Model Based Diagnosis: Failure Detection and Isolation by State Comparison with Model Behaviors 2007-01-2077

Powertrain diagnosis technology has been developed for On Board Diagnostics (OBD) legal requirements and for failure identification of electronic controllers. The diagnosis technology can be sophisticated as there are requirements for compliance with stringent OBD regulations for early repair work for failure recovery or for real time deterioration compensation. Deterioration of performance, such as piece-to-piece variation of products or aging must be detected and identified to satisfy requirements. One of the promising methodologies has long been to compare detected values with model behaviors. Calculation time was a major constraint ten years ago. Model structure and sensor properties are likewise constraints even today. Model based diagnosis technology has been studied in the literature listed in references. The residual method was selected for the model based diagnostic method. A mean value engine model was introduced as a model for the diagnosis method strategy because of shorter calculation times required for real time applications. It is not obvious if the plant is failure when single sensor output is compared to equivalent model output. So, structural redundancy is constructed in Failure Detection and Identification (FDI) part of the diagnosis system. Redundancy is one of the important concepts in the model based diagnosis. For one system, comprehensive dynamic properties and whole structure allow itself to cause redundancy of respective variables. It is possible that the values of such variables can be obtained by different means, for example, direct monitoring using a sensor, calculation using a dynamic equation and estimation by an observer. These different means lead to the redundancy of variables. Under certain conditions, one of these means may not be performed correctly, for example, sensor failure. The redundancy of the variable assures that a state can be obtained from other means, for example, calculation of a dynamic equation or utilization of an observer. However, without redundancy a single signal may not be enough to determine where faults occur. Therefore, a certain combination of signal redundancies (also called structural redundancies) is needed, and sensitivity to instant phenomena or noisy signals should be lowered. Consequently, a FDI system is proposed and applied to failures of sensors and actuators in a gasoline engine. Reliability of the FDI results for redundancy, response time for failure-to-failure identification and for resolution of the smallest detected deterioration was investigated. Some response time is required to complete fault decision making process for a model based FDI system. Failure that has shorter time cycle of failure - recovery than the response time will not be identified with the FDI system. Meanwhile, an important factor during the practical application of the FDI system is the resolution of detected deterioration. The smallest resolution needs to be properly determined. Otherwise false alarms will be activated. Computer simulation indicates that those properties depend on the degree of the structural redundancy of the FDI system, a mean value engine model and also a physical measurement of sensor properties.