The UK’s Department of Trade and Industry is investing UKP5.75m in a three year initiative aimed at promoting awareness of neural computing techniques among British companies. The programme, which is being managed by consultants Touche Ross, will combine a series of promotional activities – workshops, exhibitions, production of a range of reference literature and a hotline service – with the setting up of six special user and supplier clubs that will develop pilot neural network applications. The research is to be jointly funded by the club members, who have so far allocated UKP2m, and which has been matched by government funds. Work will be carried out on applications for the following areas: automatic identification; financial forecasting; industrial information processing; process control and monitoring; condition monitoring and sound assessment; and neurodata, which is concerned with data exploration or mining. It is hoped that some 6,000 organisations could eventually benefit from involvement with the project. Closing the perceived gap between the UK and the US and Japan, which are believed to be three years ahead, is another aim – though that is not to say that the UK is trailing in all areas. Les Wray of Stockport-based Recognition Research Ltd reckons the UK has the lead in video monitoring, for example.

Emulation

Although some of the participants develop hardware Hampshire-based Neural Technologies Ltd and Gloucester-based Micro Circuit Engineering Ltd have collaborated to develop a Neural Instruction Set Processor – the project clubs will mostly concentrate on software development. This reflects what several of the vendors argue is a more-cost effective approach to neural computing, involving the emulation of neural nodes in software. Recognition Research for example has a Windows-based package, Autonets, which can run on a personal computer. Under the government scheme, Recognition Research will jointly head the ‘neurodata’ club with software consultancy Logica Cambridge Ltd. The two are aiming to develop three demonstrators for advanced pattern recognition systems that could be used for data analysis. The systems will involve the processing of database records through a neural network, which can then reveal patterns and relationships between data. These will then be combined with other statistical and advanced graphics techniques that will enable users to visualise the model generated by the network. It is anticipated that the systems could be useful for a variety of applications including fleet management, customer credit control, plant loading, targeting direct mail campaigns, fraud detection, retail and materials analysis. One important use for the last is the use of ‘multi-phase flow analysis’ for the oil industry. This technique involves the use of a sensor which is attached to the oil pipes. The sensor feeds the network with data about the flow of material through the pipes, which is then be analysed to determine the proportions of oil to water and sand.

By Lynn Stratton

Membership fees for the club are UKP5,000. Meantime, a consortium of developers headed by AEA Technology, the trading name of the UK Atomic Energy Authority, is working on the use of neural networks for noise measuring, condition monitoring of rotating machinery, alarm detection and fault diagnosis. This is crucial for the construction of aircraft cabins and trains where high levels of noise and vibration are unacceptable, and similarly for motor vehicle design, where noise management is key for competitiveness. The bulk of the technical work involved in the project will be carried out by the Motor Industry Research Association, Southampton University’s Institute of Sound and Vibration Research and Oxford University. Club membership is UKP12,000. Research into the use of neural networks for engineering processes including fault detection and diagnosis is being headed by Surrey-based EDS-Scicon Ltd in conjunction with Newcastle University’s Department of Chemical and Process Engineering. Neural networks are useful for process contr

ol because they can interpret so-called soft sensor readings – the inferred measurements from groups of sensors; can cope with complex, variable systems; and can be used to build statistical models. EDS-Scicon and Newcastle University will use the networks in oil, chemical, bio-chemical, pharmaceutical, water, power, food, beverage, metal producing and processing, paper and glass industrial processing systems like distillation, fermentation and melting. Interested parties have the choice of full club membership for UKP10,000 or associate membership for UKP2,500. Combined talents from Leatherhead, Surrey-based ERA Technology Ltd and Kings College, London are working on neural information processing systems, in particular for monitoring, sensory, control and image analysis. Membership is open to those with a general interest in using neural techniques, those wanting more detailed involvement, and to hardware and software developers. Financial forecasting is the focus for the neural systems to be co-developed by London Business School and University College in London. They will be researching systems for predicting fluctuations in the foreign exchanges, for stock and bond valution, for commodity price prediction, for tactical asset allocation and for risk and liability management. Such systems require expertise in both financial engineeering – the choice of economic variables, correlation and use of multi-parameter data sets – as well as network engineering.

Proven

Still, they have proven themselves to be extremely effective, when compared with conventional methods particularly because of their speed. As a result, they have attracted vast government investment in the US, Japan and Europe too. Club members, which are likely to be drawn from banks, insurance companies, securities houses – and indeed any large company with extensive overseas dealings – will be able to second staff to the London Business School during development work to gain hands-on experience. Finally, there is the Axon club, jointly managed by Keele University, Micro Circuit Engineering Ltd, Neural Technologies Ltd and the UK Trade Association for Automatic Identification Manufacturers. Their aim is to develop three demonstrator systems for small to medium sized businesses. These include real time data analysis which can be used for speech, vibration and signal analysis; pattern recognition for finger print, face and signature recognition; and intelligent data management for control systems, for instance multiple sensor analysis, fault and trend detection, management statistics and transaction processing. Membership costs UKP5,000.