基于规范变量残差的化工过程微小故障检测与诊断

Canonical residual based incipient fault detection and diagnosis for chemical process

中文关键词  规范变量残差  加权平均统计量  微小故障  化工过程  贡献图  故障检测  故障诊断

英文关键词  canonical residual  weighted average statistics  incipient fault  chemical process  contribution plot  fault detection  fault diagnosis

中文摘要

      微小故障因其幅值低而易被噪声和过程扰动所掩盖, 并且会随时间慢慢演变成过程中的严重故障. 因此, 微小故障的检测和诊断变得越来越重要. 为了更有效地监测和诊断微小故障, 提出了基于规范变量残差的化工过程 微小故障检测和诊断方法. 首先, 对Hankel矩阵执行奇异值分解来获得主元和残差空间并根据过去和未来数据的差 异, 求得两个不同的规范变量残差d1, d2. 其次, 考虑数据的时间序列特性, 提出了基于规范变量残差的两个加权平 均统计量WD1, WD2及其控制限, 进行故障检测; 然后, 计算出各个统计量的归一化贡献并绘制二维贡献图, 进行故 障诊断. 最后, 在连续搅拌釜式反应器(CSTR)过程中进行两种微小故障的应用研究. 结果表明, 与传统的统计量T2, Q以及规范变量差异分析(CVDA)中统计量D相比, 基于规范变量残差的加权平均统计量WD1, WD2不仅能够及时 检测到微小故障, 而且在故障检测率和诊断率方面, 均有不同程度的提高.

英文摘要

      Incipient faults can be easily submerged by noise and process disturbance because of their low amplitudes, and will gradually evolve into serious faults in the process over time. Therefore, incipient faults detection is increasingly becoming more and more important. To monitor and diagnose incipient faults more efficiently, detection and diagnosis of incipient faults for chemical processes based on canonical residuals are proposed. First, singular value decomposition is performed on the Hankel matrix to obtain state space and residual space. Next, according to the difference of past and future data, two different canonical residuals d1, d2 are obtained. Then, considering the time series characteristics of the data, two weighted average statistics WD1, WD2 based on canonical residual and its control limit can be determined for fault detection. Then, the normalized contribution of each statistic is calculated and the two-dimensional contribution plots are drawn to perform fault diagnosis. Finally, the application of two kinds of incipient faults in the continuous stirred tank reactor (CSTR) was studied, the results show that compared with the conventional statistics T2, Q and the statistic D in CVDA those weighted average statistics based on canonical residual can not only detect incipient faults in time, but also improve the fault detection rates and fault diagnosis rates to varying degrees.