Complex Industrial Processes Research Articles

SSAE‐KPLS: A quality‐related process monitoring via integrating stacked sparse autoencoder with kernel partial least squares

AbstractKernel partial least squares (KPLS) is widely employed to address the issue of nonlinearity inherent in complex industrial processes. However, KPLS can only extract shallow features from process measurements. This paper proposes a new quality‐related process monitoring method via integrating stacked sparse autoencoder (SSAE) with KPLS (SSAE‐KPLS). First, an SSAE model is employed to exploit the nonlinearity within process variables. Through SSAE, hierarchical features are learned to extract latent representations of process variables from multiple sparse autoencoder (SAE) layers. Second, the learned hierarchical features from SSAE are used as input, and the final quality variables are used as output. A KPLS model is then built to exploit the nonlinear relationship between the hierarchical features extracted from process variables and the final product quality for process monitoring. Third, Hotelling's and statistics are employed to detect the quality‐unrelated and quality‐related faults, respectively. Finally, experiments on a numerical example and the commonly used industrial benchmark of the Tennessee Eastman process (TEP) are conducted to illustrate the efficacy and merits of the proposed SSAE‐KPLS based quality‐related process monitoring method by comparing it with other related methods.

Assessment of process operating performance with supervised probabilistic slow feature analysis

Assessment of operating performance, which is adopted to assess optimal degree for complex industrial process, has been increasing concerned in the last few years. In this study, a supervised probabilistic slow feature analysis (SPSFA) is proposed to assess the optimal degree of process performance by deep mining process variables information on the basis of the probability distribution model. First, the slow features (SFs) are modeled in state–space form as a serial correlation feature to effectively represent the potential variation of performance, and the serial correlation hiding in the performance features reveals the potential changes of performance, which helps improve cognitive depth and assessment accuracy of performance. Then, multi-order dynamic information of SFs is further combined to design the multi-order dynamic index, which accurately and timely depict the dynamic changes of the process. Hence, the proposed method provides an explicit representation of underlying driving forces of performance variations and meaningful physical interpretation of each stage from optimal to non-optimal. In addition, a sparse learning-based non-optimal factor identification method is proposed, which identifies non-optimal factor variables by reconstructing the online estimation of the Kalman filter and using Group Lasso. The feasibility and effectiveness of the proposed method is validated by an actual dense medium coal preparation process and a typical Tennessee Eastman process.

Improved Locality Preserving Projections Based on Heat-Kernel and Cosine Weights for Fault Classification in Complex Industrial Processes

Data-driven fault diagnosis techniques have been widely used in industrial processes. However, facing a large amount of high-dimensional, nonlinear, and strongly coupled process data, traditional data-driven methods achieve low diagnostic accuracy due to ignoring the structural features inside data. To overcome this problem, this article proposes an improved locality preserving projections based on the heat-kernel and cosine weight matrix named Heat-Kernel and Cosine Weights Locality Preserving Projections (HC-LPP). In HC-LPP, a novel weight matrix construction strategy is employed, where a heat-kernel function is combined with a cosine function to optimize the weight matrix between data in terms of shortening distance and angle correlation, respectively. With the new weight matrix, the proposed HC-LPP considers both the distance and the correlation of samples (the shorter the distance is, the closer the neighbors are; the smaller the angle is, the more similar the neighbors are). The dimensionality reduction process of HC-LPP can well maintain the spatial geometric structure of data. Finally, the proposed HC-LPP integrated with the AdaBoost. M2 classifier is applied to the Tennessee Eastman process and the PROcess NeTwork Optimization process for fault diagnosis performance verification. Simulation results show, the proposed HC-LPP achieves better performance in diagnostic accuracy compared with other related methods.

Novel Discriminant Locality Preserving Projection Integrated With Monte Carlo Sampling for Fault Diagnosis

In complex industrial processes, the technique of fault diagnosis has been playing an increasingly considerable role in ensuring the safety of life and property. Unfortunately, the process data of complex industrial processes have the features of high dimension. Feature extraction from high-dimensional data is promising to coping with the fault data with high dimension. Recently, one of manifold learning methods named discriminant locality preserving projection achieves excellent performance in feature extraction. However, the performance of discriminant locality preserving projection (DLPP) is subject to the problem of matrix decomposition in the denominator of the objection function caused by the small sample size (SSS) issue. To overcome this limitation, novel DLPP integrated with Monte Carlo sampling is proposed to enhance the performance of feature extraction through dimensionality reduction. In the proposed MC-DLPP, Monte Carlo sampling is first utilized to generate fault samples for each fault type. With the aid of the virtually generated fault samples, the rank of the matrix in the denominator of the objection function of DLPP increases, thus well addressing the SSS problem. The Softmax classifier is used for fault diagnosis. To test the performance of the improved DLPP-based fault diagnosis, case studies using the Tennessee Eastman process are carried out. Simulation results confirm the presented MC-DLPP achieves superior accuracy in fault diagnosis.

A Causal Model-Inspired Automatic Feature-Selection Method for Developing Data-Driven Soft Sensors in Complex Industrial Processes

The soft sensing of key performance indicators (KPIs) plays an essential role in the decision-making of complex industrial processes. Many researchers have developed data-driven soft sensors using cutting-edge machine learning (ML) or deep learning (DL) models. Moreover, feature selection is a crucial issue because a raw industrial dataset is usually high-dimensional, and not all features are conducive to the development of soft sensors. A perfect feature-selection method should not rely on hyperparameters and subsequent ML or DL models. Rather, it should be able to automatically select a subset of features for soft sensor modeling, in which each feature has a unique causal effect on industrial KPIs. Therefore, this study proposes a causal model-inspired automatic feature-selection method for the soft sensing of industrial KPIs. First, inspired by the post-nonlinear causal model, we integrate it with information theory to quantify the causal effect between each feature and the KPIs in the raw industrial dataset. After that, a novel feature-selection method is proposed to automatically select the feature with a non-zero causal effect to construct the subset of features. Finally, the constructed subset is used to develop soft sensors for the KPIs by means of an AdaBoost ensemble strategy. Experiments on two practical industrial applications confirm the effectiveness of the proposed method. In the future, this method can also be applied to other industrial processes to help develop more advanced data-driven soft sensors.

A Novel Two-Stage Fault-Detection Method Based on Constrained RVM and Integrating LDA With Minimax Probability Machine

This article presents a novel two-stage fault-detection (FD) method composed of a preclassifier and a reclassifier for complex industrial processes, where the preclassifier is developed by combining linear discriminant analysis and minimax probability machine to reduce dimensionality and classify fully separable data with low computation time. For overlapping data that cannot be separated by the preclassifier, a reclassifier is designed by constructing a constrained relevance vector machine (RVM), according to Neyman–Pearson principle, to decrease the missed alarm rate. The reclassifier has a lower computational load than traditional RVM due to the amount and dimensionality of reclassified data reduced by the first stage, thereby a balance between detection accuracy and computational burden of the whole FD method can be achieved. Finally, an industrial benchmark of Tennessee–Eastman process is utilized to verify the effectiveness of the proposed FD method.

A weighted k-nearest neighbor process monitoring method based on statistical volume pattern analysis

Process monitoring technology has developed rapidly in response to the increasing demand for safer and more reliable systems in modern process operations. Online process monitoring plays an important role not only in ensuring process safety through the timely detection of process faults but also in improving process productivity and product quality. One of the outstanding features of modern manufacturing facilities, which are large in scale and highly complex, is that the processes contain numerous variables operating under closed-loop control. Fully exploiting and utilizing the valuable information in these variables will benefit early and accurate fault detection and diagnosis of processes, minimizing downtime, increasing plant operational safety, and reducing manufacturing costs. The development of process monitoring technology is of great importance to ensure the operational safety, reliability, and economy of complex industrial processes. With the continuous development of industrial processes, the collection and use of process data are gradually increasing, and data-driven multivariate statistical process monitoring methods have developed significantly. A weighted k-nearest neighbor process monitoring method based on SPA is proposed for the problem of multimodal properties of process statistics. When the covariance difference between different modes is large, the multimodal characteristics of the original variable space are retained in the statistic feature space. By introducing weights and assigning weights to the distances between statistic samples, the proposed method can regulate the distances between statistic samples in modes with larger covariance and those with smaller covariance to the same scale, overcoming the limitations of the SPA-based process monitoring method in multimodal fault detection.

Multi‐layer features ensemble soft sensor regression model based on stacked autoencoder and vine copula

AbstractIt is crucial in industrial processes to consider key variables to ensure safe operation and high product quality. Moreover, these variables are difficult to obtain using traditional measurement methods; hence, it makes sense to develop soft sensor regression models to process the variable prediction. However, there are numerous variables integrating noisy and redundant information in complex industrial processes. Using such variables in traditional regression models may result in reducing the model's efficiency and performance. Thus, this paper proposes a multi‐layer feature ensemble soft sensor regression method using a stacked auto‐encoder (SAE) and vine copula (ESAE–VCR) to address these problems. To do so, the number of neurons in the hidden layer of the SAE is determined by the principal component analysis (PCA). The multi‐layer features of the process variables are extracted using a stacked AE, and the regression models are established for each feature layer. A linear regression ensemble method is used to combine the regression models with the multi‐layer features to obtain the final predictive model that will estimate the values of the key variables. The effectiveness and practicality of the ESAE–VCR are validated by comparing them with several common soft measurement methods in two examples. In the numerical example, the ESAE–VCR yields an accuracy of prediction (R2) of 0.9898 and a root mean square error (RMSE) of 0.1804. In the industrial example, the ESAE–VCR yields an R2 of 0.9908 and an RMSE of 0.1205.

Improved dynamic kernelPCAbased on local preserving projections and its application for electric submersible pump fault diagnosis

AbstractDynamic kernel principal component analysis (DKPCA) has been frequently implemented for nonlinear and dynamic process monitoring of complex industrial processes. However, traditional DKPCA focuses only on the global structural analysis of data sets and strongly neglects the local information, which is equally essential for process detection and identification. In this paper, an improved DKPCA, referred to as the local DKPCA (LDKPCA), is proposed based on local preserving projections (LPP) for nonlinear dynamic process fault diagnosis. The method combines the advantages of LPP and DKPCA by utilizing the local structure feature to maintain the geometric structure of the data in a unified framework. To achieve a highly comprehensive feature extraction, the local characteristics are fused in DKPCA to produce an optimization objective. The neighbouring points of the new objective function projection in the feature space are still maintained in proximity, and the variance information is retained simultaneously. For the purpose of fault detection, two statistics, known as theT2and squared prediction error (SPE) statistics, are constructed, based on the LDKPCA model, and used to monitor the latent variable space and the residual space, respectively. In addition, the sensitivity analysis is brought in for fault identification of the two statistics. Based on the experimental analysis using the shaft breakage data of an offshore oilfield electric submersible pump (ESP), the proposed method outperforms the conventional DKPCA in terms of fault monitoring performance. The experimental results demonstrate the potential of the method in nonlinear dynamic process fault diagnosis.

Special Issue on “Modelling, Monitoring, Control and Optimization for Complex Industrial Processes”

Industrial automation systems, such as chemical processes, manufacturing processes, power networks, transportation systems, sustainable energy systems, wireless sensor networks, robotic systems, and biomedical systems, are becoming more complex [...]

Advanced Statistical and Meta-Heuristic Based Optimization Fault Diagnosis Techniques in Complex Industrial Processes: A Comparative Analysis

Advanced Statistical and Meta-Heuristic Based Optimization Fault Diagnosis Techniques in Complex Industrial Processes: A Comparative Analysis

Hierarchical Causal Graph-Based Fault Root Cause Diagnosis and Propagation Path Identification for Complex Industrial Process Monitoring

Modern complex industrial processes are increasingly integrated and automated. There are numerous control loops in the process 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 paper proposes a fault root cause diagnosis and propagation path identification strategy for complex industrial process based on hierarchical causal graph (HCG). Firstly, the process is divided into several sub-blocks according to process knowledge. The monitoring model is built for fault detection. Secondly, the fault-related sub-blocks and variables are selected based on contribution plot, and the causal analysis based on transfer entropy 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 hot strip mill process (HSMP) and their comparative experiments are adopted to validate the fault diagnosis performance of the proposed framework.

Toward Robust Fault Identification of Complex Industrial Processes Using Stacked Sparse-Denoising Autoencoder With Softmax Classifier.

This article proposes a robust end-to-end deep learning-induced fault recognition scheme by stacking multiple sparse-denoising autoencoders with a Softmax classifier, called stacked spare-denoising autoencoder (SSDAE)-Softmax, for the fault identification of complex industrial processes (CIPs). Specifically, sparse denoising autoencoder (SDAE) is established by integrating a sparse AE (SAE) with a denoising AE (DAE) for the low-dimensional but intrinsic feature representation of the CIP monitoring data (CIPMD) with possible noise contamination. SSDAE-Softmax is established by stacking multiple SDAEs with a layerwise pretraining procedure, and a Softmax classifier with a global fine-tuning strategy. Furthermore, SSDAE-Softmax hyperparameters are optimized by a relatively new global optimization algorithm, referred to as the state transition algorithm (STA). Benefiting from the deep learning-based feature representation scheme with the STA-based hyperparameter optimization, the underlying intrinsic characteristics of CIPMD can be learned automatically and adaptively for accurate fault identification. A numeric simulation system, the benchmark Tennessee Eastman process (TEP), and a real industrial process, that is, the continuous casting process (CCP) from a top steel plant of China, are used to validate the performance of the proposed method. Experimental results show that the proposed SSDAE-Softmax model can effectively identify various process faults, and has stronger robustness and adaptability against the noise interference in CIPMD for the process monitoring of CIPs.

Operational control with set-points tuning - application to mobile robots

This paper proposes a novel method for the optimal tuning of set points for multiple-layered control system structure widely seen in robotics and other complex industrial processes composed of a number of subsystems. The terminal sliding mode control (SMC) is used as the low-level control strategy to ensure the stability of subsystems. When uncertainties exist, it can be shown that the deteriorated system performance will be improved by the outer loop with set points tuning. For this purpose, the learning of the new set point is designed to compensate for the effects caused by uncertainties during the system operation. At the same time, the system is proven to stay with the original set point when the compensation is introduced. A practical application to a holonomic mobile robot system is given to illustrate the presented method. Desired results have been obtained.

A novel integrated fault diagnosis method of chemical processes based on deep learning and information propagation hysteresis analysis

BackgroundTo guarantee the stationary operation of chemical plant, the anomaly symptoms of equipment should be discovered as soon as possible. Due to the limited ability of single fault diagnosis method in the increasingly complex industrial process, the integration of different methods has become a hot research direction. MethodsA three-layer fusion fault diagnosis method is proposed in this paper. This method consists of three layers. Time domain features of time series process variables are first extracted by sliding window. Then, the bidirectional long short-term memory network based on time domain feature extraction (T-BiLSTM) is formed to accomplish the online fault identification task. At last, the causalities of fault variables are obtained based on improved signed directed graph model with dynamic time warping (TD-SDG) to locate the fault sources. FindingsThe application to Tennessee Eastman (TE) shows that the average false positive rate (FPR) of T-BiLSTM is 1.98%, and the average fault diagnosis rate (FDR) is 96.6%. TD-SDG model realizes fault inference and location. This novel integrated fault diagnosis method thus can identify fault types and locate fault sources quickly based on graphical scenario object model with a high application capacity.

Life cycle assessment and potential geolocation of a multi-feedstock biorefinery: Integration of the avocado and plantain value chains in rural zones

Logistics has been identified as a hotspot when proposing biorefineries due to transport and environmental issues. This work aims to define the best location for a multi-feedstock biorefinery based on the environmental impact caused by the products transport and commercialization. Avocado and plantain residues were used as feedstocks since large amounts are produced in Colombia. The novelty of this work is addressed to (i) valorize agricultural residues of the avocado and plantain value chains by applying the biorefinery concept and (ii) define the best process location to commercialize value-added products while minimizing CO2 emissions. Biorefinery product commercialization was identified as the most contributing factor to total CO2 emissions. The total carbon footprint of the proposed multi-feedstock biorefinery with product commercialization was 2.67 tons CO2 eq/kg of avocado seed oil. This study elucidates the requirements for complex industrial process locations integrating context aspects and environmental restrictions.

A mixture of probabilistic predictable feature analysis for multi-mode dynamic process monitoring

In modern industrial processes, the demand for safety and reliability is increasingly prioritized, spawning quantities of related research on process monitoring models. It is significant for process monitoring to consider multi-mode operating conditions simultaneously. This work proposes an efficient method based on multi-mode probabilistic predictable feature analysis (MPPFA) for complex industrial process monitoring. A strategy combining deep auto-encoder (DAE) and Gaussian mixture model (GMM) is first designed to extract low-dimensional features and identify possible running modes. In line with the obtained features, original training samples are classified into corresponding Gaussian distributions with reasonable probabilities. Further, local PPFA models are established based on the separated training batches. Five monitoring statistics, including TP2, SPEP, and DIP for PPFAs, TD2 and SPED for DAE, are defined to detect abnormalities. The proposed method is first applied to a three-phase flow facility, and its superiority over comparison methods has been verified. The effectiveness of the proposed method is further demonstrated through an experiment on a practical coal pulverising system.

Root cause diagnosis for process faults based on multisensor time-series causality discovery

Root cause diagnosis (RCD) focuses on locating the critical root causes and identifying the propagation paths of the industrial process faults, which is superior to conventional fault diagnosis methods and has attracted extensive attention. However, RCD is challenging in terms of dense sensor layout and diversified interconnected subsystems in complex industrial processes, since a fault originating in one unit can propagate throughout the entire plant along with the flow of information and material, obscuring its root causes. Moreover, the previous RCD methods ignore the more crucial temporal information between multisensor time-series, suffering from the disadvantages of requiring expert knowledge and responding untimely. Therefore, a novel data-driven RCD methodology named multisensor time-series causality discovery (MTCD) is proposed for industrial processes fault diagnosis. Firstly, a temporal registration network (TRN) based on the dilated convolutional neural network (DCNN) is proposed to extract the temporal correlation between time-series and implement the time-series prediction. Ulteriorly, a permutation importance causality validation (PICV) is designed to verify the causality between time-series based on the TRN prediction results without prior knowledge or expert experience. Besides, a causal time delay discovery approach based on layerwise relevance propagation (LRP) is presented by interpreting the weights of the trained TRN. Finally, we demonstrate the effectiveness of the proposed method with a stochastic synthesis process and the Tennessee Eastman process. The experimental results show that the proposed method has better performance in both root cause localization and propagation path identification.

Pharmaceutical process optimisation: Decision support under high uncertainty

Optimisation of cell culture processes is an extremely challenging problem for the biomanufacturing industry. Limited data availability, coupled with biological complexity in modelling highly variable living cells, necessitates a decision support methodology that is performant under high levels of uncertainty. Raw materials, experimental & manufacturing facilities, and human expertise are all steeply expensive and availability is tightly constrained — planning their allocation is subject to the core uncertainties underlying the behaviour of living cells.This paper presents a novel decision support methodology for optimisation under high uncertainty. Optimisation techniques require an “objective function” that maps decision variables to the quantity being optimised, so that the decision space can be explored to find an optimum. By learning multiple types of objective function candidates with different levels of fidelity to real-world processes, our method mitigates the risk of picking a poor approximation of the objective function due to sampling effects and algorithmic randomness.Wet lab experimentation on a biomanufacturing feed optimisation problem verified that inferred candidates can successfully support domain experts in designing a new optimised feed strategy with significantly higher product yield than the current industrial control strategy. Our results indicate potential for extending our methodology to the optimisation of other complex industrial processes.

Root cause diagnosis for complex industrial process faults via spatiotemporal coalescent based time series prediction and optimized Granger causality

It is essential to conduct root cause diagnosis (RCD) to guarantee the safety of operations of chemical processes and suppress fault deterioration, yet RCD has encountered challenges in the context of the increasing complexity of industrial processes. Granger causality (GC) analysis is one of the most commonly used methods to construct process causal maps and identify root causes. However, its use is subject to a number of limitations for dynamic nonlinear industrial processes. Therefore, an optimized GC analysis is proposed to perform RCD in complex industrial processes based on the spatiotemporal coalescent-based prediction model (SCPM) and causality verification algorithm (CVA). The time series prediction model SCPM is presented as an alternative to GC’s conventional autoregressive model in order to avoid spurious regressive facing nonlinear processes, which stacks dilated convolutional neural networks (DCNN) and bidirectional gated recurrent unit (BiGRU) to coalesce time series nonlinear couplings and long-term dependence information. Moreover, considering data that may deviate from normal distribution after the fault occurs, a nonparametric causality test method CVA based on the Wilcoxon signed-rank test (WSRT) is constructed combined with SCPM to eliminate false causality discovery. Empirical results on the Tennessee Eastman process and the blast furnace process demonstrate the effectiveness of the proposed RCD method.