Reasearch On Condition Monitoring And Fault Detection Of Coal Mill Based On First Principle And Real Time Data

Coal mill is the core auxiliary device of pulverized coal preparation system, which has a significant influence on the safe and economic operation of power plant. Therefore it is necessary to study the condition monitoring and fault detection of coal mill for coal mill control and safe operation.To achieve accurate condition monitoring and diagnosis of the mill failure, a model that accurately describes the dynamic beheavior of coal mill should be established. However, coal mill is a nonlinear time-varying system with large inertia, which makes it difficult for conventional data driven method to build a precise model. Grey-box modeling method considers the first principles and real time operation data of coal mill simultaneously and results in better modeling accuracy.Conventional coal mill condition monitoring method is extended kalman filter (EKF). However in practice, due to the time-varingcharacteristic of industrial process and presence of process noise and measurement noise, the state estimate of EKF contains noise content and could be divergent, fluctuating and imprecise. This is undesirable for coal mill control and fault detection. Therefore in this paper, a more robust and precise state estimation method, moving horizon estimation (MHE), is studied.Coventional condition monitoring method considers the coal mill as a whole nolinearsystem; in this work we consider the mass balance part of coal mill as a linear time-varying system and the energy balance part of coal mill as a nonlinear system. In this way, the estimation precision is improved and the computation expense is reduced which makes it possible for online implement of MHE.Both fault data and normal operation data is required, if the pattern recognition based fault detection method is applied.In this paper, the modeling of faults and the residual based fault detection method is studied, which only requires normal operation data. The residual is analyzed using statistics method which could control false alarms and detection sensitivity, and Hotellling T2 statistic is employed as metric for fault alarm. The application of multiscalewavelet analysis to fault classification is studied.The content of this study is as follows:1?Based on the mass balance, energy balance and momentum balance of coal mill, a grey-box model of coal mill is established. The parameters of grey-box model are identified using real time data and genetic algorithm. The identified model is verified by real time data from different operating condition.2?This paper analyzes the important unmeasured parameters of coal mill and their influence on coal mill operation. Based on the analysis, EKF and MHE algorithm is studied for condition monitoring of the grey-box model. Firstly, the coal mill is considered as a nonlinear system.Through simulation, MHE proves to hve better performance than EKF.Secondly, the coal mill is divided into a linear time-varying part and a nonlinear part. MHE is applied to both of the two parts for condition monitoring, and results are tested by simulation.3?The characteristics of common mill faults are analysed and the dynamics of faults are simulated.Residual is analyzed using statistical methods to raise early faults alarms. After faults alarm, outputs residual is decomposed using multiscale wavelet analysis and the fault is classified using residual trend.