Neural Net Monitoring of the Non-Stationary Mode of an Asynchronous Motor

Artificial neural networks are increasingly used in electrical engineering. Once trained, artificial neural networks can be implemented in microprocessor equipment and used for real-time computing. Testing such algorithms on experimental data, when the structure of neural networks is selected empirically, is an urgent task. The article presents the results of real-time neural-network processing of current and voltage signals in the windings of an asynchronous motor when the power supply is turned on and off during one period of industrial frequency. In laboratory tests, the asynchronous motor was switched on through a thyristor regulator that provided power switching on and off with adjustable duty cycle (initial phase). Thus, a change in the frequency and amplitude of current and voltage was ensured during the time interval in a fraction of the supply-voltage period. In addition, signals in the engine rundown mode were analyzed. According to the results of local approximation of the signals, the frequency range was 30–50 Hz. To monitor signals in a sliding time window, the size of which did not exceed units of milliseconds, several neural networks of direct propagation were entered into the microprocessor equipment. Each of the neural networks solved a separate task: determining the local frequency, phase, amplitude, and switching moment. It is shown that the proposed approach makes it possible to control the signal frequency with an error of no more than a fraction of a hertz, while neural networks of different structures make it possible to complement each other’s results and increase the accuracy of determining signal parameters.