Russia, Ufa, Ufa State Aviation Technical University

Abstract

This paper describes a fuzzy expert system based approach of aircraft gas-turbine engine (GTE) parametric diagnostics and control. Some methods of implementation are supposed. The paper also contains a comparison of existing expert systems and general recommendations to select the proper one for gas-turbine engine diagnostic purposes.

Аннотация

В статье предлагается один из подходов к решению задач диагностики и контроля параметров авиационного газотурбинного двигателя с помощью гибридных статических и динамических нечетких экспертных систем. Предлагаются методики реализации подобных систем. Проводится сравнительный анализ и рекомендации при выборе экспертных систем для решения задач диагностики и контроля газотурбинного двигателя.

The usage of expert system increases dramatically an efficiency of the aircraft GTE diagnostics due to its ability to rapidly analyze amounts of various situational data, to make prescription about how to fix certain malfunction, to heed nonlinear and indefinite character of processes and to make general decisions about engine operation. One of perspective methods to increase an efficiency of the modem diagnostic equipment is an adaptation of some information technologies based on soft computing such as fuzzy logic systems, neural nets, genetic algorithms, dynamic hybrid expert systems, etc.

The diagnostic expert system correlates observable faults with its original reasons using one of the following methods. The former implies the usage of associative table containing relations between types of system behavior and diagnosis. The later is based on the cooperative usage of knowledge about system structure. Knowing the critical points of system structure it is possible to make suggestions about probable malfunctions matched to certain observable data.

The performance of the expert system depends primarily on expandable knowledge base which stores etalon data and rules for the problem area transformation. The modem hybrid diagnostic expert systems are capable to operate with fuzzy definitions making the diagnostic system more flexible and providing more efficient data usage, storage and acquisition.

The authors of the paper have made an attempt to analyze and to compare some general features of diagnostic hybrid expert system (including static and dynamic) which are capable to operate within partial or absolute indeterminacy.

The structure of formal static expert system includes the following components:

inference engine (IE), working memory (WM), knowledge base (KB), knowledge acquisition system, interpreter, user interface.

Unlike the static expert system, the dynamic expert system includes additional components such as the environment simulating subsystem and environment communication subsystem. The later communicates with external world by the set of sensors and controllers. Besides, the traditional elements of static expert system, namely the knowledge base and inference engine, are changed essentially in order to meet temporal component of real world's events.

The authors considered a mathematical model of P11F -300 turbo-shaft engine with the following parameters: The control routine is N₁=const, F_c=const; The control parameters are n₁ - low pressure turbine rotational speed; N₂ - high pressure turbine rotational speed; G_b - air consumption; G_m - fuel consumption; T₀- inlet air temperature; B₀ - inlet barometric pressure; F_{c/a bd} - high pressure turbine nozzle set cross-section area; F_{c/a нd} - low pressure turbine nozzle set cross-section area; T*_k, Р_k - stagnated flow temperature and air static pressure behind compressor respectively; Т*_m, P_m - stagnated flow temperature and gas static pressure behind turbine respectively.

This algorithm is based on the comparison between certain engine mathematical model and etalon defectless engine model that implies an inspection of state variables to be within acceptable thresholds. Falling outside the thresholds indicates a failure in the certain engine unit.

During intellectual technical diagnostic and control system development based on RTwork3.5 hybrid fuzzy dynamic expert system and Hugin and C-PRIZ hybrid fuzzy static expert systems the features of these program tools were considered by the developed techniques.

The GTE diagnostic expert system development process consists of six basic steps. The contents of each step is defined by the procedures required to solve technical diagnostic purposes.

As a base diagnostic algorithm the thermal gas dynamic parameters diagnostic algorithm was chosen. The dynamic parameters were chosen by the perturbation method. As result of adapting this method into dynamic expert system it is possible to detect three GTE states such as normal, near to failure and failure.

According to, if a parameter deviation is more than ±5 percents from calculated value, a GTE unit is admitted to be defective.

In order to estimate the threshold which defines normal or defective state of engine, it is supposed to use a maximum instrument error of bench test.

The references contain the estimation of such errors and the greatest one equals to 1.5%.

Thus, the parameter deviation equals to [-1.5; 1.5] will mean a normal state of engine. The parameter deviation equals to [-1.5; 5] and [1.5; 5] will mean a "near to failure" state of engine. Finally, if a parameter deviation is more than ±5, a GTE unit is defective. So, the parameter being changed, the condition vector describes an area of serviceability in the attribute space. Falling outside the limits of this area will mean a non-normal technical state of object under test. In other hand, it is possible to say that all malfunctions can be determinate by the set of attributes, i.e. each malfunction has own area, called as a diagnosis area D_i. The diagnosis system problem is to define a degree of concurrence of a current state with some diagnoses D_i.,

The linguistic variables and its membership function can be used to define diagnoses in the form of attribute conjunction. The conjunction of fuzzy membership function is know to be the search for a minimum of values: _AB(x) = min {_A (x), _B (x)}, для xX.

The C-PRIZ intellectual environment consists of the following components: the knowledge base which represents knowledge and experience for this problem area in the form of set of heuristics and rules in order to optimize logic inference algorithm; inference engine based on algorithms of casual event net generation in the functional-structural simulator; adapting unit which coordinates the usage of different knowledge set of data, knowledge and expert knowledge bases; interpreter which interprets logical inferences; scheme editor which provides object-oriented programming to solve tasks with more efficiency and flexibility; script system which consists of some independent databases, knowledge bases and expert knowledge bases and is dedicated to operate with various object and event types; user interface which provides easy-to-use means to define simulation purposes in texteditor-like system using C-PRIZ internal program- ming language called as NUT. Using this unit an user can implement concepts, formulas, script tables, fuzzy productions,etc.; calculator which performs task solving for- malized by user and written on C-PRIZ programming language. As a body of mathematics of C-PRIZ uses an intuitiotistic calculus of sentences and back chaining inferences for automatic routine generation.

The technical diagnostics and control under C-PRIZ and Hugin environments are performed according to the following technique: 1. Questionnaire development. 2. Classification of possible engine states. 3. Determination of main malfunctions of gas turbine engine. 4. Determination of diagnostic parameters. 5. Determination of knowledge base structure. 6. Implementation of diagnostic parameters. 7. Determination of concept interrelation for the knowledge base. 8. Expert rules development. 9. Problem solving strategy development. 10. Diagnostic algorithm development. 11. Deciding rule development.

RTworks3.5 environment includes an advanced frame knowledge base supporting the back-chaining (goal-driving), forward-chaining (data-driven) and scan (time-driven) inferences.

In hybrid expert system, the GTE etalon model is stored in knowledge base and can be adjusted according to the knowledge acquisition process. RTworks3.5 environment includes relative database management system providing the following data processing operations: logic sampling, relation projection, new relation creation by the certain tuple with the fuzzy attributes, etc. It should be pointed out that each type of hybrid expert system has got its own requirements for knowledge representation. As soon as hybrid system implements various types of information (such as frame, semantic net, knowledge base concept, neural net, fuzzy variable, genetic algorithm) it is rather difficult to integrate the data even within united information space of this expert system. For instance, real time hybrid expert system uses the static expert system to store the different knowledge set (C-PRIZ, for example) and the neural net to store dynamic data.

The technique of gas turbine engine diagnostic and control hybrid expert system different knowledge set design consists of the following steps: 1. Conceptual model design (formalization of problem area). 2. Selection and adaptation of diagnostic algorithm. 3. Representation of gas turbine engine diagnosis model as a set of separated concepts (knowledge sets) in the knowledge base. 4. Creation of knowledge base including rule base as a intelligence kernel control component. 5. Description of gas turbine engine diagnostic and control knowledge set in the various hybrid expert system components (database, knowledge base, expert knowledge base, graphical database, etc.). 6. Fuzzy logical software design. 7. Distribution of information flows between expert system and its various components. 8. Testing of expert system within user queries. 9. Testing of hybrid expert system.

In conclusion, it should be pointed out that the proposed approach to intellectual ТЕ diagnostics and control system development based on fuzzy static and dynamic hybrid expert systems allows the following: to facilitate adaptation of the object under agnostics; to apply different knowledge sets (rules and fuzzy rules in database, knowledge base, expert knowledge base) including different inference algorithm for efficient diagnostic problem solving; to generate powerful hybrid information environment; to use program units in order to simulate sensor activity;

The proposed GTE diagnostic system ports to the various operation systems such ; UNIX or Windows.

Comparison of basic features