Ding Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Ding Liu has a total of 158 co-authors. Here are the top co-authors by number of shared publications:
- Yin Wan
School of Automation and Information Engineering, Xi’an University of Technology, Xi’an 710048, China
15 shared publications - Jun-Chao Ren
School of Automation and Information Engineering, Xi’an University of Technology, Xi’an 710048, China
13 shared publications - Han Liu
Shaanxi Key Laboratory of Complex System Control and Intelligent Information Processing, Xian University of Technology, Xian, China
11 shared publications - Runyuan Guo
Xian University of Technology, Xian, China
9 shared publications - Youmin Zhang
Concordia University
8 shared publications - Xinyu Zhang
School of Automation and Information Engineering Xi’an University of Technology, Xi’an 710048, People’s Republic of China
5 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (22) , 53049-53071
GA-LLMSS: Text-Spatio–Temporal Graph-Guided and Cross-Modal Alignment-Enhanced Large Language Model Empowered Soft Sensor
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (22) , 53049-53071
GA-LLMSS: Text-Spatio–Temporal Graph-Guided and Cross-Modal Alignment-Enhanced Large Language Model Empowered Soft Sensor
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 34 (6) , 3171-3184
Thermal Delay Compensated in Industrial Crystal Furnaces: A Robin Boundary Control for Distributed Parameter Systems
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 34 (6) , 3171-3184
Thermal Delay Compensated in Industrial Crystal Furnaces: A Robin Boundary Control for Distributed Parameter Systems
- 25 October 2026
- Published by Elsevier in Journal of the Franklin Institute
- Vol. 363 (17)
Fuzzy-adjusted historical-input-enhanced iterative learning control for silicon single crystal growth process
- 25 October 2026
- Published by Elsevier in Journal of the Franklin Institute
- Vol. 363 (17)
Fuzzy-adjusted historical-input-enhanced iterative learning control for silicon single crystal growth process
- 1 August 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Semiconductor Manufacturing
- Vol. 39 (3) , 614-624
A Hybrid Degradation Prediction Model for the Flexible Shaft Pulling Device in Crystal Growth Furnace
Flexible shaft pulling device is a key motion device in the process of single crystal silicon preparation. The health status of the pulling device has direct influence on crystal quality. In this paper, a hybrid model is proposed for degradation prediction of the pulling device when the monitoring data is limited and with unknown statistical distribution of degradation data. Firstly, a modified filter-based degradation prediction method is designed, where a new resampling strategy is proposed to alleviate the particle degeneracy. Then, a neural network is designed to correct the potential uncertainties due to the change of the load of the pulling device. The prediction error of the filter-based method is further used to adjust the parameters of the network. Finally, experiments are conducted to verify the performance of the proposed method using both the standard dataset and data obtained from the flexible shaft pulling device.
- 1 August 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Semiconductor Manufacturing
- Vol. 39 (3) , 614-624
A Hybrid Degradation Prediction Model for the Flexible Shaft Pulling Device in Crystal Growth Furnace
Flexible shaft pulling device is a key motion device in the process of single crystal silicon preparation. The health status of the pulling device has direct influence on crystal quality. In this paper, a hybrid model is proposed for degradation prediction of the pulling device when the monitoring data is limited and with unknown statistical distribution of degradation data. Firstly, a modified filter-based degradation prediction method is designed, where a new resampling strategy is proposed to alleviate the particle degeneracy. Then, a neural network is designed to correct the potential uncertainties due to the change of the load of the pulling device. The prediction error of the filter-based method is further used to adjust the parameters of the network. Finally, experiments are conducted to verify the performance of the proposed method using both the standard dataset and data obtained from the flexible shaft pulling device.
- Open Access
Improved Langevin Surrogate-Assisted Process-Parameter Optimization for Candidate Recipe Generation in Czochralski Silicon Single Crystal Growth
To support offline process-parameter screening for Czochralski (CZ) silicon single crystal growth, this paper proposes a surrogate-assisted optimization framework based on an improved Langevin evolutionary algorithm. First, a multi-variable constrained optimization model is established, with the LSA-Transformer-predicted solid–liquid interface deformation used as the objective evaluation and with process-smoothness and physical-feasibility constraints considered. Six key process parameters–heater power, pulling rate, argon flow rate, crystal rotation speed, crucible rotation speed, and magnetic field strength–are selected as decision variables. Second, building on the classical Langevin algorithm, an adaptive inertia weight mechanism, a diversity promoter (DP) operator, and a local escaping operator (LEO) are introduced to improve global exploration and local optima escape in complex search spaces. Verification on 23 classical benchmark functions indicates that the ILEE algorithm shows competitive overall performance and achieves better or comparable results on many functions when compared with particle swarm optimization (PSO), grey wolf optimization (GWO), the original Langevin evolutionary algorithm (LEE), and other baseline algorithms. The proposed framework is then used for offline candidate recipe generation during the crystal equal-diameter growth stage (200 mm, 400 mm, 600 mm, 800 mm, and 1000 mm). The optimized candidate parameter combinations yield lower surrogate-predicted interface deformation under the given LSA-Transformer model and physical constraints. Because these values are not independent CFD or experimental measurements, the results should be interpreted as process-parameter guidance for future physical validation. This work provides a feasible surrogate-assisted offline screening framework for CZ silicon single crystal growth.
Open AccessImproved Langevin Surrogate-Assisted Process-Parameter Optimization for Candidate Recipe Generation in Czochralski Silicon Single Crystal Growth
To support offline process-parameter screening for Czochralski (CZ) silicon single crystal growth, this paper proposes a surrogate-assisted optimization framework based on an improved Langevin evolutionary algorithm. First, a multi-variable constrained optimization model is established, with the LSA-Transformer-predicted solid–liquid interface deformation used as the objective evaluation and with process-smoothness and physical-feasibility constraints considered. Six key process parameters–heater power, pulling rate, argon flow rate, crystal rotation speed, crucible rotation speed, and magnetic field strength–are selected as decision variables. Second, building on the classical Langevin algorithm, an adaptive inertia weight mechanism, a diversity promoter (DP) operator, and a local escaping operator (LEO) are introduced to improve global exploration and local optima escape in complex search spaces. Verification on 23 classical benchmark functions indicates that the ILEE algorithm shows competitive overall performance and achieves better or comparable results on many functions when compared with particle swarm optimization (PSO), grey wolf optimization (GWO), the original Langevin evolutionary algorithm (LEE), and other baseline algorithms. The proposed framework is then used for offline candidate recipe generation during the crystal equal-diameter growth stage (200 mm, 400 mm, 600 mm, 800 mm, and 1000 mm). The optimized candidate parameter combinations yield lower surrogate-predicted interface deformation under the given LSA-Transformer model and physical constraints. Because these values are not independent CFD or experimental measurements, the results should be interpreted as process-parameter guidance for future physical validation. This work provides a feasible surrogate-assisted offline screening framework for CZ silicon single crystal growth.
- Open Access
Research on V/G Value Prediction Method for Silicon Single-Crystal Growth Based on Multi-Condition Invariant Feature Extraction
In the Czochralski process of silicon single-crystal growth, the V/G value at the solid–liquid interface is a key parameter affecting intrinsic crystal defects. However, online V/G detection remains difficult because the temperature gradient G cannot be directly measured, while multi-condition distribution shifts and limited labeled data reduce the robustness of data-driven models. To address these issues, this paper proposes DWC-ISBiGNN, an adaptive multi-condition invariant feature extraction method based on the Invariant-Specific Bidirectional Graph Neural Network. The proposed method introduces dynamic sample graph construction with stage-aware global nodes to capture non-stationary process correlations, source-domain credibility weighting to suppress negative transfer, and a semi-supervised training framework combining stage-conditional alignment with teacher–student regression consistency to exploit unlabeled target-domain data. Experiments on industrial data from a 12-inch silicon single-crystal production line show that DWC-ISBiGNN achieves an RMSE of 0.0041, an MAE of 0.00285, and an of 0.9549. Compared with the original IS-BiGNN, the RMSE is reduced by 32.6%, and is increased by 5.43 percentage points. The results demonstrate that the proposed method provides an effective soft-sensing approach for V/G prediction under multiple operating conditions.Open AccessResearch on V/G Value Prediction Method for Silicon Single-Crystal Growth Based on Multi-Condition Invariant Feature Extraction
In the Czochralski process of silicon single-crystal growth, the V/G value at the solid–liquid interface is a key parameter affecting intrinsic crystal defects. However, online V/G detection remains difficult because the temperature gradient G cannot be directly measured, while multi-condition distribution shifts and limited labeled data reduce the robustness of data-driven models. To address these issues, this paper proposes DWC-ISBiGNN, an adaptive multi-condition invariant feature extraction method based on the Invariant-Specific Bidirectional Graph Neural Network. The proposed method introduces dynamic sample graph construction with stage-aware global nodes to capture non-stationary process correlations, source-domain credibility weighting to suppress negative transfer, and a semi-supervised training framework combining stage-conditional alignment with teacher–student regression consistency to exploit unlabeled target-domain data. Experiments on industrial data from a 12-inch silicon single-crystal production line show that DWC-ISBiGNN achieves an RMSE of 0.0041, an MAE of 0.00285, and an of 0.9549. Compared with the original IS-BiGNN, the RMSE is reduced by 32.6%, and is increased by 5.43 percentage points. The results demonstrate that the proposed method provides an effective soft-sensing approach for V/G prediction under multiple operating conditions. - Open Access
DIF-LSTM: A Dual Information Filtering LSTM Network for V/G Value Prediction in Czochralski Silicon Growth
In Czochralski (CZ) silicon growth, controlling the ratio of crystal growth velocity to axial temperature gradient () is critical for defect management. However, the value is difficult to measure in real-time. Furthermore, it exhibits strong multivariate coupling and extreme non-stationarity under complex thermal fields. While standard deep learning models like LSTM are used for soft sensing, they often misidentify high-frequency hardware noise as true process dynamics, causing severe error amplification in multi-step predictions. To address this, we propose a Dual Information Filtering LSTM (DIF-LSTM). It utilizes an external context-aware mechanism to screen long-term steady-state redundant information and an internal denoising gate coupled with the LSTM input to explicitly block transient high-frequency noise. Furthermore, a confidence evaluation branch and residual decay fusion ensure stable multi-step forecasting. Experimental results an industrial-scale experimental silicon single crystal furnace show DIF-LSTM achieves superior accuracy, obtaining an of 0.9935 and a Mean Squared Error of at a 3-step horizon. Even at a 9-step horizon, it maintains an of 0.9422, significantly outperforming the baseline IF-LSTM (0.8498).Open AccessDIF-LSTM: A Dual Information Filtering LSTM Network for V/G Value Prediction in Czochralski Silicon Growth
In Czochralski (CZ) silicon growth, controlling the ratio of crystal growth velocity to axial temperature gradient () is critical for defect management. However, the value is difficult to measure in real-time. Furthermore, it exhibits strong multivariate coupling and extreme non-stationarity under complex thermal fields. While standard deep learning models like LSTM are used for soft sensing, they often misidentify high-frequency hardware noise as true process dynamics, causing severe error amplification in multi-step predictions. To address this, we propose a Dual Information Filtering LSTM (DIF-LSTM). It utilizes an external context-aware mechanism to screen long-term steady-state redundant information and an internal denoising gate coupled with the LSTM input to explicitly block transient high-frequency noise. Furthermore, a confidence evaluation branch and residual decay fusion ensure stable multi-step forecasting. Experimental results an industrial-scale experimental silicon single crystal furnace show DIF-LSTM achieves superior accuracy, obtaining an of 0.9935 and a Mean Squared Error of at a 3-step horizon. Even at a 9-step horizon, it maintains an of 0.9422, significantly outperforming the baseline IF-LSTM (0.8498). - 8 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. PP (17) , 1
Towards Delay-Efficient Mobile Data Collection in WSNs via Dynamic Path Planning and Accelerated Distributed Flow Control
Mobile data collection algorithms have been extensively studied in Wireless Sensor Networks (WSNs) due to their flexibility and efficiency. However, constrained by design defects and convergence speed, existing algorithms struggle to meet stringent delay requirements. Consequently, a novel delay-efficient framework for mobile data collection in WSNs is proposed. Its core concept is as follows: First, a dynamic access path planning algorithm is designed at the physical mobility layer by analyzing the transmission coverage area of sensor nodes, with the aim of shortening the total travel distance of the mobile collector while also minimizing its total time spent accessing the sensor nodes. Second, a redundant data processing algorithm based on dynamic feature differences and adaptive thresholds has been developed to eliminate redundant data transmission burdens within the network, which reduces communication energy consumption and network congestion. Finally, a fully asynchronous accelerated distributed link flow control algorithm is proposed to further enhance data transmission efficiency. The algorithm employs an asynchronous implementation approach, enabling each sensor node to complete updates independently without being affected by other nodes. Simultaneously, the convergence properties and the convergence rate of the algorithm are rigorously analyzed from a theoretical perspective. The final simulation results demonstrate that the proposed framework achieves superior overall performance through the synergistic optimization of physical mobility, redundancy elimination, and network transmission, with its core advantages manifested in faster convergence rates, better energy efficiency, and reduced transmission latency.
- 8 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. PP (17) , 1
Towards Delay-Efficient Mobile Data Collection in WSNs via Dynamic Path Planning and Accelerated Distributed Flow Control
Mobile data collection algorithms have been extensively studied in Wireless Sensor Networks (WSNs) due to their flexibility and efficiency. However, constrained by design defects and convergence speed, existing algorithms struggle to meet stringent delay requirements. Consequently, a novel delay-efficient framework for mobile data collection in WSNs is proposed. Its core concept is as follows: First, a dynamic access path planning algorithm is designed at the physical mobility layer by analyzing the transmission coverage area of sensor nodes, with the aim of shortening the total travel distance of the mobile collector while also minimizing its total time spent accessing the sensor nodes. Second, a redundant data processing algorithm based on dynamic feature differences and adaptive thresholds has been developed to eliminate redundant data transmission burdens within the network, which reduces communication energy consumption and network congestion. Finally, a fully asynchronous accelerated distributed link flow control algorithm is proposed to further enhance data transmission efficiency. The algorithm employs an asynchronous implementation approach, enabling each sensor node to complete updates independently without being affected by other nodes. Simultaneously, the convergence properties and the convergence rate of the algorithm are rigorously analyzed from a theoretical perspective. The final simulation results demonstrate that the proposed framework achieves superior overall performance through the synergistic optimization of physical mobility, redundancy elimination, and network transmission, with its core advantages manifested in faster convergence rates, better energy efficiency, and reduced transmission latency.