In both the marine and power industries there are now a choice of off-the-shelf condition monitoring systems available that utilise artificial intelligence techniques to analyse engine performance data. These systems are proving to be a valuable aid in optimising performance and reducing down-time by assisting with maintenance planning. These systems rely on careful monitoring of an engine's performance, for instance engine speed, fuelling, boost pressure, turbine inlet pressure, turbocharger speed, and exhaust temperature. With this data, they utilise a variety of interpolation and pattern recognition algorithms to compare it with previously recorded data stored in lookup tables.
This paper describes how a neural network approach can be used as a cheap alternative for the analysis of this data, greatly reducing the need for such large lookup tables and complex pattern recognition programs. A PC-based demonstration neural network has been trained on data, generated by a digital simulation, to learn the performance characteristics of a turbocharged diesel engine. The PC is also connected to a real test engine via data acquisition cards and is able to identify a limited number of faults artificially introduced on the test engine while it is operating anywhere in its performance envelope.