Diagnostic fault detection for internal combustion engines via norm based map projections

One proven technique for monitoring health of a sealed internal combustion engine is to analyze combustion pressure cycle curves of the individual cylinders. Most techniques that are available are either overly simplistic or rely on artificial intelligence based methodologies such as neural networks. While neural network based methods can be useful, there is normally no quantitatively hard way to determine how accurately a trained neural network represents the desired goal. For this reason, neural networks have little real acceptance by industrial communities that deal with critical applications. This paper describes a technique developed for detecting combustion pressure cycle related faults in diesel engines. This method has been developed at the Pennsylvania State Universities, Applied Research Laboratories, Complex Systems Monitoring and Automation Department and applied to a fully instrumented diesel engine test bed. The new methodology utilizes pressure curve information derived from reliable and relatively inexpensive optical fiber based pressure sensors. The technique outlined in this paper uses a combination of norm based and statistical methods to develop a fault analysis map for particular internal combustion engines. A fully instrumented diesel engine test bed allows for generation of training data sets consistent with actual engine operation. Results from this technique applied to test bed data not used during development of the map show results closely match seeded fault conditions.