A review of machine learning approaches for high dimensional process monitoring

Traditional control charts have been widely used in industries due to their simplicity. However, these charts are either applied to one quality characteristic at a time or a small number of quality characteristics. Today's manufacturing processes are much more complex, and sensors are embedded throughout the processes that generate a huge amount of data in high dimensions. Traditional control charts are incapable of handling this situation while machine learning techniques are widely known for analyzing high dimensional data sets. Two general approaches reported in the literature incorporate machine learning methods into process monitoring. One approach uses artificial data to populate training data set with various potential out-of-control situations. The other approach adopts feature selection techniques to reduce data dimension. This comparative study aims to review various studies in both approaches. Pros and cons of these approaches are further discussed.