Combining vibrations and acoustics for the fault detection of marine diesel engines using neural networks and wavelets

Attempts to solve the predictive maintenance problem using numerical simulations have been already presented with some promising results, Furthermore, Neural Network methods have been used extensively on the condition monitoring either alone or through the identification of dynamic models but in a simplified manner. A basic characteristic of these methods is the assumption of the existence of accurate, either experimental or theoretical, simulations for the production of massive data with various faults for the learning purposes of the Neural Networks. On the other hand little work has been performed on the in-situ predictive maintenance of marine diesel engines especially with respect to acoustic condition monitoring, At the present paper a new method is presented for the complete modelling of the main faults of large diesel marine engines. This method combines measurements of both vibration and acoustic signals recorded at or near the operating engine and uses Wavelets as a Feature Extraction technique in order to feed a Neural Network classification system towards marine diesel engines' fault detection.