Search Results

The stochastic Fault Detection and Isolation (FDI) algorithm, known as the statistical local approach, is applied in a model-based framework to the diagnosis of component faults in the air-intake system of an automotive engine. The FDI scheme is first presented as a general methodology that permits the detection of faults in complex nonlinear systems without the need for building inverse models or numerous observers. Although sensor and actuator faults can be detected by this FDI methodology, component faults are generally more difficult to diagnose. Hence, this paper focuses on the detection and isolation of component faults for which the local approach is especially suitable. The challenge is to provide robust on-board diagnostics regardless of the inherent nonlinearities in a system and the random noise present.

This paper describes a methodology that has been developed to apply basic concepts of pattern recognition to isolate “events” in any type of time history data. The results obtained from this methodology can be used for a variety of engineering applications. In this study, it has been applied to estimate and compare the cumulative structural fatigue damage from single bump excitations versus resonance in Class 8 tractors based on consumer highway data. Using the basic concepts of pattern recognition, which include statistical methods based on correlation functions, windowing techniques and root mean square values, a similarity search has been performed to extract and classify known consequential time history traces (events) from the set of acquired data. The advantage of this model is seen in extracting events whose exact time traces are not known.

Research has been conducted to develop a methodology for the generation of driving and duty cycles for refuse vehicles in conjunction with a larger effort in the design of a hybrid-electric refuse vehicle. This methodology includes the definition of real-world data that was collected, as well as a data analysis procedure based on sequencing of the collected data into micro-trips and hydraulic cycles. The methodology then applies multi-variate statistical analysis techniques to the sequences for classification. Finally, driving and duty cycles are generated based on matching the statistical metrics and distributions of the generated cycles to the collected database. Simulated vehicle fuel economy for these cycles is also compared to measured values.