A Dynamic Failure Rate Prediction Method for Chemical Process System under Insufficient Historical Data

Because of large number of equipments, long pipelines and complex process, the chemical process system is especially prone to accidents due to component defects and equipment failures, making it tough challenging to guarantee process security. According to the relationship between process state parameters and fault conditions in chemical process, and taking into account that in some circumstance the historical fault data is insufficient to perform statistical analysis, this paper proposes a method to predict the dynamic failure rate of chemical process system based on BP (back-propagation) neural network and two parameter Weibull distribution. First, the BP neural network is applied to expand the limited amount of data of process state parameters and determine the fault states. Then combining the expanding data set, some mathematical methods are applied to determine the parameters of Weibull distribution and the failure rate function, based on which the mean failure rate can be calculated for each phase. A liquid chlorine storage system of a chemical plant is introduced to demonstrate the proposed method. Compared with the traditional method and the method that merely considers known limited amount of fault data into two-parameter Weibull distribution, the results show that the failure rate of the liquid chlorine storage system calculated by the proposed method is more consistent with the actual situation, especially more closer to the actual failure rate of the system in the early stage. Moreover, based on the expansion of historical data, this method can achieve a continuous dynamic prediction for future failure rate and failure time points, which has practical meaning for the prevision and dissolving of accident risk in chemical plants.