The focus of this PhD project is to engage in the condition-based monitoring (CBM) of synchronous machines used in automotive applications. CBM is carried out for the machine by monitoring its electrical signals, e.g., current, voltage, etc., to detect electrical and mechanical faults occurring in the machine bearings, stator, and inverter. The developed fault diagnosis functions are mounted on the electrical machine. The advantage of having these functions onboard is that the machine condition can be evaluated at all times, even during operations. Secondly, the economic losses that result from the downtime of vehicles can be reduced by planning the maintenance of the machine in advance. The following faults are considered in the five-phase synchronous machine known as the Boost Recuperation Machine or BRM: 1. open-circuit fault in the stator windings; 2. open-circuit fault in the inverter MOSFET; and 3. inner and outer race fault in the bearing. The following signals are used at a 1 kHz sampling rate for fault detection: • E-Drive DC current; • E-Drive DC voltage; • speed; and • torque. The training dataset for deep-learning 1D convolutional neural network was collected from the PLECS simulation of the machine, the experimental data from the measurement of machines on the test bench and the data generated used a generative adversarial network (GAN). The task has been successfully completed with a fault classification accuracy well above 90%. The methodology has been developed for a speed range from 500 rpm to 8000 rpm and a torque range from -20 Nm to 40 Nm.
Russell Sabir (Thu,) studied this question.