Our research projects advance the future of intelligent manufacturing through innovations in advanced sensing, industrial AI, sensor fusion, and digital twin technologies. Although our primary focus is on manufacturing, the developed methodologies are generalizable to a broad range of applications, including shipbuilding, infrastructure health monitoring, leak detection, fire monitoring, and wearable health monitoring.
AI Framework for LPBF Monitoring and Control (Sponsor: SDSURF)
Laser Powder Bed Fusion (LPBF) is transforming metal additive manufacturing by enabling the production of highly complex, high-performance metal components. However, achieving consistent part quality remains challenging due to the complex interactions among process parameters, material behavior, and defect formation. This project develops an intelligent manufacturing framework that integrates multi-modal in-situ sensing, sensor fusion, physics-informed machine learning, and digital twin technologies to monitor the LPBF process, predict defect formation, and optimize process parameters in real time. By combining data-driven intelligence with physics-based process understanding, the framework enables predictive quality assurance and closed-loop process control, ultimately advancing autonomous additive manufacturing with improved quality, productivity, and process reliability.
Agentic AI-Enabled Manufacturing Intelligence for Autonomous CNC Machining (Sponsor: KITECH/MOTIE)
This project develops an intelligent manufacturing framework for the autonomous optimization and control of automotive-parts machining processes. We develop Edge AI technologies that use multimodal sensor data to monitor machining conditions and predict tool wear and process anomalies in real time. The framework integrates multimodal sensing, Digital Thread, Digital Twin, and robotic vision/VLM to provide comprehensive information about the machining process. We also develop Agentic AI technologies that integrate and interpret information from these heterogeneous sources and make informed decisions for process diagnosis, optimization, and execution. MLOps supports continuous model validation, deployment, monitoring, and retraining.
AI-Based Intelligent Real-Time Weld Inspection Framework for the Shipbuilding Industry
This project aims to develop an AI-based intelligent comprehensive weld inspection framework (Weld Monitoring AI Framework, WM-AIF) for smart welding quality management in the shipbuilding industry. WM-AIF integrates multimodal in-situ sensor data with sensor fusion, condition-aware, and physics-informed AI to predict weld quality, defects, and microstructure in real time. The framework is designed to provide reliable inspection across varying welding conditions and processes. The technology will support 100% weld inspection, early defect detection, real-time microstructure prediction, reduced rework and inspection costs, and improved welding quality and reliability.
AI-Driven Image-Based Materials Characterization (Sponsor: NSF)
Accurate materials characterization is essential for understanding microstructure–property relationships and accelerating materials development. This project develops an AI-driven image analysis framework for automated characterization of pores, grain boundaries, and other microstructural features from cross-sectional images. It also introduces a machine learning approach that infers three-dimensional contact area distributions directly from two-dimensional images, providing a fast and scalable alternative to conventional 3D characterization. The framework is designed to enable high-throughput, robust, and data-driven materials characterization for advanced manufacturing.
Controllable Data Augmentation (Sponsor: NSF)
Developing reliable AI models for manufacturing often requires large experimental datasets, yet acquiring sufficient data is expensive, time-consuming, and sometimes impractical. This project develops an AI-driven controllable data augmentation framework based on diffusion generative models to overcome this challenge. Unlike conventional augmentation techniques, the proposed framework generates realistic microstructure images for user-specified processing conditions, enabling the synthesis of experimental data beyond measured conditions. This capability significantly reduces the need for costly experiments while improving the accuracy, robustness, and generalizability of AI models for advanced manufacturing.
Cutting Tool Wear Prediction using Sensor Fusion and Digital Twin (Sponsor: KITECH)
Cutting tool wear is one of the primary factors affecting machining quality, productivity, and manufacturing cost. However, accurately predicting tool wear remains challenging because it depends on complex interactions among cutting conditions, tool geometry, workpiece material, and thermal-mechanical behavior. This project develops an intelligent framework that integrates multi-modal in-situ sensing, sensor fusion, physics-based digital twin modeling, and machine learning to monitor the cutting process and predict tool wear in real time. By combining experimental sensor data with digital twin simulations, the framework enhances prediction accuracy and generalizability while reducing the need for extensive experimental datasets. The ultimate goal is to enable predictive tool condition monitoring, optimize machining processes, and improve manufacturing efficiency and tool life.
Process Optimization of Sintering-Assisted Additive Manufacturing using Machine Learning (Sponsor: NSF)
Optimizing process parameters in sintering-assisted additive manufacturing is critical for achieving the desired density, microstructure, and mechanical properties of the final part. This project develops an AI-driven process optimization framework that integrates three machine learning models for powder spreading, binder jetting, and sintering. The models are coupled to predict process evolution across multiple manufacturing stages and optimize input parameters, such as powder characteristics, rolling speed, layer thickness, and sintering conditions, for desired material performance. By combining experimental and virtual manufacturing datasets with AI-based microstructure analysis and generative models, the framework enables efficient inverse process design while significantly reducing experimental effort. The developed methodology is generalizable and can be extended to optimize a wide range of manufacturing processes.
Physics-Informed Machine Learning for Pipe Leakage Detection and Prediction (Sponsor: COMPA, DOT)
Water and hazardous-liquid pipeline leakage is a major challenge for aging infrastructure, yet existing monitoring systems are primarily limited to detecting leaks after they occur and often suffer from poor robustness in noisy field environments. This project develops an AI-based smart sensing framework that integrates acoustic emission (AE) and vibration sensing with physics-informed machine learning for real-time leak detection, localization, and prediction. By combining edge computing, high-frequency signal processing, and physics-guided AI, the framework enables reliable leak monitoring under real-world operating conditions, including noisy environments and diverse pipeline materials and fluids. The ultimate goal is to improve the safety, reliability, and sustainability of water distribution systems and hazardous-liquid transport infrastructure.
AI-Based Silent Speech Recognition Using Wearable EMG/EEG
Silent Speech Recognition (SSR) enables communication without audible speech by decoding non-acoustic physiological signals, offering transformative opportunities for medical assistive technologies and human–machine interaction. However, existing SSR systems are often limited to small vocabularies, task-specific models, and labor-intensive data collection procedures. This project develops an AI-driven multimodal speech decoding framework that integrates EEG, EMG, sensor fusion, few-shot learning, and language models to recognize words and sentences from continuous physiological signals. By eliminating the need for manual word segmentation and improving learning from limited training data, the framework aims to achieve accurate, robust, and generalizable silent speech recognition for real-world applications.
Prompt Detection and Classification of Early-Stage Fires using Innovative Acoustic Field Sensors and Deep Learning