Fault detection using probabilistic prediction and data fusion on a bulk good system

Industrial companies worldwide want to keep a steady performance of their processes. In order to ensure this, a company usually implements a continuous condition monitoring for their machines. Ideally, the company analyzes the resulting data and obtains an insight on the failure causes of the machines. Unfortunately, the data itself cannot always discover the hidden causes for a fault in the machine, due either to the big amount of data or its complexity. This paper applies probabilistic prediction and data fusion techniques for fault detection on a bulk good system. The data acquisition from the bulk good system is implemented using the OPC Unified Architecture (OPC-UA) Machine-to-Machine Communication Protocol. OPC-UA collects data from the automation platform, and stores it in batches. Each batch contains all system features. Firstly, the system data is analyzed by means of a centralized approach. For that purpose, the probabilistic methods Naïve Bayes and Full Bayes are implemented. Furthermore, a decentralized approach is implemented using a two-step method. The first step gathers the health status of the main components by means of a local analysis. The second step fuses the results of each component, in order to obtain an overall status. The results show that the data fusion approach improves the performance of the fault detection algorithm.