Causality Analysis of Industrial Process Based on Improved Convergent Cross Mapping

The equipment of an industrial process is interconnected and coupled with each other. The abnormality of one equipment often causes multiple alarms. How to analyze the real cause of failure from a wide range of alarm messages is an urgent problem to be solved in modern industrial process. Causality analysis will be conducive to assess the causal relationships between devices, trace the path of disturbance propagation, and locate the root cause of the failure. In this paper, a causality analysis framework based on convergent cross mapping is proposed. In the proposed framework, the Cao algorithm is first used to determine the phase space embedding dimension, and then convergent cross mapping is executed and a hypothesis test method is employed to detect whether the convergence condition is satisfied, finally the time delay between variables is detected to distinguish between unidirectional strong coupling causality and bidirectional causality. The effectiveness of the algorithm is verified by a numerical case firstly, and then the algorithm is applied to the continuous stirred tank heating process to analyze the causal relationship between the key variables in the process.