Search

In industrial process monitoring, the long-term stationary features play an important role in representing essential statistical information. However, the autoencoder-based methods extract the deep features by achieving the numerical approximation of the original data, which may lead to the destruction of the hidden stationary information. To solve this problem, a cointegration stacked autoencoder model based on stationary features reconstruction is proposed in this paper to maintain long-term equilibrium relationships during model training. First, a cointegration analysis model is constructed to extract the stationary features hidden in the non-stationary data. Based on this, a cointegration stacked autoencoder is designed to reconstruct the extracted stationary features and the original data simultaneously. In addition, the monitoring statistics for both deep and stationary features are integrated by Bayesian inference criterion. By reconstructing the stationary features, the proposed network is able to retain the beneficial relationship among the non-stationary variables. Finally, the fault detection performance of the proposed method is verified in two cases.

Due to the complex and variable working conditions of manufacturing processes as well as the interconnection and coupling loops, faults of multiple processes, subsystems or control loops may occur at the same time or cascade successively, compound fault becomes the norm. After the compound fault occurs, it may be characterized by the abnormal sensor data, the image or video data. Most of the traditional methods use a single type of sensor data for fault diagnosis, which may affect the diagnosis performance. To solve this problem, considering the heterogeneity of compound fault data, a multitask learning-based collaborative modeling method is proposed for compound fault diagnosis. Specifically, the heterogeneous compound fault data is mapped into a common subspace, of which a dynamic combination coefficient is reasonably set considering the different contributions of unstructured and structured data. Moreover, considering the correlations among heterogeneous data, the low-rank constraint trace norm is introduced into multitask learning, and an attention-based feature fusion network is designed for compound fault diagnosis. Finally, two cases on the actual manufacturing process, hot rolling process, are conducted to evaluate the effectiveness of the proposed method. The experimental results demonstrate that the new collaborative modeling scheme of heterogeneous data can better perform compound fault diagnosis tasks than the existing state-of-the-art algorithms.

The distribution of actual industrial process data is complex, and variations in data distribution caused by equipment wear and changing operating conditions can easily lead to model mismatch, presenting a severe challenge to fault diagnosis methods that assume data follows a Gaussian distribution. In this context, we propose a novel fault detection method based on generative models in this paper. Firstly, historical data are used to train a denoising diffusion probabilistic model (DDPM) to generate data. Secondly, both the training set and the generated data are input to an autoencoder, and a data evaluation metric is constructed to filter high-quality out of distribution features. Subsequently, positive and negative sample pairs are constructed based on these features, and an improved supervised contrastive learning detection model is designed to extract unique features of normal data under the supervision of virtual fault samples. Finally, the effectiveness and superiority of the proposed method are validated through the Tennessee Eastman simulation process.

Modern manufacturing processes typically comprise multiple subprocesses, and the temporal-spatial features of process variables are difficult to extract, which bring significant challenges for traditional quality-related fault diagnosis. To solve this problem, we propose a fault detection model driven by graph attention networks integrated gated recurrent unit-canonical variable analysis (GATRU-CVA). First, the knowledge of domain experts and historical data are used to construct a subblock knowledge graph. Next, a spatial feature extractor for the graph attention network (GAT) is constructed for global variables. In addition, the subblock knowledge graph is used to divide the global spatial features into subblocks, and the corresponding temporal feature extractor is constructed. Then, taking the process dynamic characteristic into account, CVA is used to model the process based on the spatial-temporal features, upon which the corresponding statistics are calculated. The threshold is determined by the kernel density estimation (KDE) method. Finally, actual production data of the hot strip mill process (HSMP) are used to validate the proposed model. The results show that the proposed method has a 97% correct monitoring rate (CMR) for HSMP, which is better than the other comparative fault detection methods.

This brief proposes a novel degradation model and remaining useful life prediction (RUL) prediction framework to quantify the aleatory, epistemic and measurement uncertainty of complex devices. First, an uncertain random process-based degradation model is established to describe the implicit degradation features of the device. Then, the prior parameters are identified by the stochastic uncertain maximum likelihood estimation (SUMLE) method. Afterward, a similarity-based weighted SUMLE method is proposed to update the uncertain parameter and combined with the Kalman filter to update the implicit degradation states. Finally, the approximate analytical expression for the probability density function of RUL under triple uncertainty is derived to achieve RUL prediction dynamically. The effectiveness of the proposed method is validated by the GaAs laser degradation dataset.

Operating performance assessment (OPA) of process industries plays a significant role in improving the product quality and pursuing the best economic benefits. With the continuous improvement of industrial intelligence, massive data has been collected. Variable time-delay issue is brought by the different spatial and temporal distributions of production units, which affects the actual causal relationship of collected data and assessment accuracy. Therefore, a spatiotemporal synergetic operating performance assessment (SSOPA) method is proposed in this paper. First, a reconstructed correlation matrix is carefully crafted in both time and space, leveraging the intricate multi-correlated spatiotemporal-delay (MCSD) parameters. To precisely quantify the multi-correlated relationships, a mixed gray relation analysis (MGRA) is employed. Subsequently, a predatory search-based genetic algorithm (PSGA) is utilized to meticulously search for the optimal MCSD parameters. Once these parameters are determined, an OPA model is formulated for each subsystem, drawing upon aligned process data and a distributed dissimilarity analysis (DISSIM). Finally, based on the subsystem-level OPA, a comprehensive assessment of the global operating performance level is conducted. The feasibility and effectiveness of the proposed method are verified by the plant-wide hot strip mill process (PHSMP).

The problem of adaptive fault-tolerant fixed-time control is studied for a class of strict-feedback nonlinear systems with multiple faults and multiple disturbances. We incorporate the tuning functions and the command filter technique into the controller design. Note that the former reduces the order of the controller, and the latter simplifies the derivation and calculation process and avoids the high calculation cost and differential explosion. In addition, since our proposed strategy is based on fixed-time control, it can ensure that the closed-loop system keeps the practically fixed-time stable, and the tracking error converges to a small neighborhood near the origin within fixed-time. As for the unknown items in the systems, fuzzy logic systems are used to approximate them. The simulation results show that the proposed control strategy is effective even if there exist actuator faults and multiple disturbances.

In the process industry production, the online sensing of process performance is very important for the optimization and control of the manufacturing process. However, the information island is formed by long processes and multiple systems of complex production processes. The process data are characterized by high dimensional heterogeneity, nonlinearity, and strong coupling, and the offline assay of process performance is characterized by high discretization and irregular sampling period. In order to solve the above problems, a cloud-edge collaborative soft sensing framework for multiperformance indicators prediction of manufacturing processes with nonregular sampling is proposed. Also, some experiments are carried out with the actual hot strip rolling process, which realizes the joint real-time sensing of the three performance indicators of yield strength (YS), tensile strength (TS), and elongation (EL) with good accuracy.

The accurate perception of key quality variables in complex manufacturing processes is a necessary prerequisite for achieving system optimization control and ensuring the safe and stable operation of the system. Consider that the manufacturing process has complex characteristics, such as multioperating conditions, time-varying behavior, nonlinearity, and scarcity of labeled samples, making accurate perception of process quality more difficult. This article proposes a semi-supervised quality soft sensing method under multiple operating conditions for complex product manufacturing processes. First, a dual k-means operating condition identification method assisted by deep metric learning (DML) is proposed, aiming to achieve a fine-grained partition of online data into specific operating conditions and the recognition of new operating conditions. Second, an online selective perception strategy is employed, wherein different approaches are taken based on whether the samples belong to existing operating conditions. In the construction of the base model, both spatial and temporal information among the samples are simultaneously considered, and the results are fused through an adaptive fusion strategy. Furthermore, considering the scarcity of labeled samples in real industrial processes, a high-confidence pseudolabel sample construction method based on just-in-time learning (JITL) and genetic algorithms (GAs) is proposed. This approach aims to obtain high-confidence pseudolabel samples to augment the modeling dataset. Finally, the effectiveness of the proposed method is validated through a real manufacturing process, the float glass production process.

In order to emphasize the fault-related information and to reduce the redundant information simultaneously, an optimized distributed cointegration analysis (CA) monitoring approach is proposed in this article for large-scale nonstationary processes. In the optimized distributed strategy, fault-related nonstationary variables are selected by further analyzing different faulty datasets. In addition, the subblocks with high partial correlations are fused to obtain the optimized blocks with less redundant information. On this basis, CA methods are constructed for fault-related nonstationary blocks to explore long-term equilibrium relationships. Finally, the monitoring results of each block are integrated by the Bayesian inference criterion (BIC). With the fault-related variables selection and partial correlation fusion strategy, the optimized blocks are able to characterize the different faulty modes and to retain beneficial nonstationary information. The performance of the proposed approach is compared to other process monitoring methods on the benchmark simulated vinyl acetate monomer (VAM) process.

The dynamic performance metrics of manipulators are decreased when actuator faults happen. To ensure the response speed and improve the tracking accuracy under actuator fault, the complementary sliding mode (CSM) controller is designed in this article. Different from the general sliding mode control (SMC), there are two sliding surfaces in the CSM controller design process, and the mathematical complexity of the controller is simplified through the relationship of the two complementary sliding surfaces, which helps to reduce the response time. Also, the structural parameter changes and disturbance existed in the real manipulator system are also considered in this paper by introducing uncertainty item during modeling, thus the robustness of the controller is improved. In the experimental part, the control effect of the CSM proposed is compared with that of the conventional sliding mode method. Normal conditions and actuator constant deviation fault of the manipulator are studied. The results show that the CSM has superiority in control accuracy and response speed. Note to Practitioners—In recent years, the control problem of manipulators under complex conditions has attracted particular attention. The key to the design of the controller lies in its stability while keeping the tracking performance. Many complex auxiliary methods are proposed to guarantee this, but a few can achieve acceptable performance. This article proposed the complementary sliding mode controller design method, in which two sliding surfaces are used to simplify the proof in the controller design. We choose the error variables to construct the sliding surface, so the CSM could handle several actuator faults very well, which enhances the adaptive ability of the algorithm.

Root-cause diagnosis of quality-related faults plays a crucial role in ensuring stable product quality and high-efficient production for modern manufacturing processes. However, there exists complex nonlinear and dynamic sequential characteristics for process and quality data before and after faults, which contain important fault information. How to fully explore and use this information to locate the root-cause and identify the propagation path is a hot topic. Thus, a new nonlinear dynamic Granger causality analysis method is developed for root-cause diagnosis, which will provide timely and useful reference information to take reasonable measures for field engineers. First, an optimal variable division based on minimal redundancy maximal relevance algorithm is presented to get the quality-related variables. Then, a new attention-based random disturbance gated recurrent unit is designed for nonlinear dynamic Granger causality analysis, aiming at locating the root-cause and identifying the propagation path of quality-related faults. A typical manufacturing process, the hot strip mill process is used for verification. The results show the practicability of the proposed method, and its strong robustness to noise. Note to Practitioners—Reasonable and feasible root-cause diagnosis technology will provide timely reference information for field operators to take maintenance measures. However, the nonlinear and dynamic characteristics of time-series data before and after quality-related faults have brought great challenges to the traditional root-cause diagnosis techniques. To address this issue, a practical nonlinear dynamic Granger causality analysis method is proposed, which is purely data driven without knowing complex mechanism knowledge. It will provide feasible solutions for safety monitoring of hot strip mill process as the representative of manufacturing processes.

Driven by the need for remaining useful life prediction of degraded motor systems with feedback controllers, a real-time updated Wiener stochastic process is adopted to model the performance degradation of motor systems. First, a closed-loop performance index of the motor system is derived incorporating the multiple slow time-varying characteristic of motor parameters. On this basis, the drift coefficient and diffusion coefficient of the Wiener degradation model are updated to obtain the prior maximum likelihood function with available historical data. It is followed by the iterative optimization of nonlinear feature parameter in the Wiener degradation model with taking the prior maximum likelihood function as the cost equation. The effectiveness of proposed remaining useful life prediction architecture for closed-loop motor systems is demonstrated by motor systems.

Plant-wide process monitoring and fault diagnosis are key technologies to ensure the high-efficiency and safety of the production. This article pays close attention to the new characteristics in modern large-scale sequential industries, including long process, multiunit cooperation, and strong coupling among subblocks. A two-layer distributed fault diagnosis method based on feature transfer is proposed. By characterizing the coupling correlation between subblocks and imposing the feature heredity of adjacent subblocks into plant-wide process modeling, a distributed fault diagnosis framework at the subblock layer and the process layer is developed. First, the process decomposition is performed and we obtain several sequential subblocks. For each subblock, the local tangent space alignment (LTSA) algorithm is employed to perform the nonlinear dimension reduction. Then, canonical correlation analysis (CCA) is used to characterize the coupling relationships among subblocks and the internal unique features of each subblock and the external features related to other subblocks are extracted by CCA. In the feature extraction process, we use the external features as the input of the next adjacent subblock to model the chain feature transfer effect between subblocks. On this basis, the internal and external monitoring statistics for each subblock are constructed and the corresponding control limits are determined by kernel density estimation (KDE). In this way, the subblock layer monitoring model is developed, where the internal fluctuations and underlying propagation tendency of anomalies can be monitored in a fine-scale manner. The plant-wide process monitoring results are reached based on all the monitoring submodels via Bayesian fusion. Finally, the internal and external statistics-based contribution plots are designed for fault diagnosis and the fault-relevant variables can be located. The proposed framework is testified on a real hot strip mill process and the results show its effectiveness.

Quality prediction is important for precise control of industrial processes and improvements of product quality. The nonlinear and dynamic features of time-series data can be effectively extracted by the appropriate soft sensor models for quality prediction. However, the temporal features may be often considered by the conventional methods, ignoring the important spatial features, which makes the soft sensor models not have stronger generalization abilities because of insufficient feature information for quality prediction. Moreover, due to the unstable transmission signals, equipment failures, and sensor packet losses, missing values maybe present in the industrial data, which influences the accuracy of soft sensor models. To overcome those problems, a novel spatiotemporal industrial soft sensor modeling scheme is designed for quality prediction. To be specific, the generate adversarial imputation network (GAIN) is first applied for data imputation. Then, the k-nearest neighbor mutual information (kNNMI) is used for constructing the adjacency matrix of graph attention network (GAT) with the purpose of adaptive measuring the spatial correlations among process variables at different moments. After that, the temporal attention (TA) mechanism is introduced for enhancing the temporal feature extraction capability of minimal gated unit (MGU), aiming at improving the quality prediction performance. Finally, sufficient simulation experiments are conducted by a typical industrial process, hot rolling process (HRP), to demonstrate the superiority of the proposed scheme compared with some classical and advanced algorithms.

This article is concerned with the optimal fault detection issues for discrete-time nonlinear systems with the aid of Takagi–Sugeno fuzzy dynamic modeling technique. To this end, in the first part of this article, the nonlinear system is formulated in the time-varying fuzzy manner, and based on it, a unified fault detection approach is developed by solving a multiobjective optimization problem. In this sense, the optimal tradeoff between fault detectability and robustness against unknown inputs is ensured by solving the Riccati equation. Meanwhile, a fuzzy fault detection approach is studied in the second part of this article based on piecewise-fuzzy Lyapunov functions, which is realized by solving linear matrix inequalities. Two examples are given at the end of this article to demonstrate the proposed approaches.

This article proposes a novel transferable manifold projection embedded dictionary learning (TMPDL)-based scheme with domain transfer for multimode process (MP) monitoring, where the new modes in evolving scenarios can be rapidly modeled. Considering that only new measurements are necessary for updating the model parameters, the proposed method elevates engineering applicability. First, in order to quantitatively analyze the discrepancy between the new and previous modes, the common features are extracted by TMPDL, upon which new modes can be modeled using domain transfer, saving storage resources and ensuring scalability. Then, the corresponding optimization process is fully discussed, which incorporates feature selection and extraction to select specific features for updating while enhancing the interpretability of the model. Concurrently, consistency and independence constraints are imposed on dictionary learning (DL), which makes the features extracted by the proposed method more discriminative. Finally, the monitoring model is developed by feature reconstruction error (FRE), which can derive monitoring results prior to mode identification. Experiments on the real hot strip mill process (HSMP) reveal that the fault detection ability of TMPDL is highly robust against MP, achieving a 94.8% monitoring accuracy rate for the newly arriving mode.

Large-scale manufacturing processes are usually made up of multiple interrelated and distributed continuously subprocesses, which are transmitted and connected by information and quality flow. The characteristics of long processes, quality heritability between subprocesses, and dynamic-coupled variables bring severe challenges to conventional quality-related fault diagnosis. Against this background, a novel distributed diagnosis framework for quality-related faults is proposed in this article. First, the sequential manufacturing process is decomposed into multiple subprocesses based on mechanism knowledge. Second, a novel dual-attention quality-driven autoencoder method is designed as the model for local fault diagnosis. Deep nonlinear features are extracted under quality supervision; meanwhile, the dynamic information and the different correlations among variables are also considered. Then, based on the tandem structure of the manufacturing process, multiple dual-attention quality-driven autoencoder models corresponding to each subprocess are constructed and stacked into a distributed model. Bayesian inference is used to build global monitoring statistics. Moreover, after faults occur, the intervariable attention weights are achieved to identify faulty variables. Finally, the effectiveness and advantages of the proposed framework are demonstrated via a practical large-scale sequential manufacturing process, the hot strip mill process.

Due to the interconnected characteristics between subsystems and the strong correlation within subsystems, the monitoring of plant-wide processes has become a challenging problem, especially for tandem plant-wide processes that exist in various industrial fields, such as petrochemicals, metallurgy, and sewage treatment. In this article, a novel spatio-temporal monitoring method is proposed for the hot strip mill (HSM) process, a typical tandem industrial process. First, the plant-wide process is divided into different subblocks based on the tandem structure. Then, a distributed conditional variational recurrent autoencoder-based process monitoring method is proposed to build the local latent variable model of each subsystem using relevant dynamic features extracted from the previous subsystem. The latent distributions and reconstructed errors are used to design local monitoring statistics for local process monitoring. A global monitoring statistic is established by deep support vector data description to monitor the whole process. Finally, the effectiveness and superiority of the proposed method are demonstrated by a HSM process case, which shows better monitoring performance compared to the existing methods.

For a category of complex equipment with multiphase degradation characters subject to changes in internal mechanisms or environmental stresses, existing remaining useful life (RUL) prediction methods only consider aleatory uncertainty but ignore epistemic uncertainty caused by the lack of data and insufficient knowledge. For quantifying the aleatory and epistemic uncertainty of multiphase degradation equipment, this article proposes a novel multiphase degradation model and RUL prediction method based on the uncertain random process. First, a 2-D multiphase degradation model is presented in probability space and uncertainty space to describe multiphase degradation patterns with aleatory and epistemic uncertainty. Second, the two-stage empirical estimation method and stochastic uncertainty maximum likelihood estimation (SUMLE) are integrated for change point identification and parameter estimation. Meanwhile, similarity-based weighted SUMLE is projected to update the model. Then, taking into account the heterogeneity between individuals and the randomness of the degradation state at the change point, the analytical expression of RUL is derived under the concept of first hitting time. Finally, the effectiveness of the proposed method is verified through numerical simulation and practical cases.

This article takes into account the problem of adaptive fixed-time control for nonlinear systems in a strict form via finite-time command-filtered backstepping. Our presented control scheme gives consideration to rapidity by using fixed-time control and finite-time command filter, and its prime control objective is to ensure that the system output can be guided from any initial conditions to go after an ideal variable. Meanwhile, this control strategy makes sure that all the system states and other signals are bounded at finite time, and its convergence time does not have any connection with the system initial conditions and is determined by the design parameters. At every step in the design process, a fuzzy logic system is introduced to approximate the unknown part, and a finite-time command filter is also utilized so as to avoid the issue of derivation of virtual control laws and complex calculation. At last, we verify that the presented scheme is viable through a simulation example.

Process monitoring is crucial for guaranteeing product quality and process safety in large-scale processes. The large-scale processes typically consist of different subprocesses or operation units. The quality fluctuations that occur in the upstream operation units are inherited into the downstream ones with the quality flow, as there are many complex relationships of nonlinear and strong coupling among subprocesses or operation units in series. However, traditional monitoring strategies often ignore or even fragment the quality inheritance among operation units, which reduces the accuracy and performance of process monitoring. To solve the problem mentioned above, a novel process monitoring framework based on quality inheritance for large-scale processes is proposed in this article. First, a priori knowledge is used to divide the complex large-scale processes into multiple subblocks according to the product processing order. Second, mutual information (MI) is used to evaluate the correlations of variables and design assessment indicators for them. Then, a new local monitoring model based on the parallel deep variational information bottleneck (DVIB)-variational autoencoder (VAE)-minimal redundancy maximal relevance (mRMR) is constructed. For coupling subblocks in series, the parallel DVIB-VAE can precisely extract features and the mRMR method can eliminate the effects of redundant information. Finally, Bayesian inference is used to obtain the distributed monitoring results for large-scale processes from both local and global perspectives. The superiority of the proposed method is verified by the real hot strip mill process (HSMP).

High-precision prediction of material properties is an advanced direction of hot rolled product technology in manufacturing processes. However, complex disturbance usually produces wrong or missing data in the acquisition and entry stages. The long process and the diverse operating conditions bring great challenges to constructing soft sensors for the hot rolling process. To address this problem, a new time-series soft sensor model framework is proposed for mechanical properties prediction. First, in order to process the raw dataset, the isolation forest algorithm is used to remove the outliers, and the random forest algorithm is used to fill in missing values. Second, the auxiliary variables are extracted by the combination of the XGBoost algorithm and prior mechanistic knowledge. Then, a dual-stage attention-convolutional neural network and bidirectional gated recurrent unit (DA-CBGRU)-sequence-to-sequence (Seq2Seq) time-series model based on a two-stage attention mechanism is designed for soft sensor modeling. In the encoder stage, external factors are introduced by an entity embedding layer. A combination structure of bidirectional gated recurrent units (GRUs) and convolutional layers is designed for extracting and learning the input features adaptively with the input attention mechanism. In the decoder stage, the long-term time dependence of the time series is learned by the temporal attention mechanism. Finally, the reliability of the proposed sensor model is compared and verified by the hot rolling process. The results indicate that the proposed method has an accuracy of over 95% in predicting the mechanical properties of a real hot-rolling process dataset, and has a better performance compared to other comparative soft sensor prediction methods.

Quality prediction plays a crucial role in improving the product quality and economic benefit for modern process industries. The high-dimensional characteristics of process variables and the nonlinear dynamic behaviors of sequential data make the conventional soft sensor modeling approaches encounter difficulties. Toward this end, a novel nonlinear dynamic soft sensor modeling scheme is designed for quality prediction. It will offer a timely reference for judging product quality and adjusting production plans. Specifically, the maximum information coefficient is introduced for measuring the interdependencies between process and quality variables, and an evaluative criterion is designed for quality-related variable selection. Subsequently, a novel bidirectional minimal gated unit structure is proposed for nonlinear dynamic soft sensor modeling, of which the historical and future temporal information of production and quality features is fully used and integrated for quality prediction. In the end, the effectiveness and practicality of the proposed scheme is demonstrated via a challenging process industry, the hot rolling process (HRP). The simulation results indicate that the proposed method is superior to other competitive methods.

With the development of modern information technology, the collection, storage, and transmission of information in the process industry have been gaining popularity. However, the massive streaming industrial data obtained in real time have some nonideal characteristics, such as lack of labels and missing values, which greatly increase the difficulty of process monitoring in process industry. Therefore, a robust semi-supervised fault classification method is proposed in this article. First, Wasserstein generative adversarial network (WGAN) and enhanced minimal gated unit (EMGU) are integrated to complete the missing data imputation of the incomplete unlabeled streaming industrial data, and then a semi-supervised ladder network (SLN) is trained with the imputed unlabeled data and complete labeled data for fault classification. A case study on the hot rolling process (HRP) demonstrates that the proposed method shows outstanding modeling and classification performance in lack of labeled data and missing data, compared with the other state-of-art deep learning methods.

Process monitoring is important for ensuring the safe running of process industries. From the perspective of pattern recognition, fault detection can be regarded as a binary classification problem, while fault classification can be regarded as a multi-class classification issue. As the extension of one-class classifier, multi-sphere support vector data description (MSVDD) are oriented to vector data and cannot deal with tensor data directly. Moreover, MSVDD cannot deal with feature selection and parameter optimization synchronously. In order to deal with above problems, a faut classification method based on multi class support tensor data description (MSTDD) is proposed in this paper. Based on sooty tern optimization algorithm (STOA) and beetle antennae search (BAS), a hybrid synchronization optimization (HSO) strategy is used to select features and optimize parameters at the same time. Application to the Tennessee Eastman process (TEP) verifies the classification performance of HSO-MSTDD comparing to MSTDD.

Modern complex industrial processes are increasingly integrated and automated. There are numerous control loops in the processing system, and the closed-loop feedback makes units strongly coupled. This makes the faults propagate and evolve with the material and energy flow networks. Thus, it poses a great challenge to the fault root cause diagnosis and propagation path identification for industrial processes. This article proposes a fault root cause diagnosis and propagation path identification strategy for complex industrial processes based on the hierarchical causal graph (HCG). First, the process is divided into several subblocks according to process knowledge. The monitoring model is built for fault detection. Second, the fault-related subblocks and variables are selected based on the contribution plot, and the causal analysis based on transfer entropy (TE) is performed on them to construct HCG. In this way, the causal connections can be significantly reduced. Subsequently, a propagation path identification method based on causal feature long short-term memory (CF-LSTM) is designed. Finally, real data from the hot strip mill process (HSMP) and their comparative experiments are adopted to validate the fault diagnosis performance of the proposed framework.

Remaining useful life (RUL) prediction based on the degradation model is vital for the maintenance and management of complex equipment. It is necessary to describe the aleatory uncertainty in internal changes during equipment degradation and quantify the epistemic uncertainty caused by a lack of sufficient knowledge simultaneously. This article proposes a novel framework for the degradation model and RUL prediction with aleatory and epistemic uncertainties. First, the probability theory is combined with the uncertainty theory to establish a stochastic uncertain degradation model. Following that, a new stochastic uncertain maximum likelihood estimation (SUMLE) method is proposed based on probability function and uncertainty distribution to identify model parameters. The Bayesian inference is adopted to update the model parameters in the current time. Afterward, the above dual-source uncertainty is incorporated into RUL prediction. The distribution function of the RUL is derived, which can be updated in real-time according to the arrival of new degradation observations. Finally, the experimental studies on gallium arsenide (GaAs) laser and gyroscope degradation data are conducted to interpret the superiority of the proposed method over based on the stochastic process or uncertain process separately for predicting the RUL.

In the field of fault detection, the nonstationary characteristics caused by external disturbances of wind turbine (WT) and other reasons can mask the fault signals, while the inconsistent data distribution between training data and test data due to equipment loss and other reasons can lead to model mismatch problems, both of which can lead to the degradation of fault detection performance. In order to solve the above problems, a novel adaptive fault detection framework is proposed in this work. First, the stationary features of nonstationary variables are extracted based on short-time Fourier analysis, after which the features are combined with the stationary variables. Second, isolation-based anomaly detection using nearest-neighbor ensembles (iNNE) is introduced as monitoring metrics for designing the statistic for the slow feature analysis (SFA) method. Then, the differences between online normal data and training data are calculated, the model update factor and update strategy are designed, and an adaptive fault detection framework based on short-time Fourier transform-SFA (STFT-SFA) is proposed. Finally, the effectiveness of the proposed fault detection framework is verified using the Tennessee Eastman process (TEP) and actual WT data. The results show that the proposed STFT-SFA method has a 94.0% correct monitoring rate (CMR) for TEP failures, and the proposed adaptive STFT-SFA method has a 95.6% CMR for WT failures, which are better than other comparative fault detection methods.

Nonlinear characteristic widely exists in industrial processes. Many approaches based on kernel methods and machine learning have been developed for nonlinear process monitoring. However, the fault isolation for nonlinear processes has rarely been studied in previous works. In this paper, a process monitoring and fault isolation framework is proposed for nonlinear processes using variational autoencoder (VAE) model. First, based on the probability graph model of VAE, a uniform monitoring index can be calculated by the probability density of observation variables. Then, the fault variables are estimated with normal variables by a missing value estimation method. The optimal fault variable set can be searched by branch and bound (BAB) algorithm. The proposed method can resolve the "smearing effects" problem existing in traditional fault isolation methods. Finally, a numerical case and a hot strip mill process case are used to verified the proposed method. [ABSTRACT FROM AUTHOR]

Process monitoring and fault diagnosis is the key to ensuring process safety and improving product quality. In traditional studies, process safety indicator related fault detection and evaluation have not been fully resolved. To address this problem, a method of maximum information coefficient and slow feature analysis is presented in this paper. The relevance between process and safety variables is calculated by using the maximum information coefficient, and the process variables are divided into strongly safety-related block, weakly safety-related block and safety-unrelated block. Then, the distributed local monitors are established in each block using slow feature analysis method which can identify the process dynamic anomalies. Meanwhile, a new decision fusion method based on logical analysis is proposed to process monitoring and fault grade assessment. Finally, the proposed method is practiced with the Tennessee Eastman Process, and the effectiveness and superiority are proved by comparing with other methods.

Accurate estimation of the remaining useful life (RUL) and health state for rollers is of great significance to hot rolling production. It can provide decision support for roller management so as to improve the productivity of the hot rolling process. In addition, the RUL prediction for rollers is helpful in transitioning from the current regular maintenance strategy to conditional-based maintenance. Therefore, a new method that can extract coarse-grained and fine-grained features from batch data to predict the RUL of the rollers is proposed in this paper. Firstly, a new deep learning network architecture based on recurrent neural networks that can make full use of the extracted coarsegrained fine-grained features to estimate the heath indicator (HI) is developed, where the HI is able to indicate the health state of the roller. Following that, a state-space model is constructed to describe the HI, and the probabilistic distribution of RUL can be estimated by extrapolating the HI degradation model to a predefined failure threshold. Finally, application to a hot strip mill is given to verify the effectiveness of the proposed methods using data collected from an industrial site, and the relatively low RMSE and MAE values demonstrate its advantages compared with some other popular deep learning methods.

Probabilistic models, which can model the process noise and can handle the problem of missing data in the probabilistic framework, recently have been got much attention in process monitoring and fault diagnosis area. This paper presents a new probabilistic methodology for fault detection and diagnosis in nonlinear processes using a variational autoencoders (VAEs) models. Two statistic index, based on the probability density distribution of measure variables and latent structure variable, are built to monitoring fault. Then a probabilistic contribution analysis method, based on the concept of missing variable estimation, is proposed for fault diagnosis. The performance of fault detection and diagnosis is demonstrated through its application for the monitoring of Tennessee Eastman (TE) industrial process, and the effectiveness is verified.

Compared with existing process monitoring approaches, a joint data-driven method using knowledge propagation based on manifold clustering is proposed for fault detection, which utilises historical data containing knowledge information (labelled data). The main contributions of this work are as follows: 1) two transformation matrices are derived based on manifold learning and clustering method; 2) different from conventional data-driven fault detection method, knowledge propagation based on manifold clustering is used to extract the features of unlabelled data; and 3) according to extracted features, the fault detection approach is proposed. The proposed method is applied to Tennessee Eastman (TE) process. The simulation results indicate that the proposed monitoring scheme can effectively monitor the working conditions of the process and identify fault types.

In modern industry, timely and accurate fault diagnosis plays an important role in satisfying the demands of production safety and stability of production quality. This paper dedicates on propagation path identification of faults in industrial processes, which will offer a feasible technology or solution to take corrective and timely maintenance measures for field engineers. Specifically, a recurrent neural networks-based Granger causality analysis approach is developed, which has sufficiently considered the nonlinear and dynamic relationships among time series after faults happen. Finally, we validate our approach on a typical industrial process, finishing mill process, to demonstrate the efficiency of the proposed scheme.

In integrated circuit manufacturing industry, etch process is a complex nonlinear batch process. Al stack etch is the penultimate layer of dry etch. Based on the specific steps of the recipe, it has the multiphase characteristic and the can exhibit significantly different behaviors over different phases. However, conventional fault detection methods cannot effectively monitor Al stack etch process due to nonlinear and multiphase characteristics. Moreover, they are usually modeled by normal process data. In Al stack etch process, fault process data can be obtained from the datalog of equipments. In order to utilize these data, a novel nonlinear fault detection method called fault-related multistage principal polynomial analysis (FMPPA) is proposed in this work. FMPPA is efficient to deal with nonlinearity of the multiphase batch process. Furthermore, it can make full use of fault data by decomposing original feature space into three subspaces. FMPPA is applied to monitoring the Al stack etch process. Simulation results demonstrate that FMPPA is superior to other methods.

Nowadays, many industrial processes generate large amounts of multimode data, which generally have a natural tensor structure, causing some faults invisible with traditional process monitoring (PM) and fault diagnosis (FD) methods. Tensor decomposition (TD) is a more practical approach for its effectiveness in solving high dimensionality problems as well as indicating the links between different modes. This paper proposes a common and individual feature extraction method based on TD, which identifies and separates the common and individual features from multimode data. The newly proposed approach is applied to a typical multimode hot strip mill process (HSMP), where common and individual feature for all steel products are existing. The final results indicate that the proposed approach can accurately detect and identify different faults in the HSMP.

A Routh table test for stability of commensurate fractional degree polynomials and their commensurate fractional order systems is presented via an auxiliary integer degree polynomial. The presented Routh test is a classical Routh table test on the auxiliary integer degree polynomial derived from and for the commensurate fractional degree polynomial. The theoretical proof of this proposed approach is provided by utilizing Argument principle and Cauchy index. Illustrative examples show efficiency of the presented approach for stability test of commensurate fractional degree polynomials and commensurate fractional order systems. So far, only one Routh-type test approach [1] is available for the commensurate fractional degree polynomials in the literature. Thus, this classical Routh-type test approach and the one in [1] both can be applied to stability analysis and design for the fractional order systems, while the one presented in this paper is easy for peoples, who are familiar with the classical Routh table test, to use.

Smart manufacturing in the context of Industry 4.0 presents new characteristics, such as large-scale, complex sub-processes cooperation, and integrated information management, which makes comprehensive performance-driven process monitoring difficult to be achieved in practice. Cloud-edge-device collaboration brings new perspectives for modern industries. This paper presents a performance-driven process monitoring framework in cloud–edge-device collaborative architecture, including a spatio-temporal information analytics based method and its application to a hot strip mill process (HSMP). Firstly, an integrated performance-driven process monitoring framework is proposed, where different monitoring tasks are deployed in the device, edge, and cloud layers, respectively. Secondly, a new performance-related spatio-temporal information analytics based fault detection and localization method is developed. In detail, temporal kernel canonical correlation analysis (CCA) is employed to extract the canonical correlation features with nonlinearities in the temporal domains, and spatial CCA is designed to form the spatial features. Based on the spatio-temporal features, the equipment operation condition monitoring model is developed in the device layer for real-time fault localization, and the process monitoring model is deployed in the edge for fault detection. Furthermore, to give a high-level evaluation of the operating status of the process, a qualitative trend analysis based operation risks evaluation model is deployed in the cloud. Different from the traditional architectures, the data processing capability of each layer is fully utilized in the proposed one, and the function is relevant to the enterprise management system. It provides a topological basis with clear priorities for subsequent process regulation and optimization. Finally, an actual HSMP is introduced to testify the performance of the framework. From the results, the fault detection rates are improved in comparison with traditional methods, and the fault localization results are consistent with the actual situations.

Thickness and flatness are important quality indicators for strip. It is important that the rapid and accurate prediction of the exit thickness and flatness for the optimal control of the hot strip rolling process. Due to the fast and long rolling process, there are time delays, non-linearity and strong coupling among the variables, which cause difficulties in the establishment of prediction models. In this paper, the variables related to thickness and flatness are selected by analyzing the rolling process mechanism and data. Based on the data related to the rolling quality, a rolling exit thickness and flatness joint prediction model combined multi-indicator Transformer with embedded sliding window (SW-MTrans) is proposed. First, a sliding window is embedded into the input layer of the model in order to address the effect of the time delay among variables. Then a Transformer network is improved to achieve accurate prediction of thickness and flatness simultaneously. It is verified that the proposed method can predict the thickness and flatness at the same time with higher prediction accuracy and generalization ability compared with other methods through actual production data. The mean absolute error (MAE) for thickness prediction was reduced by 19.37% and MAE for flatness prediction was reduced by 14.03% compared to the existing prediction model.

This paper investigates adaptive neural control methods for robotic manipulators, subject to uncertain plant dynamics and constraints on the joint position. The barrier Lyapunov function is employed to guarantee that the joint constraints are not violated, in which the Moore–Penrose pseudo-inverse term is used in the control design. To handle the unmodeled dynamics, the neural network (NN) is adopted to approximate the uncertain dynamics. The NN control based on full-state feedback for robots is proposed when all states of the closed loop are known. Subsequently, only the robot joint is measurable in practice; output feedback control is designed with a high-gain observer to estimate unmeasurable states. Through the Lyapunov stability analysis, system stability is achieved with the proposed control, and the system output achieves convergence without violation of the joint constraints. Simulation is conducted to approve the feasibility and superiority of the proposed NN control. [ABSTRACT FROM AUTHOR]

This paper investigates the data-driven realization of predictive controllers, and its application to fault-tolerant control. With the aid of subspace identification method, the subspace predictor is first derived in the closed-loop setup. To cope with the change in the process, a recursive algorithm is adopted to recursive update the subspace matrices. The data-driven predictive controller is designed by seeking the solution of the quadratic programming problem, without the information of the predesigned controller. Then, it is applied to fault-tolerant control by implementing the recursive subspace predictor. In the end, the effectiveness of the proposed method is verified by a case study on the three-tank system.

Deep learning has been obtained extensive attention in many fields. In this paper, a fault detection based on deep belief network (DBN) method is proposed for nonlinear processes. Then the industrial processes abnormal monitoring is realized by test statistics, which is built by feature variables and residual variables produced by DBN. The Tennessee-Eastman (TE) process have been used to evaluate the efficiency of the proposed method.

Modern hot strip mill process (HSMP) is increasingly characterized by producing small-batch, multi-specification and high-value-added products. Associated with such suffered complexities, performing valid process monitoring/fault diagnosis (PM-FD) is becoming a challenging task of ensuring process safety and product quality. In this paper, a new PM-FD approach based on principal component regression (PCR) is proposed for quality-relevant fault detection and diagnosis of HSMP. Firstly, the historical multi-batch process and quality variables data sets (three-dimensional) should be appropriately transferred into the availably applied two-dimensional data sets. Then, the presented approach could orthogonally project the process variable space into the quality-relevant part from the -irrelevant part. Next, when there comes a new measurement regarding a new batch from an unknown specification of the strip, it is automatically assigned into its preferential model by the prediction power of PCR integrated with Bayes inference. In depth, the detection and diagnosis are continued by the presented PCR based monitoring scheme. To the end, the new proposed scheme would be practiced with real HSMP data, where the individual steps as well as the complete framework were extensively tested. [ABSTRACT FROM AUTHOR]

This paper proposes two different approaches to the problem of modelling the key performance indicators (KPIs) in a hot strip mill process (HSMP). The first approach takes into consideration the inherent dynamics of the process and develops a nonlinear iterative dynamic partial least squares (PLS) model. The second approach takes into consideration the multibatch property of HSMP to develop an appropriate multibatch modelling framework. Both approaches can be integrated into the dynamic model framework. The proposed framework is examined using industrial data obtained from a steel plant, where it provides a decent model for this process. However, there exists the potential for improvements including dealing with interbatch variability.

Hot strip mill process (HSMP) plays a pivotal role in steel manufacturing industry, but involves significant complexity. Several faults could cause the decreasing evaluation of the key performance indicators (KPIs). Partial least squares (PLS) model has been popularly accepted for KPI-monitoring tasks, whereas some drawbacks have been reported such as high false alarm rate and strict limitation of Gaussian distribution. In this paper, a new scheme is designed without any distributional priority. The process information is extracted by the independent component analysis (ICA) and principal component analysis (PCA) one after another to obtain the Non-Gaussianity and Gaussianity rooted in process variables. Then the correlation canonical analysis (CCA), a classic tool of analyzing the correlation of two data sets, will be utilized to incorporate the process information and KPIs. Finally, two KPI-related indices are formed respectively, which are both bounded by key density estimation based approach. In the end, application of the new approach in a real steel plant will be demonstrated, where the comparison with PLS based results is covered. [ABSTRACT FROM AUTHOR]

A fault detection, isolation, and estimation approach is proposed in this paper based on Interactive Multimodel (IMM) fusion filtering and Strong Tracking Filtering (STF) for asynchronous multisensors dynamic systems. Time-varying fault is considered and a candidate fault model is built by augmenting the unknown fault amplitude directly into the system state for each kind of possible fault mode. By doing this, the dilemma of predetermining the fault extent as model design parameters in traditional IMM-based approaches is avoided. After that, the time-varying fault amplitude is estimated based on STF using its strong ability to track abrupt changes and robustness against model uncertainties. Through fusing information from multiple sensors, the performance of fault detection, isolation, and estimation is approved. Finally, a numerical simulation is performed to demonstrate the feasibility and effectiveness of the proposed method.

Focusing on quality-related complex industrial process performance monitoring, a novel multimode process monitoring method is proposed in this paper. Firstly, principal component space clustering is implemented under the guidance of quality variables. Through extraction of model tags, clustering information of original training data can be acquired. Secondly, according to multimode characteristics of process data, the monitoring model integrated Gaussian mixture model with total projection to latent structures is effective after building the covariance description form. The multimode total projection to latent structures (MTPLS) model is the foundation of problem solving about quality-related monitoring for multimode processes. Then, a comprehensive statistics index is defined which is based on the posterior probability of the monitored samples belonging to each Gaussian component in the Bayesian theory. After that, a combined index is constructed for process monitoring. Finally, motivated by the application of traditional contribution plot in fault diagnosis, a gradient contribution rate is applied for analyzing the variation of variable contribution rate along samples. Our method can ensure the implementation of online fault monitoring and diagnosis for multimode processes. Performances of the whole proposed scheme are verified in a real industrial, hot strip mill process (HSMP) compared with some existing methods.

An asynchronous RUL fusion estimation algorithm is presented for the hidden degradation process with multiple asynchronous monitoring sensors based on multisource information fusion. Firstly, a state-space type model is established by modeling the stochastic degradation as a Wiener process and transforming asynchronous indirectly observations in the fusion period to the fusion time. The statistical characteristics of involved noises and their correlations are analyzed. Secondly, the estimate of the hidden degradation state is obtained by applying Kalman filtering with correlated noises to the established state-space model, where the synchronized observations are fused. Also, the unknown model parameters are recursively identified based on the Expectation-Maximization (EM) algorithm with the Generic Algorithm (GA) adopted to solve the maximization problem. Finally, the probability distribution of RUL is obtained using the fused degradation state estimation and the updated identification result of the model parameters. Simulation results show that the proposed fusion method has better performance than the RUL estimation with single sensor.

Abstract In this paper, a practical technology or solution of quality-related fault diagnosis is provided for nonlinear and dynamic process. Unlike traditional data-based fault diagnosis methods, the alternative approach is focused more on identifying the propagation path that combines diagnostic information and process knowledge. The new method addresses the quality-related fault detection issue with developed nonlinear dynamic latent variable model for extracting nonlinear latent variables that exhibit dynamic correlations, then the advantage of relative reconstruction based contribution approach is followed to analyze the potential root-cause variables. Meanwhile, a new partitioned Bayesian network methodology is proposed for propagation path identification of quality-related faults. Finally, the whole proposed framework is applied to a real hot strip mill process, where the effectiveness is further demonstrated from real industrial data. [ABSTRACT FROM AUTHOR]