Elsevier

Applied Energy

Volume 86, Issue 1, January 2009, Pages 9-17
Applied Energy

Development and multi-utility of an ANN model for an industrial gas turbine

https://doi.org/10.1016/j.apenergy.2008.03.018Get rights and content

Abstract

Demonstration of different utilities for industrial use of an artificial neural network (ANN) model for a gas turbine has been reported in this paper. The ANN model was constructed with the multi-layer feed-forward network type and trained with operational data using back-propagation. The results showed that operational and performance parameters of the gas turbine, including identification of anti-icing mode, can be predicted with good accuracy for varying local ambient conditions. Different possible applications of this ANN model were also demonstrated. These include instantaneous gas turbine performance estimation through a graphical user interface and extrapolation beyond the range of training data.

Introduction

The power and combined heat- and power (CHP) sectors all over the world have gone through major transformations over the last two decades due to rising competitiveness for deregulation of electricity and more stringent laws for environmental protection. Increasing the reliability, availability and maintainability of existing plants is a present need to improve the performance and minimize the environmental impact. Conventional simulation and performance monitoring as well as maintenance schedules need significant improvement to cope with this present need. Advanced simulation tools and condition monitoring systems are thus very crucial for modern plants. Some simulation and condition monitoring tools use heat- and mass balance programs [1], which often rely on physical and thermodynamic laws. These are very powerful regarding thermodynamic studies and the estimation of parameters (efficiency, power output etc.). However, they require significant expertise and man-hours for monitoring the condition of the plant and locating faults. Several artificial intelligence (AI) based tools are available [2], [3]. ANN is one such AI based tool suitable for e.g. simulation and condition monitoring of power and CHP plants. ANN is a computer algorithm that can learn patterns through proper training, i.e. adaptive [4]. Because of this feature, these are often well suited for modelling complex and non-linear processes of real life. Feasibility of ANN applications at the process/component level [5], [6], [7] as well as at the plant/system level [8], [9], [10] has been reported by several authors. These are considered to be modern ‘high value, low cost’ IT-based ‘intelligent’ tools to substitute for conventional simulation and condition monitoring tools. Thus ANN has been demonstrated to be a useful technique for accurate and realistic modelling of real life plants aiming at better prediction of performance and environmental impacts.

Researchers at the division of Thermal Power Engineering at Lund University has carried out several investigations [1], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21] in this field and ANN has shown to be a good candidate for fault diagnosis, process identification and modelling of non-linear systems in the energy field and this work is a continuing study of those previous studies. In this work a demonstration project has been reported in collaboration with Lunds Energi, Lund, Sweden. Operational data from a gas turbine based cogeneration unit at Gunnesbo, Lund, Sweden has been supplied by Lunds Energi. Researchers at the division of Thermal Power Engineering were assigned with the task of developing an ANN model with this data to demonstrate the multi-utility of such a model for real life plants. The novelty of this study lies in that a complete cycle, from data acquisition, data screening, ANN training and evaluation of developed ANN model, development of a graphical user interface and delivery of a finished product to the plant, has been completed. Real life aspects, such as anti-icing operation, have also been taken into account. In previous studies simulation data was often used to establish the use of ANN for modelling of power plant systems rather than focusing on delivering an utilizable product.

It was decided that initially a very simple model would be developed with local ambient conditions (pressure, temperature and relative humidity) as only inputs to the model. Output parameters were decided on the basis of availability of plant data and real life needs. The utility of this model was demonstrated with respect to very accurate prediction of the operational parameters with the variation of local ambient conditions and extrapolation beyond the range of training data. Developed model can be used for continuous condition monitoring of the plant by comparing the actual sensor data during operation of the plant with the predicted data from the model. Moreover, all the input data (local ambient conditions) may be assumed without operating the plant and an estimation of expected performance of the plant can be done before actually starting, which means that the model is even useful for offline applications. The plant has to be operated with anti-icing measures during simultaneous cold and humid ambient conditions to avoid the very harmful possibility of ice formation on the aerofoil blades in the compressor. Switching over to anti-icing and coming back to normal mode of operation has also been modelled with ANN and integrated to develop a unified ANN system for all possible modes of operation. Finally an easy-to-use graphical user interface has been developed so that the model can be used by any person for monitoring, simulation or learning of the plant basics.

Section snippets

Brief basics of ANN

ANN is a simulation tool that mimics the neural structure of the human brain [4]. The brain basically learns from experience. In contrary to traditional mathematical models, which are programmed, ANN learns the relations between selected inputs and outputs through an iterative process called training.

ANN consists of a number of interconnected artificial neurons with linear or non-linear transfer functions and is well capable of predicting non-linear behaviour of a system. The multi-layer

Brief description of the gas turbine and CHP plant

The schematics and main operating parameters of the CHP plant, with its gas turbine and heat recovery unit for the district heating system, is shown with a screenshot from an operator station in Fig. 1. The gas turbine was installed in 1991 by ABB STAL, now Siemens Industrial Turbomachinery. It was originally a GT10A but was rebuilt to GT10B in that same year. The GT10 (now known as SGT600) series gas turbines are lightweight industrial gas turbines. These are designed and developed to

ANN model development

The overall objective of this work was to demonstrate the practical usefulness of ANN modelling for an existing gas turbine in a cogeneration plant. Scopes for using ANN for multi-utility objectives for a real life plant with modest cost and effort were demonstrated by this work. For the experienced ANN modeller approximately one month is required for data collection and filtering, system studies and training of the neural networks. Operational data was obtained from the plant owner and used

Development of the graphical user interface

The idea of developing a graphical user interface (GUI) for the ANN model was to make it user-friendly so that any person without the formal knowledge of either the plant or ANN can also use the model to estimate plant performance. The GUI was created in an Excel environment as most people are familiar with this software. As the training of the ANN was over, the weights of this final ANN was used in visual basic to generate functions for the prediction of output parameters for a given set of

ANN extrapolation capability

Another utility of the developed model was shown for the extrapolation of predictions even beyond the range of training data. The collection of input data for varying ambient conditions can be difficult as it might have to be recorded over the whole year. If the prediction could be done with the trained ANN for those ambient conditions for which data was not available the ANN model could be more useful. This feature has been tested for this developed model. The range of ambient pressure and

Conclusions

Artificial neural network is found to be a useful tool for prediction of gas turbine performance if it can be trained properly with operational data. This is demonstrated by very high prediction accuracy of the developed ANN model. Developed ANN model may have several utility. It may be used for offline simulation of gas turbine performance or online condition monitoring of the gas turbine for early detection of faults or degradation. It may also be used for sensor validation purposes.

In this

References (24)

  • S. De et al.

    Development of an artificial neural network model for the steam process of a coal biomass co-fired combined heat and power (CHP) plant in Sweden

    Energy

    (2007)
  • Olausson P. On the selection of methods and tools for analysis of heat and power plants, doctoral discussion, Lund...
  • E. Rich et al.

    Artificial intelligence

    (1991)
  • S. Russell et al.

    Artificial intelligence: A modern approach

    (1995)
  • S. Haykin

    Neural networks, a comprehensive foundation

    (1999)
  • DePold H, Douglas Gass F. The application of expert systems and neural networks to gas turbine prognostics and...
  • Embrechts M, Schweizerhof A, Bushman M, Sabatella M. Neural network modelling of turbofan parameters. In: ASME turbo...
  • Bettocchi R, Spina P, Torella G. Gas turbine health indices determination using neural networks. In: ASME turbo expo,...
  • R. Perryman et al.

    Condition monitoring and predictive analysis of combined heat and power systems

    (1994)
  • C. Boccaletti et al.

    A neural network simulator of a gas turbine with a waste heat recovery section

    J Eng Gas Turbine Power

    (2001)
  • G. Ferrettl et al.

    Estimation of NOX emissions in thermal power plants using neural network

    J Eng Gas Turbine Power

    (2001)
  • Mesbahi E, Assadi M, Torisson T, Lindquist T. A unique correction technique for evaporative gas turbine (EvGT)...
  • Cited by (127)

    • Recent advancements in sustainable aviation fuels

      2023, Progress in Aerospace Sciences
    View all citing articles on Scopus
    View full text