: Today, almost all the variables of a process can be measured using sensors. With the development of the technology in manufacturing sensors and the ever growing use of alarm systems in industry, improving the performance of the alarm systems is of great importance. The main responsibility of an alarm system is to inform the operator about the occurrence of an abnormal situation in the process. This situation is studied with comparing the value of a variable in normal case and during operation. When the value of a variable exceeds its normal case value, operator is informed of this situation by an activated alarm. Ideally, operator should receive one and only one alarm for each abnormality. This is almost never the case. Operator receives alarms even in normal operation that are falsely activated. Furthermore, many alarms may be activated due to a single cause, called nuisance alarms. This high number of false and nuisance alarms will disturb the operator and lead to unreliability to the alarm system. Some of the most important factors in measuring the performance of an alarm system are probability of the false alarm, probability of the missed alarm and detection delay. Among various methods to improve an alarm system, using filters to process signal data is commonly used in industry. Moving Average filters and Exponentially Weighted Moving Average filters are the most common filters in this case. In this thesis, the effect of these filters on the performance evaluation factors is discussed. Moreover, the use of these filters introduces delay in detecting faults. This detection delay is calculated in this work. An algorithm to design an Exponentially Weighted Moving Average filter to achieve the value of the parameters to reach to a desirable alarm system is proposed. Key Words: Alarm systems, Alarm system evaluation factors, Discrete-time filter, Moving Average filter, Exponentially Weighted Moving Average filter
