Progressive (die) stamping is a widely used sheet metal forming process in the manufacturing of automobiles and consumer electronics. Current literature suggests limited knowledge of online health monitoring and fault detection capability. This paper aims to propose a data-driven, empirical approach for diagnosing potential abnormal health conditions of the progressive stamping machine. The proposed signal features and selected novelty detection algorithm work together for fault diagnosis and compensate for the lack of theoretical knowledge of this high-volume high-speed production process. Novelty detection, usually approached within the framework of one-class classification, demonstrates luminous potential applications in machine system health monitoring. During progressive stamping process, there are abundant normal operating data while it is unclear how complex machine health factors together affect the feature signals. The task is to recognize the abnormality by recognizing the significance of how testing data differ from training data by forming a decision boundary from normal behaving training data only. Through literature review, One-Class Support Vector Machine (OCSVM) is selected as the main algorithm for this application. Through applying kernel tricks, the hyperplane can be used to classify nonlinear machine health indictors. OCSVM also utilizes sparse supporting vectors to form a computationally efficient decision boundary and can therefore be used for online fault detections. In this study, the proposed novelty detection approach is explained and successfully implemented based on industrial data from a progressive stamping machine. The paper discusses the promising results and potential limitations.

This content is only available via PDF.
You do not currently have access to this content.