A multi-nets ANN model for real-time performance-based automatic fault diagnosis of industrial gas turbine engines

When a robust mathematical model of a process equipment is available, model-based diagnostic methods can be used to identify the occurrence of faults in a system. However, these methods are less effective when the non-linearity, complexity, and modeling uncertainties of the system increase. In recent years, a new discipline, known as artificial intelligence-based methods, has emerged, which allows the behavior of the system to be studied using operational data. While single-learner artificial neural network (ANN)-based models demonstrate a satisfactory level of capability in assessing the health of gas turbines, this article investigated the application of a multiple networks artificial neural network (multi-nets ANN) model using a multiple-views multiple learners approach to provide a real-time performance-based automatic fault detection (AFD) system in gas turbine engines. Towards this end, a number of key performance variables, which are commonly measurable on most industrial gas turbine engines, were monitored, and their associated ANNs were trained for healthy conditions. Two back-propagation training algorithms, namely the Levenberg–Marquardt and Bayesian regularization algorithms, and the k-fold cross-validation technique, were employed to train the optimal networks using a training data set. Using the trained multi-nets ANN model, two case studies were conducted pertaining to the detection of drop in compressor flow capacity and compressor fouling in an industrial 18.7-MW twin-shaft gas turbine engine. The ability of all the trained networks of the multi-nets model to detect these faulty conditions was investigated. The results obtained showed that among the four trained networks in the multi-nets model, the associated network for gas generator rotational speed was able to track these incidents earlier. The findings demonstrated the capabilities and performance of the proposed multi-nets model with regard to AFD in industrial gas turbines. © 2017, The Brazilian Society of Mechanical Sciences and Engineering.