Identification and Monitoring of Automotive Engines

The objective of this paper is to extend and refine the nonlinear canonical variate analysis (NLCVA) methods developed in the previous work for system identification and monitoring of automotive engines. The use of additional refinements in the nonlinear modeling are developed including the use of more general bases of nonlinear functions. One such refinement in the NLCVA System identification is the selection of basis functions using the method of Leaps and Bounds with the Akaike information criterion AIC. Delay estimation procedures are used to considerably reduce the state order of the identified engine models. This also considerably reduces the number of estimated parameters that directly affects the identified model accuracy. This increased accuracy also affects the ability to monitor changes or faults in dynamic engine characteristics. A further objective of this paper is the development and use of nonlinear monitoring methods as extensions of several previously used linear CVA monitoring procedures. For the case of linear Gaussian systems, these monitoring methods have optimal properties in detecting faults or system changes in terms of the general maximum likelihood method. In the nonlinear case, departures from optimality are investigated, but the procedure is shown to still work quite effectively for detecting and identifying system faults and changes.