Implementing a novel fault prognosis technique based on nonlinear fault observer and online parameters estimation

This research presents a methodology for predicting errors of parameters, as the algorithm tries to monitor the parameters in order to maintain or replace them when needed to avoid excessive expenses. The presented implementation mechanism is based on monitoring parameters according to a specific number of batches and each batch consists of a number of iterations, which in turn are a number of samples. The proposed algorithm involves designing a new nonlinear observer and writing a secondary algorithm for parameter estimation based on the online nonlinear recursive least squares algorithm associated with the observer states. In addition, the algorithm presents an attempt to find a relationship between the error states and the state of the parameters by creating a new function to determine the weight of the error according to four components; parameter changes, output residuals, output errors and the error diagnosed by the new observer. The algorithm also includes introducing the probability form of the weights using the kernel density function for the average and maximum weights for each batch. Finally, relying on the results, it is possible to take the appropriate decision to maintain or change the parameters as shown a non-linear direct current motor model case study