Improved Machine Learning for Multiclass Fault Classification in Industrial Processes

Multiclass fault classification in complex processes is challenging due to many classes, nonlinear dynamics, overlapping fault signatures, and expanding fault taxonomies. Traditional machine learning models often struggle in such settings. The goal of this paper is to develop a model-agnostic, extensible framework. The proposed methodology aims to boost any base classifier via optimization, interval-based feature selection, and intelligent binary decomposition. By restructuring a multiclass task into hierarchies of binary subproblems and linking each boundary to automatically selected statistical features, the developed method improves diagnostic accuracy and generalization. Experimental results on a large-scale dataset demonstrate improved performance compared to existing methods, achieving a high accuracy rate. Although the approach increases the computation time, the notable improvements in accuracy make the balance between precision and computation time advantageous for real-world use.