Neural-network-based sensor validation for gas turbine test bed analysis

Abstract

Abstract

This paper presents the use of neural networks for steady state sensor validation of gas turbines. The application is particularly suitable for test bed data analysis. Sensor failures are frequent in gas turbines and their effect on the accuracy and reliability of the performance analysis can be dramatic. Both test bed analysis and analysis of on-wing engines can be strongly affected by a measurement bias. The problem of isolation of faulty sensors can be complicated by the presence of actual engine faults that are still unknown at this stage of the analysis. Auto-associative neural networks have been used together with a minimization technique to perform multiple sensor failure detection and isolation of a low-bypass-ratio turbofan engine. Neural networks have been shown to be able to isolate faulty sensors successfully, even in the presence of unknown engine faults. Direct consequence of the application of neural networks is that the following estimation of the engine component performance parameters can be carried out separately, after rejection of biased measurements.