基于动态模拟的精馏塔故障诊断

Fault diagnosis of rectifying tower using unsteady stage model

提出了一种基于动态模拟的精馏塔故障诊断方法。该方法从现场采集数据,同时利用动态模拟获得这些数据的模拟值,将二者之差作为状态监测目标。为减少监测变量个数,将这些残差针对模型误差进行校正,并经标准化后形成T^2统计量及其阈值。如果该统计量低于阈值,则说明装置运行正常;否则,说明装置出现了故障,此时启动故障诊断算法。故障诊断通过动态模型的在线参数估计完成,这些参数代表了设备内在故障原因和未知的外界干扰,从这些参数的变化过程就可以分析出故障的原因。参数估计算法基于参数对残差的灵敏度分析来构建,以残差的平方和为优化目标,通过非线性最小二乘法来实现。将该方法应用到了田纳西-伊斯曼仿真流程(TEP)中的汽提塔,分析了故障1时T^2统计量的变化,并给出了故障7时的塔底进料损失参数变化。应用结果表明,该方法具有较为灵敏的监测特征,并可以较为准确地给出故障的原因。

We develop a novel model-based fault detection and diagnosis method for rectifying tower. The method collects sampling data from equipments while producing their simulation values through dynamic simulation, and then uses their deviations as state inspection index. To reduce number of these variables, they are corrected with model error, and standardized into one T2 statistic. If this statistic is within its threshold, equipment is indicated running normally; otherwise, equipment is indicated as out of work, and fault diagnosis procedure is activated thereafter. Diagnosis runs as a kind of parameter estimation, and these parameters represent inner malfunction reason in equipments and unknown disturbs from outside. Estimation algorithm is formed according to the sensitivity of each parameter on inspection index, and is realized via nonlinear least square method with quadratic sum of inspection deviations as optimization objective. The distillation model is built on the mass and heat balance, and is simulated using two tier approach. This method is applied to the stripping tower of Tennessee Eastman Process （TEP）, and the T2 variation under fault 1 and pressure loss coefficient change in bottom feed are analyzed. Case studies show that our method has characters as follows： （1） fault inspection is sensitive; （2） computation load of fault inspection is low; （3） the diagnosis result is more meaningful.