Helicopter main rotor fault diagnosis by using GA-and PSO-based classifiers

Abstract

This paper presents an improvement in the recognition of faulty signals, encountered in the case of the Gazelle helicopter's main rotor, using Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) methods. The main focus is on the distinction between faulty and healthy signals and then between the three subclasses of faulty signals, i.e. faulty bearings, joints problem and mechanical loosening. This research work is divided into three parts. The first part approaches the two above-mentioned classes of signals at the same time, and, to this purpose, the Linear Discriminant Analysis (LDA), Non Linear Discriminant Analysis (NLDA) and Back-propagation Neural Network (BPNN) are used. In the second and third part of the paper, GA and PSO are employed for optimizing the hyperplanes and hypersurfaces which separate the above-mentioned classes of signals, as well as the architecture and connection weights of a neural network (NN). Real data are used, which correspond to the vibration signals measured during periodic technical inspections, and are characterized by amplitudes and frequencies typical of the eight highest peaks of the Welch spectrum. The results obtained confirm the validity of the above-mentioned approaches and comparable favorably with those of other multivariate methods. The GA-or PSO-based neural networks diagnosis can therefore be established for helicopter computers so that faults can be detected.