallest ball size s, and radial dimension (I value). Column headings (Column F clearance RC) and the random sampling onward) implement the relationship (7) in the steps mechanism from specified distributions (O and of computation. The simulation progresses as I). follows: • Once the model is coded and set up (here the • Identify the sequence of processing steps, Excel® model of Figure 3), conduct pilot runs to control and noise factors, any information on estimate the variance of the response (here p or randomness and its nature (the associated yield) and then the required sample size (length distributions). Determine the relationships of the simulation run). 86 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects • Simulate for the desired sample; collect data and complete output analysis. The total count of “ball unavailable” indicates ring size mismatches that could not be automatically assembled into a bearing using the k standard • Replicate runs with identical seeds to reduce assorted ball sizes provided. This estimates p and variance of the yield (p) estimated (Law and the manual work needed to complete the job. The Kelton 2000). simulation model was validated using physical inner/outer ring production lots of size 300 each at Results of one round of simulated assembly with the prevailing grinding variability (σ) level (5 inputs from worksheet cells C26, D26, E26, etc., micron) and a target RC of 20 micron. Hand appear in Column K—“Micron Size of Balls assembly produced about 10% mismatches (too Required.” Note that in practice calculated ball sizes small or too large rings) that could not deliver the are rounded down to a whole number standard size target RC using 12 different standard sized balls at for conservative (slightly larger) clearance resulting hand (cf. Table 2). For given σO and σI for grinding, a in the assembled bearing. If a feasible size of balls is sample size of 1,000 simulated assemblies provided found, its size is noted. Otherwise the a conservative 2-digit precision of yield estimate, corresponding row (i.e., the simulated O and I pair sufficient to illustrate the utility of simulation. put up for assembly) reports “ball unavailable”. Table 2 Numerical estimation of yield p (fraction of bearings automatically assembled) using the standard normal cdf ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 87 Experiments could be now run by treating process ball sizes must be precision-made to exact specs or machining variances σO and σI as experimental (10th or better in micron). Manual sorting and variables with nominal sizes μO and μI and assembly using matched outer and inner rings and engineering parameters RC, s and k held at specified balls is the traditional fallback. But this results in levels. Plots of the output data will graphically errors and the consequent unwanted variation in indicate the effect of variances σO and σI on the bearing characteristics. Some operations are, process output p (yield)—the fraction of rings therefore, automated. The common one in bearing submitted for assembly that could be finished into production is assembly. complete bearings. This exercise showed how the Still, many questions remain about the estimation of effect of comptemplated process changes may be the visible and hidden COQ and the economic quantitatively estimated. Simulation is a well- optimality of technology interventions. Some of known method to use here (Law and Kelton 2000). these questions may be probed by Monte Carlo simulation. Examples are: A Strategic Application of the Assembly Simulation Model Created A plant typically confronts questions for which quantitative answers are not easily found to guide strategic changes such as adopting new technology. Answers are often sought based on the tacit (experiential and intuitive) knowledge of senior management, and the machinists and quality control staff with practical hands-on experience. However, industry now generally appreciates that reduction in variability of parts and the final product or service should be a key target for an enterprise. For ball bearings, part size variability affects the final radial clearance achieved, which affects it life, performance and production cost. Managers also want consistency and, therefore, look to locate the “problem” stages in the shop processes that lead to high variability of output. As found here, high variance of dimensions, for instance, raises rework and scrap, and thus chokes throughput. Importantly, it raises hidden costs. Due to the inescapable variability in track grinding and the stringency demanded in reaching the target clearance in each bearing assembled, a key not-sovisible cost in bearing production is the requirement of large variety of balls. Each of these 88 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 • Quantitative projection of the impact of high machining variability on the variety of standard balls (k) needed to complete the final assembly at high yield—p, the proportion of acceptable bearing assemblies produced. • Effect of Radial Clearance (RC) on utilization of balls. • Combined impact of high variability of inner and outer rings on yield. • Incentives for tightening machining tolerances at key process steps, and at input—the raw ring grinding stage. • Quantification of the extent of manual re-work given specified levels of Cp/Cpk at different processing steps. • Cost optimization of the complete bearings manufacturing process based on quantified estimates of scrap, rework and capital cost of work centers. • Business development—help determine plant capabilities required to move into making superior quality ball bearings used in machine tools, aerospace and similar applications. In the following paragraphs we examine one such question. Justifying Six Sigma Projects Estimating the Variety Required in Ball Sizes Some inferences may be easily drawn from Table 3 Given Cpk of Ring Grinding Machines and Figure 4. As ring grinding variability Earlier, we had hinted that low Cp/Cpk will lead to represented here by σO and σI improves (i.e., extra manual work, larger variety required in ball standard deviations σO and σI reduce in value), sizes, as well as possible degradation of bearing production yield improves. This implies performance of bearings that are near-marginal. drop of rework and possible stoppage of scrapping Low Cp/Cpk, i.e., high natural variability or process unmatched rings. will also lead to extra manual work (re-work) to find matching ring pairs that an automated assembly Figure 4 displays another important effect of machine would reject. Such matters are intuitively variance reduction. As grinding σ rises so does the known to most bearing manufacturers. They believe variety of balls required to assure the correct that if variation in track grinding, for instance, is bearing assembly. Engineering considerations reduced, fewer varieties in the standardized sizes of dictate that balls must fit into the grooves as well as balls would be required, considerable waste could leave the required clearance RC within the bearing. be reduced and the manual assembly operation Hence, given a ball size, wider the ring size done using inner and outer rings rejected by the variability σ, larger will be the number of trials with automatic assembly could even be eliminated. different rings needed to complete the assembly. However, the theoretical derivation of the link The quantitative relationship is not difficult to infer between grinding variance and the extra ball sizes here. When costs of automated and manual required (see Figure 2) is non trivial. assembly are known and so is the extra cost of producing an extra variety of ball, one may work out Using numerical or analytical (where feasible) the trade off to determine the optimum grinding models or Monte Carlo simulation, the assembly machine capability or σ or the corresponding Cpk. process may be studied to relate the statistical The appropriateness of technology upgradation variance of track grinding to ball size variety. While may thus be found. this study is attempted, one can restrict the answers such that the designed (target) radial clearance is always maintained. This relationship, determined by the Monte Carlo simulator is shown in Table 3 and Figure 3. Ring (OR and IR) lot size was 1,000 for each simulation. The inference that can be immediately drawn is that yield (% of bearings correctly assembled in one pass) goes up as ring grinding σ (representing σO and/or σI) goes down. Figure 4 displays the distribution of ball sizes required determined by simulation for certain prestated grinding variability (σO and σI). Other scenarios may be similarly evaluated. ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 89 Figure 4 Ball size distributions for different grinding σ 90 RC = 20, σ = 1 RC = 20, σ = 2 RC = 20, σ = 3 RC = 20, σ = 4 RC = 20, σ = 5 RC = 20, σ = 6 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects Increasing RC will shift the outer ring dia outward if the same assortment of balls is to be used. Alternatively, smaller balls will be more frequently required to maintain yield. Incentives for tightening machining tolerances may also be similarly evaluated. The same may be done for any contemplated improvement in Cp/Cpk at grinding. In fact, simulation may be extended backward where raw rings are received form outsourced machine shops. This provides a basis to set incoming specs. Figure 5 Finishing balls before inspection (adopted from Reference The Manufacturing of a Ball Bearing) Table 3 Yield (p) and ball sizes required as function of grinding variability Radial Clearance = 15 micron Radial Clearance = 20 micron Grinding variability (std dev) p with 11 balls Distinct ball sizes required for 100% yield Grinding variability (std dev) p with 11 balls Distinct ball sizes required for 100% yield σ=1 100% 7 σ=1 100% 6 σ=2 99% 10 σ=2 100% 10 σ=3 96% 13 σ=3 99% 15 σ=4 90% 19 σ=4 94% 17 σ=5 83% >21 σ=5 89% >20 σ=6 74% >22 σ=6 79% >22 For new business development, to enter superior quantified information. This will enable bearing markets, the required grinding σ hence management select best intervention options on the machining capabilities (Cpk) may be similarly basis of reliable estimates of the gains such as determined. In fact, due to variability, a part of production yield improvement. Found this way, the current production may already qualify to be sold as financial returns become significantly more certain high precision bearings. Many other uses may be and measurable. made of the methods and tools illustrated here. The greatest of all is that such methods produce ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 91 Outline of a Procedure to Economically Justify reduction in does not come free, it frequently Variance Reduction requires a step jump in grinding technology as it is Balls are what make the bearing bear the load and affected by tools, grinding speed, dressing, material ensure friction-free movement for the life of the being ground, operator skills, coolant temperature, b e a r i n g . B a l l s a re , t h e re f o re , c a re f u l ly spindle vibration, etc. Thus, the benefits of reducing manufactured, starting with thick steel wire, cold must be economically more than the cost of variance heading, cut into pieces and then smashed between reduction. We outline this analysis as follows: two steel dies (The Manufacture of a Ball bearing 2009). Then, the flash is removed and balls are heat- Let the automatic assembly cost of a bearing be a. treated to make them very hard, then tempered to Assume that the % yield of the automatic assembly make them tough. Finishing requires grinding operation is p. Therefore, the proportion of manual between grinding wheels and then lapped with very rework required to complete assembling all inner fine abrasive slurry to polish them for several hours and outer rings manufactured is (1 - p). Let the to reach correct dia and mirror-like finish (Figure rework cost be ca or ca, with the assumption that 5). Each ball type takes 6 to 10 hours to finish. A the factor c ≥ 1. Therefore, the total assembly cost of medium precision ball cannot be out of round more a bearing will be (Gryna et al 2008, Chapter 2) than 25 millionth of an inch while high speed precision bearing balls are allowed only five- Total assembly cost = a (fraction automatically millionth of an inch roundness variation. Therefore, assembled) + ca (fraction manually assembled) to change ball size, set up has to be changed with much care and effort and the process must be = a p + ca (1 - p) stabilized before production begins. As shown = ca - p(c-1)a (12) above, as ball dia variety increases, so does the required number of set ups, each set up reducing Now, as seen in the sections above, p is a function of production capacity. σ and k. Figure 6 illustrates this relationship for an example in which radial clearance RC has been As indicated in earlier sections, in bearing assembly, assumed to be 20 micron and inner and outer mean the critical process characteristic is the standard track diameters are as shown in Table 3. deviation (σ) of track dia produced by ring grinding. This characteristic determines the yield of quality bearings and the extent of rework or scrap generated. And, lower the yield (p) in automatic assembly, higher will be the share of manually assembled bearings, both a slower and a more costly process. As we saw, higher the , higher will be the variety (k) required to complete the assembly automatically (Table 3 and Figure 4). Generally speaking, due to the additional ball manufacturing set ups required, ball and hence bearing production cost rises as the variety of ball size increases. But, 92 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects Figure 6 Dependency of yield of automated assembly on σ and ball variety k Some observations from Figure 6 are straight- variety of balls. This cost is a “hidden” COQ forward. For a given variety (k) of balls available, as component of production cost and is determined, σ (i.e., track grinding variability) increases, p falls. among other things, by the length of the set up time. Also, to increase p at a given grinding variability (σ) It is incurred when the ball grinding machine is level, k must be raised, i.e., a wider variety in ball reset to make a different size ball, assuming that the sizes must be available to increase the yield of plant is operating in a “sold out” market and can sell automatic assembly. From data generated by all it can produce. This hidden cost has three numerical modeling, it is possible to empirically components: (a) the cost of resources expended on relate p to process parameters and k. Therefore, physically changing the actual setup; (b) the loss when this relationship is known, one can estimate due to production lost during the switchover; and (c) the total assembly cost using (12) given any values the extra cost of storing and managing the stock of of and k. Using cost accounting methods such as an extra variety of balls. Alternatively, if balls are Activity Based Costing the relevant costs may be outsourced, it will lead to purchasing and stocking estimated. For illustration we used the numerical an additional size ball. This information too is framework used in producing Table 3 to develop an quantifiable. Thus, given ring grinding variability empirical model relating p to σ and k as follows. Such and ball variety k (ignoring material cost) we find empirical models (Gujarati and Sangeetha 2007) are helpful when the direct theoretical derivation of Total (ball + assembly cost) = ca - p(c-1)a + Cost of the required relationships is not possible. Thus, making k different ball sizes (14) following Jianxin and Tseng (1999) and Al-Omiri and Drury (2007), where p is given by (13). Expression (14) is a 2 p = 0.693689 + 0.088342 k – 0.10595 σ - 0.00413 k + 2 0.00232 σ + 0.005933 k function of and k. (13) The only cost that is missing so far is the cost of 2 Model (13) has a R of 0.80. The other cost that one reducing or the track grinding variability. Grinding needs to quantify is that of producing an additional variance (σ2) is a function of technology and the ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 93 tightness with which the grinding process is interventions by DMAIC (Pyzdek 2000). controlled by operators (operator skill). Furthermore, a reduction of σ will improve bearing performance and thus create intangible benefits of However, as outlined above, it is possible to quantify supplying superior quality bearings. Such benefits the cost savings resulting from reducing σ for a will be additional and their value and returns will be given volume of annual bearing production. It is strategic (Pyzdek 2000; Evans and Lindsay 2005). possible to estimate the NPV of accumulated yearly savings and consequently how much investment The actual costs at the plant in question cannot be can be justified in ring grinding precision shown due to proprietary reasons. However, improvement to benefit business. This can be subsequent to this study, based on the insights projected for a planned period of selling those gained and incentives estimated based on delays, particular bearings. The startling surprise is that if a production lost due to set up changes in ball plant produces 10 million bearings annually and manufacture and the potential to upgrade products, one is able to reduce cost of production by 2 the company decided that it now had sufficient basis cents/bearing, assuming that the business runs for to launch a full-blown Six Sigma® project involving 7 years, an NPV in excess of USD 1 million will multi-factor Orthogonal Array experiments to pin materialize at nominal interest rates. Such sums can down factors that could raise p, the company's first- j u st i fy si g n i fi c a n t p ro c e ss i m p rove m e n t pass bearings yield. Conclusions Every plant manager aims to reduce defects. procedure to quantify hidden losses—here due to However, few in the supervisory or engineering staff the extra sets of bearing balls that must be are able to formally quantify the cost of poor quality precision-manufactured, stocked and used to raise even where monetary values (losses) are suspected yield in automated bearing assembly. The method to be large. The result is the perpetuation of status employs numerical as well as Monte Carlo models, quo, unless a new facility with superior all done in Excel®, and it results in quantitative technologies is proposed and justified. In a estimates of yield, manual work required and the manufacturing set up the “tip of the variety required in ball bearing balls before any iceberg”—scraps, rework and warrantee commitment needs to be made to alter plan service—are generally visible and reported equipment, workforce or facilities. monthly or yearly. But, the hidden factory that runs alongside of the brick-and–mortar factory or other Numerous other questions in strategizing hidden costs of poor quality (COQ) typically evade manufacturing could also be tackled by such estimation. Still, without the benefits monetized, analytical procedures. Typical instances of these are management will not be interested in large scale listed in Section 6. However, before one undertakes interventions such as Six Sigma, even if the such a study to initiate Six Sigma® DMAIC (see methodology has worked wonders at scores of Pyzdek 2000 or Evans 2005), all COQ (cost of organizations worldwide. This paper illustrates a quality) components must be drilled into and 94 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects targeted for quantification. This would be the most capabilities proved sufficient for the plant sensible way to initiate priority action using DMAIC, management to comprehend the steps in the done best at the “D” (Define) stage. analysis and see how the conclusions were reached. This became a stepping stone to raise many related In this instance it was particularly gratifying to questions about alternative ways to cut cost and quantitatively affirm the intuitive assertions of impact profit. One such initiative justified was the plant management. Their reactions became upgradation of vendor management. Another was instrumental in delving deeper to locate tacit to reset the settings in the automated facilities opportunities that existed for raising profits. toward the gradual removal of manual work. A significant issue tossed up was the engineering ® Excel models were deliberately used in this work to optimization of the target RC (radial clearance) serve as easy-to-use decision support tools usable values for its close interplay with bearing at the plant level, for many such decisions are often performance and the variety of balls required in made locally without the sophistication consultants automatic assembly. This constituted the charter of ® typically engage in. Excel 's graphic and statistical a separate Six Sigma® project. References • Al-Omiri M and Drury Colin (2007). A Survey of • Juran's Quality Planning and Analysis for Systems in UK Organisations, Management Enterprise Quality, 5th ed., Tata McGraw-Hill. • Volume-1/Ball-Bearing.html, accessed April 25, Tseng Mitchel M (1999). A Pragmatic Approach 2009. to Product Costing based on Standard Time Bhat U Narayan (2008). An Introduction to Estimates, International Journal of Operations Queueing Theory: Modeling and Analysis in and Production Management, 735-755. • Law A M and Kelton W D (2000). Simulation Modeling and Analysis, 3rd ed., McGraw-Hill. Evans J R and Lindsay W M (2005). An • Improvement, Thomson . • Rice J A (2007). Mathematical Statistics and Data Analysis, 3rd ed., Thomson • Sharma Poonam (2009). Report on MBA Project, Statistical Techniques, Quality Engineering, 17: Vinod Gupta School of 309-315. Kharagpur, India. Gitlow H S, Oppenheim A J, Oppenheim R and rd Introduction to Statistical Quality Control, 4 ed., Wiley. Gijo E R (2005). Improving Process Capability of Manufacturing Process by Application of Montgomery D C (2005). th I n t ro d u c t i o n t o Si x Si g m a & P ro ce s s • Basic Econometrics, Tata McGraw-Hill.Jianxin Jio and Technology, Birkhouser. • Gujarati D N and Sangeetha (2007). Ball Bearing, http://www.madehow.com/ Applications Series, Statistics for Industry and • Gryna F M, Chua R C H and Defeo J A (2008). facts influencing the Choice of Product Costing Accounting Research, Vol 18(4), Dec, 399-424. • • • Management, IIT SKF Bearings Handbook (2009). http:// www. Levine D M (2005). Quality Management, 3 ed., who-sells-it.com/r/skf-bearing-handbook.html, Tata McGraw-Hill. accessed April 25, 2009. ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects 95 • • Taugchi G and Clausing Don (1990). Robust Acknowledgements Quality, Harvard Business Review, January- The author thanks Harsh Sachdev, Joydeep February. Sengupta, Jyoti Mukherji and Tapan Mondal of Tata The Manufacturing of a Ball Bearing, Bearings. Together they provided a wealth of http://www.bearingsindustry.com/manufactu practical shop floor knowledge and economic ring.pdf, accessed April 25, 2009. insights into bearing manufacturing. Poonam Sharma provided data collection assistance. Dr. Tapan Bagchi is the Director of Shirpur campus of NMIMS university. He was was recenctly awarded Doctor of Science by IIT Kharagpur. His research interests have been in the area of quality engineering and production. Dr. Bagchi has published prolifically. Prior to his current responsibility, he has served in professorial and academic leadership capacities in IIT Kharagpur, NITIE and S.P. Jain Institute of Management and Research. 96 ISSN: 0971-1023 NMIMS Management Review Volume: April - May 2012 Justifying Six Sigma Projects