This paper deals with an identification procedure of gas turbine nonlinear models for monitoring and diagnostic systems. Introduction of a special time variable into a conventional thermodynamic model helps to create a model of the engine with a variable deterioration level. To identify this model, registration data of great volume and different gas turbine deterioration severity can be attracted. This ensures high accuracy of the identified model as well as quality of a baseline function that can simply be extracted from the model.

The main focus of this paper is reliable fault recognition for gas turbines. Gas path models are employed to describe different faults of variable severity. To recognize them, two methods are used and examined in the paper. The first method is based on the Bayesian recognition while the second applies neural networks. The recognition process for the both methods is simulated numerously under the conditions of random measurement errors, and diagnosis errors are fixed. The objectives are to verify the methods statistically, adjust them, and compare the networks recognition errors with the Bayesian recognition ones. To make the accuracy analysis more general, the paper compares the methods for two fault classification variants and different gas turbine operating conditions.

Main focus of this paper is a reliable fault identification of gas turbines. The paper examines the methods of gas turbine parametric diagnosing. The nonlinear thermodynamic model describes different gradually developing faults. To identify them, the method operation is simulated in the conditions of random measurement errors, the correct and incorrect diagnostic decisions are fixed, and corresponding averaged probabilities are computed. The objectives are to verify the methods statistically, adjust them, and choose the best one which ensures the higher probability of fault correct identification. Besides the method comparison, the paper also considers a problem of classifying the various gas turbine faults. Different classification variants are presented to make more general the analysis of diagnosing process. A generalized fault classification which unites the fault descriptions from different operational regimes is also proposed and analyzed. It conserves a trustworthiness level of the previous regime dependent classification and makes the diagnosing be much more universal.

Gas turbine fault isolation or identification generally uses a model-based classification of gas path faults. This classification is not too exact because of model errors. The present paper looks at the possibility to create a fault class on the basis of gas turbine real data containing cycles of a compressor fouling and washing. The concerned data-driven fouling class formation is realized in the space of deviations of measured gas path quantities. Analyzing deviation plots for different fouling cycles, we have confirmed identifiability of the fouling. In order to draw sound conclusions, the analysis was conducted for two gas turbines of different application.

Gas path fault localization algorithms based on the pattern recognition theory are an important component of gas turbine monitoring systems. To simulate random measurement errors (noise) in description of fault classes, these algorithms usually involve theoretical random number distributions, like the Gaussian probability density function. A level of the simulated noise is determined on the basis of known information on typical maximum errors of different gas path sensors. However, not measurements themselves but their deviations from an engine baseline are input parameters for diagnostic algorithms. These deviations computed for real data have other error components in addition to simulated measurement inaccuracy. In this way, simulated and real deviation errors differ by an amplitude and distribution. Consequently, with such simulation, the performance of a diagnostic algorithm is poorly estimated, and therefore, the conclusion on algorithm efficiency may be wrong. To understand better noise peculiarities, plots of deviations of real measurements are tracked in the present paper. Additionally, possible deviation errors are surely analyzed analytically. To make noise presentation more realistic, it is proposed to extract random errors from real deviations and to integrate these errors in fault description. Finally, the effect of the new noise representation mode on gas turbine diagnosis reliability is estimated.

Temperatures measured in a hot section of gas turbines are very important for a gas path analysis. A suite of parallel thermocouples are usually installed in the same gas path station in order to compute a filtered and averaged temperature quantity for its further use in control and diagnostic systems. However, in spite of the preliminary treatment, the resulting quantity is not completely free from errors. To eliminate or reduce the errors, the present paper analyzes anomalies in the behaviour of each thermocouple of an industrial gas turbine engine. To that end, time graphs of both measured magnitudes themselves and their deviations from reference magnitudes are plotted. In order to draw sound conclusions, the analysis is conducted on a large volume of the data collected for three particular engines.

Fault identification algorithms based on measured gas path variables constitute an important component of a gas turbine engine condition monitoring system. In addition to gas path faults diagnosis, these algorithms are capable to identify malfunctions of sensors and an engine control system. The fault identification algorithms widely use pattern recognition techniques, in particular, different artificial neural networks. Since monitoring system efficiency depends on accuracy of all system's components, the most exact mathematical technique should be chosen for every component. To recognize gas turbine faults, a specific network type, multilayer perceptron (MLP), is mostly applied. However, other network type, probabilistic neural network (PNN), can be applied as well. It uses a probabilistic measure to recognize the faults. In the present paper, the PNN is firstly tailored to a gas turbine diagnosis application and then compared with the MLP. The comparison has shown that both networks yield practically equal accuracy. The PNN is recommended for real gas turbine monitoring systems because, in addition to a diagnostic decision, this network provides confidence estimation for this decision.

In gas turbine engine condition monitoring systems, diagnostic algorithms based on measured gas path variables constitute an important component. Not only gas path faults are diagnosed by these algorithms, but also malfunctions of sensors and an engine control system can be identified with gas path measurements. Many gas path diagnostic algorithms use pattern classification techniques. In particular, a specific neural network, Multilayer Perceptron (MLP), is mostly applied. Unfortunately, the MLP cannot provide confidence estimation for its diagnostic decisions. However, there are techniques that classify patterns on the basis of probability. For example, Parzen Window and K-Nearest Neighbor methods compute probabilities of the considered classes estimating their probability densities. Thus, every diagnosis made is accompanied by its probability that is a very useful property for real gas turbine diagnosis. In the present paper, these two techniques are compared with the MLP in order to determine the technique that provides the best diagnostic accuracy on average for all possible gas turbine faults. The mentioned advantage of the Parzen Windows and K-Nearest Neighbors is also taken into account.

The parametric diagnostics of gas turbine engines has been improved in the last decades due to computer technology development and better analysis methods such as artificial neural networks. It has demonstrated to be a very powerful tool providing an insight into an actual engine health condition and predicting possible future failures. On the basis of a thermodynamic model that relates monitored variables with operating conditions and fault parameters, it is possible to obtain healthy and faulted engine performances. This model allows calculating deviations between actual and baseline engine performances. Based on the deviations computed for all monitored variables, the diagnosis is made by pattern recognition techniques. These deviations include errors due to measurement uncertainty and model inadequacy. Since an engine operating point changes, the deviation errors change as well, resulting in varying diagnostic inaccuracy. In the present paper, two hypotheses on how the errors influence engine diagnosability at varying operating points are first investigated on simulated data and then verified with real information.

Diagnostic algorithms that use gas path measurements and models are capable to diagnose not only different faults of the gas path itself but also malfunctions of measurement and control systems. Since the variety of gas turbine fault conditions is great, they are joined in a limited number of classes. Different principles to create these classes are known and there are many fault classifications in practice. In an investigation stage, it is difficult to predict what classification will then be applied in a real monitoring system therefore the investigators usually experiment with different fault types and fault numbers. The present paper proposes the approach that allows simple creating multiple classification variations including complex and realistic fault classes. This approach also permits easy changing between the variations and the pattern recognition techniques applied for each variation. As a result of application of each technique to each classification, probability of correct diagnostic decisions and execution time are determined. They are criteria of diagnosis efficiency. In this way, the approach allows studying the influence of fault classification on diagnosis efficiency. The paper performs such a study for a power plant for natural gas pipelines. Twelve classification variations are analyzed with the use of three recognition techniques: Multi-Layer Perceptron, Radial Basis Network and Probabilistic Neural Network. Additionally, a new boundary for fault severity is proposed and investigated by comparing three boundary options.