Statistical process control (SPC) is a method of monitoring and improving the quality and performance of a process by using statistical tools and techniques. SPC can help you detect and prevent problems, reduce variation, and increase efficiency. But how can you ensure that your SPC process is statistically valid? Here are some tips to follow.
Control charts are graphs that display the variation in a process over time and compare it to the expected or desired limits. They can help you identify common and special causes of variation, as well as trends, shifts, and cycles. However, not all control charts are suitable for every type of data or process. You need to choose the right control chart based on the nature and distribution of your data, the number and frequency of samples, and the purpose of your analysis. For example, if you have continuous data that are normally distributed, you can use an X-bar and R chart or an X-bar and S chart. If you have discrete data that are binomial or Poisson distributed, you can use a p-chart or a c-chart.
Other common charts include the X and Moving Range chart for continuous individuals data and the U chart for Poisson distribution with variable denominator. Also consider special charts for rare events, such as the G chart, which is common in healthcare.
Before you apply any SPC technique, you need to make sure that your data are reliable, accurate, and representative of your process. You can validate your data by checking for errors, outliers, missing values, and inconsistencies. You can also use descriptive statistics, such as mean, median, mode, standard deviation, and range, to summarize and explore your data. Additionally, you can use graphical methods, such as histograms, boxplots, scatterplots, and normal probability plots, to visualize and assess the distribution and variation of your data. If your data do not meet the assumptions or requirements of your chosen control chart, you may need to transform, standardize, or stratify your data.
Control limits are the boundaries that define the acceptable range of variation in a process. They are usually calculated from the historical or baseline data of the process, using statistical formulas that depend on the type of control chart. Control limits are different from specification limits, which are the requirements or expectations of the customer or the product. Control limits reflect the inherent variation in the process, while specification limits reflect the desired variation in the outcome. You should establish your control limits based on the data, not on the specifications. However, you can compare your control limits to your specification limits to evaluate the capability and performance of your process.
Once you have your control chart set up, you can start monitoring and analyzing your process by plotting data points and looking for any patterns or signals that indicate your process is out of control or unstable. Common signals include a point outside the control limits, a run of seven or more points on one side of the center line, a run of seven or more points that are increasing or decreasing, a run of 14 or more points that alternate up and down, two out of three points beyond two-thirds of the distance from the center line to the control limit, four out of five points beyond one-third of the distance from the center line to the control limit, and eight out of eight points not within one standard deviation of the center line. If any of these signals are detected, you should investigate potential causes and take corrective actions to bring your process back to control. Additionally, it is important to document and record your findings and actions.
There are different sets of special cause rules, with Western Electric and Nelson being common options. In choosing which set of rules to use, balance detection power with false positive rate. Wheeler has a practical perspective on this topic. I’m addition to checking for special cause signals in your baseline data as you set up your chart, it’s also important to give consideration to sample size within and between subgroups. Attribute charts can have too many values at the extremes or bounded control limits, which indicates the need to adjust the rational subgrouping strategy or choose a more appropriate measure and/or chart.
SPC is not a one-time activity, but a continuous improvement process. You should regularly review and update your control chart and your process based on the changes in your data, your environment, and your goals. You should also evaluate the effectiveness and efficiency of your SPC process by measuring and monitoring the key performance indicators (KPIs) of your process, such as defect rate, cycle time, yield, and customer satisfaction. You should use the feedback and data from your SPC process to identify and implement opportunities for improvement and innovation.
SPC is a powerful and proven method of ensuring quality and excellence in any process. By following these tips, you can ensure that your SPC process is statistically valid and robust.
This section was obviously not written by an expert. It’s all fluff. Review your control chart after you test each improvement intervention to assess the effectiveness. If your process is stable, establish a review process such as weekly, monthly or quarterly, depending on risk levels. If you have a software like SQC Pack, you can automate notifications of special cause on your control charts.
Control charts are an extremely easy predictive analysis tool, once you have experience identifying the type of data/distribution and understanding of the special cause rules. The main problem you will encounter is lack of competency for developing appropriate measures, which requires advanced statistical thinking.