Batch Process Fault Research Articles

A generalized zero-shot semantic learning model for batch process fault diagnosis

Abstract In industrial monitoring, although zero-shot learning successfully solves the problem of diagnosing unseen faults, it is difficult to diagnose both unseen and seen faults. Motivated by this, we propose a generalized zero-shot semantic learning fault diagnosis model for batch processes called joint low-rank manifold distributional semantic embedding and multimodal variational autoencoder (JLMDSE-mVAE). Firstly, joint low-rank representation and manifold learning makes the training samples map to the low-rank space, which obtains the global-local features of the samples while reducing the redundancy in the inputs for the training model; Secondly, the bias of human-defined semantic attributes is corrected by predicting the attribute error rate; Then, fault samples and corrected semantic vectors are embedded into the consistency space, in which the samples are reconstructed using the multimodal variational autoencoder to fully integrate the cross-modal information, meanwhile, Barlow matrix is designed to measure the consistency between the fault samples and the attribute vectors, the higher the consistency, the higher the learning efficiency of attribute classifiers; Finally, the generalized zero-shot fault diagnosis experiments are designed and conducted on the Penicillin fermentation process and the Semiconductor etching process to validate the effectiveness, the results show that the proposed model is indeed possible to diagnose target faults without their samples.

An adaptive strategy for time‐varying batch process fault prediction based on stochastic configuration network

AbstractFault prediction ensures safe and stable production, and cuts maintenance costs. Due to the changing operating conditions that lead to the changes in the characteristics of industrial processes, there is a need to monitor the fault state of batch processes in real‐time and to accurately predict fault trends. An adaptive slow feature analysis‐neighborhood preserving embedding‐improved stochastic configuration network (SFA‐NPE‐ISCN) algorithm for batch process fault prediction is proposed. Firstly, SFA is used to extract the time‐varying features of process data and establish the update index of the NPE model. Then, to extract local nearest‐neighbor features and reconstruct them by the NPE model with adaptive update capability, square prediction error (SPE) statistics are constructed as fault state features based on the reconstructed error. Further, the hunter‐prey optimization (HPO) algorithm optimizes the weights and biases in the stochastic configuration network, and the singular value decomposition (SVD) and QR decomposition of column rotation are introduced to solve the ill‐posed problem of SCN and obtain the prediction model of ISCN. Finally, the obtained statistics SPE is formed into a time series, and the ISCN model is used to predict the process state trend. The effectiveness of the proposed algorithm is verified by case studies of industrial‐scale penicillin fermentation processes and the Hot strip mill process.

Cycle temporal algorithm-based multivariate statistical methods for fault diagnosis in chemical processes

Multivariate statistical process monitoring methods are often used in chemical process fault diagnosis. In this article, (I) the cycle temporal algorithm (CTA) combined with the dynamic kernel principal component analysis (DKPCA) and the multiway dynamic kernel principal component analysis (MDKPCA) fault detection algorithms are proposed, which are used for continuous and batch process fault detections, respectively. In addition, (II) a fault variable identification model based on reconstructed-based contribution (RBC) model that paves the way for determining the cause of the fault are proposed. The proposed fault diagnosis model was applied to Tennessee Eastman (TE) process and penicillin fermentation process for fault diagnosis. And compare with other fault diagnosis methods. The results show that the proposed method has better detection effects than other methods. Finally, the reconstruction-based contribution (RBC) model method is used to accurately locate the root cause of the fault and determine the fault path.

Efficient batch process monitoring based on random nonlinear feature analysis

AbstractEfficient batch process monitoring is a challenging task because of the large data scale, the strong process nonlinearity, and the complicated local behaviours. To handle this issue, this paper presents a random nonlinear feature analysis method, called two‐level localized ensemble random Fourier feature analysis (TLERFFA), for effective batch process fault detection. Considering the large data scale of the nonlinear batch process, the random Fourier feature analysis (RFFA) is introduced to develop the statistical model, which is more efficient than the traditional kernel modelling method in terms of computation loads. Furthermore, for overcoming the model uncertainty resulting from the random mapping weights, the ensemble learning based on Bayesian inference is used to integrate multiply RFFA models with different weight settings. To further mine the rich local information of batch process data, a two‐level localization strategy is designed from the perspectives of the variables and the statistics. At the first level, the monitored variables are divided into several sub‐groups according to their maximal information coefficients, while at the second level, the local monitoring statistics are constructed based on the temporal–spatial neighbour analysis. Lastly, one case study on the benchmark process is given to demonstrate the effectiveness of the proposed methods.

Deep Convolutional Feature-Based Probabilistic SVDD Method for Monitoring Incipient Faults of Batch Process

Support vector data description (SVDD) has been widely applied to batch process fault detection. However, it often performs poorly, especially when incipient faults occur, because it only considers the shallow data feature and omits the probabilistic information of features. In order to provide better monitoring performance on incipient faults in batch processes, an improved SVDD method, called deep probabilistic SVDD (DPSVDD), is proposed in this work by integrating the convolutional autoencoder and the probability-related monitoring indices. For mining the hidden data features effectively, a deep convolutional features extraction network is designed by a convolutional autoencoder, where the encoder outputs and the reconstruction errors are used as the monitor features. Furthermore, the probability distribution changes of these features are evaluated by the Kullback-Leibler (KL) divergence so that the probability-related monitoring indices are developed for indicating the process status. The applications to the benchmark penicillin fermentation process demonstrate that the proposed method has a better monitoring performance on the incipient faults in comparison to the traditional SVDD methods.

A sequential resampling approach for imbalanced batch process fault detection in semiconductor manufacturing

Fault detection is one of the most important research topics to guarantee safe operation and product quality consistency especially in the batch process of semiconductor manufacturing. However, the imbalanced fault data bring great challenges to extract the high nonlinearity and inherently time-varying dynamics of the batch process. Motivated by these, we propose a sequential oversampling discrimination approach for imbalanced batch process fault detection. Especially, different from the traditional oversampling methods, which extract temporal features from the whole process, we transform a whole batch sequence into multiple fixed-length sequences each batch by a sliding window, to extract the robust time-varying dynamics features. Then, an oversampling neural network is performed to balance both sequences of minority and majority classes. The needed sequences of the minority class are generated by an improved combination model of variational auto-encoder and generative adversarial network. Finally, a simplified sequential neural network is learned by the balanced-class sequences to perform the discrimination. We conduct extensive experiments based on two datasets of semiconductor manufacturing. One is a benchmark dataset and the other is a dataset from a real production line. The results achieved significant improvement, compared with other state-of-art fault detection methods and oversampling techniques.

Dynamic nonlinear batch process fault detection and identification based on two‐directional dynamic kernel slow feature analysis

AbstractThe batch process generally covers high nonlinearity and two‐directional dynamics: time‐wise dynamics, which correspond to inherently time‐varying dynamics resulting from the slowly varying underlying driving forces within each batch duration; and batch‐wise dynamics, which are associated with different operating modes among different batches. However, most existing dynamic nonlinear monitoring methods cannot extract the slowly varying underlying driving forces of the nonlinear batch process and rarely tackle the batch‐wise dynamic characteristics among batch runs. In order to address these issues, a new monitoring scheme based on two‐directional dynamic kernel slow feature analysis (TDKSFA) is developed by combining kernel SFA with a global modelling strategy. In the TDKSFA method, kernel SFA is integrated with the ARMAX time series model to mine the nonlinear and time‐wise dynamic properties within a batch run due to its capability of extracting the slowly varying underlying driving forces. Furthermore, the global modelling strategy is presented to handle the batch‐wise dynamics among batches by calculating the total average kernel matrix of all training batches. After the slow features are extracted, Hotelling's T2 and SPE statistics are built to detect faults. To solve the issue of fault variable nonlinear identification, a novel nonlinear contribution plot inspired by the pseudo‐sample variable projection trajectories in the TDKSFA model is further proposed to identify fault variables. Finally, the feasibility and effectiveness of the TDKSFA‐based fault diagnosis strategy is demonstrated through a numerical system and the penicillin fermentation process.

Batch process fault detection for multi-stage broad learning system

In the real industrial production process, some minor faults are difficult to be detected by multivariate statistical analysis methods with mean and variance as detection indicators due to the aging equipment and catalyst deactivation. With structural characteristics, deep neural networks can better extract data features to detect such faults. However, most deep learning models contain a large number of connection parameters between layers, which causes the training time-consuming and thus makes it difficult to achieve a fast-online response. The Broad Learning System (BLS) network structure is expanded without a retraining process and thus saves a lot of training time. Considering that different stages of the batch production process have different production characteristics, we use the Affinity Propagation (AP) algorithm to separate the different stages of the production process. This paper conducts research on a multi-stage process monitoring framework that integrates AP and the BLS. Compared with other monitoring models, the monitoring results in the penicillin fermentation process have verified the superiority of the AP-BLS model.

Enhanced high‐order information extraction for multiphase batch process fault monitoring

AbstractConventional independent component analysis (ICA) monitoring methods extract the feature information of process data by selecting more important independent components (ICs), which discard a small part of ICs that may contain useful information for faults, leading to unsatisfactory monitoring results. However, when the number of sampling points is greater than that of process variables, the ICA monitoring model does not work well. To address the aforementioned problems, a novel monitoring method, multiphase enhanced high‐order information extraction (MEHOIE), is proposed in this paper. The entire production process was first divided into several steady phases and transition phases by the affinity propagation (AP) phase partitioning method. The enhanced high‐order information extraction (EHOIE) model was then built in each phase for fault monitoring. Finally, the algorithm was applied in the penicillin simulation platform and industrial microbial pharmaceutical process. The flexibility and superiority of this algorithm were verified by comparing it with other conventional methods.

Phase partition and identification based on kernel entropy component analysis and multi-class support vector machines-fireworks algorithm for multi-phase batch process fault diagnosis

For the characteristics of nonlinear and multi-phase in the batch process, a self-adaptive multi-phase batch process fault diagnosis method is proposed in this paper. Firstly, kernel entropy component analysis (KECA) method is used to achieve multi-phase partition adaptively, which makes the process data mapped into the high-dimensional feature space and then constructs the core entropy and the angular structure similarity. Then a multi-phase KECA failure monitoring model is developed by using the angular structure similarity as the statistic, which is based on the partitioned phases and the effective failure features by the KECA feature extraction method. A multi-phase batch process fault diagnosis method, which applies the multi-class support vector machines (MSVM) and fireworks algorithm (FWA), is proposed to recognize each sub-phase fault diagnosis automatically. The effectiveness and advantages of the proposed multi-phase fault diagnosis method are illustrated with a case study on a fed-batch penicillin fermentation process.

KECA Similarity-Based Monitoring and Diagnosis of Faults in Multi-Phase Batch Processes.

Multiple phases with phase to phase transitions are important characteristics of many batch processes. The linear characteristics between phases are taken into consideration in the traditional algorithms while nonlinearities are neglected, which can lead to inaccuracy and inefficiency in monitoring. The focus of this paper is nonlinear multi-phase batch processes. A similarity metric is defined based on kernel entropy component analysis (KECA). A KECA similarity-based method is proposed for phase division and fault monitoring. First, nonlinear characteristics can be extracted in feature space via performing KECA on each preprocessed time-slice data matrix. Then phase division is achieved with the similarity variation of the extracted feature information. Then, a series of KECA models and slide-KECA models are established for steady and transitions phases respectively, which can reflect the diversity of transitional characteristics objectively and preferably deal with the stage-transition monitoring problem in multistage batch processes. Next, in order to overcome the problem that the traditional contribution plot cannot be applied to the kernel mapping space, a nonlinear contribution plot diagnosis algorithm is proposed, which is easier, more intuitive and implementable compared with the traditional one. Finally, simulations are performed on penicillin fermentation and industrial application. Specifically, the proposed method detects the abnormal agitation power and the abnormal substrate supply at 47 h and 86 h, respectively. Compared with traditional methods, it has better real-time performance and higher efficiency. Results demonstrate the ability of the proposed method to detect faults accurately and effectively in practice.

Batch process fault detection and identification based on discriminant global preserving kernel slow feature analysis

As an attractive nonlinear dynamic data analysis tool, global preserving kernel slow feature analysis (GKSFA) has achieved great success in extracting the high nonlinearity and inherently time-varying dynamics of batch process. However, GKSFA is an unsupervised feature extraction method and lacks the ability to utilize batch process class label information, which may not offer the most effective means for dealing with batch process monitoring. To overcome this problem, we propose a novel batch process monitoring method based on the modified GKSFA, referred to as discriminant global preserving kernel slow feature analysis (DGKSFA), by closely integrating discriminant analysis and GKSFA. The proposed DGKSFA method can extract discriminant feature of batch process as well as preserve global and local geometrical structure information of observed data. For the purpose of fault detection, a monitoring statistic is constructed based on the distance between the optimal kernel feature vectors of test data and normal data. To tackle the challenging issue of nonlinear fault variable identification, a new nonlinear contribution plot method is also developed to help identifying the fault variable after a fault is detected, which is derived from the idea of variable pseudo-sample trajectory projection in DGKSFA nonlinear biplot. Simulation results conducted on a numerical nonlinear dynamic system and the benchmark fed-batch penicillin fermentation process demonstrate that the proposed process monitoring and fault diagnosis approach can effectively detect fault and distinguish fault variables from normal variables.

Feature extraction for batch process monitoring and fault detection via simultaneous data scaling and training of tensor based models

This contribution presents a novel framework for process monitoring and fault detection of batch processes. The batch platform is key for the full scale production of specialty chemicals, active pharmaceutical ingredients, biochemical products and other high added value products, where flexibility is required due to the relatively limited life span of the products and/or the diversity on big product portfolios of relatively small scale. Traditional approaches for fault identification in batch processes require unfolding the third order structure of the data (variables × time × batches). In this contribution a tensor based approach is considered because it applies a more comprehensive multilinear analysis of the data. The novel approach allows to impose independence relations through knowledge based structural sparsity constraints on the tensor decomposition. Additionally, a novel methodology is presented for simultaneous data scaling and model training. This allows finding the optimal data scale to exploit the variability present in all directions (i.e., from every variable). The combination of the two proposed methods results in an improved framework to extract interpretable features from the data. The application of the proposed novel framework to the dynamic simulation of the fed-batch penicillin production (Pensim) allows to evaluate its advantages over traditional methods for multivariate statistical process monitoring (MSPM). Some identified advantages of the novel framework are a better distribution on the approximation of the data with lower noise propagation and bias, as well as a higher interpretability of the extracted features and the fault detection. This results in an enhanced capability for monitoring batch processes and fault detection.

Discriminant diffusion maps based k‐nearest‐neighbour for batch process fault detection

In this paper, a novel discriminant diffusion maps based k‐nearest‐neighbour rule (DDM‐kNN) is developed for the high‐dimensional data of batch process and nonlinear characteristics of the data, and the traditional multivariate statistical process control and monitoring methods is less effective. Firstly, a discriminant kernel parameter is applied to the framework of the diffusion maps, and the Gaussian kernel width is selected from the within‐class width and the between‐class width according to discriminating sample class labels, which can make kernel function effectively extract data correlation features and exactly describe the structure characteristics of data original space in the low‐dimensional feature. Subsequently, the adapted kNN rule is applied to the low‐dimensional manifold feature space for fault detection. The effectiveness of DDM for the performance of data dimension reduction and feature extraction is verified in 3 arm spiral test data experiments compared with other dimensionality reduction methods, and successfully shows the high‐dimensional data in the low‐dimensional space and optimally preserves the original intrinsic nonlinear structure of the dataset. In addition, DDM‐kNN is applied to penicillin fermentation process monitoring and fault detection, and results also verify the effectiveness of the proposed method by integrating discriminant diffusion maps with k‐nearest‐neighbour rule.

Quality-Relevant Batch Process Fault Detection Using a Multiway Multi-Subspace CVA Method

For batch process fault detection, regular data-driven methods cannot distinguish quality-irrelevant faults from quality-relevant faults. To solve such problem, we propose a multiway multi-subspace canonical variate analysis (MMCVA) method for the batch processes. First, the combination of batch-wise unfolding and variable-wise unfolding is adopted to unfold the three-way process and quality data in to two-way data. Then, we use CVA to project the process and quality data spaces to three subspaces, a process-quality correlated subspace, a quality-uncorrelated process subspace, and a process-uncorrelated quality subspace. Fault detection statistics are developed based on the three subspaces. The proposed MMCVA method is capable of indicating the normality or abnormality of the quality variables, while detecting a process fault. The simulation results of a fed-batch penicillin fermentation process illustrate the effectiveness of the proposed method.

Monitoring of within batch and batch-to-batch dynamics using adaptive LASSO

The output of a batch process under within batch operation control, and batch-to-batch feedback control can be represented by a two-dimensional auto-regressive moving average (2D ARMA) time series. However, it is not easy to identify this time series under close-loop conditions. In this paper, an adaptive least absolute shrinkage and selection operator (LASSO) method is used to identify the orders and coefficients of this 2D time series. The estimated coefficients of 2D time series model as inputs are to calculate the stability indexes by inner-matrix method. These indexes can be monitored by traditional control charts. On basis of this, the faults are detected by applying on these stability indices. The simulation results show that these stability indexes are sensitive to the faults in 2D batch processes, verifying the effectiveness of the proposed method. Furthermore, the detection capabilities are much superior when adaptive LASSO was used instead of other identification techniques.

Quality relevant nonlinear batch process performance monitoring using a kernel based multiway non-Gaussian latent subspace projection approach

Multiway kernel partial least squares method (MKPLS) has recently been developed for monitoring the operational performance of nonlinear batch or semi-batch processes. It has strong capability to handle batch trajectories and nonlinear process dynamics, which cannot be effectively dealt with by traditional multiway partial least squares (MPLS) technique. However, MKPLS method may not be effective in capturing significant non-Gaussian features of batch processes because only the second-order statistics instead of higher-order statistics are taken into account in the underlying model. On the other hand, multiway kernel independent component analysis (MKICA) has been proposed for nonlinear batch process monitoring and fault detection. Different from MKPLS, MKICA can extract not only nonlinear but also non-Gaussian features through maximizing the higher-order statistic of negentropy instead of second-order statistic of covariance within the high-dimensional kernel space. Nevertheless, MKICA based process monitoring approaches may not be well suited in many batch processes because only process measurement variables are utilized while quality variables are not considered in the multivariate models. In this paper, a novel multiway kernel based quality relevant non-Gaussian latent subspace projection (MKQNGLSP) approach is proposed in order to monitor the operational performance of batch processes with nonlinear and non-Gaussian dynamics by combining measurement and quality variables. First, both process measurement and quality variables are projected onto high-dimensional nonlinear kernel feature spaces, respectively. Then, the multidimensional latent directions within kernel feature subspaces corresponding to measurement and quality variables are concurrently searched for so that the maximized mutual information between the measurement and quality spaces is obtained. The I2 and SPE monitoring indices within the extracted latent subspaces are further defined to capture batch process faults resulting in abnormal product quality. The proposed MKQNGLSP method is applied to a fed-batch penicillin fermentation process and the operational performance monitoring results demonstrate the superiority of the developed method as apposed to the MKPLS based process monitoring approach.

An ARMA Model Integrated MPCA Approach to Fault Prediction in Batch Processes

To on-line forecast faults in batch processes, we introduced integrated models combining MPCA and ARMA. Time series of T2 and SPE statistics obtained from MPCA models based on variable cross-correlations are employed to build univariate ARMA models in terms of their auto-correlations, predicting the h-step-ahead values implying process dynamic behaviors. The integrated models not only take advantage of both powerful data compression and prominent prediction ability, but also improve the performance of MPCA based process monitoring as well as trend forecasting, which are of great significance to ensure safe and smooth running of batch processes. Finally, experimental studies consisting in fed-batch penicillin benchmark problems demonstrate the effectiveness and potentials of the contributions.

Using warping information for batch process monitoring and fault classification

This paper discusses how to use the warping information obtained after batch synchronization for process monitoring and fault classification. The warping information can be used for i) building unsupervised control charts or ii) fault classification when a rich faulty batches database is available. Data from realistic simulations of a fermentation process of the Saccharomyces cerevisiae cultivation are used to illustrate the proposal.

Multiway discrete hidden Markov model‐based approach for dynamic batch process monitoring and fault classification

AbstractA new multiway discrete hidden Markov model (MDHMM)‐based approach is proposed in this article for fault detection and classification in complex batch or semibatch process with inherent dynamics and system uncertainty. The probabilistic inference along the state transitions in MDHMM can effectively extract the dynamic and stochastic patterns in the process operation. Furthermore, the used multiway analysis is able to transform the three‐dimensional (3‐D) data matrices into 2‐D measurement‐state data sets for hidden Markov model estimation and state path optimization. The proposed MDHMM approach is applied to fed‐batch penicillin fermentation process and compared to the conventional multiway principal component analysis (MPCA) and multiway dynamic principal component analysis (MDPCA) methods in three faulty scenarios. The monitoring results demonstrate that the MDHMM approach is superior to both the MPCA and MDPCA methods in terms of fault detection and false alarm rates. In addition, the supervised MDHMM approach is able to classify different types of process faults with high fidelity. © 2011 American Institute of Chemical Engineers AIChE J, 2012