Featured Product
This Week in Quality Digest Live
Lean Features
Gleb Tsipursky
Belief that innovation is geographically bound to office spaces is challenged by empirical evidence
Jamie Fernandes
From design to inspection to supply chain management, AI is transforming manufacturing
James Chan
Start the transition to preventive maintenance
Mark Rosenthal
The intersection between Toyota kata and VSM
Erin Vogen
Eight steps to simplify the process

More Features

Lean News
New video in the NIST ‘Heroes’ series
Embrace mistakes as valuable opportunities for improvement
Introducing solutions to improve production performance
Helping organizations improve quality and performance
Quality doesn’t have to sacrifice efficiency
Weighing supply and customer satisfaction
Specifically designed for defense and aerospace CNC machining and manufacturing
From excess inventory and nonvalue work to $2 million in cost savings
Tactics aim to improve job quality and retain a high-performing workforce

More News

Lean

Control Charts in Manufacturing: Are They Still Relevant?

Part 2 of our series on SPC in a digital era

Published: Wednesday, November 8, 2023 - 11:03

We are one year away from the 100th anniversary of the creation of the control chart: Walter Shewhart created the control chart in 1924 as an aid to Western Electric’s manufacturing operations. Since it’s almost prehistoric, is it now time to leave the control chart technique—that started out using pen and paper—to the past? Or, in this digital era, is the control chart still relevant to enable manufacturers to improve their competitive position by improving quality and productivity, and reducing waste?

Read on to see the story of two plants. Some key words to look out for:
• Predictable process
• Actionable insights
• Improvement
• Cost savings
• Waste


The annotated control chart above is from the Tokai Rikka plant in Japan more than 40 years ago.

The two plants

Plants 7 and 11 report their manufacturing performance results and improvement plans to the division’s technical unit at the end of each month. The plant managers have little in common, and one of their differences is seen in the word control. The manager of Plant 11 attended a W. Edwards Deming four-day seminar in 1991, which goes some way in explaining this difference:
• In Plant 7, an “in-control process” signifies that things are OK, with the process having produced to specification and no issues in production.
• In Plant 11, an “in-control process” is said to be a predictable process.

Could this apparently trivial difference influence the performance and profitability of either plant? To explore this question, we focus on product characteristic 16 in product AFP07-1, a product with good margins that has recently been introduced into both plants. Characteristic 16 comes predominantly from the most expensive raw material used in the production of AFP07-1, which is added as part of a mixing step before further downstream processing.

To start, we look at the first two weeks of data of the new month, knowing that the meeting with the technical unit takes place two weeks later. To see how both plants do, read on.

Plant 7, where “in control” means “in spec”: Weeks 1 and 2

In Plant 7, the nominal batch size is 2.25 tons of finished product. By the end of the second week of this new month, 85 batches had been produced, totaling just over 191 tons of product. Figure 1 plots the batch measurements for characteristic 16 in time sequence. Also shown are the specification limits of 62 to 71.5 and the target value of 66. The target value was placed slightly below the midpoint of the specifications to keep to a minimum the addition rate of the expensive raw material.


Figure 1: Time-series plot of product characteristic 16’s data over two weeks in Plant 7.

Batches 15, 30 and 54 were out of spec and had not yet been reworked, meaning that the total amount of finished product made after two weeks was 191 tons minus three batches. By this point in the current month, the production plan had targeted 200.3 tons—89 in-spec batches—leaving Plant 7 uncomfortably behind the clock.

With too much out of spec, and the production quota not met, in Plant 7’s daily language this process was clearly not in control. Leading into the weekly meeting, two key questions to address were:
Question 1: How could production be accelerated in the next two weeks to meet the monthly quota?
Question 2: What caused the three out-of-spec batches, and what could be done about it?

Question 1 was easily answered: Some maintenance activities could be postponed until the next month, freeing up time on the production line.

Question 2 had no immediate answer, so the production manager instructed one of the shift leaders, Kim, to (a) find the root cause for the three out-of-spec batches and (b) propose how to deal with it. Kim was asked to report back in two days.

Kim brought in the various support functions to investigate the specific batches 15, 30, and 54. A recently updated cause-and-effect diagram was used to drive the investigation. Among other points, raw material records were checked, rework usage was checked, the production log book was checked, automation data that trended key process parameters were checked, measurement records in the lab were checked, and storage and handling conditions in the warehouse were checked. Kim asked the production operators if they recalled anything different when these batches had been produced. Not one of the points investigated offered a valid explanation for the three out-of-spec batches.

Two days later, a disenchanted Kim updated her boss in the daily review meeting: Nothing concrete had been found to explain the out-of-spec batches. The meeting wasn’t the best, but Kim was comforted by the fact that her shift had three days off. Production continued unabated, with further investigation held off for the time being.

Plant 11, where “in control” means “predictable”: Weeks 1 and 2

Plant 11 has the same specifications (62 to 71.5) and target value (66) as Plant 7 for characteristic 16. Plant 11 has a standard batch size of 2.6 tons and had restarted operations in this new month after a 10-day shutdown for annual maintenance. By midday Friday of Week 2, 80 batches had been produced, totaling 208 tons of product. Friday afternoon and the weekend were required to finish off some maintenance tasks. 208 tons output was a little shy of the planned 213 tons due to some teething problems on the first day of production after the shutdown.

All 80 produced batches were in spec for product characteristic 16. While this might have been the end of the story in Plant 7, it wasn’t in Plant 11 because the question of process predictability for characteristic 16 had not yet been discussed.

Plant 11 had produced a summary schematic of the key features of predictable and unpredictable processes, as shown in Figure 2. (The control charts use the data from Figure 7 in Shewhart’s 1931 book Economic Control of Quality of Manufactured Product.)


Figure 2: Plant 11’s summary of predictable and unpredictable processes

For Weeks 1 and 2, Figure 3 shows the control chart for product characteristic 16’s data, with a control chart for individual values used because the data occur one-value-per-batch. The red limits in Figure 3 come from the 80 batch data and how to calculate the limits for this type of chart is described in “Process Capability: What It Is and How It Helps, Part 2.” For charts of individual values, Plant 11 uses the term natural process limits instead of the more traditional control limits. (See postscripts at the end for a brief discussion of terminology.)

Figure 3: Control chart of product characteristic 16’s data over Weeks 1 and 2 in Plant 11.

There are no specification limits—the voice of the customer—on Figure 3 because the control chart represents the voice of the process. Learning to differentiate these two voices is fundamental to successful control chart use.

For the two weeks under study, this process does not display predictable behavior over time. An abundance of points are circled in red and have a “1” or “2” next to them, signaling that one or more changes in the process have occurred.

Plant 11 routinely applies two detection rules to detect process changes (i.e., unpredictable process behavior).

A “1” signals any single value falling outside a natural process limit, noting that this signal can occur as an isolated point or as part of a “2.”

A “2” is used whenever nine or more points in a row are on one side of the control chart’s central line, noting that the consecutive nature of this signal implies a prolonged change in the process.

Detection rules provide guidance on when to look for a cause of process change and when not to. Signals in the data therefore provide a green light to further interpret the control chart, asking when, and for how long, did the change, or changes, in the process happen?

Discovering the cause of a process change presents an opportunity to learn about the process and to generate actionable insights. The follow-through—taking action—is the essential ingredient in control chart use. This ingredient enables process improvement, which is illustrated and discussed below.

Beth, a member of the plant’s technology group leading this work, studied Figure 3 on the Friday afternoon, aiming to best pinpoint when the changes in the process had likely occurred over the last two weeks. One of her “go-to approaches” is the rule of thumb to look for a potential process change when the running record last crossed the control chart’s (green) central line. This is illustrated in Figure 4.


Figure 4: Annotations to Figure 3 to illustrate how the plant technologist sought to understand the signals of process unpredictability.

Figure 4 allows for the data to be investigated in three groups: Batches 1-21, 22-52 and 53-80. Beth examined these three groups for internal predictability, or consistency, by creating three sets of limits, as shown in Figure 5.


Figure 5: Control chart reflecting the three groups of batches identified by Beth.

Thinking she might be on to something, Beth followed through and discovered that, for product AFP07-1’s key raw material contributing most of characteristic 16, batches from three different suppliers had been used in the last two weeks of production. The first change of supplier occurred at Batch 22, and the second change at Batch 55. The timing of these two changes of supplier material is in close agreement with the insight Beth had gained.

Beth investigated further, aiming to make the insight actionable. She found that the density of the material varied significantly between supplier yet, in production, the density value used to control the automatic dosing of the ingredient had been the target value stated in the raw material’s supplier specification document.

Beth packaged this new information into Figure 6’s time-series plot using a line break and different colors to highlight this “supplier effect” (i.e., the change at Batches 22 and 55). She liked this graph, using it to present and communicate the information to colleagues.


Figure 6: Time-series plot organized by supplier of the key raw material.

This finding on the densities was instrumental because the equipment used to dose this raw material as part of the mixing process works on a volumetric dosing principle. The product formulation, however, doesn’t ask for volume, instead stipulating an addition level by mass. For the dosing step to achieve the correct mass, the density of the material is one of the critical inputs since Density = Mass/Volume. Beth knew that if the volume is dosed on target, but the material density varies significantly, the added mass will also fluctuate significantly, causing increased variation in characteristic 16, exactly as seen in the charts above. 

The agreed action was to measure the density of each supplied batch of the key raw material before use and input the measured value into the automation system, allowing for the necessary adjustments to dosing parameter settings. This action would better control this cause of variation to eliminate the identified “supplier effect.”

Plant 11, where “in control” means “predictable”: Week 3

The action to eliminate the “supplier density effect” was implemented over the weekend, and in the course of Week 42 batches were produced using material from two of the three suppliers. With 109 tons of in-spec product made in the week, Plant 11 was back on track to meet the quota for the month.

Figure 7 shows the control chart covering Weeks 1, 2, and 3 of the current month, with the points color-coded by raw-material supplier. Respecting the knowledge that a process change had occurred at the end of Week 2, a new set of natural process limits is calculated using all 42 data from Week 3.


Figure 7: Control chart of Plant 11 with limits calculated for Weeks 1 and 2, and Week 3.

In Plant 11’s meeting at the end of Week 3, Beth updated the team that (a) the process was currently predictable and (b) the quota for the week had been met. When challenged on the first two batches of Week 3—Batches 81 and 82 are above the upper limit, and therefore signals—Beth responded that the measured material density value hadn’t been entered in the automation system’s computer (meaning the target value from the raw material specification was used by default as a control input to the dosing process). With this, the importance of controlling this “supplier density effect” became even clearer. The team acknowledged Beth’s work.

Beth agreed to provide an update in the following week (Week 4) before the meeting with the technical unit. Before looking at Week 4, we return to Plant 7.

Plant 7, where “in control” means “in spec”: Weeks 3 and 4 and the monthly meeting

After her three days off, Kim and her shift returned to work. They were happy to hear that, so far in Week 3, no out-of-spec batches had been produced for characteristic 16. This positive trend continued until the end of the week, by which time 52 batches had been produced to successfully meet the week’s revised output of 117 tons (52 x 2.25 = 117). This output included use of the three rework batches from Weeks 1 and 2.

Kim prepared Figure 8 for the end-of-week meeting to confirm that things were back on track, and the process, in Plant 7’s language, was back in control. As expected, this meeting was more pleasant than the one a week earlier.


Figure 8: Time-series plot of product characteristic 16’s data over Week 3 in Plant 7.

Unfortunately, the good news of Week 3 didn’t last long. Even though all 44 planned batches were made in Week 4, two out-of-spec batches were identified on the last day of production, leaving the output for the month shy of the month’s quota. Week 4’s chart for characteristic 16 is shown in Figure 9. The process, in Plant 7’s description, was again out of control due to the two out-of-spec batches and the indirect effects of failing to meet the quota.


Figure 9: Time-series plot of product characteristic 16’s data over Week 4 in Plant 7.

It doesn’t take much imagination to realize that the manager of Plant 7 didn’t have the happiest of times in the technical division’s monthly meeting: Quality was below target, costs were above target, and the quota for the month hadn’t been met. There was no celebratory beer for the plant manager this time.

The plant manager knew things wouldn’t be easy in the next month. The problem of the out-of-spec batches in Weeks 1 and 2 hadn’t been resolved, and the recently postponed maintenance activities would have to be carried out in the next couple of weeks, eating up time to produce. A special meeting was arranged for the next morning.

After further discussion of Plant 11, we’ll see whether control charts could have been helpful to Plant 7.

Plant 11, where “in control” means “predictable”: Week 4

In Week 4, all 37 batches needed to meet the month’s production quota were made to specification. The control chart for these new data, and the data from Weeks 1 to 3, is shown in Figure 10. The data used to calculate the limits in Figure 10 are shown (batches 83 to 122). These data represent the process with the improved control of raw material density.


Figure 10: Control chart of product characteristic 16’s data from Weeks 1 to 4. The limits are calculated using the Week 3 data minus the first two points.

The limits from the Week 3 data are appropriately extended into Week 4 to assess if the process remained predictable. This wasn’t the case because batch 132’s value is above the upper process limit. After some investigation, it was discovered that an operator returning to work after vacation had inadvertently returned the density value for the dosing step to its target value. He hadn’t been informed of the new procedure. In the production of batches 131 and 132, the target density value for the dosing step had been used rather than the material’s measured value. The operator had been updated in the course of his shift and things got back on track. All production shift leaders agreed to recheck that all operators in their teams had understood and were applying the change in operating procedure. Having accounted for the control chart signal in Week 4’s data, Beth was able to communicate that the process had otherwise been predictable in Week 4 when operated as per the new control procedure.

Plant 11, where “in control” means “predictable”: The technical unit’s monthly meeting

The plant manager’s message was based on the summary below, with the savings opportunity related to product characteristic 16.

If asked to give detail to the savings, the plant manager was prepared with a simplified version of Figure 11. An improvement at the dosing step had been identified, enabling a lower, and more optimal, process average for characteristic 16. This improvement enables a more optimal recipe formulation to decrease the cost of materials per unit of product manufactured. (Note: The technology group had confirmed that a lowering of the target would not contravene legislation nor customer expectations of the product.)


Figure 11: Representation of the improvement opportunity in Plant 11.

With Plant 11 allowing for 1-sigma of “safety space” to buffer any unexpected process changes, a reduction in the average of ~0.44 units of characteristic 16 is possible: Current target-potential target = 66.0−65.56 = 0.44.

The cost-saving calculation is shown below. Plant 11 was urged to go ahead with this improvement.

When asked by a young member of the technical unit’s team why his plant uses control charts, the manager of Plant 11 responded simply that he’d learned a long time ago that a predictable process is a fantastic enabler of knowledge development and better quality at lower cost.

As expected, the meeting went well. The plant manager went home content.

Plant 7: How could the control chart have helped?

Even though Plant 7 didn’t use a control chart, we can. Figure 12 shows a control chart of the month’s data next to a histogram on which the natural process limits are found as well as the specification limits and the process target value.


Figure 12: Control chart of Plant 7’s data over Weeks 1 to 4 and accompanying histogram. (All 181 data are used to calculate the natural process limits.)

What can we learn from Figure 12?

Control chart: The process displayed predictable behavior over the course of the month, i.e., no signals of change in the process.

Histogram: The range of expected output from this predictable process is wider than the specification range; in other words, the process is not capable of routinely meeting the specifications for product characteristic 16.

The best estimate of the out-of-specification rate for this predictable but incapable process is just under 3%:


While Plant 7 appeared to be surprised when the out-of-spec batches were produced, we learn from Figure 12 that the occasional out-of-spec batch was to be expected. Listening to the voice of the process, the question was when, and not if, an out-of-spec batch would be produced. Why? Because the lower natural process limit is clearly below the lower specification limit.

A second lesson from Figure 12’s control chart is that Plant 7 took a wasteful path with the “root cause analysis” of specific batches. This path was ineffective because there was nothing unique, or special, about the out-of-spec batches (see Plant 11’s guidance in Figure 2). To eliminate the occurrence of out-of-spec product, a better way would have been to study all the batches as a precursor to some fundamental change in the process.

Moving forward a couple of weeks into the next month, an engineer in Plant 7 discovered that the mixing time had been shortened to meet the line’s specified hourly throughput. It was in this process step that the key raw material for characteristic 16 was added.

Further investigation revealed that the batch mixing time had been shortened to 60% of its original setting. Hypothesizing that a too-short mixing time could be a factor in the excessive variation seen in Plant 7’s data, past records were reviewed and some trials run with more extensive within-batch sampling. These data showed reduced variation with increased mixing time. The outcome was to lengthen the mixing time to its original setting, with the variation in the product measurements clearly reduced.

Following this fundamental process change, Plant 7 had a good run of producing to specification. The hourly throughput target was also revised to properly reflect the time needed for the mixing step. The change to the mixing time at least meant that the plant manager had a better time of it in the following monthly meetings.

Summary

We asked if control charts are still relevant in today’s digital world. The answer is a resounding yes, with the continued relevance of control charts summarized below, guided by the key words found at the start of this article.

Many manage their processes without control charts. If you are one of these, why not give this almost prehistoric statistical technique a chance?

Our next article will discuss the use of SPC with high-speed data collection systems. If you are drowning in data, like many are these days, this article might be a game changer.

Postscript

Gaining acceptance to the use of control charts

The usefulness of control charts has been thoroughly proven over the years. This doesn’t mean, however, that they are well understood, that their use is widespread, that detractors are few, or that gaining acceptance to their use is easy. A good discussion of the kind of resistance control charts can meet, and how to overcome this resistance, is in “Ken the Grithead—An Improvement Story.”

The word control

The word control in itself is a potential obstacle to successful control chart usage because “in control” is commonly understood to mean product is in spec and/or things are OK. This is something completely different from what Shewhart meant when he created the control chart.

As was the case in Plant 11, some have advocated different terminology to facilitate better understanding. Sophronia Ward proposed, and uses, the name process behavior chart instead of control chart. David Chambers proposed and used the name natural process limits instead of control limits (for individual values). Donald J. Wheeler has popularized the two, and other terms such as predictable process, over the last two decades. See “The New Terminology” by Wheeler for a more detailed discussion on terminology intended to overcome the obstacles caused by the word control.

Human input in the digital era: Making sense of the signals

Changes in a process do not always occur at the same point in time as signals on the control chart. Judging when a change most likely occurred requires human judgement. The ensuing investigation to identify the cause of the change also requires human input. The key to improvement in Plant 11 was Beth’s interpretation of Figure 3.

If Japan Can, Why Can’t We?

To see how control charts and “in control processes” can help to improve quality and productivity, take a look at the last 16 minutes of the NBC documentary If Japan Can, Why Can’t We?

Please share your thoughts on this topic: Do you agree, or disagree, with our ideas? Are you practicing the ideas discussed above?

Discuss

About The Authors

Scott A. Hindle’s picture

Scott A. Hindle

Scott A. Hindle has been using data to study and improve processes, and actively working in the field of SPC, for close to 15 years.

Douglas C. Fair’s picture

Douglas C. Fair

A quality professional with more than 35 years of experience in manufacturing, analytics, and statistical applications, Douglas C. Fair is the former chief operating officer at InfinityQS International, an SPC software company. At InfinityQS, he spent 25 years helping manufacturers around the world deploy SPC and benefit from its use. 

Fair holds a bachelor’s degree in industrial statistics from the University of Tennessee, and a Six Sigma Black Belt from the University of Wisconsin. He’s a regular contributor to various quality magazines and has co-authored two books on industrial statistics: Innovative Control Charting (ASQ Quality Press, 1998) and Quality Management in Health Care (Jones and Bartlett Publishing, 2004).

Comments

Control charts

Great article! Thank you,

PB

Pierre's comment

Thanks!

almost always

Great article - I wish I had wrote it first! I see these same tampering and missed signal errors with almost every client.  They are bent on finding an easy quick automated way to do this - but mis the learnings that you share. Great job. Shehart and Deming, I think, would agree. Leaders put dots on paper....

Still a place for the 'old'

Thanks, Tim, for the note and for confirming you still see a valued place for this under-used approach to working with data and processes. Hope too you see similar value in the articles we still have to come in this series.

SPC - Smart Performance Charts

Shewhart charts have been around for 100 years. Why aren't they used everywhere? Based on my analysis of 1,000s and of improvement posters over the last eight years, it's because Excel line and bar charts are too available. They still represent 75-85% of improvement poster charts.

But control charts are so useful, everywhere. Ryan on my staff uses them to track email delivery rates and PPC (Pay Per Click) performance to save money and increase conversions and open rates. Nurses use them to track patient falls and medication errors. They're not just for manufacturing (a common misconception/excuse).

As you point out, Shewhart called them "Control" Charts. And no one likes to be controlled. Dr. Wheeler calls them process behaviour charts. 

I know it's almost impossible to change jargon, but I got to thinking: Everyone wants to improve their performance--golf scores, walking/running steps, etc. What if we called them "performance" charts?

And since they went to college and took statistics, why not "Smart Performance Charts" or SPC?

When I learned SPC 30 years ago, we did charts by hand. Frankly, I knew it would be next to impossible to get people in the phone company to use them. Now it can be done easily and accurately with affordable SPC software. Automation to the rescue.

We have to stop admiring the problem, buy some software and start improving performance across every industry. Let's create hassle-free healthcare, a hassle-free world.

"Excel Line and Bar Charts

"Excel Line and Bar Charts are too available"... 

This is true, but there is an increasing trend in usage of the JMP software package (www.jmp.com). 

JMP is primarly a desktop tool but there is also a private/public cloud version as well (see demo at public.jmp.com)

In fact, the authors used JMP to produce the charts we see in Parts 2,3 and 4 of this article! 

SPC charts

Thanks for the comments and observations. Great you are also an advocate. We didn’t use the word “smart” but this word is inherent to this article and how it proposes to use data and people’s time and energy.

Our focus in this series to date is manufacturing but we know great uses exist outside this narrow focus area. We might squeeze an article in on a different topic before the series reaches its end.

Control Charts: Are they still relevant in manufacturing Part 2.

G'day Scott and Douglas,

Yes indeed, they ares still used. But as mentioned in Part 1 graphic / image used, the Shewhart Control Chart and at Tokai Rikka have the U/LCL in dotted lines, never full lines as the lines show very well and your process behavior exapmples, it varies.

Secondly and kind of worse is placing the U/LSL lines (Full of course) on the Control Chart. As Shewhart, Deming, Ishikawa, Juran, Wheeler, AIAG "SPC" text told and tell us, Control Charts Economic CL - dotted; Histogram U/LSL - Full.

How the process behavs then and subsequent capabilities can be so determined meaning Stability before Capability.

My joy was seeing little Ishikawa Basic C&E Diagrams on the Clayton Melbourne Australia, Opama Japan and Smyna Tenn USA NIssan Plants and vehicle Control Chart "Outliers" and then again in the Toyota Hi-Ace plants, using CC and HIstograms as I stated.

When the process was really unstable they showed me very well, Dr IShikawa's 3rd Type of Cause Analysis (Introduction to QC ~1962) the Process Classification Cause & Effect Diagram. As all work happens through a process, it is very useful.

Another joy is seeing CC used in banks, insurance companies and in defense procurements / supply charts. As Mr Toyoda and his engineers told the Cadilac VP after they toured the robot Cadliac line and she presented and told us the story at the Quality Conference in Canada. He queried her on solving problems and so few people to solve them, plus so many robotocs etc. He said that if you do not control your processes, with your people, 'I fear you will only make what you do, badly, quicker'. GM had to write-off so much of their then Captial Investment. 

Other uses of SPC

Hi Mike, Doug and I know that control charts have great potential outside of manufacturing - thanks for reinforcing this in your comments.

A huge risk for control charts and their potential to add value to Operations isn't whether the control chart limit lines are dotted or not but whether they are confused with spec limits - the old story of confusion around the voice of the customer and the voice of the process.