Statistical process control consists of tools for quality control common not only in industrial areas. Statistical process monitoring and control detect process shifts caused by an assignable cause and eventually proposes a corrective action before many nonconforming parts are manufactured. The detection is typically made on process output sampled according to a certain sampling plan. Because of limited sampling and measuring capacities, the number of such samples can be much smaller than the number of produced outputs, which can lead to large delays between the occurrence of the assignable cause and its detection and consequently to additional costs connected with scrapped parts or non-detected faults. Fortunately, besides the measurements made on the sampled process output, in-process data are becoming increasingly available, which can enable to detect of a problem immediately after it appears. With rapid ICT development, much larger process data can be collected and stored with many variables and large sample sizes.

The majority of deployed solutions are however based on univariate methods that require different assumptions to be met. In the new situation, the univariate setting becomes inconvenient and the probability of assumption violation rapidly increases. AI and machine learning principles help to overcome those issues via multivariate approaches enabled by the increasing data availability. The experiment will define a process to be controlled, define and analyze the post-process and eventually also the in-process data, propose AI monitoring methods tailored to the particular situation, implement the methods, and perform their preliminary assessment on available historical data.

This has been presented to interested representatives of selected SMEs, potential overlaps were be discussed and a proposal of potential use-cases were outlined. Not only the benefits of AI process monitoring and control, but also their issues, potential pitfalls, and barriers were be demonstrated and discussed.