Research on modular cloud transformation fault diagnosis for mechanical equipment considering the simultaneous long-tailed and high-dimensional factors.

Nowadays, the joint progress of artificial intelligence technology, sensor technology, and data analysis technology has made it possible to perform large-scale and complex mechanical equipment health monitoring and fault diagnosis. Nevertheless, equipment monitoring data often exhibits typical characteristics of industrial big data, such as large data volume, high dimensionality, and sparse high-value samples. To address the issue of the curse of dimensionality and long-tailed distribution in fault diagnosis, this study proposes a novel intelligent modular fault diagnosis framework named Cloud Augmentation and Cloud Selection (CACS) that utilizes the feature cloud-based transform to achieve fault sample augmentation and feature dimension reduction systematically. Firstly, in the augmentation module of CACS, the original fault dataset is successively input into an improved cloud-based generative module to obtain cloud models that can characterize feature consistency and overall trends. Notably, sample augmentation is achieved through hierarchical sampling of generative cloud droplets. Secondly, in the reduction module of CACS, a symmetric Bidirectional Kullback-Leibler (BKL) divergence is proposed to measure inter-class discrepancy between feature clouds, and feature dimension reduction is achieved by solving the Frobenius norm (F-norm) of the discrepancy matrix. Eventually, the suggested modular method is validated on high-dimensional and long-tailed fault datasets of gears and bearings. Experimental results illustrate the method's excellent diagnostic performance.