Dr. Nguyen Duc Huy, an AI expert at FHM.QAI, FPT Software, FPT Corporation, and his colleagues have recently published two research papers in international journals, focusing on artificial intelligence (AI) applications and language models.
In CNC machining, classifying cutting tool wear requires AI solutions that deliver not only high accuracy but also computational efficiency and practical deployability. To address this challenge, the research team explored two approaches: developing a lightweight AI model and leveraging compact large language models (Compact LLMs).
The first study, “Parallel feature fusion with mixed normalization for efficient tool wear classification in CNC milling: HyMixNet,” was published in Engineering Research Express.
The research introduced HyMixNet, an AI model with just 0.56 million parameters, designed to automatically classify cutting tool wear. The model architecture balances accuracy with computational efficiency, supporting tool wear monitoring and predictive maintenance in resource-constrained manufacturing environments.
The second study, “Tool wear classification using compact language models with multi-scale filter bank extractor and low-rank adaptation,” was published in Signal, Image and Video Processing.
The research evaluated Compact LLMs ranging from 0.5 billion to 1.7 billion parameters, combining a multi-scale feature extractor with Low-Rank Adaptation (LoRA), a technique that fine-tunes models using a relatively small number of additional parameters.
The proposed approach achieved 94.96% accuracy and an F1-score of 94.89% in testing, demonstrating the potential of compact language models for industrial AI applications.
Dr. Nguyen Duc Huy, FHM.QAI, FPT Software, FPT Corporation.Across both studies, Dr. Nguyen Duc Huy played a leading role, directly contributing to methodology design, model architecture development, data analysis, experimentation, fine-tuning, and manuscript preparation.
In particular, he personally developed HyMixNet and the LoRA-integrated language model for cutting tool wear classification in CNC machining.
Reflecting on the research process, Dr. Huy explained that the team focused not only on model accuracy but also on real-world applicability.
One of the biggest challenges was obtaining sufficient, reliable data. To address this, the team established contact with a professor in Belgium, who provided access to a standardized dataset for model development and experimentation.
Through these two approaches, the studies have expanded the potential applications of AI in manufacturing tool wear monitoring while further strengthening the international research presence of FPT's Quy Nhon AI Center.
From Project Management to Coordinating an “AI Team”
Research That Expands Organizational Capabilities
Dr. Nguyen Duc Huy earned his PhD in System Optimization and Reliability from the University of Technology of Troyes, France. The discipline combines applied mathematics, operations research, computer science, and industrial and systems engineering.
After completing his doctorate and returning to Vietnam in 2019, Dr. Huy connected with FPT through friends in the Francophone community, opening an opportunity to pursue advanced research in large language models, AI, and their applications.
“When FPT established an AI center in Quy Nhon, I saw it as an opportunity and decided to join,” Dr. Huy recalled.
At FPT, Dr. Huy works as a data expert and AI researcher while also serving as a Technical Lead. His responsibilities include defining technical approaches, designing model architectures, mentoring team members, and directly developing AI solutions for real-world applications.
Dr. Huy recalled that one of his close friends from his doctoral research group in France specialized in mechanical engineering.
His friend had long been interested in applying AI to complex engineering challenges in real-world manufacturing, particularly CNC machining and cutting tool wear monitoring.
“I realized this was an incredibly interesting interdisciplinary challenge. It offered significant academic value, with many new algorithmic problems to explore, while also having enormous practical potential to help factories, including FPT's partners, optimize costs and prevent risks,” Dr. Huy shared.
This shared enthusiasm and commitment motivated the team to embark on the research project.
The greatest challenge throughout the research process was data collection and preparation.
In CNC milling, setting up machinery and sensors to continuously capture machining signals and measure tool wear in a standardized and reliable manner is extremely costly and complex, requiring highly specialized equipment.
To overcome this obstacle, the team expanded international research collaboration and connected with a leading professor in Belgium.
With the professor's valuable support, Dr. Huy and his colleagues gained access to a standardized, high-quality dataset suitable for specialized research.
According to Dr. Huy, the research has also generated significant benefits for his professional work, particularly by expanding advanced AI capabilities into industrial applications.
Working with real-time sensor signal processing has enabled Dr. Huy and his team to develop additional technical capabilities and solution frameworks that can be applied to future predictive maintenance projects at FPT.
“At the same time, personally designing new model architectures and completing research papers that meet international publication standards has helped me maintain a rigorous research mindset and stay up to date with the latest technologies. This allows me to inspire younger engineers and guide their academic and technical development,” Dr. Huy concluded, highlighting the value of research and development in strengthening both technical expertise and team leadership.