Data Driven Methods for Analysis and Design of Industrial Alarm Systems
An alarm system is an integral part of process monitoring and safety. A poorly configured alarm system can sometimes cause more harm than good, by introducing many false and nuisance alarms for the operators. Various standards on alarm system management and rationalization suggest many configuration methods to help in improving the overall performance of alarm monitoring systems. In this thesis, the problem of analyzing and designing alarm systems for both single-and multi-mode processes is
... de processes is considered. A design procedure of a multivariate alarm system for multi-mode processes is developed. A hidden Markov model based modeling approach is adopted to capture the multi-modality of data and the mode-reachability constraints of a multi-mode process. A monitoring index utilizing the proposed two-step Viterbi algorithm is developed, and for fault isolation, reconstruction based contribution plots are used. The utility of delay-timers in improving existing univariate alarm systems for multi-mode processes is studied. A mathematical model is developed to calculate analytical expressions for different performance indices (the false alarm rate, missed alarm rate, and expected detection delay). A particle swarm optimization based method is proposed for designing delay-timers, while satisfying the constraints on the performance indices and delay-timer lengths for various modes of the operation of a process. The analysis and design of time-deadbands for univariate alarm systems ii is also considered in this thesis. In particular, a Markov chain process based mathematical model is developed to capture the time-deadband configurations for single mode processes. Analytical expressions for the performance indices are calculated, and design procedures based on process data and alarm data are developed.