Neural networks for gas turbine fault identification: multilayer perceptron or radial basis network
Loboda I., Feldshteyn Ya, Ponomaryov V. Neural networks for gas turbine fault identification: multilayer perceptron or radial basis network / International Journal of Turbo & Jet Engines, Walter de Gruyter GmbH und Co. KG, Germany, 2012, Vol.29, Issue 1, pp. 37-48.
Efficiency of gas turbine condition monitoring systems depends on quality of diagnostic analysis at all its stages such as feature extraction (from raw input data), fault detection, fault identification, and prognosis. Fault identification algorithms based on the gas path analysis may be considered as an important and sophisticated component of these systems. These algorithms widely use pattern recognition techniques, mostly different artificial neural networks. In order to choose the best technique, the present paper compares two network types: a multilayer perceptron and a radial basis network. The first network is being commonly applied to recognize gas turbine faults. However, some studies note high recognition capabilities of the second network.
For the purpose of the comparison, both networks were included into a special testing procedure that computes for each network the true positive rate that is the probability of a correct diagnosis. Networks were first tuned and then compared using this criterion. Same procedure input data were fed to both networks during the comparison. However, to draw firm conclusions on the networks’ applicability, comparative calculations were repeated with different variations of these data. In particular, two engines that differ in an application and gas path structure were chosen as a test case. By way of summing up comparison results, the conclusion is that the radial basis network is a little more accurate than the perceptron, however the former needs much more available computer memory and computation time.
Link to the publication »ASME Paper No. GT2011-46752
Link to the publication »Gas turbine fault diagnosis using probabilistic neural networks
Loboda I., Olivares Robles M.A. Gas turbine fault diagnosis using probabilistic neural networks / International Journal of Turbo & Jet Engines, Walter de Gruyter GmbH und Co. KG, Germany, May 2015, Vol.32, Issue 2, pp.175-192.
Efficiency of gas turbine monitoring systems primarily depends on the accuracy of employed algorithms, in particular, pattern classification techniques for diagnosing gas path faults. In recent investigations many techniques have been applied to classify gas path faults, but recommendations for selecting the best technique for real monitoring systems are still insufficient and often contradictory.
In our previous work, three classification techniques were compared under different conditions of gas turbine diagnosis. The comparative analysis has shown that all these techniques yield practically the same accuracy for each comparison case.
The present contribution considers a new classification technique, Probabilistic Neural Network (PNN), and we compare it with the techniques previously examined. The results for all comparison cases show that the PNN is not inferior to the other techniques. We recommend choosing the PNN for real monitoring systems because it has an important advantage of providing confidence estimation for every diagnostic decision made.
Link to the publication »Computation and monitoring of the deviations of gas turbine unmeasured parameters
L. A. Miro Zarate and I. Loboda, Computation and monitoring of the deviations of gas turbine unmeasured parameters, ASME Turbo Expo 2015: International Technical Congress “Power for Land Sea & Air”, Montreal, Canada, June 15-19, 2015, 10p., ASME Paper No. GT2015-43862.
One of the principle purposes of gas turbine diagnostics is the estimation and monitoring of important unmeasured quantities such as engine thrust, shaft power, and engine component efficiencies. There are simple methods that allow computing the unmeasured parameters using measured variables and gas turbine thermodynamics. However, these parameters are not good diagnostic indices because they strongly depend on engine operating conditions but in a less degree are influenced by engine degradation and faults.
In the case of measured gas path variables, deviations between measurements and an engine steady state baseline were found to be good indicators of engine health. In this paper, the deviation computation and monitoring are extended to the unmeasured parameters. To verify this idea, the deviations of compressor and turbine efficiencies as well as a high pressure turbine inlet temperature are examined. Deviation computations were performed at steady states for both baseline and faulty engine conditions using a nonlinear thermodynamic model and real data. These computational experiments validate the utility of the deviations of unmeasured variables for gas turbine monitoring and diagnostics.
The thermodynamic model is used in this paper only to generate data, and the proposed algorithm for computing the deviations of unmeasured parameter can be considered to be a data-driven technique. This is why the algorithm is not affected by inaccuracies of a physics-based model, is not exigent to computer resources, and can be used in on-line monitoring systems.
Link to the publication »Gas turbine fault classification using probability density estimation
Loboda Igor. Gas turbine fault classification using probability density estimation, ASME Turbo Expo 2014: International Technical Congress “Power for Land Sea & Air”, Dusseldorf, Germany, June 16-20, 2014, 13p., ASME Paper No. GT2014-27265.
Diagnostics is an important aspect of a condition based maintenance program. To develop an effective gas turbine monitoring system in short time, the recommendations on how to optimally design every system algorithm are required. This paper deals with choosing a proper fault classification technique for gas turbine monitoring systems.
To classify gas path faults, different artificial neural networks are typically employed. Among them the Multilayer Perceptron (MLP) is the mostly used. Some comparative studies referred to in the introduction show that the MLP and some other techniques yield practically the same classification accuracy on average for all faults. That is why in addition to the average accuracy, more criteria to choose the best technique are required. Since techniques like Probabilistic Neural Network (PNN), Parzen Window (PW) and k-Nearest Neighbor (K-NN) provide a confidence probability for every diagnostic decision, the presence of this important property can be such a criterion. The confidence probability in these techniques is computed through estimating a probability density for patterns of each concerned fault class.
The present study compares all mentioned techniques and their variations using as criteria both the average accuracy and availability of the confidence probability. To compute them for each technique, a special testing procedure simulates numerous diagnosis cycles corresponding to different fault classes and fault severities. In addition to the criteria themselves, criteria imprecision due to a finite number of the diagnosis cycles is computed and involved into selecting the best technique.
Link to the publication »A more realistic scheme of deviation error representation for gas turbine diagnostics
Loboda I., Yepifanov S., Feldshteyn Ya. A more realistic scheme of deviation error representation for gas turbine diagnostics / International Journal of Turbo & Jet Engines, Walter de Gruyter GmbH und Co. KG, Germany, 2013, Vol.30, Issue 2, pp. 179-189.
Gas turbine diagnostic algorithms widely use fault simulation schemes, in which measurement errors are usually given by theoretical random number distributions, like the Gaussian probability density function. The scatter of simulated noise is determined on the basis of known information on maximum errors for every sensor type. Such simulation differs from real diagnosis because instead of measurements themselves the diagnostic algorithms work with their deviations from an engine baseline. In addition to simulated measurement inaccuracy, the deviations computed for real data have other error components. In this way, simulated and real deviation errors differ by amplitude and distribution. As a result, simulation-based investigations might result in too optimistic conclusions on gas turbine diagnosis reliability.
To understand error features, deviations of real measurements are analyzed in the present paper. To make error presentation more realistic, it is proposed to extract an error component from real deviations and to integrate it in fault description. Finally, the effect of the new noise representation mode on diagnostic reliability is estimated. It is shown that the reliability change due to inexact error simulation can be significant.
Link to the publication »ASME Paper No. GT2012-69368
Link to the publication »