基于ESMMP算法的工业过程故障检测

Fault Detection Based on Ensemble Statistic Multi-manifold Projection Algorithm for Industrial Process

针对多流形投影(multi-manifold projection, MMP)算法忽略原始数据高阶统计量信息的问题,将统计量模式分析(statistic pattern analysis, SPA)算法引入MMP算法中,提出集成统计量多流形投影(ensemble statistic multi-manifold projection, ESMMP)算法。首先,对SPA算法中的窗口宽度参数设置不同的值,计算原始数据的均值、方差、偏度和峰度等低阶和高阶统计量来构建多个统计量空间,避免了单一窗口宽度参数难以适用于不同类型故障数据的局限性;然后,采用MMP算法在各统计量空间建立T;和SPE统计量进行监控;最后,通过贝叶斯策略融合各子模型的监控结果。数值例子和TE过程的仿真结果验证了所提算法在工业过程故障检测方面的有效性和优越性。

Aiming at the problem that multi-manifold projection(MMP) algorithm ignores high-order statistic information of original data, statistic pattern analysis(SPA) algorithm is introduced into MMP, and ensemble statistic multi-manifold projection(ESMMP) algorithm is proposed. Firstly, the window width parameter in SPA algorithm is set as different values, so a series of statistic spaces of original data can be constructed by calculating low-order and high-order statistics such as mean value, variance, skewness and kurtosis of original data, which avoids the limitation that single window width parameter cannot be applied to different types of fault data. Then,MMP algorithm is used to establish T;and SPE statistics in the statistic spaces for monitoring. Finally, the monitoring results of each sub-model are fused by Bayesian strategy. The simulation results of the numerical example and TE process verify the effectiveness and superiority of the proposed algorithm in fault detection of industrial process.