The design and development consultancy Cambridge Consultants Ltd based, naturally enough, in Cambridge recently held an open day at which the message from its information engineering division was that it was working on new computing technologies of future importance for industry. Indeed, the company’s managing director Dr Paul Auton stressed that Cambridge Consultants was not a high-tech, nor a research-based, company. Rather it was an innovative consultancy geared to business needs and the application and development of appropriate technology. Last year the company produced a turnover of UKP11m, 70% of which was derived from the UK with the rest being generated on the continent. Arthur D Little Cambridge Consultants which is part of the US consultancy group Arthur D Little, now has a staff of 260 including 170 qualified engineers and scientists and functions as Arthur D Little’s laboratory base in Europe. To this end it works closely with management consultancies throughout the continent. It has a collaborative agreement with DG Conseil in Paris and co-operates with Attexor in Switzerland, while in Germany it is moving its offices and laboratories from Offenberg to Wiesbaden where it is joining its parent company Arthur D Little. Cambridge Consultants is mainly involved with the defence, telecommunications, healthcare, food, phar-ceuticals and fabricated industrial products sectors, and its projects range from a few days of counselling to large assignments involving teams of 20 for several years. A recent large technology-related project of which Cambridge Consultants is proud is the development for Shell Research Ltd of a fully automated system for screening crop protection agents. The integrated system links up Shell’s DEC VAXes (which hold the database for the screening lines’ compounds) with the local supervisory IBM ATs. The screening lines themselves automatically collate, spray, label and sort up to 10 targets every 90 seconds. Bespoke software, written in C underXenix, enables chemicals and target plants to be tracked throughout the system, checking and recording events for each plant at every stage of the screening process. According to Cambridge Consultants, more than 40 man-years of effort went into the system’s development, including 250 man-weeks of software development to produce the 35,000 lines of code that operate the machines and another 170 man-weeks of effort to interface the chemical and biological data. The contract was worth several million pounds of which software development accounted for approximately 25% with the majority of the cost going on the building of the actual hardware. The company is now looking for further contracts for customised laboratory automation packages. As regards Cambridge Consultants’ information engineering activities, the company was eager to show itself at the forefront of industrial applications in what it termed forward-looking technologies.
The first of these on demonstration was neural networking. Such networks link large numbers of relatively simple computing elements together to perform pattern recognition tasks and have been used in application areas such as voice recognition, yield management, credit scoring, financial trading and process control. Cambridge Consultants’ own demonstration concerned signal classification. Using Nestor Inc’s Nestor Learning System, Cambridge Consultants had developed an application in a matter of weeks to look for sonar signals. The application was designed to help pipeline operators survey miles of sea-bed oil and gas pipelines. By taking a sample sonar reading of the seabed every 6, the application with a 94% correct judgement could find a suspended gas pipe that would be prone to cracking. The Nestor Learning System has only been available since February but Cambridge Consultants are already negotiating contracts to integrate neural networks with computer systems for defence. It is also talking to a well-known UK oil company about the sort of pipeline inspection system it was demonstrating on its Open Day. Cambridge Con
sultants has also developed its own real-time knowledge-based system toolkit called MUSE. The tools were originally developed for the Royal Aircraft Establishment which wanted real-time expert systems that could be downloaded to a target machine in a helicopter. The system consists of a development environment hosted by a Sun workstation, on which prototypes can be developed and partially tested. Software, requiring 4Mb, can then be downloaded from this system to a stand-alone system such as a 68000-based computer. Using MUSE, Cambridge Consultants, with the Cambridge University Engineering Department, the glue company Loctite and the Cambridge-based robotics company Altek have investigated (thanks to funds from the Science & Engineering Research Council) how to control the glue for surface mount components on printed circuit boards. Over the past two years the project has developed a knowledge base for the application and the appropriate sensing technologies, including vision-based sensing of glue droplet size and profile. The knowledge based application is at its pilot stage at the moment, but collaborative plans are afoot to commercialise it before the end of the year. Meanwhile, Cambridge Consultants’ Human Factors Consultancy had an interface for equity traders on show that acts as an intelligent assistant for the trader, identifying real offers from spoof bids, and prioritising a list of suggested brokers for the deal. A variety of bespoke designs on this theme have apparently already been sold in the US. Finally, Cambridge Consultants had a type of technology on display that is still nascent in commercial terms: genetic algorithms. Genetic algorithms These algorithms use a search technique that mimics biological evolution hence their name. The technique starts by randomly seeding the screen search space with a number of individuals which take values or strengths depending on where in the search space they are located. The individuals are then allowed to breed and form new generations. Following the evolutionary pattern, clusters of successful individuals form, representing the optimised evolution of whichever variables the individuals were meant to be representing. The original inputs which are encoded on binary strings apparently manage the variables or individuals as DNA would do, applying a mutation operator to generate new values. At the moment this is a technology looking for problems to solve. It can be used to balance the load of phone networks or, since it is noise tolerant, it can be used in medical image registration to correct distortions in before and after X-ray images when a patient has moved position. The tool demonstrating interactive experimentation with genetic algorithms was written in C++ using object-oriented techniques and is used purely as a research tool at the moment. It runs on Sun workstations but can theoretically be ported to other machines. A mouse is used to draw a function on screen, with size of population, mutation probability and so on controlled by the user.
