Model-based fault diagnostics of nonlinear systems using the features of the phase space response

• Presentation of a new approach for model-based diagnostics of nonlinear systems.
• Feature extraction using nonlinear dynamic analysis tools for estimation of system parameters.
• Introduction of new features for the characterization of periodic and multi-periodic nonlinear response.
• Application of an artificial neural network as a tool for mapping the features space to the parameters space.
• Implementation of the diagnostics algorithm to a nonlinear pendulum setup.

A novel algorithm is presented in this paper for the diagnostics of nonlinear systems based on the idea of estimating the parameters of the system using features of the nonlinear response. The method combines statistical features of the nonlinear response and capabilities of artificial neural networks in data fitting with the objective of estimating the parameters of a defective nonlinear system. New features extracted from the density distribution of position and velocity signals are introduced to characterize the complex topologies of the phase plane response in periodic and multi-periodic domains. A nonlinear pendulum is used for experimental validation of the procedure. The results show that, with appropriately selected features of the response, the parameters of the nonlinear system can be estimated with an acceptable accuracy.