Analysis And Design Of Data-driven Multi-variables Alarm System

Alarm system plays an important role in ensuring product quality,equipment running normally and the safety of person and property.A well-designed alarm system provides operators with timely and valuable information.However,there are a large number of invalid alarms and consequential alarms in the current alarm system,which seriously affects the reliability of the alarm system.This thesis focus on the two main reasons for generating too many alarms,alarm threshold incorrectly designed and the coupling between multiple variables,and summarize the existing research,identify the deficiencies and improve them.For incorrectly design the alarm threshold,this thesis studies the characteristics of process variables under time-invariant conditions and time-varying conditions,derives the formula for calculating the false alarm rate and missed alarm rate under various conditions and presents the direct method for optimal alarm threshold under different conditions.The design steps and calculation of the optimal threshold of the alarm system are given under the condition of considering the moving average filter.The feasibility and effectiveness of the method are verified by simulation experiments.To addressed the problem of coupling between multiple variables,this thesis use the time delay estimation method,and propose an improved method of constructing the topology of the process to find the root alarms.First,using the alarm data to build a causal matrix between multiple variables.Then using the process data to create another causal matrix for auxiliary correction.Combined with known process knowledge,a more comprehensive and accurate process topology can be established finally.This topology can locate alarm roots intuitively and eliminate consequential alarms in time.Two case studies based on TE process prove the practicability and accuracy of the method.