Automatic fault diagnosis and location in CSI-fed brushless DC motor drives using neuro-fuzzy systems

The present research involves the development of independent intelligent tools to diagnose and locate different faults of the permanent-magnet brushless DC motor drive. The current source inverter (CSI) fed drive considered in this work is used for electric power steering in small and medium-size vehicles. Adaptive Neuro-Fuzzy Inference Systems (ANFIS) are utilized to automate the diagnostic process by being trained based on indices derived at different healthy and faulty operating conditions.; Performance characteristics of the drive system are obtained under normal operation using a discrete-time lumped-parameter network model. The model is modified to accommodate different types of electrical faults in the system in order to obtain faulty performance. Normal and faulty performances are compared to select one or more characteristic waveforms to be monitored in the time domain. Diagnostic indices are derived from the characteristic waveform(s) in the frequency domain after being processed by a suitable digital signal processing (DSP) algorithm. Candidate DSP techniques include Discrete Fourier Transform (DFT), Short-Time Fourier Transform (STFT), and Continuous Wavelet Transform (CWT). Different characteristic waveforms processed by appropriate DSP tools are used to detect different faults. The research focuses on electrical faults of the drive system including open-circuits across inverter switches, open-phase, insulation failure, and stator winding inter-turn short circuits.; Data sets, which are used to train diagnostic ANFIS, contain diagnostic indices extracted from simulated characteristic waveforms covering wide ranges of operational parameters. Normal operation cases include perfect and imperfect commutations as well as noisy operation in order to prevent ANFIS from issuing false-fault alarms under such conditions. Diagnostic ANFIS are tested within the training range; however, testing points do not explicitly belong to the training data sets. Testing results show acceptable performance of the developed ANFIS in diagnosing the considered faults.; Experimental measurements are taken in the lab under some faults in order to verify modeling results. Good agreement between simulated and measured waveforms is achieved. Implementation of the proposed systems should be straightforward once a fixed-point processor is loaded with the designated DSP and ANFIS-evaluation routines.