What is manufacturing process stabilization?

 The characteristics of a "stable" process

"Keeping the production process stable" is one of the most important tasks in quality management, but what kind of production process is stable? The understanding of those who have not studied statistical process control is that a process is stable if it consistently produces acceptable products, which is a misconception.

From a process control point of view, we need to strictly control all aspects of the production process, including man, machine, material, method, environment, and measurement, and analyze, improve, and monitor the major sources of variation. Whenever possible, we also need to evaluate whether the process is truly stable in terms of the variation of key product characteristics of the process output over time.

As defined by ISO 22514-1, a stable process is one that is affected only by random factors and can be considered to be under statistical control.

The following points must be clarified:

• Stable process, the output distribution of eigenvalues is not necessarily normally distributed, nor can it be assumed that a process whose eigenvalue output is normally distributed is necessarily stable, and whether the process is stable or not has no relationship with whether the output eigenvalues are normally distributed or not, as shown in the figure below, as defined by ISO 22514-1, both A1 and A2 models are stable processes, while the C1 model, even if the output distribution is normally distributed, is also an unstable process;
• A stable process does not mean that the dispersion of the output distribution is large or small, nor does it mean that the center of the process must be on the target for the process to be stable, it only means that statistical tools can be used to predict the variance.
• For some processes, the mean of the output distribution is subject to systematic changes over time, such as tool wear, such processes are unstable according to the definition of the international standard ISO 22514-1. (e.g. c3,c4 models, even if the products are within the specification limits, the process is unstable).

How do you determine if a process is "stable"?

According to the AIAG SPC manual, the stability of a process can be determined using a control chart, which lists 8 commonly used rules for determining control chart variances. These rules for determining that a process is out of control are based on a normal distribution model, i.e., the A1 model mentioned in "How to Choose a Distribution Model in Process Capability Studies? The A1 model is mentioned in "How to Rationally Choose a Distribution Model in Process Capability Studies?

However, these rules are inappropriate for engineers analyzing control charts after the fact, because if the amount of data is large enough, the stability rules will certainly be violated. (This is because the control limits are determined based on a 99.73% probability of non-intervention, i.e., on average, even if the process is an A1 model, there will be about 27 points where the control limits are exceeded when 10,000 data sets are extracted).

So what are we to make of the use of these discriminant rules? These rules are based on the rule that "small probability events do not occur, and if a small probability event is observed, the process is considered to have changed". Therefore, the use of these rules is only recommended at the operator's site, and the use of these rules is based on the risk of omissions and false alarms that we can afford to decide.

The AIAG APC manual does not provide a clear method for determining the stability of the process for the engineer to subsequently analyze.

The definition of "process under control" is given in DIN 55350, which states that a process is under statistical control if the values of the key product characteristics are under control (the same definition is given in ISO 22514-1). A statistically controlled process is one in which the parameters of the eigenvalue distribution remain nearly constant, or change only in a known manner, or do not exceed a known limit, as shown in the figure below. Even if the parameters of the eigenvalue distribution change systematically over time, the process can be considered "controlled" if the changes are predictable and the control is more from the operator's point of view.

So how do you determine after the fact whether the process is "stable" or not? In the Q-DAS software solution it is recommended to look at the following two things:

1. Whether a single value exceeds the specification limits;
2. The number of points on the control chart that exceed the control limits of the process without exceeding the limits of the random discrete bands of the binomial distribution. (For example, for 10,000 subgroups, the normal number of points exceeding the control limits is 27, with a 95% probability of being between 17 and 38).