An artificial neural network-based diagnostic methodology for gas turbine path analysis—part II: case study

The reliability of gas path components (compressor, burners and turbines) of a gas turbine is generally high, when compared with those of other systems. However, in case of forced stops, downtime is usually high, with a relatively low availability. The purpose of conditions monitoring and faults diagnostics is to detect, isolate and evaluate (i.e., to estimate quantitatively the extent) defects within a system. One effective technique could provide a significant improvement in economic performance, reduce operating costs and maintenance, increase the availability and improve the level of safety achieved. However, conventional analytical techniques such as gas path analysis and its variants are limited in their engine diagnostic, due to several reasons, including their inability to effectively operate in the presence of noise measures, to distinguish anomalies of component from a failure sensor, to preserve the linearity in the relations between parameters of gas turbines and to manage the sensors range to achieve accurate diagnosis. In this paper, the approach of a diagnostic scenario to detect faults in the gas path of a gas turbine has been presented. The model provides a large-scale integration of artificial neural networks designed to detect, isolate and evaluate failures during the operating conditions. The engine measurements are considered as input for the model, such as the speed, pressure, temperature and fuel flow rate. The output supplies any changes in the sensor or in the efficiency levels and flow rate, in the event of fault components. The diagnostic method has the ability to evaluate both anomalies of multiple components or multiple sensors, within the range of operating points. In the case of components failures, the system provides diagnostic changes in efficiency and flow rate, which can be interpreted to determine the nature of the physical problem. The technique has been applied in different operating conditions by comparing the results obtained with the solutions provided by linear and nonlinear analysis.