Detection of open switch faults in cascaded multilevel inverters using artificial intelligence techniques

Inverters are power electronic circuits that convert direct current source to alternating current source. With their widespread use in industrial applications such as adjustable speed drive applications, uninterruptible power supplies, renewable energy systems, etc. where variable voltages are required, identifying faults occurring in voltage source inverters has become an important issue for researchers. In this thesis, single and multiple open switch circuit faults in a three-phase five-level voltage source inverter were considered and the phase in which each fault was located and the faulty switch were determined. By simulating the inverter circuit in the MATLAB/Simulink environment, the average and effective values of the output phase currents and the ratios of these values were recorded. A data set was created for classification techniques using these obtained features. The load dependency problem was eliminated by adding the average/effective ratios of the phase currents. Three different classification models such as support vector machines, k-nearest neighbors and long-short term memory were used and the performance of each model was analyzed separately. With the proposed approach, unlike existing studies, fault detection and classification success was achieved in multi-level inverters. Simulation results show that the presented fault detection and classification methods provide high accuracy in predicting single, double and triple switch faults.