IRT and ANN Based System Identification Schemes for Model Based Fault Detection of Nonlinear Systems

Faults in electrical power transmission systems can cause system failures and even may cause explosions. It is desired to remove a faulty component immediately to prevent further damage to the system and the environment. To address this, a typical technique is devised based on thermal imaging and various artificial neural network algorithms. The faulty element or component will radiate or emit higher energy when compared to normal or healthy conditions, because of the higher current flow rates. The thermal image taken from such defective part of the power system will be more highlighted in the image in contrast with the normal cool background. This drastic change in the grey level values in contrast with the healthy power system's picture hints or predicts a fault in that region. To support this methodology, an exhaustive simulation is implemented and demonstrated using thermal image processing and self-learning neural network algorithms and the simulation results are compared. This analysis is performed through various types of ANN techniques, and comparisons are established between them to report the network with best prediction results on a typical 'Step' dataset and 'Realistic' dataset. A similar low score mean square error MSE is exhibited with these models, and the R-square values are closer to the best score of one in all algorithms discussed. A better graph is obtained in Levenberg-Marquadrdt training Algorithm where most of the predictions fall alongside the target; however, Bayesian regularization gives a better plot than LM with the best fit is being obtained at lesser number of iterations that is in less time.