Fault diagnosis of BLDC drive using advanced adaptive network-based fuzzy inference system

Brushless direct current (BLDC) motor is widely used in many applications for its high reliability and simplicity. The motor fault-tolerant control is important for its continuous operation even under the fault condition. The fault diagnosis should be achieved to implement the fault-tolerant control method. In our proposed work, short-circuit (SC) and open-circuit (OC) faults are detected in brushless direct current (BLDC) motor by using advanced adaptive network-based fuzzy inference system (AANFIS). AANFIS is a combination of social ski driver (SSD) and advanced adaptive network-based fuzzy inference system (ANFIS) for optimal feature selection and training process. There is no need to give entire data to the first (training) layer of ANFIS which reduces operational complexity, rapid learning capacity without expert knowledge as well as less training period and also SSD is used for the reduction of time complexity and fast convergence; thus, the reason for combining the ANFIS with SSD named as AANFIS is used for better output. Initially, BLDC drive speed is measured by signal speed oscillation. Discrete wavelet transform is used to extract the residual signal from BLDC. The amplitude and frequency of residual signal denote the open-circuit and short-circuit faults. Optimal feature selection will be performed by SSD in training stage of ANFIS. Fault diagnosis is performed through the layer of ANFIS structure and detected results are the outcomes of ANFIS. The proposed work will be implemented in Matlab/Simulink working platform. Resultant parameters like BLDC drive voltage and current analysis, fault diagnosis of BLDC drive with OC and SC, motor speed, torque and detection accuracy are evaluated. The power loss and accuracy are compared with ANN existing technique. The power loss of the existing ANN is 5.8% and the proposed method of AANFIS is 3% and the accuracy of the existing ANN is 97% and proposed method of AANFIS is 98.5%. The proposed method achieves better performance than the existing technique.