In Limits of Statistical Process Control in China, experienced consultant Brad Pritts described his observations over the years. Below is his advice to use statistical tools to improve production processes.
—–
What I do when I work with companies, whether US or Chinese, is the following approach. I learned most of this from books by Dr. Don Wheeler, a statistician and disciple of Deming. I strongly recommend anything written by Wheeler to people who want to use statistics to improve quality. While I am a skeptic about Xbar/R, I am an absolute believer in the value of statistics for process improvement. Wheeler’s books do a good job of explaining variation without unnecessary complexity.
First, determine the capability of your measurement system. This is done using gage repeatability and reproducibility studies, as well as determining bias and linearity. No measurement system is perfect. You need to know that your measurement system is effective relative to the degree of variation your process has, and the tolerances you need to achieve.
One mentor I worked with emphasized the importance of this point by stating emphatically that if you changed a manufacturing process based on unproven measurements, you were guilty of engineering malpractice — tinkering blindly.
Many times it’s necessary to spend quite a bit of work improving the measurement process before tackling the production process itself.
Second, determine whether your process is stable in the short run. Take a sample of, say, 20 or 30 consecutive pieces. Chart the characteristics as a simple run chart. Look for unnatural patterns in the data – trends up or down, sudden shifts, etc. Calculate the short term process capability.
Often this step alone will show you where process improvement is needed. Sometimes it will be very simple things — tightening up loose fittings in a machine, making an adjustment to a fixture. Other times it may be very difficult – for example, it may be necessary to redesign and rebuild a stamping die… or replace a machine with a better technology machine. These may happen but only with major management commitment.
Third, use the knowledge of the process to establish inspection frequencies in the control plan. Highly capable processes may only need to be checked at first piece setup while troublesome processes may need 100% inspection.
Finally, where warranted, do ongoing longer term studies to see what is changing over the long haul — due to machine and tool wear, changes in operators, different batches of raw material, etc.
One amusing project I was involved with was an assembly operation. Simple statistical analysis helped us in an unusual way. This product had a 32 person assembly workforce. Most of the workers had been newly recruited, as this operation was a new job to this company. Turnover was high.
We charted the defect and rework rates and found a striking pattern in defects which were clustered on occasional Mondays. With a bit of discussion we found that the defects resulted from a practice of hiring several new workers at a time, and starting them all on the line on Monday mornings, overtaxing the supervisors’ abilities to train and monitor their work. We changed the hiring practices to bring new people in on a staggered basis. No new people on Monday morning! Tuesday, bring in one new worker on a half-day shift (starting mid-day), then progressing to a full shift on Wednesday; if multiple workers were needed a second new worker could be added Thursday in the same pattern. Defects and customer complaints fell by about two thirds!
While I am proud of my education and experience with complex statistics, the fact is that many of my biggest success stories come from very simple, practical moves like the story above.
Conclusions:
- Reducing variability is a key to improving quality and profit.
- To understand and reduce variability, statistics are a powerful tool.
- Use the simplest statistics first — counts, totals, average measurements, run charts. Sometimes these are all that are needed, and you’ll have a much easier time explaining your results to others.
- Confirm (and if necessary, improve) your measurement processes first. Only then will you be able to effectively measure and improve the actual production process.
- Assess short term and long term process capability, and use this knowledge to set up your inspection program and improvement priorities.
- If ongoing Shewhart style x-bar and r charts are needed, either automate them, or plan on extensive supervisory and worker training, handholding, and management if you expect them to work.
Particularly critical is establishing an environment where workers are willing to honestly report problems. This is difficult in many companies, American or Chinese. If you can’t get to this situation don’t waste your time on SPC.