Machine Learning for Diagnosis of Event Synchronization Faults in Discrete Manufacturing Systems

Abstract

Common in discrete manufacturing, timed event systems often have strict synchronization requirements for healthy operation. Discrete event system methods have been used as mathematical tools to detect known faults, but do not scale well for problems with extensive variability in the normal class. A hybridized discrete event and data-driven method is suggested to supplement fault diagnosis in the case where failure patterns are not known in advance. A unique fault diagnosis framework consisting of signal data from programmable logic controllers, a Timed Petri Net of the normal process behavior, and machine learning algorithms is presented to improve fault diagnosis of timed event systems. Various supervised and unsupervised machine learning algorithms are explored as the methodology is implemented to a case study in semiconductor manufacturing. State-of-the-art classifiers such as artificial neural networks, support vector machines, and random forests are implemented and compared for handling multi-fault diagnosis using programmable logic controller signal data. For unsupervised learning, classifiers based on principal component analysis utilizing major and minor principal components are compared for anomaly detection. The rule-based random forest and extreme random forest classifiers achieve excellent performance with a precision and recall score of 0.96 for multi-fault classification. Additionally, the unsupervised learning approach yields anomaly detection rates of 98% with false alarms under 3% with a training set 99% smaller than the supervised learning classifiers. These results obtained on a real use case are promising to enable prognostic tools in industrial automation systems in the future

Machine Learning in Manufacturing towards Industry 4.0: From ‘For Now’ to ‘Four-Know’

“…The primary goal of unsupervised learning is to identify hidden and interesting patterns in unlabeled data. In terms of principles, there are three types of unsupervised tasks: Dimension Reduction [118,119], Clustering [120], and Association Rules [121]. Many aspects of unsupervised learning can be beneficial in manufacturing applications.…”

“…Principal component analysis (PCA) [118]: The main idea of PCA is to minimize the number of interrelated variables in a dataset while preserving as much of the dataset's inherent variance as possible. A new set of variables, called principal components (PCs), are generated; these are uncorrelated and sorted such that the first few variables retain the majority of the variance included in all of the original variables.…”