Case-Based Diagnosis in the principal component space. Application to injection

In this paper, a Case-Base approach for fault detection and diagnosis of faulty sensors in an injection moulding process is presented. First, a statistical model of the Normal Operating Conditions (NOC) is built using the Multiway Principal Component Analysis (MPCA), which grants dimensionality reduction while keeping the most significant information. After that, a Case-Based Reasoning (CBR) approach is applied over the MPCA space for fault detection and diagnosis purposes. This combination guarantees that the CBR attributes will be uncorrelated, which is an important requirement. Results obtained show that the methodology reduced the number of false alarms, while no missed detections were added to the process. Additionally, it has been found that determining the fault typology occurring in the process presented a best performance. Finally, this combination has led to a better process understanding without a deep a priori knowledge of the process being studied.