4 December 2024 in Materials Genome Engineering Advances
Machine learning (ML) techniques have made enormous progress in the field of materials science. However, many conventional ML algorithms operate as “black‐boxes”, lacking transparency in revealing explicit relationships between material features and target properties. To address this, the development of interpretable ML models is essential to drive further advancements in AI‐driven materials discovery. In this study, we present an interpretable framework that combines traditional machine learning with symbolic regression, using Janus III–VI vdW heterostructures as a case study. This approach enables fast and accurate predictions of stability and electronic structure. Our results demonstrate that the prediction accuracy using the classification model for stability, based on formation energy, reaches 0.960. On the other hand, the R2, MAE, and RMSE value using the regression model for electronic structure prediction, based on band gap, achieves 0.927, 0.113, and 0.141 on the testing set, respectively. Additionally, we identify a universal interpretable descriptor comprising five simple parameters that reveals the underlying physical relationships between the candidate heterostructures and their band gaps. This descriptor not only delivers high accuracy in band gap prediction but also provides explicit physical insight into the material properties.
10 December 2024 in Ironmaking & Steelmaking: Processes, Products and Applications
Industrial production process includes complex and high-risk operational procedures. Maintaining process safety and stability is essential to prevent casualties, equipment damage, and asset loss. Traditional process safety management (PSM) on production site heavily relies on manual inspections and video surveillance which inevitably overlook some hidden risks. Therefore, this paper proposes a novel multi-flow integration PSM framework to overcome the trouble. Firstly, risk factors of the material flow (MF) and energy flow (EF) are systematically analyzed and organized based on the 4M1E method. The wireless sensor network (WSN) is established by deploying multiple sensors which facilitate the data flow (DF), including data perception, collection, transmission and aggregation of different risk variables. Additionally, the analysis model to extract the information flow (IF) is obtained by comparing prediction performance of different deep learning (DL) models. Finally, response strategies against potential dangers are formulated and the control flow (CF) across production layers are achieved relying on the circulation of safety directives, the execution of control measures and the feedback. The effectiveness of this framework is verified in a steel continuous casting scenario through the acquisition of DF, the extraction of IF and the loop of CF. Results indicates that the Bi-LSTM model can achieve the most outstanding prediction performance with relative root mean square error of 4.2428%, root mean square error of 0.0551 and R-square of 0.9816. This study can aid in advancing the digitization and intelligence of the PSM and providing a practical research perspective and application mode.
31 August 2026 in SSRN Electronic Journal
Intelligent safety monitoring and accurate risk assessment for large-scale manufacturing equipment is essential to avoid catastrophic accidents. However, due to the multi-dimensional attributes of risks and the uncertainty of their evolution, existing methods lack high interpretability and efficient response capacity for identifying hazardous states. To address this, we propose an interpretable risk assessment framework by integrating spatial-temporal graph neural network (STGNN) modeling and fuzzy inference broad learning, named ST-FIBL. First, the dilated causal convolution-based temporal convolution module is designed to extract evolution features from single node, while an dynamic attention-based graph convolution module captures spatial dependency among coupled sensor nodes. The temporal and spatial embedding features are integrated in an adaptive fusion module. Then, a fuzzy inference broad learning system that combines semantic risk rules with the incremental update mechanism is proposed to improve the interpretability and efficiency of state identification. We further employ a level-aware loss optimization strategy to address the severe class imbalance and ordinal constraints. Experiments on continuous casting equipment safety monitoring dataset validate the superiority of ST-FIBL. Compared with baseline models, it achieves the best overall performance with the highest recall of 92.21% and F1-score of 90.34%. Notably, the recall rate of high-risk samples reaches 98.18%. The dominant fuzzy risk rules and activated graph nodes can provide traceable interpretable evidence chains of accountability for safety decision-making. Benefiting from lightweight parameter scale and fast inference speed, ST-FIBL offers a practical and effective paradigm for risk-sensitive safety monitoring for large-scale manufacturing equipment.
1 June 2026 in SSRN Electronic Journal
Accurate risk assessment in process manufacturing scenario is essential to avoid catastrophic infrastructure accidents. However, due to the multi-dimensional attributes of risks and the uncertainty of their evolution, existing methods lack high interpretability and efficient response capacity for identifying hazardous states. To address this, we propose an interpretable risk assessment framework by integrating spatial-temporal graph neural network (GNN) modeling and fuzzy inference broad learning, named ST-FIBL. The framework first represents multi-sensor data as a dynamic graph. Dilated causal convolution-based Temporal Convolutional Network (TCN) extracts evolution features from each monitoring node, while an attention-based graph convolution module captures spatial dependency among coupled sensors. The temporal and spatial embedding features are integrated in an adaptive fusion module. Then, a fuzzy inference broad learning system that combines semantic risk rules with the incremental update mechanism is designed to improve the interpretability and efficiency of state identification. We further employ a level-aware loss optimization strategy to address the severe class imbalance and ordinal constraints. Experiments on continuous casting safety monitoring dataset validate the superiority of ST-FIBL. Compared with baseline models, it achieves the best overall performance with the highest recall of 92.21% and F1-score of 90.34%. Notably, the recall rate of high-risk samples reaches 98.18%. The dominant fuzzy risk rules and activated graph nodes can provide traceable interpretable evidence chains of accountability for safety decision-making. Benefiting from lightweight parameter scale and fast inference speed, ST-FIBL offers a practical and effective paradigm for risk-sensitive infrastructure safety maintenance in Industry 4.0.