Open-Switch Fault Diagnosis of a Three-Phase Voltage-Source Inverter Using Machine Learning

An IGBT that stops switching (open-switch fault, e.g. a broken gate driver or a lifted bond wire) does not trip a drive at once, but one half-wave of a phase current disappears and the motor runs with a DC current and torque ripple. This project finds the open IGBT(s) of a two-level three-phase voltage-source inverter from the three phase currents: healthy, 6 single and 15 double open-switch faults (22 classes). ALL DATA ARE SIMULATED: a switching-level inverter model written for the project (SPWM, dead time, correct diode logic, R-L and motor-equivalent loads, current-sensor errors) generates 3,850 cases and is cross-checked with the circuit simulator ngspice. The classical normalised average-current (Park-vector) method with a decision table (baseline) is compared with a Random Forest on 36 normalised-current features and a small 1-D CNN on one fundamental period - on operating points that were never seen in training, with sensor noise, and with the diagnosis time after a fault. A GNU Octave cross-check (run), a complete MATLAB version (written, not run) and a web application with five databases are included. Everything runs on a student laptop CPU in about 4-10 minutes.

Category: Electrical and Electronics Engineering (Power Electronics, Electric Drives, Fault Diagnosis, Circuit Simulation, Machine Learning, Deep Learning, ngspice, GNU Octave, MATLAB, Python)

Electrical & Electronics Engineering Project EE07 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.