February 21, 2019
Careful consideration of some fundamental aspects of Statistical Process Control (SPC) can go a long way toward determining whether or not manufacturers are able to effectively prevent problems and control their production processes.
When applied properly, SPC identifies significant changes to a process. These can be changes that are still within specification—but are statistically different than where the process was previously running. By identifying the changes, personnel can identify what caused the change and potentially improve the process or prevent the production of inferior products.
1) What is the Purpose of SPC?
A misunderstanding of the purpose of SPC will frequently derail well-intentioned efforts to implement it. Some manufacturing personnel falsely believe that SPC charts indicate whether products will meet specifications. Rather, SPC charts tell us when the system has changed, so we can quickly identify the causes and potentially prevent an issue or make an improvement. (The appropriate method to assess whether the products will meet specification consistently is Process Capability Analysis, provided that the process is stable).
2) What is the Relationship between Control Limits and Specification Limits?
Control limits represent the range over which we expect statistics (such as process averages or standard deviations) to fall. When these process statistics fall outside the control limits, we conclude that a process change has occurred. On the other hand, specification limits are limits of conformance pertaining to individual values such as a part dimension or material properties. Control limits should only be determined by actual process data whereas specification limits are determined by an engineer, designer, or customer. Manually “assigning” control limits should only be permitted when based upon the process data, otherwise the control chart will result in inappropriate chart signals or lack of appropriate chart signals. Finally, specification limits do not belong on control charts.
3) Does my Sampling Scheme Provide Rational Samples?
Traditional control charts (e.g. Xbar / R or Xbar / S) assume that rational samples are obtained. In short, this means that the measurements within the sample come from a single distribution and the within-sample variation is indicative of the common cause variation (between sample variation) in the system. However, many types of production processes do not lend themselves to rational sampling, such as sampling from multiple cavities or filling heads where a statistical difference may exist between the locations. Often, continuous and batch processes also fail to produce rational samples for traditional SPC. Before attempting to use traditional control charts, rational samples must be verified.
4) What Types of Control Charts do I Need to Control my Process?
Many manufacturing professionals are only familiar with the control charts of the 1920’s, the popular Xbar / R charts and X / MR charts. These charts were invented when production systems were far more primitive, but often, these charts are inappropriate at managing and identifying changes that occur in modern day manufacturing systems. There is a large variety of SPC charts available for efficiently monitoring complex equipment with multiple cavities, spindles, or filling heads, and handling special issues such as tool wear, or detecting small shifts with individual measurements. In many situations relying on traditional charts will mask the process changes we aim to detect.
Click for more info: How to choose a control chart for my application
5) What Sample Size is Appropriate?
As with any statistical method, the sample size has a significant impact on the performance of SPC charts. Essentially, small sample sizes (such as Individuals charts where n=1) lack the power to detect small process changes. On the other hand, excessively large sample sizes will result in control charts that frequently indicate instability even when the process change is not of practical concern. Since the purpose of control charts are to notify us promptly when a practically significant process change has occurred, it’s important to determine a sample size that provides statistical signals when they are desired and minimizes unnecessary signals.
6) Do we Still Need Range Charts?
The use of Range charts to monitor process variation dates back to the 1920’s when SPC first came into use and simple calculations were of high practical value. However, the continued reliance on Range charts today is unwise. The problem with ranges is that they discard most of the data and rely on the least representative data points. For example, if a sample of 10 measurements is taken, the range only uses two of the 10 values—and it ignores the other 8 measurements! The two it does use are the most subject to variation and error. While some practitioners only replace the Range chart with Standard Deviation charts for large sample sizes, it’s really time (2013!) to use the best estimator of variation (Standard Deviation) because, with the advent of computers, the need for calculation simplification has ceased.
7) When is Normality of the Data a Concern?
Traditional charts of averages (Xbar charts) from a stable process follow a predictable pattern due to the Central Limit Theorem. Thus, control limits may be computed using a standard approach since the resulting limits estimate the expected variation in the plotted averages, provided the process remains stable. However, when charts of individuals measurements are utilized, many practitioners and professionals believe that it is not necessary to account for any natural non-normality (such as skewness) in the data when computing control limits. Unfortunately, most manufacturing data is not bell-shaped and control limits (limits of expected variation) ought to account for what is expected from the process. Otherwise, we will routinely mis-assign common cause variation for special cause variation. The erroneous assumption that data always follows a normal distribution also leads to highly erroneous process capability estimates.
8) What do we do When the Control Chart Identifies a Change has Occurred?
When a control chart identifies a change, an appropriate response is for personnel to determine the causes of the change. Instead, when a control chart shows a change, operators or supervisors often try to make an adjustment—without knowing what to adjust or by how much to adjust. Taking such action without first identifying the cause of the change can actually increase the variation. If personnel resist knee-jerk responses and investigate the source of the change, then something has been learned—and random (and potentially harmful) adjustments can be avoided. To promote this action, operators cannot be blamed when processes become unstable!
9) What Characteristics Should I Monitor on Control Charts?
Plants often monitor characteristics that are not important and neglect to monitor key parameters that are important. The reason is that they do not always understand which inputs and characteristics affect the product most and should be controlled. To enhance or verify existing engineering and manufacturing knowledge, efficient and effective methods such as Design of Experiments, ANOVA, and Regression may be used to help determine which characteristics should be controlled and to specify the required degree of control. This in turn will enable an appropriate sample size to be utilized to control the process most efficiently while minimizing risks.
10) Are Adequate Measurement Systems in Place to Support SPC?
Since the purpose of SPC is to detect process changes, it is imperative that our measurement systems are capable of measuring accurately and precisely. Frequently, measurement systems are utilized without verifying their adequacy and we end up reacting to “noise” or not reacting when we should due to excessive measurement error. Measurement Studies to assess precision (Gage Repeatability and Reproducibility) should be performed before utilizing the data for SPC or other purposes. Accuracy and Linearity studies are also important. If measurement deficiencies are present, they should be addressed before trusting methods that use the data.
This article only suggests a few additional benefits of SPC data, but there are many others. Hopefully, readers will think about taking full advantage of SPC data to prevent problems, quickly resolve issues, and fully understand operations so that their companies satisfy customers while realizing maximum profitability.
Steven Wachs, Principal Statistician
Allise Wachs, Ph.D.
Integral Concepts, Inc.
Integral Concepts provides consulting services and training in the application of quantitative methods to understand, predict, and optimize product designs, manufacturing operations, and product reliability.
About DataNet Quality Systems
DataNet Quality Systems empowers manufacturers to improve products, processes, and profitability through real-time statistical software solutions. The company’s vision is to deliver trusted and capable technology solutions that allow manufacturers to create the highest quality product for the lowest possible cost. DataNet’s flagship product, WinSPC, provides statistical decision-making at the point of production and delivers real-time, actionable information to where it is needed most. With over 2500 customers worldwide and distributors across the globe, DataNet is dedicated to delivering a high level of customer service and support, shop-floor expertise, and training in the areas of Continuous Improvement, Six Sigma, and Lean Manufacturing services.