Componente Sociocultural

IntroductIon

The initial development of the Toyota production system and the related technologies of lean manufacturing in Japan followed the introduction of statistical process control (SPC) to Japanese industry so closely that they are often considered contemporaneous. Certainly, they are com- pletely aligned and the differences between the two approaches are more complementary than conflicting. This is one reason for the recent growth in Western industry of combined programs, sometimes called “lean six sigma.” This is not a new concept. It is more a reunification of historic companions. Lean manufacturing and statistical process improvement are made for one another.

tHe Power of statIstIcal QualIty

comBIned wItH lean manufacturIng

As everyone who participated in Western manufacturing during the mid- 1970s knows, the early adoption of these two practices in many Japanese industries resulted in improved product quality at the same time that manufacturing costs went down. The devastating competitive effect was that General Motors and others found themselves competing with com- panies such as Toyota, who became quickly able to deliver better products at lower prices. That is essentially the same thing competitors of Gilbarco

and Exxon Butyl Polymers experienced (see Chapter 1). General Motors knew what Toyota was doing, but could not understand how to match that performance. As we adopt lean in the process industries, we will also want to have the natural companion capabilities that come from statistical process control.

Just as lean principles have changed the paradigm of traditional manufac- turing practices, SPC has changed the conceptual model of product qual- ity and process consistency. Statistical process control shifts the emphasis away from the lagging indicator of postproduction inspection toward the leading indicator of assessing and controlling the inherent variation of manufacturing processes. The new theoretical construct is that, when a manufacturing process is inherently capable of successfully producing a product and that process is controlled to operate without variation from that capability, the quality of the product will assuredly be good. In this way, product quality can be controlled by proactive measures to improve the process and control variation rather than through reactive responses to accommodate poor results.

Thus, SPC has two essential components. The first is assessing and improving the inherent capability of the process and the second is oper- ating the process without variation from that capability. Statistical pro- cess control provides engineers a proactive opportunity to improve the process and it provides teams at the frontline a proactive opportunity to operate the process without variation. Like lean, SPC enables everyone to make a meaningful contribution toward achieving the lean ideal of per- fect production.

statistical methods in the Process Industries

Adaptation of statistical methods from mechanical manufacturing for use in the process industries is straightforward. With the exception of some modest computational differences to account for liquid effects common in process manufacturing and virtually unknown in discrete production, such as autocorrelation of material commingled within a single reactor, statistical methods are applied in very similar ways in both industries. For this reason, Exxon Mobil and other leading process manufacturers have been widely practicing SPC in liquid operations for more than 30 years. Indeed, because liquid manufacturers have special problems with process stability and consistency, it is clear that, once again, we have potentially more to gain from these methods than mechanical manufacturers do.

case study: managing variation

The fundamental concept and benefit of assessing and controlling “varia- tion” as a means of managing the performance of manufacturing processes are not always self-evident. Therefore, an example is often useful to intro- duce this topic. One of the best examples of controlling variation came early in my career while I was leading a team that made instrument panel safety pads for General Motors. For reasons of styling and appearance, the instrument panel pads were covered using vacuum-formed vinyl sheets. In the original design, the flat sheet of vinyl prior to vacuum forming was spec- ified as 0.040 inches thick with a tolerance of plus or minus 0.004 inches.

The engineering purpose for this specification was to accommodate the stretching and thinning that occurred as the flat sheets were vacuum formed to the shape of the instrument panel. By starting with a target thickness of 0.040 inches, we believed we could ensure that the final thickness in all locations on the instrument panel would be greater than 0.020 inches. However, variation in the thickness of the starting sheet had a great impact on the production results.

The relatively thicker portions of the sheet heated more slowly and there- fore exited from the heating oven at a lower temperature than the thinner portions. During the forming operation, the portions of the sheet that were both thicker and cooler were naturally less elastic than the thinner and hotter portions. Therefore, the portions of the sheet that were originally thicker stretched far less than the portions that were originally thinner. In fact, the existence of relatively cool and relatively inelastic places on the sheet caused the thinner and hotter places to stretch even more than they otherwise would have.

We went through several iterations of reducing the thickness variation in the flat vinyl sheet. Each time that we improved the variation in the sheet, the performance difference in heating, stretching, and forming also improved. By the time we were able to produce vinyl sheets with a thickness variation of less than 0.001 inch, the sheets formed into a finished product with a uniform thickness. This new consistency in our process allowed us to reduce the target starting thickness for the flat sheet before forming from 0.040 to 0.030 inches and still meet the target for producing formed instru- ment panel covers with a minimum thickness of 0.020 inches.

In fact, we met the minimum thickness target for the finished product more frequently with a thinner and more consistent starting material than we did previously with a thicker material. We did not change any charac- teristic of the material or of the product; however, by improving the con- sistency of the process, we were able to reduce total material consumption by 25%. Further, the now more consistent material also performed better in the field. The instrument panel pad in its final form continued to benefit from the lack of variation in thickness as it responded to the environmental heating and cooling experienced by the cars during their lifetimes.

Key idea: Improved consistency, or reduced variation, and nothing

more allowed a significant reduction in cost accompanied by a signifi- cant improvement in product quality.

The existence of natural process variation has been long recognized in industrial quality practices. It is apparent in product specifications that provide for a tolerance around the optimum measurement. For example, the vinyl sheet was specified to have a thickness of 0.040 inches plus or minus 0.004 inches. This tolerance implied that the product designer anticipated a manufacturing process that would produce up to 10% natu- ral variation on either side of the optimum result.

Unfortunately, prior to SPC there was no practical method to assess and manage variation as an attribute of individual production systems. The designer of instrument panel pads had no specific knowledge of the natu- ral variation of our sheet extruders. Ten percent variation on either side of the optimum seemed like a reasonable target. Still more unfortunately, when natural variation was approached or considered to be an unman- aged and perhaps unmanageable attribute of production, it did not invite the focused improvement that is possible today.

Basic statistical concepts

The most basic element in the industrial practice of statistical methods is process capability. The origin of this concept is generally credited to W. Edwards Deming. Deming’s essential idea is that manufacturing processes are inherently variable, but variation in the process is both measurable and predictable; more importantly, once known, natural variation is manageable. In this way, SPC gives manufacturing leaders the ability to control an attribute of our business that we previously only experienced.

Practically speaking, in manufacturing, the importance of natural pro- cess variation becomes evident when the statistically predicted output of a production process is compared to the specification of the products to be produced by that process. For example, variation in thickness measured in thousandths of an inch, which was very important in the production of vinyl sheets, probably has no importance at all in the production of marine anchor chains. Considered in this way, it becomes clear that the control of process variation that is appropriate to one product is likely to

be inappropriate to the other. The value of managing variation in the pro- duction process is product specific.

Key idea: Deming taught us that the inherent consistency of indus-

trial processes could be analyzed statistically to create a mathemati- cal model that will accurately predict the natural performance of each process. By comparing the natural performance of a process with the requirements of the products produced by that process, it is possible to describe precisely the capability of the process to produce the prod- uct successfully. Once the inherent capability of a process is known in detail, it is possible to improve the operating factors that have an impact on that capability and it is possible to manage the operation of the pro- cess effectively to sustain that capability. That is the essence of statistical process control.

six sigma

Because this practice depends upon statistical analysis, mathematical notation is used to compare process variation with product requirements. Statistical practice reports variation around a mean value in terms of stan- dard deviations, designated with the Greek letter sigma (σ). Statistically, nearly all of the natural variation of a process (99.9997%) is contained within three standard deviations on both sides of the mean value. Thus, nearly all of the natural variation of a process will be contained within a total range of six standard deviations or within a total range of six sigma.

A process that is adjusted so that the central or mean value of the output is aligned with the optimum value of the product specification is described as a “centered process.” When a centered process also has a range of statis- tically predicted natural variation that is the same as the tolerance range of the product specification, all of the natural output of that process is contained within the product specification. The process is naturally and routinely capable of satisfactorily producing that product. According to recent industry practice, this process is a “capable” process and the situa- tion is described as “six-sigma quality.”

In that situation, the primary source of defective products will come from conditions or events that are not normal to the process. These abnor- mal events are described as “special cause” defects. The task of improving the fundamental process is to create a process that is naturally capable of

producing specific products. The task of improving process operations is to protect the process from special causes that disrupt the natural capability of the process.

Using Deming’s methods, it is possible to create a mathematical model of process operations that predicts with great precision the capability of each manufacturing process to produce each product. Further, once the statistical model of process variation exists, it is possible by experiment or experience to identify the sources within each process that produce the variation. By proactively identifying and removing or improving the sources of natural variation, it is possible to make the process inherently more capable.

Furthermore, once the statistical model of the process exists, it is pos- sible to identify periods when the process is not performing according to its natural capability. Identification of such periods makes it possible to know with precision when nonsystemic or special cause variation is influ- encing the performance. Thus, special cause variation can also be proac- tively identified and removed as it occurs. As special causes of variation are removed, the natural capability of the process is not improved, but the immediate performance of the process can be improved.

The statistical model of the process created by SPC practice thus enables two powerful new approaches to process and performance improvement that could not otherwise exist. Through targeted experimentation sup- ported by statistical analysis, it is possible to identify the sources and impact of natural variation within the process. By removing natural varia- tion, it becomes possible to improve the inherent capability and the long- term performance of the process. Similarly, by continuous comparison of actual performance to the statistically predicted performance, it becomes possible for operators to identify in real time when special causes are affecting the production. With nearly perfect correlation between cause and effect, it is possible to identify and eliminate special causes of varia- tion in a way that improves immediate performance. Permanent removal of recurrent special causes can also improve the effective natural capabil- ity of the process.

This understanding and action make proactive quality improvement possible. Previously, we inspected the finished product and reactively adjusted the process. In statistical practice, we can proactively analyze and improve the process and we can continuously monitor process per- formance to protect it from abnormal events.

Process Improvement before statistical analysis

Very few manufacturing processes are completely incapable of producing good products. Most industrial processes produce quite a few good prod- ucts. In the nature of random variation, those “pretty good” processes have periods of great performance and periods of poor performance—all within their natural capabilities. As a result, prior to SPC, when we oper- ated a process that generally produced 98% good products, we were happy when the success rate was periodically 100% and sad when it periodically dropped to 96%. We did not recognize that both the good and bad results coexisted within the natural capabilities of the process.

For this reason, we did not have a well-structured approach to pro- cess improvement. Engineers and managers generally assumed that the existence of good periods demonstrated that the process we had created was a good process, and therefore the bad periods must be the result of something other than our good process. As a result, for processes that periodically but not always produced great performance or, as we would now describe them, processes that were nearly capable, most improvement efforts prior to Deming focused on searching for and removing special cause problems.

Much of the search to identify special causes focused on operator per- formance, which often had nothing at all to do with the natural varia- tion of the process. Second only to operator performance, improvement efforts depended largely on engineers or operators adjusting the process to recenter the mean value of the output. Without understanding that the variation we were seeking to affect was inherent to the process, seeking improved operator performance often had no impact and recentering the range often made the situation worse. We periodically found and resolved a true special cause or a source of natural variation, but these activities were unstructured and infrequent.

As a result, improvement prior to SPC was much slower and much less successful than is now acceptable. Improvement efforts that reac- tively focus only on operator performance, process adjustment, and spe- cial cause elimination are unlikely to enable manufacturing to approach the state of perfect (or defect-free) operations because they ineffectively address only one of two possible sources of defective products: special causes. Improvement of the inherent capability of the process is rare in that situation.

Process Improvement using statistical analysis

Because each process has an inherent capability that can be precisely meas- ured and meaningfully related to the specifications of the product to be produced, SPC enables us to create a useful analytical tool for structured and widespread process improvement. In simple terms, this tool is a sta- tistical model that describes the expected natural performance of the pro- cess. By comparing the statistically predicted performance of a process to the specification limits for the products to be produced, it is possible to predict precisely the extent of success that each process will have in rou- tinely manufacturing each product. To the extent that the process capabil- ity demonstrated by this analysis is not sufficient to meet the needs of the product, more detailed experimentation and analysis can be conducted to identify and remove the sources of variation inherent in the process so that the inherent capability increases.

case study: using a Predictive model

In Suncor’s extraction area, we routinely experienced periods when the process did not satisfactorily separate the sand from the bitumen. As a result, excessive sand was delivered to the upgrader along with the bitu- men. This caused erosion in our pipes and furnaces. For a long time, we lacked a predictive model of the extraction process. Bitumen quality was determined by an inspection conducted after the bitumen was produced. In a classical approach to product quality, when we found that the bitumen contained too much sand, we adjusted the operation of the extraction unit, typically by running it more slowly. Operating slowly improved product quality, but it constrained capacity.

After we adopted statistical methods, we found that the extraction unit was inherently incapable of routinely producing bitumen without sand at full production rates. The natural variation of our process periodically exceeded the range of the product specification. Knowing this, we conducted a more detailed analysis and determined that the source of this variation in product quality was determined largely by the consistency with which we intro- duced feed into the separation cells. When the rate of feed into the cells surged up, the amount of sand in the bitumen surged up as well.

We had thought the separation cells to be the site where the discrete operations of our mine transformed into the continuous processing of synthetic crude oil production. However, this analysis taught us that the change to continuous flow had to occur as material entered the separation process, rather than within that process. With that knowledge, we were

able to improve the inherent capability of the process to operate at full rates while producing satisfactory product.

operational Improvement with statistical analysis

In addition to providing a tool for improving the process, the statistical model of process capability can be utilized as a tool for ongoing assessment and improvement of real-time performance. By comparing current per- formance to the predicted normal performance, it is possible for frontline teams to unambiguously identify when the process is not performing as expected. At that point, the teams also know with some certainty that the process is subject to a nonsystemic or special cause of variation. Because this comparison and assessment are happening in real time, the frontline