Recently, diagnostic approaches based on Artificial Intelligence have become very attractive. In particular Neural Networks (NNs) seem to have suitable characteristics for gas turbine diagnostics. This paper deals with the activities carried out for: 1. selecting the most appropriate NN structure for gas turbine diagnostics; 2. developing a NN for the detection, isolation and assessment of single and combined causes of performance degradation in a two shaft industrial gas turbine; 3. testing both the NN performance in recognizing causes of performance degradation and robustness in presence of scarce and/or wrong input data. The data used in all these phases in order to train and test the NN have been generated using a non-linear Cycle Program. So, the Cycle Program becomes a data generator, which may be integrated with data derived from field experience, while the diagnostic function is performed by the NN.
展开▼
