Mahalanobis-Taguchi System for Multiclass Classification of Steel Plates Fault

Abstract

Fault identification is fundamental to condition monitoring. An identification method for a single fault is unbalanced as there are usually multiple possible failures involved when considering a system. This paper presents a method for applying the Mahalanobis-Taguchi system (MTS) in a multiclass problem space. MTS provides a means of extracting information in a multidimensional system and integrating information from different variables into a single composite metric. MTS is used to construct reference scales by creating individual measurement scales for each class. These measurement scales are based on the Mahalanobis distance (MD) for each sample. Orthogonal arrays (OA) and signal-to-noise (SN) ratio are used to identify variables of importance and these variables are used to construct a reduced model of the measurement scale. By reducing the dimensionality of the problem, less variables are tracked which reduces the cost of the system monitoring. A classification threshold based on 1.5 sigma shift from the centre of the measurement scales was utilised for each class. In order to evaluate the effectiveness of the method presented, a case on multiple fault class of manufacturing a steel plate is studied, and results indicate the practicality of the method in industrial applications.