Multi-Fault Discrimination with Fault Model and Periodic Residual

This work deals with the issue of faults classification on line for non linear Hessenberg systems whose faults are not isolable in steady state conditions. A stepwise procedure is proposed. The first step consists of the parameters estimation associated with one fault model when abnormal conditions are detected. As second step, a residual is calculated considering the identified model together with a periodic auxiliary input to the system. The auxiliary signal added to the system allows the discrimination of two faults scenarios via features of the residual. The novel procedure allows: (1) the identification of the fault parameters, if only one occurs and (2) the evaluation of some boundaries for the faults parameters, if two faults occurs simultaneously. To show the viability of the method, the leaks diagnosis in pipeline is considered as example.