A Knowledge-Based System for the Detection and Diagnosis of Out-of-Control Events in Manufacturing Processes

In this paper, we discuss a system which combines statistical process control principles and knowledge of the process to automatically arrive at a comprehensive detection and diagnosis of out-of-control conditions in a manufacturing process. This approach consists of capturing data from the process and passing selected signals from it through a two level decision-making system. The first level of this system employs nonlinear filtering techniques to detect three features (peaks, steps, and ramps) of the input signals. These features are examined to produce a set of out-of-control events. The second level of the process applies a ruleset to each event using a backward chaining algorithm to attempt to diagnose a process cause that led to the event. Status reports of diagnosed and undiagnosed events are generated by the system. A detailed description of the entire system and some discussion of its use in an actual aluminum rolling mill will be presented.