国家自然科学基金项目 (61873226, 61803327)；河北省自然 科学基金项目(F2017203304, F2019203090).
The National Natural Science Foundation of China(61873226, 61803327);Natural Science Foundation of Hebei Province(F2017203304, F2019203090).
针对平方和双指数加权移动平均(SS-DEWMA)图数据监控结果受其参数影响较大的问题, 本文提出一种SS-DEWMA图的多目标优化(MO-SS-DEWMA图)数据监控方法, 并将该方法用于非线性系统的输出传感器故障检测. 首先, 基于系统输出和控制输入数据,采用复合嵌入式均方根容积卡尔曼滤波器(CESCKF)对系统状态进行估计, 并产生残差; 其次, 通过对SS-DEWMA图的平滑参数和控制宽度与残差评价(数据监控)指标漏报率(MDR)和误报率(FAR)关系的分析, 以MDR和FAR同时最小为优化目标, 利用多目标粒子群优化(MO-PSO)算法对平滑参数和控制宽度进行离线优化; 其中, 采用小波分析算法对SS-DEWMA 图进行多尺度分解、阈值去噪和重构, 削弱噪声对系统的影响, 再将优化后的SS-DEWMA图(MO-SS-DEWMA图)输出值与阈值比较、在线判断输出传感器是否存在故障. 最后, 针对伺服电机驱动的连铸结晶器振动位移跟踪系统进行仿真验证, 并与现有残差评价方法对比, 结果表明, 本文所提出方法能够更精准地检测输出位移传感器故障, 并能有效降低故障检测的漏报率和误报率.
In view of the problem that the monitoring results of the sum of squares double exponentially weighted moving average (SS-DEWMA) chart will be greatly affected by its parameter, a multi-objective optimization of SS-DEWMA chart (MO-SS-DEWMA chart) data monitoring method is proposed and applied to detect output sensor fault of nonlinear systems in this paper. Firstly, based on the system output and control input data, a composite embedded square-root cubature Kalman filter (CESCKF) is used to estimate the states of the nonlinear systems and generate residuals. Secondly, by analyzing the relationship between smooth parameter and control widths of SS-DEWMA chart and residual error assessment (data monitoring) MDR and FAR, and taking the MDR and FAR as the minimum cost objective functions, a multi-objective particle swarm optimization (MO-PSO) algorithm is employed to offline optimize the control width and smoothing parameter. In which, in order to reduce the influence of noise on the system, a wavelet analysis algorithm is used for multi-scale decomposition, threshold de-noising and reconstruction of the SS-DEWMA chart, and the output values of the optimized SS-DEWMA chart are compared with the threshold value to detect the sensor fault online. Finally, the simulation is carried out on the displacement tracking system of continuous casting mold driven by servo motor to verify the proposed method. By comparing with the existing residual evaluation methods, such as SS-DEWMA chart, the simulation results show that the displacement sensor fault can be detected by the proposed method more accurately than by other existing methods, as well as the MDR and FAR can be reduced by the proposed method.