Journal of Process Control最新文献

This manuscript addresses the problem of leveraging thermal images for modeling and feedback control, specifically tailored for terminal quality control of batch processes. The primary objective, common in many batch processes, is to produce products with quality variables aligning with user specifications, available for measurement only at batch termination, precluding the direct use of classical control strategies. Furthermore, in many instances, traditional online sensors such as thermocouples may not be available, but instead spectral inputs like thermal images or acoustic data may be more readily available for feedback control. The challenge is to not only use the non-traditional sensor data for building a dynamic model but also to use that model for terminal quality control. The proposed approach involves a multi-layered modeling strategy. Initially, a dimensionality reduction technique is employed to condense the high-dimensional image into a set of representative outputs. Subsequently, subspace identification (SSID) is applied to develop a Linear Time-Invariant (LTI) State Space (SS) model between the inputs and the reduced outputs. Finally, a Partial Least Squares (PLS) model is constructed linking the terminal states of a batch (identified using SSID) with the product qualities obtained for that specific batch. This model is then incorporated into a Model Predictive Control (MPC) formulation. The effectiveness of the MPC is illustrated by showcasing its capability to generate products of high quality by deploying the MPC on a bi-axial lab-scale rotational molding setup.

In real industrial processes, the rapid and accurate acquisition of quality variables is essential. Therefore, this paper proposes a pruned tree-structured temporal convolutional network (PT-TCN) for efficient and accurate variables prediction. First, a novel tree network is developed, utilizing dilated causal convolution blocks as nodes to avoid the loss of local information. Each node extracts distinct local information, and by concatenating all tree nodes, the network can capture a comprehensive range of temporal scales. Then, to avoid the increased complexity caused by the tree structure, we design an online two-stage pruning strategy to compress the tree network without introducing additional computations. During the training process, blocks are initially pruned based on the correlation assessment between quality variables and tree nodes. Subsequently, weight normalization layers are employed to evaluate the importance of output channels in blocks, thereby enabling intra-block channel pruning. The effectiveness of PT-TCN is verified on Tennessee Eastman benchmark process. In addition, experiments on the real zinc flotation process demonstrate that the proposed PT-TCN improves in $R^2$ and MAE by 1.32% and 1.26% respectively in predicting quality variables, and it can reduce 91.8% parameters of the initial tree-structured TCN without sacrificing accuracy.

Quality variable prediction is crucial for improving product quality and ensuring safety for industrial processes. Recently, researchers have explored the application of graph neural networks (GNNs) for this task, leveraging process knowledge encoded in graphs. GNN-based methods have demonstrated high prediction accuracy and partial interpretability. However, these methods typically consider only one type of prior graph and fail to utilize the multi-view prior graphs that coexist in the same process. This knowledge bias prevents effective representation learning about process dynamics, leading to inconsistencies with true process dynamics and overfitting. Thus. their practical applications are limited, especially under scenarios of limited data availability. To address this, a multi-view graph convolutional network with information short (MVGCN-IS) framework is proposed. MVGCN-IS comprises three key components: multi-view graph utilization, multi-view graph fusion, and information shortcut. First, multi-view prior graphs are integrated through multiple pre-trained preliminary GCNs to extract view-specific node representations. Then, a multi-view fusion module aggregates node representations from different views into unified unit representations, capturing comprehensive process structural information. Finally, an information shortcut extracts measurement representations and integrates detailed process measurement data to further enhance model performance. The proposed MVGCN-IS framework is validated on a benzene alkylation process and a debutanizer column process, with a special focus on model reliability under small data scenarios. Experimental results demonstrate the superior prediction accuracy and improved reliability of MVGCN-IS, validating its effectiveness in representation learning and capturing process dynamics.

With gradual deepening of a low-carbon transition of energy, the application of the multi-microgrid system (MMS) is becoming more and more popular. The internal carbon pricing mechanism is a key issue in realizing low carbon of the MMS. In order to fully utilize the advantages of energy mutual benefit and collaborative optimization, a real-time carbon trading model with cost minimization is established in the day-ahead market, in which the shadow price is taken as the optimal internal carbon price and the proposed distributed algorithm protects microgrids’ privacy. Furthermore, for the purpose of amending the deviation of carbon emission between the actual and target values, we design an automated process control (APC) strategy to adjust the real-time carbon price. And then a dual-objective problem is portrayed that balances cost and carbon emission deviation minimization in the intra-day market, and it is transformed into a single-objective constrained problem to be solved. Total cost and carbon emission decrease by 4.03% and 6.17% respectively in the solution. The results show that the proposed models have great performance of cost savings and carbon reduction for the MMS.

In this paper, an actuator fault and state interval estimation method for a class of nonlinear systems is proposed by integrating observer design and zonotope analysis. For the considered systems, we present a novel unknown input observer structure with broad applications. The design procedure is based on $H_{\infty }$ method to decrease the influence of unknown but bounded process disturbances and measurement noise. Moreover, a novel interval estimation method is presented based on zonotope analysis to obtain tighter intervals. Numerical simulations of a quadruple-tank system are conducted to assess the performance of the proposed approach.

Establishing effective soft sensors relies on feature representation that is capable of capturing critical information. Stacked AutoEncoder (SAE) is able to capture the intricate structures of data characterized by high dimensionality and strong non-linearity by extracting abstract features layer by layer, making it widely used. However, the pretraining process of SAE is unsupervised, which means the features extracted cannot leverage label information to provide more actionable insights for downstream tasks. To extract more valuable feature representation, a new quality-driven dynamic weighted SAE (QD-SAE) is proposed in this paper. By incorporating supervised information dominated by the quality variable into the learned features during the pretraining of the SAE and weighting the abstract features layer by layer, the features that are beneficial to the prediction task are thus focused. In QD-SAE, the supervised information is computed by an improved attention score. In the initial state of the supervised fine-tuning process, the weighted features compose the hidden layers of the entire network. Finally, a benchmark function case and a real complex industrial process case are used to verify the effectiveness and advantages of QD-SAE. The experimental analyses demonstrate that the soft sensor constructed by the QD-SAE can predict the output variable with high accuracy and outperforms the conventional neural networks.

The traditional Smith Predictor (SP) is restricted to dealing with stable plants. In this paper, a slight modification of the SP is proposed in order to control unstable plants: systems of any order but with one unstable pole are tackled. In fact, many modifications to the SP can be found in the literature dealing with this kind of system, but none, to our best knowledge, have the simplicity of the structure here proposed. Only one or two gains are added to the traditional SP structure to achieve the stabilization of this kind of unstable system. A simple relation states the necessary and sufficient condition guaranteeing the existence of stabilizing gains, in terms of the location of the unstable pole and the size of the delay term. The range of values for the gains solving the problem is characterized. In addition, the tracking of setpoints and disturbance rejections are analysed. Some numerical examples are presented to illustrate the effectiveness of the proposed strategy, as well as one real-time example.

Model predictive control (MPC) allows for dealing with multivariable interactions, known future changes and dynamic satisfaction of constraints. Standard MPC has a cost function that aims at keeping selected controlled variables at constant setpoints. This work considers systems where the steady-state optimal active constraints change during operation. This situation is not handled optimally by standard MPC which uses fixed controlled variables for the unconstrained degrees of freedom. We propose a simple framework that detects the constraint changes and updates the controlled variables accordingly. The unconstrained controlled variables are chosen to be the reduced cost gradients, which when controlled to zero minimizes the steady-state economic cost. In this paper, the nullspace method for self-optimizing control is used to estimate the cost gradient using a static combination of the measurements. This estimated gradient is also used for detecting the current set of active constraints, which in particular allows for giving up constraints that were previously active. The proposed framework, here referred to as “region-based MPC”, is shown to be optimal for linear constrained systems with a quadratic economic cost function, and it allows for good economic performance in nonlinear systems in a neighborhood of the considered design points.

Recently, deep learning algorithms have been successfully applied to industrial fault detection because they are better at automatically extracting complex features and processing high-dimensional data than traditional methods. However, most existing deep learning-based fault detection methods only concentrate on extracting features from industrial process data without considering the crucial long-term temporal features and higher-order statistical information. To address this challenge, we proposed a novel enhanced higher-order pooling-based network (EHOPN) for industrial fault detection. First, the data pre-processing of the network is presented to capture the dynamic features of time-series process data and unify the high-dimensional data scale. Second, the EHOPN utilizes channel and temporal second-order pooling techniques to gather temporal and channel statistics information, facilitating the backbone network’s ability to capture complex inter-dependencies and long-term dynamics. Additionally, the high-order feature aggregation module is presented to aggregate global and local features, enhancing the network’s generalization ability. The proposed industrial fault detection approach is evaluated on the Tennessee Eastman benchmark and a real-world heavy-plate production process. Experimental results show that the proposed method is significantly better than comparison models in four evaluation metrics: accuracy, precision, recall, and F1-score, further proving the effectiveness of EHOPN.

Alarm floods are leading issues that compromise the efficiency of industrial alarm systems and are identified as major causes of many industrial accidents. As an advanced technique to handle alarm floods, sequence alignment based similarity analysis has been developed to match alarm flood sequences, and thus can help with further root cause identification and early warning of alarm floods. However, existing methods based on biological sequence alignment algorithms ignore the relations between alarm occurrences, and thus may cause incorrect matches or mismatches of alarms when comparing two flood sequences. Accordingly, this paper proposes a new alarm flood similarity analysis method based on global vectors and Move–Split–Merge (MSM) distance. The contributions are mainly twofold: (1) An alarm encoding model based on modified global vectors is devised to convert alarm sequences into numerical vectors that reflect the correlations of alarms; (2) a similarity analysis method based on the modified MSM distance is proposed for comparison of encoded alarm flood sequences of unequal lengths. The effectiveness of the proposed method is demonstrated through a case study with a publicly accessible industrial model for Vinyl Acetate Monomer.