Comparative analysis of fuzzy classifier and ANN with histogram features for defect detection and classification in planetary gearbox

• Published in: Applied soft computing 2021-07, Vol.106, p.107306, Article 107306
• Main Authors: Shaul Hameed, Syed, Muralidharan, V., Ane, Bernadetta Kwintiana
• Format: Article
• Language: English
• Subjects: ANN; Decision tree; Defect detection; Fuzzy classifier; Histogram features
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2021.107306

The planetary gearbox plays a vital role in many heavy-duty power transmission systems. It is essential to monitor such systems for smooth and continuous operations to anticipate machine downtime, production loss and to schedule maintenance. Vibration-based condition monitoring of mechanical systems has been gaining momentum among other methods due to its robustness. Generally, condition monitoring of planetary gearbox forms a classification problem. This study aims to make a comparative analysis between Fuzzy classifier and Artificial Neural Network (ANN) in terms of classification ability with histogram features. Here defect detection in planetary gearbox in different conditions such as gearbox in healthy condition (Healthy), gearbox with sun gear defect (SGD), gearbox with planet gear defect (PGD), gearbox with ring gear defect (RGD) and gearbox with sun and planet gear defect (SPGD) are considered. Analysis is performed using vibration signals acquired for different conditions of the planetary gearbox introducing one-fault-at-a-time. From the acquired vibration signals, histogram features are extracted and using decision tree algorithm the predominant features can be selected. Afterwards, a set of rules are formed from the selected features and given as input to the fuzzy classifier as well as ANN in order to evaluate the classification capability of each method. By running the experiment on 500 samples, the output show that both algorithms are found to have high accuracy in detecting defect and classifying the defect condition in planetary gearbox. •Histogram feature extraction method is proposed for extracting meaningful information from the vibration signals of planetary gearbox.•The vibration signals are analyzed with ANN and Fuzzy Logic algorithms.•The ANN algorithms out-performs with 99.8% classification accuracy compared to Fuzzy classifier.