Neural networks on predictive maintenance of turbomachinery

Predictive maintenance is becoming a widespread methodology. On-line monitoring evaluation of machinery mechanical condition is justified by low cost and high reliability of data acquisition systems. The use of artificial intelligent techniques is a suitable solution for handling the large amount of information that is generated. This paper introduces a computer program for on-line turbomachinery monitoring which is running at an oil plant and involves the analysis of vibration data. The introduction of neural networks associated to fuzzy systems will permit to verify the input data quality, to improve the: alarms and to detect features that will permit to develop a reliable diagnostic. Copyright (C) 1998 IFAC.