Fault Diagnosis in Industries: How to Improve the Health Assessment of Rotating Machinery

Rotating machinery like pumps, compressors, and engines, are widespread in industries. A failure in one or more of their components, e.g., rolling bearings and gearboxes, may cause system breakdowns or even catastrophic events. Hence, fault diagnosis of rotating machinery is achieving great attention from both the industrial and academic fields. After collecting signals from sensors mounted on different locations, different condition monitoring techniques, e.g., vibration analysis and acoustic emissions, are usually applied to get accurate information on machinery health conditions. The most challenging step in the fault diagnosis process is feature extraction from collected signals, which describe the fault behavior synthetically while retaining as much information as possible. In intelligent fault diagnosis, the extracted features feed Machine Learning algorithms for fault classification. However, given the high number of variables that result from feature extraction, dimensionality reduction should be performed to improve the classification accuracy and training times of intelligent methods. In the literature, the problem of dimensionality reduction is faced through feature selection and feature fusion. These methods are often combined in different ways, resulting in a broad range of possible solutions. In this paper, a bibliometric analysis is conducted to discover trends over time and identify the connections between the different activities of the fault diagnosis process using dimensionality reduction techniques. The present study's main result lies in identifying the most promising frameworks for industrial applications of fault diagnosis using dimensionality reduction.

Automated summary

This paper conducts a bibliometric analysis to study the application of dimensionality reduction techniques in fault diagnosis. Fault diagnosis is a focus of attention in both the industrial and academic fields, and feature extraction from collected signals is the most challenging step in the process, where dimensionality reduction can improve classification accuracy and training time.