近些年,频繁发生的产品召回事件,愈发体现出产品质量追溯的重要性。然而,传统的质量追溯方法缺乏对于造成质量问题的关键因素的有效分析,难以满足实际工业生产的需求。针对这一问题,本文提出一种稀疏偏最小二乘-典型相关分析（SPLS-CCA）的产品质量智能追溯方法,实现了影响产品质量的关键过程变量的有效分析与识别,从而为后续提高产品质量提供了依据。这一方法对于企业的产品质量把控与经济效益的提高有着较大的意义。

Process monitoring is a challenging task for modern industrial processes which are commonly nonstationary in nature, revealing typical non-Gaussian characteristics. Nowadays, data-driven based fault detection methods have drawn increasing attention, most of which work under an assumption that the process is subject to Gaussian distribution. But in practice, the underlying non-Gaussian characteristics may be typical in the complex process, which cannot be properly enclosed by a statistical model with a close confidence region and thus may be insensitive to fault detection. Hence, it is necessary to explore and separate the underlying Gaussian and non-Gaussian distributions in fine-grain. In this work, a Gaussian and non-Gaussian subspace decomposition method is proposed by designing a variant of stationary subspace analysis (VSSA) for nonstationary process monitoring. First, the whole time-wise nonstationary process can be neatly converted to condition-wise slices. Then, a Monte Carlo sampling based VSSA technique is designed to separate Gaussian and non-Gaussian subspaces from each other, which focuses on analyzing sample distribution rather than time series properties. Here the Gaussian subspace, which is readily characterized by a statistical model, is used for revealing similar condition slices and affiliate them into the same condition mode. And two monitoring statistics are developed to explore the Gaussian and non-Gaussian distribution structures, thus providing fine-grained distribution analytics and promoting monitoring performance. The feasibility and performance of the proposed method are demonstrated on a real thermal power plant process. © 2020 IEEE.

Efficient fault root cause diagnosis is essential to ensure the production safety of industrial processes. The existing root cause diagnosis models can be summarized as linear methods and nonlinear methods. Linear methods cannot handle nonlinear processes well, while nonlinear methods usually require pairwise calculations between variables, which are complex and difficult to apply in real time. To address the above issues, a method for root cause diagnosis of nonlinear processes, termed sparse adjacency forecasting (SAF), is proposed in this paper. SAF is a causal inference method based on the idea of Granger causality. While forecasting time series, it constructs an adjacency matrix to synthesize the process information and the interaction of different variables. By adding sparse constraints to the adjacency matrix, the predictive effects between variables are reflected, and the causality is captured. This method only needs to model once to obtain the causal relationship between all variables, which avoids multiple modeling and improves diagnosis efficiency. Besides, in order to solve the nonlinear problem, multiple nonlinear random feature nodes are introduced for time series prediction. Two cases are adopted to verify the causal inference and root cause diagnosis performance of the proposed method, including a numerical case and the Tennessee Eastman benchmark process. © 2021 The Authors. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0)

The wide spread of dissimilarity analysis (DISSIM) gives rise to many useful process monitoring models. However, it is only limited to stationary processes. Reliable DISSIM based monitoring methods will encounter challenges of nonstationary behaviors including the time-varying mean or variance. In this work, a stationary subspace decomposition based DISSIM (SSD-DISSIM) model is developed to detect incipient faults sensitively for nonstationary industrial processes. As faults may disappear in time-varying process variations, the key is how to extract information with stationary characteristics from the complex process data. To eliminate the interference caused by the mixed nonstationary signals, the extraction of the stationary components is first conducted by projecting data into a low-dimensional subspace. As a type of distribution-based method, the DISSIM is combined to monitor the extracted stationary components in terms of not only the mean and variance, but also the correlations and distribution. Thus the proposed model can overcome the limitation of DISSIM that arises in nonstationary processes and enhance the sensitivity and reliability of the monitoring. The method effectiveness is demonstrated through the real thermal power plant example. © 2020 IEEE.

Time delay widely exists among industrial process variables, which may lead to invalid description of systems and reduce the model accuracy for soft sensor, process monitoring etc. Therefore, it is crucial to estimate time delay for improving model accuracy. In traditional research, time delay estimation (TDE) algorithm for two variables is fully investigated, but little research pays attention to multivariate. The method of pairwise comparison may not only cause high calculation complexity but also cut off the correlation between multivariate. In this work, a new TDE algorithm is proposed to estimate time delay between multivariate, which extracts dynamic latent variable to represent the process using improved dynamic inner PCA algorithm (DiPCA). Dynamic latent variable extracts the autocorrelation and cross-correlation between process variables. Take it as a standard and analyze the time delay, so that the influence of multivariate variables could be comprehensively considered. Since there may be non-linear variables, the significant factor (SF) is used for variable selection, and the multivariate variables are divided into multiple linear subgroups, which makes the analysis result more reasonable. The effectiveness is illustrated with two numerical examples and the Tennessee Eastman process. © 2020 Technical Committee on Control Theory, Chinese Association of Automation.