Adaptive multispace adjustable sparse filtering: A sparse feature learning method for intelligent fault diagnosis of rotating machinery. (April 2023)

• Record Type: Journal Article
• Authors: Zhang, Guowei; Kong, Xianguang; Du, Jingli; Wang, Jinrui; Yang, Shengkang; Ma, Hongbo
• Is Part Of: Engineering applications of artificial intelligence. Volume 120(2023)
• Publication Date: 2023-04
• Subjects: Intelligent fault diagnosis -- Rotating machinery -- Sparse filtering -- Attention mechanism -- Matrix pseudo-norm; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2023.105847
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Fault diagnosis based on artificial intelligence methods is a promising tool to eliminate reliance on a priori knowledge. Sparsity is an increasingly important topic in the field of artificial intelligence in recent years, but the impact of sparsity on intelligence fault diagnosis of rotating machinery is still rarely explored. Therefore, we propose an intelligent fault diagnosis method based on sparse feature learning called adaptive multispace adjustable sparse filtering (AMSASF). Firstly, the multispace sparse filtering is proposed to automatically capture rich and complementary features under multiple spaces by combining four classical sparse measures. Secondly, the attention mechanism is designed to adaptively assign different importance to different sparse spaces to improve the robustness of the algorithm. Finally, the possible effect of sparsity on the inter-class distance is analysed, and the sparsity is adjusted to increase the inter-class distance using matrix pseudo-norm to obtain more discriminative features. Meanwhile, the characteristics of the objective equation based on the matrix pseudo-norm reconstruction for sparse optimisation are discussed. The proposed method was extensively experimented on a self-generated bearing dataset and a public dataset from Case Western Reserve University, with prediction accuracies of 98.32% and 99.67%, respectively, using only 1% training samples. Highlights: The MSSF is proposed to capture features under multiple spaces by combining four classical sparse measures. The attention mechanism is designed to adaptively assign different importance to different sparse spaces. The sparsity is adjusted to increase the inter-class distance using matrix pseudo-norm. The features of the objective equation based on the matrix pseudo-norm reconstruction for sparse optimisation are discussed.