Diagnosing bearing fault location, size, and rotational speed with entropy variables using extreme learning machine

• Published in: Journal of the Brazilian Society of Mechanical Sciences and Engineering Vol. 46; no. 1
• Main Authors: Akcan, Eyyüp, Kuncan, Melih, Kaplan, Kaplan, Kaya, Yılmaz
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 2024; Springer Nature B.V
• Subjects: Artificial neural networks; Bearings; Condition monitoring; Datasets; Engineering; Entropy; Fault location; Feature extraction; Machine learning; Mechanical Engineering; Neural networks; Rotating machinery; Rotating machines; Technical Paper; Vibration analysis; Vibration monitoring; Bearing failure; Vibration signals classification; Entropy variants; ELM
• ISSN: 1678-5878; 1806-3691
• DOI: 10.1007/s40430-023-04567-2

Bearings play a crucial role in transmitting motion in rotating machines and are considered fundamental equipment. Any errors occurring in these machines can lead to a reduction in mobility and complete machine failure if not addressed promptly. Condition monitoring of bearings through the utilization of vibration information is a widely researched and advanced field. Analyzing irregularities in vibration data using entropy methods enables the extraction of valuable information that characterizes the health status of bearings. In accordance with this purpose, vibration signals were collected from artificially defective bearings in special dimensions, using a dedicated experimental test setup. Three different scenarios were considered for evaluating the proposed model performance. Data set 1 encompassed bearing signals collected at various speeds (1500, 1740, 1800, 1860, and 2100 RPM). Data set 2 consisted of vibration signals using different fault location (ball, inner, and outer ring faults), while data set 3 comprised bearing vibration signals with faults of varying sizes (0.15 cm, 0.5 cm, 0.9 cm) under the same speed. For feature extraction from bearing vibration signals, 18 distinct entropy methods were employed in all experiments. The extracted entropy features were utilized as inputs for the extreme learning machine (ELM) model. ELM offers a fast and efficient approach for training neural networks, making it a valuable tool in various machine learning applications. The experiment conducted using all features achieved an accuracy rate ranging from 98.48% to 100%. To assess the individual effectiveness of entropy features, separate trials were conducted for each feature. Fuzzy entropy demonstrated the highest success rates in data sets 1 and 2, while the slope entropy feature exhibited superior performance in data set 3. The proposed approach has been compared with relevant studies in the literature, and its significant results have been duly acknowledged. This comparison further affirms the efficacy of the proposed approach and highlights its potential contribution to the field.