Interaction-based Fault Detection and Classification Using Randomly Masked 1D Convolutional Neural Network

Abstract

Fault detection and classification (FDC) refers to a data analysis technique that classifies normal and abnormal situations. Data for FDC is being collected in various processes such as manufacturing processes, and various algorithms such as machine learning and deep learning are used in FDC. In particular, convolutional neural network (CNN) models have the advantage of automatically extracting data features and are widely used in analysis, which has a limitation that it is difficult to directly reflect the relationship between variables in the model in multivariate time series data. This work proposes a randomly masked convolutional neural network (RM-CNN) model that enhances FDC performance by adding intervariable relationships to a 1D CNN algorithm. The proposed methodology can directly reflect the relationship between variables in the model to improve FDC performance and identify variables that have a lot of influence. The experiments were conducted by comparison with a typical 1D CNN model, and evaluation indicators used Accuracy, Precision, Recall, and F1 score. As a result of the experiment, the proposed methodology performed better than the comparative models.