This paper focuses on the time series prediction problem for underflow concentration of deep cone thickener. It is commonly used in the industrial sedimentation process. In this paper, we introduce a dual attention neural network method to model both spatial and temporal features of the data collected from multiple sensors in the thickener to predict underflow concentration. The concentration is the key factor for future mining process. This model includes encoder and decoder. Their function is to capture spatial and temporal importance separately from input data, and output more accurate prediction. We also consider the domain knowledge in modeling process. Several supplementary constructed features are examined to enhance the final prediction accuracy in addition to the raw data from sensors. To test the feasibility and efficiency of this method, we select an industrial case based on Industrial Internet of Things (IIoT). This Tailings Thickener is from FLSmidth with multiple sensors. The comparative results support this method has favorable prediction accuracy, which is more than 10% lower than other time series prediction models in some common error indices. We also try to interpret our method with additional ablation experiments for different features and attention mechanisms. By employing mean absolute error index to evaluate the models, experimental result reports that enhanced features and dual-attention modules reduce error of fitting ~5% and ~11%, respectively.
4 October 2024 in IEEE Transactions on Industrial Informatics
In the field of mineral processing, controlling the paste thickener is a highly challenging and critical task because of the high complexity, incomplete observation space, and excessive environmental noises. In this article, we propose an offline-data-driven controlling strategy to optimize the operational indices in the thickening system based on offline reinforcement learning (RL). Compared to common RL methods that rely on online interactive training, our approach ensures the safety of the production process by training the controller solely using offline datasets, thereby avoiding dangerous online exploration. In terms of offline dataset collection, this study utilizes the prior knowledge of the thickening mechanism to design a proportional–integral–derivative controller as the behavior policy to collect operational trajectories as the offline dataset. In addition, to tackle a critical issue in controlling the thickening system: constrained observation space, this article analyzes the dynamical properties of the thickening system and introduces a novel offline RL algorithm, temporal batch-constrained Q-learning (TBCQ). The algorithm and associated model framework are specifically developed for controlling partially observed Markov decision processes. The TBCQ and trained policy are evaluated in both a simulated thickening environment and a real industrial paste thickener in a copper mine. The real-world experiments demonstrate that the proposed controller outperforms the baselines and effectively reduces the tracking error of underflow concentration by over 12%. The successful application of our pipeline in paste thickener also offers an innovative perspective on addressing optimization problems in complex industrial systems: performing offline RL on a dataset sampled from a suboptimal policy.
19 September 2022 in IEEE Transactions on Industrial Informatics
Periodic Jump processes commonly occur in complex industrial systems. As the systems vary dynamically between different stages, learning their dynamics in an unified model, so as to forecast and simulation accurately is challenging. In this study, we propose autonomous jump ordinary differential equation net (AJ-ODENet) to learn the continuous-time periodic jump system. The model consists of several Hierarchical ODENets (H-ODENets) and a stage transition predictor. Each H-ODENet is an advanced version of ordinary differential equations network to individually learn specific dynamics in each stage from irregularly sampled sequence data. The stage transition predictor realizes autonomous stage transition during open-loop simulation. Furthermore, an encoder–decoder framework built on AJ-ODENet is employed on a real cooling system of data center to simulate some variables in runtime. With multivariate data given, such as server power and environmental temperature, the model can simulate the working patterns as in reality, and the relative error of the predicted energy consumption is within 5%. Furthermore, based on the model, we infer the optimal cooling temperature settings under different heat loads. The simulation results indicate that 6%–25% of cooling energy consumption can be optimized.
10 July 2025 in IEEE Transactions on Intelligent Transportation Systems
This paper explores the application of Cyber-physical Internet (CPI) in prefabricated construction logistics to enhance module shipment efficiency through a new scheduling framework. Drawing parallels between the TCP/IP model’s data transmission process and the physical shipment of construction modules, the study identifies inefficiencies in current logistics practices, including obstructed information sharing and collaboration among practitioners. To address these challenges, the paper proposes a suite of CPI protocols, like the internet protocols, to standardize information sharing, scheduling & routing rules among logistics practitioners and nodes. Based on the CPI protocols, a hierarchical and decentralized shipment decision framework is proposed to govern how the routing decisions and shipment scheduling decisions are made at each logistics node. A set of numerical experiments is conducted based on a real-life shipment case of construction modules in the Greater Bay Area to exhibit the great efficiency and resilience of the proposed protocol-based decision framework. And a case study is designed to show how the proposed protocols influence the decision process. The study’s contributions are threefold: demonstrating CPI application in a logistics scenario, developing protocols for efficient information sharing, and proposing a new decision framework for resilient and timely scheduling in complex logistics networks.
4 March 2025 in IEEE/CAA Journal of Automatica Sinica
In high-risk industrial environments like nuclear power plants, precise defect identification and localization are essential for maintaining production stability and safety. However, the complexity of such a harsh environment leads to significant variations in the shape and size of the defects. To address this challenge, we propose the multivariate time series segmentation network (MSSN), which adopts a multiscale convolutional network with multi-stage and depth-separable convolutions for efficient feature extraction through variable-length templates. To tackle the classification difficulty caused by structural signal variance, MSSN employs logarithmic normalization to adjust instance distributions. Furthermore, it integrates classification with smoothing loss functions to accurately identify defect segments amid similar structural and defect signal subsequences. Our algorithm evaluated on both the Mackey-Glass dataset and industrial dataset achieves over 95% localization and demonstrates the capture capability on the synthetic dataset. In a nuclear plant's heat transfer tube dataset, it captures 90% of defect instances with 75% middle localization F1 score.