基于统计特征的不等长间歇过程故障诊断研究

为了提高不等长间歇过程故障诊断的性能, 同时降低算法的复杂度, 提出了一种基于统计特征的不等长间歇过程故障诊断算法。首先计算每个不等长批次的均值、方差、偏度、峭度和任意两个变量间的欧氏距离, 并将这些统计特征组合成一个等长的特征向量; 然后运用主元分析(PCA)进行过程监视。半导体工业实例的仿真结果表明, 与传统的多向主元分析(MPCA)方法相比, 基于统计特征的不等长间歇过程故障诊断算法的故障诊断率提高15%, 故障检测时间减少了0. 002 s, 因此该算法具有很好的故障诊断性能。

Fault diagnosis for uneven-length batch processes based on statistic features

In order to improve the fault diagnosis performance of the uneven-length batch processes, and decrease the complexity of the algorithm, this paper presented fault diagnosis method based on statistic features for uneven-length batch processes. Firstly, it calculated the means, variance, skewness, kurtosis and the Euclidean distance between two variables for each uneven-length batch. Secondly, it combined these statistic features into an even-length feature vector. Lastly, it used principal component analysis (PCA) to the feature vectors for monitoring the batch processes. The monitoring results of an industrial example show that compared with traditional MPCA, the fault diagnosis method based on statistic features for uneven-length batch processes increases 15% of the fault diagnosis rate and reduces 0. 002 second of the fault diagnosis time, so it has good fault detection performance.