Fujitsu Ltd has announced an artificial intelligence system claimed to combine the benefits of neural networks and of expert systems: NeuroLib/L is a hierarchical neural network library developed in conjunction with the Eshell/SB inferencing expert system development tool that can take data in real-time from sensors such as measuring and control devices and process the data as a human would: the system is pitched at applications such as fault diagnosis systems using spectral or sensor data, for label or defect detection in a production line and for control of plant operation and features of NeuroLib/L include a four-layer network of 1,000 neurones combined with the inferencing capability of Eshell/SB, which enables an optimal plant control system with trend analysis to be performed by neural network, with decisions on the basis of the analysed data being performed by the Eshell inferencing system; the neural network learns through the recognition of patterns and programs written in Fortran or C can make use of the NeuroLib/L neural network library; connection with the Fujitsu FMR series NeuroSim/L personal computer neural network learning system is possible; the system will be available on Fujitsu’s A series Unix workstations from mid-December at $5,600 and the company is hoping to sell 300 of the systems over the next three years.