Induction motors are often used in critical applications such as nuclear plants, aerospace and military applications, where the reliability must be at high standards so three-phase induction motors are the "workhorses" of industry. These motors are exposed to a wide variety of environments and conditions. These factors, coupled with the natural aging process of any machine, make the motor subject to faults which, if undetected, may lead to serious machine failures. From the scrupulous review of the related work it is observed that neuro-fuzzy and neural network based fault detection schemes are performed well for large machines and they are not only expensive but also complex. In this invention RBF-MLP neural network based fault-detection scheme has been developed which overcome the limitations of the present schemes in the sense that, they are costly, applicable for large motors, furthermore many design parameters are requested and especially concerning to long time operating machines, these parameters cannot be available easily. As compared to existing schemes, proposed scheme is simple, accurate, reliable and economical. Most common faults i.e. inter turn short, eccentricity and both simultaneously are selected for demonstration. Simple statistical parameters of stator current which obtained from custom designed 2 HP, three phase 50 Hz induction motor, are calculated and using Principal Component Analysis (PCA), suitable inputs are chosen for network (Dimensionality Reduction). Systematic design procedure of NN based classifier is developed. Finally network is trained and tested rigorously and various performance measures such as MSE, NMSE, MAE, Correlation coefficient and classification accuracy are compared. Designed scheme must suitable for the real world applications hence the network is tested for the robustness to the uniform and Gaussian noise. The invention is further disclosed with the help of figure 1 which shows block diagram of proposed scheme, Figure 2 shows flow chart of design of NN based classifier, Figure3(a) shows Typical scattered plot, Figure 3(b) shows variation of performance measures with number of PCs as inputs, Figure 4(a) shows effect of competitive rule on convergence of training and CV MSE, Figure 4(b) shows effect of metric on convergence on training and CV MSE and Figure 5(a) shows variation of training and CV MSE with cluster centers Figure 5(b) shows variation of average minimum MSE with number of PES in hidden layer, Figure 6(a) shows effect of different transfer functions on convergence of training /CV MSE., Figure 6(b) shows variation of average minimum MSE with error criterion.
展开▼
