FEATURE SELECTION ALGORITHM TO IMPROVE THE EFFICIENCY OF SPACE TECHNIC ELEMENTS DIAGNOSTICS

Authors proposed a hybrid feature selection algorithm to improve the quality of diagnostics of the electrical part of a permanent magnet synchronous motor using machine learning methods. The developed algorithm has been tested on data obtained as a result of studies of a laboratory sample of a permanent magnet synchronous motor with a load from a pump unit. Comparative estimates of the accuracy and performance of machine learning methods obtained, which make it possible to implement the presented algorithms in embedded on-board diagnostic systems. The diagnostic parameters and features most sensitive to electrical faults of a permanent magnet synchronous motor have been determined. The influence of correlated features and features of low significance on the quality of diagnostics was determined. The extraction of key features made it possible to increase the accuracy and speed of diagnostics while reducing the size of the training sample, memory and computing resources when using several machine learning methods.