Details

Autor(en) / Beteiligte

• Afia, Adel
• Gougam, Fawzi
• Rahmoune, Chemseddine
• Touzout, Walid
• Ouelmokhtar, Hand
• Benazzouz, Djamel

Titel

Gearbox Fault Diagnosis Using REMD, EO and Machine Learning Classifiers

Ist Teil von

• Journal of Vibration Engineering & Technologies, 2024-03, Vol.12 (3), p.4673-4697

Ort / Verlag

Singapore: Springer Nature Singapore

Erscheinungsjahr

2024

Link zum Volltext

Quelle

SpringerLink (Online service)

Beschreibungen/Notizen

• Gearboxes are critical equipment in many industrial applications such as machine manufacturing, petrochemical industry, renewable energy, etc. However, due to their complex structure and regularly harsh working environment, gearboxes are inevitably prone to a variety of faults and defects during operation. Therefore, intelligent condition monitoring techniques are crucially important for early gear and bearing fault recognition and detection to avoid any industrial failure due to machine breakdowns. In this paper, an intelligent algorithm for gear and bearing fault diagnosis is suggested based on several approaches mainly: robust empirical mode decomposition (REMD), time domain features are used for the feature extraction step, while equilibrium optimizer (EO) in the feature selection. For feature classification, random forest (RF), ensemble tree (ET) and nearest neighbors (KNN) are chosen as classifiers. REMD is used to alleviate the mode mixing problem by monitoring the sifting process and selecting the optimal iteration number. EO is a recent optimization approach based on the laws of physical theory in nature. EO reduces the high-dimensional data problem, by filtering redundant features, and increasing model generalization efficiency by avoiding the over-fitting curse. The proposed approach is applied to real-time vibration signals from a healthy gearbox and four different faulty gear and bearing conditions. According to our approach, data signals are decomposed by REMD to several intrinsic mode functions (IMFs). Thereafter, time-domain features are computed for each IMF to construct the feature matrix for every gear and bearing health status. After that, EO is applied to every matrix in the feature selection step. Finally, RF, ET and KNN are used to calculate classification accuracy and give the confusion matrix. Compared to several feature selection techniques, experimental results prove the efficiency of the proposed approach in detecting, identifying, and classifying all gear and bearing defects even under different operating modes.