As one of the protection layers, alarm systems play an important role in process safety. The performance of these systems faces various challenges in process industries, such as false alarms, missed alarms, repetitive alarms, and detection delays. In this thesis, the aim is to present algorithms to improve the performance of the alarm system for one of the types of faults, called momentary faults. This fault has a short time duration and can appear individually, periodically, and randomly. Revealing this fault is facing various challenges. Phenomena such as outliers, noise, and oscillation can make the decision-making of the alarm system challenging. To improve the performance of the alarm system for the detection of momentary faults, this thesis presents the up-down counter algorithm for single momentary faults and the counter and latch algorithm for periodic momentary faults. In detecting a single momentary fault, it is shown that the up-down counter algorithm has more accuracy and less delay than the two widely used methods in alarm systems, i.e. filter and delay timer, respectively. In the detection of periodic momentary faults with the help of the counter and latch algorithm, according to the periodic behavior of the fault, the combination of the up-down counter has been used to have a low delay, and the latch has been used for the continuity of the alarm signal. Among the common features of both algorithms is simplicity in design and implementation. The method of detecting the fault in the presented algorithms is to use the change in data distribution. In these methods, it is assumed that the data received from the system in a healthy and faulty state have a specific distribution, and when a fault occurs, the data distribution changes from a healthy state to a faulty one. The mathematical description of the mentioned algorithms is presented using the Markov chain theory and the analytical relations of the alarm system performance eva‎luation criteria, including false alarm rate, missed alarm rate, and detection delay are derived for them. The relationships obtained are verified with the help of Monte Carlo simulation and the process of designing the parameters of each algorithm is presented with an example. The improvement of the performance of the alarm system using the proposed algorithms has been shown compared to some existing methods.

مرضيه كمالي , ايمان ايزدي نجف آبادي

مريم ذكري , فريد شيخ الاسلام