Loboda Igor

Problems Of Gas Turbine Diagnostic Model Identification On Maintenance Data.

Abstract – The paper is devoted to the problems of gas turbine model identification on the real data of automatic diagnosing system. Influence of the data inaccuracy on identification results and possible sources of the inaccuracy are examined.

Automatic diagnosing system is an integral part of modern gas turbine power plants (gas turbines), and is applied not only for maintenance needs but for aims of gas turbine development ant tests too. The algorithms of fault detection based on treatment of gas path measured parameters (gas pressures and temperatures, rotor rotation speeds, fuel consumption, …) are considered as principal and most complex. Above the gas path the control system and measurement system may be diagnosed too on these parameters [1,2] without gas turbine shut down and disassembling.

Difficulties of gas path parameter diagnosing consist of the simultaneous influence of any fault on all measured parameters and the influence of gas turbine regime too.

With taking into account of these difficulties the special state parameters are induced to be sensitive to certain faults and insensitive to another faults and factors. Diagnostic models of different type are served for a connection of the measured parameters and the state parameters. One of these models, which is used in current investigations, is static multi-regime nonlinear mathematical model of gas path:

                 (1)

where the vector of measured gaspath parameters  depends on the vector of engine regime and atmospheric conditions  and the vector of state parameters . In the capacity of state parameters the parameters of gas turbine component characteristics are chosen. These characteristics are able to move and deform the component characteristics and, as a result, to simulate engine failures.

A model identification procedure was applied for the state parameters calculation. Estimations  are determined as the result of distance minimization between model values  and measured values :

.  (2)

Identification quality is connected with the accuracy achieved on gas path parameters and state parameters, and depends on many factors including input information condition.

Maintenance parameter registration by automatic diagnosing system of the gas turbine for driving of natural gas pumping unit is used as a source of the input information. Diagnostic data bases of two gas turbine (let us call them as GT-1 and GT-2) was selected. The bases include parameter mean numbers registered with one hour frequency. The data base of GT-1 includes the period of axial compressor cleaning, which represents an important information about compressor fouling.

Two objects are pursued: at first, to check the identification procedure functioning and to estimate the possibility of the procedure application in gas turbine automatic diagnosing systems for aims of monitoring and diagnosing; at second, to analyze sources of possible obstacles standing in the way of diagnostics progress.

The following parameter structure was selected: temperature tН and pressure PН of atmospheric air, fuel consumption Gf and power turbine rotation speed nPT (vector ); 7 regular measured parameters (vector ); 6 estimated parameters of gas turbine component characteristics – two fault indices (a gas consumption parameter and an efficiency parameter) for every of principle mechanic components: axial compressor, gas generator turbine, power turbine (vector ).

It is famous that the most common cause of performance deterioration is a compressor fouling. That is why the fouling influence on measured and estimated parameters was a first aim of analysis. The data extract was formed from 1880 registration points until and after the compressor cleaning and divided on 46 samples of 80 points (the sample displacement is 40 points). During the calculations the maintenance diagnosing process was imitated: the model identification was repeated consequently for every sample and correspondent consequent estimations  were found.

Initial deviations  of measured parameters, determined on the first identification step, are shown in Fig.1. The deviations are not so great and mainly do not exceed the level of 2%. The parameter jump after the cleaning (points 917-918) and total parameter degradation due to fouling are well distinguished. Thus, the model (1) satisfies the requirements for a norm function in gas path monitoring algorithms of the automatic diagnosing system.

The estimations , found as a result of the model identification and presented in Fig.2, show us that:

a) parameters dGC и dηC of compressor characteristics correctly reflect the compressor clearing influence (after sample number 21) and the following compressor state degradation, that is why the identification procedure suitability for fault detection algorithms is confirmed;

b) accidental fluctuations in the estimations  induced by correspondent fluctuations in the deviations  are still significantly great and may mask the fault effects.

So, the next growth of monitoring and diagnosing trustworthiness depends of the results of analyzing and elimination of the fluctuations in .

Great consideration was given to study this problem. Table forms and graphic means for viewing and visualization of data bases, special diagnostic archives (period of registration – 1 second) and their playing means, and additional verifying calculations were applied. Numerous parameters plot comparisons were fulfilled for detection of the fluctuation causes. In the result of this search the cases of measurement faults, of abnormal functioning of variable inlet guide vane, anti-icing system defects were found.

Let us concern, for example, the case of fluctuation increasing in deviations  of GT-1 after 2.04.00 (Fig.3). It may be seen that the fluctuations were synchronous and appeared and disappeared several times, therefore only the factors, acting on the gas turbine regime or on the model behaviour, are possible.

Originally were suggested the following versions of the deviation sources

·  the distance growth between the gas turbine parameters and the model parameters in the new region of operational regimes in which the gas turbine did not operate previously;

·  additional factors above the model arguments which affect the gas turbine but are not took into account in the model;

·  errors of model inlet parameters tН, PН, Gf, nPT.

For the real course detection the special document in MathCAD was developed as an instrument of real data visualization and analyzing on a background of the norm (model).

Accurate analyzing of all possible parameter plots in this document shown that a) the model adequately reflects the gas turbine behaviour; b) the fluctuations have appeared on these same gas turbine regimes where previously they were absent; c) the fluctuations are not timely synchronized with the regime changes.  In that way, the first version was rejected.

Other possible courses were examined by the way of analyzing of timely synchronized parameter plots and were rejected too for exception of an influence of parameter tin (air temperature in the gas turbineinlet).

Comparison of tin plot with tн plot, and with these same plots of GT-2, and with deviations  had shown that a) on the time interval after 2.04.00 the errors of parameter tin on the level of +5° take place without any doubts (Fig. 4); b) analyzed fluctuation in  are fully synchronized with these temperature errors.

The next task was to understand the mechanism of temperature error influence and to confirm it by calculations.

According our opinion, the error influence on the gas turbine is connected with angle rebuilding of variable guide vane and occurs according to the following scheme: [increasing of temperature tin due to the errors] → [drop of calculated value of the corrected rotation speed nT corr] → [reducing of inlet guide vane angle φIGV by the control system] → [gas flow decreasing and correspondent power drop below the setting level] → [feeding the additional fuel by the control system to reach the power setting] → [correspondent changes of gas path parameters due to the compressor state change and the regime raising].

Fig. 1. Gas path parameter deviations	Fig. 2. State parameter estimations
Fig. 3. Appearance of fluctuations	Fig. 4. Temperature error

In the fulfilled calculations the next information was used: correction formulas for gas path parameters, the plots from data bases with real dependencies between these parameters, data of influence of angle φIGV displacement from special gas turbine diagnostic tests.

Also calculated values had not concur with measured ones, the common picture was the same: gas path temperatures increased and the pressures decreased. Thus, the version of temperature tin responsibility for fluctuation increasing in deviations  is confirmed.

For comparison the data base of GT-2 was analyzed too. It is established that the canal tin operates correctly and a level of “jumps” in the deviations  is significantly low.

The problem becomes complicated due to high level of reached model accuracy (errors in gas path parameters are about 1%).  It is clear that next  accuracy increasing will require much more effort, but without them the progress is impossible neither in the development of identification procedure nor  in monitoring and diagnosing algorithms real application.