ESPRIT-TLS and ANN Genetic Algorithm combination for an accurate discrimination of fault bearing in induction machines

Abstract

Fault detection is a strategy that can be easily implemented. Indeed, electrical rotating machines have played and still play an important role in many applications where very often continuity of operation is crucial (aircraft, ships, electric vehicles, industries, etc.). In the case of electromechanical systems, several failures can occur and especially in inaccessible locations (bearing faults, rotor bar breakage, misalignment, eccentricity, cracks, or gear breakage).To ensure acceptable levels of reliability and safety, effective diagnostic methods (at the earliest stage of fault occurrence), fault monitoring, and fault handling are mandatory to avoid any production downtime or loss and to reduce additional repair costs. The detection of these faults by MCSA (Motor Current Signature Analysis) and Principal Component Analysis (PCA) has been widely explored and applied. These techniques are mainly based on the analysis of the stator current by advanced signal processing algorithms to extract useful information for the detection and characterization of defects and their accurate classification. The remarkable limitations of these approaches have prompted researchers to improve their accuracy and to enhance their complexity. In this work, we propose by study the application of ANN-GA (Artificial Neural Networks - Genetic Algorithm) combined with ESPRIT method variants for efficient faults recognizing in real-time. Computer simulations in MATLAB demonstrated that ESPRIT TLS (Estimation of Signal Parameters via Rotational Invariant Techniques Total Least Square) variant allows satisfactory precision in discriminating bearing fault even with a noisy signal. Moreover, this algorithm is suitable for application in dataset preparation and in ANN training for the development of a classification model. According to study finding, Genetic Algorithm optimize ANN architecture for identifying each fault type with very good accuracy in time or frequency domains.