Methods of Pattern Extraction and Interval Prediction for Equipment Maintenance

Abstract

Maintenance interval is one of the most important indexes in equipment maintenance strategy. In the traditional planned maintenance strategy, maintenance interval is often predicted by the way of statistical theory. This method lacks flexibility and can not adjust maintenance intervals according to the actual situation of maintenance. It will easily lead to an under-maintenance or over-maintenance. In order to carry out individual maintenance, in this paper, we use BP Neural Network to predict dynamically maintenance intervals. At first, we extract a lot of maintenance models which exist in the historical maintenance data, and then use these models to train the BP neural network, finally use the trained BP neural network to predict the maintenance interval according to the equipment maintenance model. This method considered the past maintenance factors and made maintenance interval better. The experiment shows that this method achieved 27.1% average relative error of patterns. The dynamic maintenance interval makes the amendment of maintenance interval more scientific for individual strategy.