Adaptive Management Strategies Towards Improving The Validity Of Industrail Alarm Systems

Due to the imperfection of alarm systems, the operators are heavily burdened with huge alarm messages that beyond their processing capactity. Therefore, an efficient alarm system is very important for safe operations of the plant. It becomes an important issue to improve the effectiveness of alarm systems by establishing adaptive alarm management strategies with respect to the changing production environment.In responsible to real-time characteristicsand unreasonable alarm setting caused by uncertainty information of industrial process alarm systems, this research improves the fault diagnosis and localization of alarm systemsby alarm correlation analysis concerning alarm information filtering and classificationbefore creating reasonable alarm optimization strategiesto further improve the performance of alarm management systems.The main contents and organization of the thesis are presented as follows.Firstly, an analysis approach to causal and temporal relations of alarm time sequencesis established. The causal alarm correlation is analyzed based on the alarm data, and a fuzzy weighted association alarm rules mining method combing fuzzy sets and Apriori algorithms is proposed. Additionally, alarm propagation paths are established taking advantage of alarm correlation knowledge expressions of time fuzzy Petri nets.Secondly, ananalysis method of real-time alarm state priorities is suggested.Underrandom disturbances, aquantitative alarm risk assessment model and aMarkov alarm risk state transition processare presented. In order to predict the future risk alarm state transition probability, the method of particle filter is used to estimate the dynamic Markov process state transition matrix and alarm state transition time.The alarm priority is determined by the seriousness of real-time states.Thirdly, adaptive management strategies of nuisance alarms are proposed. The different types of online nuisance alarms are determined based on alarm intervals and alarm durations of historical alarm data, creating adaptive calculations of alarm delay timers or alarm deadbands. An ARMA model is established to forecast adaptive alarm deadbands, which is aiming to deal with high frequency alarms. The alarm delays are updated by adjusting alarm time intervals which is used to deal with low frequency alarms. Distinguishing the different alarm generation mechanism, the proposed optimization method can reduce the number of nuisance alarms or eliminate nuisance alarms.Fourthly, anadaptive management strategy of control system alarm thresholds is established. A stochastic optimization constrained approachis introduced to achieve the optimized control retreat values.The variance constraint relationship between control variables and output variablesis determined by solving LQG benchmarks. According to the mandatory and optimization constraints, two levels of constraint involving hard and soft constraints are suggested to adjust the alarm limits, which can guarantee the feasibility and operability of the process outputstogether with the consistency of the double-layer alarm limits.At last, the alarm system of a practical industial process is analyzed and optimized. The effectiveness of the proposed methods is validated through extracting causal alarm association rules, establishing alarm paths and applying adaptive management strategies against nuisance alarms based on historical production data.