USE OF NEURAL NETWORKS IN NUCLEAR-POWER-PLANTS

The application of neural networks, alone or in conjunction with other advanced technologies (expert systems, fuzzy logic, and/or genetic al gorithms), to nuclear power plants has the potential to enhance the sa fety, reliability, and operability of these systems. The work describe d here deals with these power plants or parts of these plants that can be isolated. Typically, the measured variables from the plants are an alog variables that must be sampled and normalized to expected peak va lues before they are introduced into neural networks. Often data must be processed to put it into a form more acceptable to the neural netwo rk (e.g., a fast Fourier transformation of the time-series data to pro duce a spectral plot of data). The neural networks are usually simulat ed on modern high-speed personal computers or work stations that carry out the calculations serially. However, it is possible to implement n eural networks using specially designed microchips where the network c alculations are carried out in parallel, thereby providing virtually i nstantaneous outputs (microsecond response times) for each set of inpu ts. Specific applications described include: transient identification, plant-wide monitoring, analysis of vibrations, and monitoring of perf ormance and efficiency.