Robust Residual Generation for Model-Based Fault Diagnostic Systems

点击翻译按钮获取中文摘要

9523087 Danai This research project deals with the development of innovative approaches for robust fault diagnosis in machines such as automotive engines. The approach involves feedback training and pattern classification. Pattern classification is adopted from both residual transformation and fault signature estimation. Fault signatures are estimated via the multi-valued influence matrix (MVIM) method which also provides measures of uniqueness and repeatability of fault signatures. In residual transformation, a time-delay connectionist net is trained iteratively on a batch of measurement - fault data. The learning algorithm adjusts the parameters of the time-delay net such that the next set of transformed residuals will have better detection and diagnostic performance within the range of training set. The residual generation method is experimentally evaluated at the center for Automotive Research of the Ohio State University. The experiments include sensor and actuator faults and are carried out over a broad range of engine operating conditions.***