Improving quality of the faults’ classification models for electromechanical systems’ diagnostics

Abstract

The research is part of the current work carried out at BSTU «VOENMEH» named after D.F. Ustinov with the finan-cial support of the Ministry of Science and Higher Education of the Russian Federation for the design and creation of high-resource electric pumping units for aviation, transport, and space technology. Recommendations are given on the hardware and algorithmic support of the on-board system for diagnosing the technical condition of spacecraft electromechanical units. Ground testing of the proposed solutions was carried out in the course of experimental studies of an electric pump unit laboratory sample. The advantages of the hybrid feature selection algorithm for improving the accuracy and speed of diagnostics with a feedforward artificial neural network with a significant decrease in the number of input values are shown. The quantities that are sensitive to changes in the state of the electrical parts of electromechanical systems have been determined. Key words Diagnostics, electric motor, machine learning, feature selection, classification.