Condition monitoring is the term used to describe a combination of techniques for implementing a condition-based maintenance strategy on industrial machinery. Data such as vibration levels (both overall and in terms of frequency spectra), temperature, oil analysis etc, are acquired from plant, and analysed to determine the condition of that plant at the time of measurement. Software packages provide a graphical display of the data, and most provide some form of diagnostic tools to assist engineers in performing data analysis. Rule-based expert systems are available to perform machinery defect diagnosis, with varying degrees of automation and human interaction. However, such systems have inherent problems, such as their ability to deal successfully only with clearly defined problems within a narrow band of parameters, and their inability to cope with contradictory, incomplete, or noisy data -just the type of data found in many real-world applications. This paper describes the implementation of an off-line condition monitoring system at Blyth Power Station, one of the stations owned by National Power in the United Kingdom. It explains the application area and the type of data acquired. The paper then goes on to describe the neural network models which have been developed to analyse condition monitoring data, both at Blyth Power Station, and by the Neural Applications Group at Brunei University. Transactions on Information and Communications Technologies vol 6, © 1994 WIT Press, www.witpress.com, ISSN 1743-3517
Read full abstract