Thanh Son Pham
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Thanh Son Pham has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
- Duc Huy Nguyen
Faculty of Information Technology, Dong A University, Danang, Vietnam
2 shared publications - Nguyen Duy Minh Phan
Faculty of Mechanical Engineering, The University of Danang - University of Technology and Education, Danang, Viet Nam
2 shared publications - Hoang Thuan Tran
Center of Electrical Engineering, Duy Tan University, Danang, 550000, Vietnam
1 shared publication - Tran Le Thang Dong
Duy Tan University
1 shared publication - Truc Thanh Tran
FPT University
1 shared publication - Vuong Anh Trung
Air Def Air Force Acad, Fac Aviat Technol, Hanoi, Vietnam
1 shared publication
Latest Publications
- 27 September 2026
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 20 (13) , 660
Tool wear classification using compact language models with multi-scale filter bank extractor and low-rank adaptation
- 27 September 2026
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 20 (13) , 660
Tool wear classification using compact language models with multi-scale filter bank extractor and low-rank adaptation
- 26 June 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (13) , 135402
Parallel feature fusion with mixed normalization for efficient tool wear classification in CNC milling: HyMixNet
This paper presents HyMixNet, a lightweight deep learning architecture for automated tool wear classification in CNC milling processes using image-based data. The proposed model combines parallel feature mixer blocks with internal residual connections, a global skip connection to help preserve low-level spatial details across the network, and a hybrid normalization strategy that combines GroupNorm and BatchNorm at different stages. These design choices aim to enhance feature representation while keeping the overall parameter count low. With only 0.56 million parameters, HyMixNet is the most compact among the evaluated architectures, requiring 1.55 × fewer parameters than SqueezeNet, 2.71 × fewer than ShuffleNetV2, and 3.18 × fewer than MobileNetV3. Despite this compact design, the model achieves an accuracy of 0.928 , a macro recall of 0.843 , a macro F 1-score of 0.863 , and a weighted F 1-score of 0.925 , surpassing all baselines in minority-class detection metrics. These results highlight the potential of HyMixNet for real-time predictive maintenance and intelligent monitoring applications in resource-constrained industrial environments.
- 26 June 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (13) , 135402
Parallel feature fusion with mixed normalization for efficient tool wear classification in CNC milling: HyMixNet
This paper presents HyMixNet, a lightweight deep learning architecture for automated tool wear classification in CNC milling processes using image-based data. The proposed model combines parallel feature mixer blocks with internal residual connections, a global skip connection to help preserve low-level spatial details across the network, and a hybrid normalization strategy that combines GroupNorm and BatchNorm at different stages. These design choices aim to enhance feature representation while keeping the overall parameter count low. With only 0.56 million parameters, HyMixNet is the most compact among the evaluated architectures, requiring 1.55 × fewer parameters than SqueezeNet, 2.71 × fewer than ShuffleNetV2, and 3.18 × fewer than MobileNetV3. Despite this compact design, the model achieves an accuracy of 0.928 , a macro recall of 0.843 , a macro F 1-score of 0.863 , and a weighted F 1-score of 0.925 , surpassing all baselines in minority-class detection metrics. These results highlight the potential of HyMixNet for real-time predictive maintenance and intelligent monitoring applications in resource-constrained industrial environments.
- Open Access
Learning-Inspired Fuzzy Logic Algorithms for Enhanced Control of Oscillatory Systems
The transportation of sensitive equipment often suffers from vibrations caused by terrain, weather, and motion speed, leading to inefficiencies and potential damage. To address this challenge, this paper explores an intelligent control framework leveraging fuzzy logic, a foundational AI technique, to suppress oscillations in suspension systems. Inspired by learning based methodologies, the proposed approach utilizes fuzzy inference and Gaussian membership functions to emulate adaptive, human like decision making. By minimizing the need for explicit mathematical models, the method demonstrates robustness in both linear and nonlinear systems. Experimental validation highlights the controllers ability to adapt to varying suspension lengths, reducing oscillation amplitudes and improving stability under dynamic conditions. This research bridges the gap between traditional control systems and learning inspired techniques, offering a scalable, data efficient solution for modern transportation challenges
Open AccessLearning-Inspired Fuzzy Logic Algorithms for Enhanced Control of Oscillatory Systems
The transportation of sensitive equipment often suffers from vibrations caused by terrain, weather, and motion speed, leading to inefficiencies and potential damage. To address this challenge, this paper explores an intelligent control framework leveraging fuzzy logic, a foundational AI technique, to suppress oscillations in suspension systems. Inspired by learning based methodologies, the proposed approach utilizes fuzzy inference and Gaussian membership functions to emulate adaptive, human like decision making. By minimizing the need for explicit mathematical models, the method demonstrates robustness in both linear and nonlinear systems. Experimental validation highlights the controllers ability to adapt to varying suspension lengths, reducing oscillation amplitudes and improving stability under dynamic conditions. This research bridges the gap between traditional control systems and learning inspired techniques, offering a scalable, data efficient solution for modern transportation challenges