基于滑动窗和多块卷积自编码器的故障检测

Fault Detection Based on Sliding Window and Multiblock Convolutional Autoencoders

为了进一步提升故障检测性能,充分挖掘时序和隐含特征信息,提出一种基于卷积自编码器的故障检测方法。在对原始信息集进行建模的基础上增加了对累计信息与变化率信息的建模,以增强对隐含信息的挖掘;对重构的3个信息集进行滑动窗采样,基于卷积自编码器进行时序特征提取和建模;将卷积自编码器的决策结果进行贝叶斯融合得到统计量,并用核密度估计的方法确定控制限从而进行故障检测。将该方法进行数值仿真并应用于TE过程,仿真结果验证了所提方法的有效性和检测性能。

In order to further improve the fault detection performance and fully mine the timing and hidden feature information,a fault detection method based on convolutional auto encoder is proposed.On the basis of modeling the original information set,the modeling of cumulative information and rate of change information is added to enhance the mining of implicit information;The three reconstructed information sets are sampled by sliding windows,and time series feature extraction and modeling are performed based on convolutional auto encoders.Bayesian fusion of the decision results of the convolutional auto encoder is performed to obtain the statistics,and the control limit is determined by the method of kernel density estimation for fault detection.The method is numerically simulated and applied in TE process,and the simulation results confirm the effectiveness and detection performance.