动态内全潜结构投影的空间扩展故障检测方法

动态内偏最小二乘(DiPLS)方法是基于数据驱动的潜结构投影的动态扩展算法, 用于动态特征提取和关键 性能指标预测. 在大型装备系统中, 传感器采集的当前时刻样本受历史样本的影响且可能包含较大噪声. 在动态特 征提取中, 因DiPLS算法未按降序提取主成分, 导致残差空间仍存在较大变异, 动态和静态信息难以有效分离, 影响 故障检测性能. 为此, 本文提出了一种基于动态内全潜结构投影的故障检测方法(DiTPLS). 首先, 使用动态内偏最小 二乘方法和向量自回归模型建立动态模型并检测故障, 用于捕捉质量相关动态信息; 基于结构化动态主成分分析 算法建立一种改进的动态潜在变量模型, 用于残差分解, 提取质量无关的动态信息和静态信息, 并构造合适的统计 量进行故障检测. 数值仿真和田纳西–伊斯曼过程实验验证了DiTPLS算法的有效性.

Space expansion fault detection method based on dynamic inner total projection to latent structures

Dynamic inner partial least squares (DiPLS) is a dynamic extension algorithm based on the data-driven latent structure projection (PLS), which is used for dynamic feature extraction and key performance index prediction. In large equipment systems, the current moment samples collected by sensors are affected by historical samples, and may contain large noise. In the dynamic feature extraction, because the DiPLS algorithm does not extract the main components in descending order, there is still large variation in the residual space. It is difficult to effectively separate the dynamic and static information, which affects the fault detection performance. As such, a fault detection method based on the dynamic inner total PLS (DiTPLS) is proposed. Firstly, the dynamic internal partial least squares method and vector autoregressive (VAR) model are used to establish a dynamic model and detect fault, which is used to capture the quality-related dynamic information. An improved dynamic latent variable model (DLV) is established based on the structured dynamic principal component analysis (DPCA) algorithm for residual decomposition to extract the quality-independent dynamic and static information, and to construct appropriate statistics for fault detection. Numerical simulations and Tennessee-Eastman (TE) process experiments verify the effectiveness of the DiTPLS algorithm.