Identification Of Root Causes For Industrial Alarms Based On Probabilistic Graph Network Model

Security issues is a top concern for modern process industries.In the modern process industry towards a large-scale,complex,intensive and sophisticated development trend,scientific and reasonable alarm systems for the maintenance of process safety,product quality and operational efficiency are of great importance.Alarm floods caused by poor design,non-comprehensive maintenance and deficient strategies of alarm system,poses a serious challenge to the security for modern process industries,which has attracted a lots of attention in both academia and industry.Aiming at the issue of alarm floods,the thesis combines clustering analysis and root-cause analysis for alarms to develop a solution with a certain practical value to solve it under multivariate framework.After considering alarm floods as alarm flood sequences,the thesis propose a similarity measure using the Euclidean distance of Discrete Fourier Transform(DFT)power spectrum of alarm flood sequences.Combining the measure with Unweighted Pair Group Method with Arithmetic mean(UPGMA),a DFT-based clustering analysis for alarm flood sequences is developed to discover inherent patterns and rules of alarm flood sequences and further find potential alarm root-causes under the abnormal situations.In order to establish causal network model more accurately and reduce the computational cost,a unit division method for process plants is employed divided the process into different units.For each unit,a hierarchy division method for process variables is utilized to assist to design network structure,and then the probabilistic graph network model is used to build the causal network sub-graph.Once all sub-graphs for all units are constructed,the overall causal network model for the whole process is obtained by integrating all sub-graphs.For further identifying the alarm root causes,root-cause analysis is conduct over the overall causal network model,then a Probabilistic Graph network Model(PGM)-based root cause identification is formed to determine root-cause and abnonnity propagation path.