Data analysis and processing for the system reliability neural network based on expert judgment

The article presents a data analysis and processing for tuning artificial neural network (ANN) of the anthrop technical system reliability, based on the opinions of experts. In general, the system reliability parameters are functions of operands – physical values – like time to failure, time between failures, duration times of specific reliability or operational states, number of failures in a time interval (event frequencies). These values are easier to be determined by an expert – operator with long year experience – than probabilistic model parameters. It is suggested that they be used in elicitation, for example linguistic estimates of the shares of reliability system elements in the system failure frequency. The numerical – linguistic elicitation of these opinions was carried out, which turned out to be uncorrelated and not suitable for tuning the network. Data processing method was used with the appropriate adopted analytic hierarchy process (AHP) geometric scale and matrix approximation method evaluations (logarithmic least squares method). Correlation analyses were performed for received input and output data of network and error of data processing method was determined. The results are shown in the example of elicitation and data correlation analyses for tuning the reliability neural network of the ship propulsion system.