Data Driven Based Alarm Analysis In Process Industry

The process correlation information indicates the interaction among the different variables. In this paper, the data-driven interpretative structural modeling method and correlation based alarm thresholds optimization are integrated. In alarm systems, when the alarm thresholds are changed, the corresponding alarm series will be also changed, leading to the different relationship of the alarms, so the objective function is established by correlation consistency of the cross correlation coefficient between the process data and the alarm data for the correlated variables. During this procedure, the process correlation information and the selection of variables are determined by the Interpretative Structural Modeling method.The contents of this paper are as follows:(1) Studied on the data-driven Interpretative Structural Modeling method. For the industrial process measurement data, the adjacency matrix and the reachability matrix are generated by partial correlation coefficient, describing the relationships among the variables.(2) The alarm thresholds optimization method which is based on correlation analysis is also be studied in this paper. With the parameters of the process data estimated by non-parametric kernel density estimation, the Pearson coefficients are obtained and compared with the correlation coefficients of process measurement data to establish the analytical expression for the thresholds optimization. And the particle swarm optimization algorithm is used to find the optimal solutions.(3) In this paper, a multi-variable thresholds optimization method is proposed based on the interpretative structural modeling method and the correlation based thresholds optimization. A cased study of Tennessee Eastman process is used to demonstrate the efficiency and effectiveness of the proposed method. Compared with the 3σ method and univariate alarm threshold optimization method, this proposed method can determine several thresholds at the same time with the minimum difference between the process data correlations and the alarm data correlations, and it has the less false and missed alarms.