Bayesian Network Building for Diagnosis in Industrial Domain based on Expert Knowledge and Unitary Traceability Data

This paper presents the CBNB (Causal Bayesian Networks Building) algorithm for the causal Bayesian Network construction. This algorithm is designed for diagnosis models in the industrial domain It uses expert knowledge and operates process and product traceability data. The first phase of this algorithm consists of exploiting expert knowledge and properties of the application domain for allocating the variables at different levels of causality. This phase results in a cascade arrangement of the system's variables starting with the root causes and ending with the ultimate effects passing through one or more intermediate levels. In the second phase based on the unitary traceability data, the CBNB algorithm then allows to determine the causal relationships existing between variables. We provide the necessary assumptions and the theoretical justifications for the proposed algorithm. We have conducted empirical studies assessing the ability of the algorithm to provide the true network from synthetic data on three benchmarks whose nodes are arranged in cascade. The results of comparative analysis have shown that the CBNB algorithm outperforms GS and MMHC, two state-of-the art structural learning algorithms in terms of ability to rebuild the true network. (C) 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.