Article

An expert system concept for diagnosis and monitoring of gas turbine combustion chambers

Instituto Superior Tecnico, Mechanical Engineering, Av. Rovisco Pais, 1049 Lisbon, Portugal; Aristotle University of Thessaloniki, 2461 Kozani, Greece; All Russian Thermal Engineering Institute, 095 Moscow, Russia; University of Aveiro, 234 Aveiro, Portugal
Applied Thermal Engineering 12/2005; DOI:10.1016/j.applthermaleng.2005.04.020

ABSTRACT In this paper, the main principles of operation, the conceptual design and the development of an expert system for fault diagnosis and monitoring of gas turbine combustion chambers are presented. The concept of the gas turbine chamber expert system is based on the monitoring of the spatial and temporal distribution of the heat flux inside the combustion chamber and the simultaneous comparison of respective readings of diagnostic variables with values obtained through numerical model simulation of different situations which lead to the malfunction of combustion chamber and deterioration of its performance. The demonstration of the expert system will be performed at the VTI (All Union Energy Institute, Russia) gas turbine combustion chamber test facility.

0 0
 · 
2 Bookmarks
 · 
144 Views
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: In the modern world, computing is essential in all aspects of manufacturing activity. Computers have brought to life terms like artificial intelligence, and have played a critical role in reinvention of manufacturing industry. In continuing quest to decrease the interval time between conceptualization of a product, infor-mation technology has been fused with manufacturing practice. This paper describes the use of an expert system shell to develop a knowledge-based system (KBS) for an automatic wire bonding machine in the hi-tech semiconductor industry. The main aim of the KBS is to diagnose and solve the problem of an automatic wire bond machine. The diagnosis method was based on maintenance requirements such as measurement, machine, and material. In the semiconductor industry, production equipment and machine have depended heavily on the use of human expertise for maintenance and it is costly. Without a knowledge-based system, their experience will be lost when they are unavailable or resign. With the developed KBS, the diagnosis process for the automatic wire bond machine is standardized and accuracy will be increased compared to the conventional way. Therefore, the quality of products that are produced will improve. The constrain values for KBS are based on the design data and experience of the engineer. The KBS is to improve bonding quality by reducing the production yield loss.
    Env. Sci. 01/2008; 32:235-244.
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: Esta tesis estudia los problemas relacionados con la alta dimensionalidad de los datos en un contexto científico de teledetección, con el fin de estimar perfiles de temperatura en el interior de nubes gaseosas a alta temperatura (como es el caso de una llama). El objetivo principal es identificar los problemas de las técnicas existentes en este contexto práctico y proporcionar soluciones. Para ello se realiza una introducción a los retos presentes en los datos de alta dimensionalidad, y al área de minería de datos que es actualmente la más activa en el estudio y tratamiento de este tipo de datos. La reducción de dimensionalidad aparece como un proceso necesario para solventar algunos de los retos planteados y mejorar el rendimiento de los algoritmos de aprendizaje. El resto del trabajo está dividido principalmente en dos partes. Cada una de estas partes desarrolla un camino alternativo para reducir la dimensionalidad de los datos y solucionar así los problemas relacionados con la alta dimensionalidad en el contexto de teledetección. En el primero de ellos, el trabajo se centra en la selección de características no supervisada para buscar la información relevante a la aplicación. El principal problema en la selección de características es la imposibilidad de realizar una búsqueda exhaustiva debido al gran número de posibles soluciones. Por esto, se propone el uso de conocimiento previo específico de la aplicación física a tratar, para guiar el proceso de selección. Los resultados obtenidos muestran que esta solución mejora los resultados en un entorno de selección no supervisado, o frente a la ausencia de selección. La segunda parte de esta tesis se centra en la reducción de dimensionalidad desde un punto de vista de extracción de características. En ella se trata de abordar uno de los problemas principales relacionados con la alta dimensionalidad, la multicolinealidad, buscando extraer de un modo supervisado los conjuntos de datos que mantienen un comportamiento similar u homogéneo. Esto va a permitir diferenciar diferentes grupos de datos y, lograr con esta división, aplicar modelos de estimación especificos para los diferentes grupos. La aproximación se basa en estructuras de grafos para incluir la información local de los datos, lo cual es muy útil en nuestra aplicación. Esta solución muestra mejoras significativas en los resultados obtenidos, a la vez que permite obtener estimaciones precisas para los nuevos casos. Además, también posee una interpretación física y ayudará a un mejor entendimiento de la aplicación estudiada.--------------------------------------------------------------------------------- This thesis studies some of the problems related with high dimensional data in a scientific context, pursuing the estimation of temperature profiles inside a hot gas cloud at high temperature (as it occurs inside a flame). The main objective is to identify the main disadvantages of the actual techniques in this practical context and to provide solutions to them. For that purpose we introduce currently known challenges related to high dimensional data, and to data mining field which is the most active regarding the study and processing of this type of data. The dimensionality reduction appears as an important step to solve some of the established challenges and to improve the performance of machine learning algorithms. The work is mainly divided into two parts. Each one of them develops an alternative to reduce the dimensionality of the data solving some of the problems related to high dimensional data in a remote sensing environment. The first one, focuses on unsupervised feature selection to search for relevant information to the application. The main problem in feature selection is the impossibility to do an exhaustive search due to the huge number of possible solutions. Thus, we propose to use specific physical previous knowledge to guide the selection process. The obtained results show that this solutions improves the results obtained in an unsupervised framework or against non-selection. The second part of the thesis is focused in dimensionality reduction from a feature extraction point of view. In it, we try to solve one of the problems related with high dimensionality data, the multicollinearity. For that purpose we extract, in a supervised mode, subsets of data which have similar behavior or are homogeneous. This allows to find out different groups of data and, with this division, to apply specific estimation models for the different discovered groups. This dimensionality reduction approach is based on graph structures which is useful to include local similarity information about the data, which is extremely useful in our application. This solution shows significant improvements and allows better accuracy for new samples. Furthermore, it also has a physical interpretation and it enables a better understanding of the studied application.
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: Condition monitoring and fault diagnosis are of fundamental importance for many industrial systems. In the last decade, substantial research efforts have been made on the surveillance and diagnosis systems for different types of equipment, with the approach of integrating information technologies and intelligent computing methods. This paper presents the conceptual design of a distributed information system of condition monitoring and fault diagnosis for a growing number of gas turbine-based power generation systems. Each individual information system that monitors a specific gas turbine system, locally deployed in a power plant, is linked to another information system, deployed at the manufacturer's site, which oversees all the gas turbine systems in parallel. The systems are constructed on the basis of COM components, which are conceptually separated into three tiers. Subsequently, this paper proceeds to present a generic business domain model with components encapsulating physical entities of interest. Finally, this paper addresses the interactions among components, which considerably affect the performance of the system including efficiency and effectiveness. It has been identified that both asynchronous and synchronous communication mechanisms are required for exchanging information for different scenarios. COM+ services are also required for supporting object pooling, transaction coordination, and security control.
    Information Systems. 01/2007;

Full-text (4 Sources)

View
410 Downloads
Available from
Nov 30, 2012