Due to technological advancements, many manufacturing processes are now real-time monitored through sensors that provide continuous signals of the process parameters rather than providing simpler point observations of the process response. Signals (profiles) obtained through these sensors can reveal important information about the quality of the process being monitored. In this work, we propose a general predictive control framework for on-line process quality monitoring where data is available in the form of a profile. The proposed framework is an integration of ideas from classical on-line process control and advanced machine learning techniques, namely, Random Forests. The proposed framework has the advantages of being more interpretable compared to other methods found in the literature, and has the flexibility to include several commonly used transformations of the signal as features. In addition, abnormal out of control signal characteristics of the process known from experience by operators can be easily incorporated in the random forest technique. An illustration of the proposed framework applied to the case of a tapping manufacturing process is provided. Model comparison results show a superior performance of the proposed framework over previously proposed monitoring methods for the considered tapping process. From a receiver operating characteristic curve analysis, it was found that an area under the curve (AUC) of 0.923 was achieved by the proposed model compared to an AUC of 0.867 for the Generalized Variance model proposed in the literature.
All Science Journal Classification (ASJC) codes
- Strategy and Management
- Management Science and Operations Research
- Industrial and Manufacturing Engineering