Diagnostic fault detection for internal combustion engines via pressure curve reconstruction

One proven technique for monitoring the health of a sealed internal combustion engine is to analyze the combustion pressure cycle curves of the individual cylinders. Current techniques for doing this require a pressure sensor mounted directly in the combustion chamber. This necessitates maintenance and design considerations that may be unacceptable especially on legacy systems. This paper describes a non-invasive technique developed for monitoring combustion pressure cycle related faults. This method has been developed and tested on a diesel engine test bed at Penn State University's Applied Research Laboratoq (ARL), Condition Based Maintenance Department. The diesel engine test bed was used to gather all forms of data under different engine operating conditions. Using crdshaft angular velocity data fiom a high-resolution encoder, a trained neural network is used to reconstruct the combustion pressure cycle curves. These reconstructed combustion pressure curves are then passed into another trained neural network for fault detection analysis.