A Multi-stage Optimized Fault Diagnosis Model for Imbalanced Fault Data in Manufacturing Process

Abstract

Fault diagnosis of equipment in the manufacturing process is particularly important, and it is an indispensable link in intelligent manufacturing. The fault data generated by the equipment during the manufacturing process is often imbalanced. However, modeling and training on imbalanced dataset will result in a very high rate of misclassification. In order to solve this problem, and improve the accuracy of fault diagnosis in the manufacturing process. This paper proposes a multi-stage optimizatied fault diagnosis model based on bayesian optimization, synthetic minority oversampling technique(SMOTE) and stagewise additive modeling using a multi-class exponential loss function(SAMME), namely the BSS model. The multi-stage optimized fault diagnosis model proposed in this paper improves the diagnosis accuracy of imbalanced fault dataset from the aspects of dataset processing, model building and model training. Finally, the results of ultrasonic flowmeter failure diagnosis show that among the various evaluation indicators, the multi-stage optimized fault diagnosis model proposed in this paper is better than the traditional single machine learning model.