Embedded Files
The following is a brief list of some main projects I have accomplished throughout my PhD, for the full list of publications and research areas please explore the Publications page.
The following is a brief list of some main projects I have accomplished throughout my PhD, for the full list of publications and research areas please explore the Publications page.
Building Smart Manufacturing Research Infrastructure
Building Smart Manufacturing Research Infrastructure
A prominent accomplishment of my graduate school journey has been the design and commissioning of the Future Factories lab. I helped grow and establish this lab by building upon the infrastructure available at the McNair Aerospace Research Center at UofSC. This lab started out as one industrial robot and has evolved to a five-robot assembly cell equipped with a full conveyor system, electrical cabinet, PLCs, and on-premise servers.
Collaborators: West Virginia University SM Lab, AI Institute at the University of South Carolina
Within the realm of smart manufacturing, manufacturing systems themselves have been categorized based on their level of intelligence and autonomy. Systems have evolved from automated to autonomous and now to cognitive manufacturing systems. I have contributed in cementing the field of cognitive manufacturing through establishing a formal definition and the research trends accompanying this emerging topic. From a systematic analysis, which encompasses (1) what cognition affects in manufacturing, (2) how it is achieved (technologies/approaches), and (3) why it matters (capabilities/outcomes), the resulting definition was: intelligent cyber-physical manufacturing capable of perception, decision making, and reacting by utilizing information obtained throughout the whole product life cycle.
Time Series Analytics
Time Series Analytics
Collaborators: West Virginia University SM Lab, Siemens Digital Industries
A major focus of my work has been the deployment of advanced time series forecasting and classification models into manufacturing systems. Unlike many theoretical studies, I validated my frameworks in real-world assembly lines, addressing challenges such as high-dimensionality, noise, and heterogeneity in sensor data. My publications present both methodological innovations and benchmarked deployments, showing how classification models can detect defects. I also demonstrated an end-to-end forecasting application for robotic assembly monitoring, showcasing how state-of-the-art algorithms can be seamlessly integrated with DTs for increased manufacturing safety.
TIMESE~1.MOVModern manufacturing is changing fast thanks to Industry 4.0, where factories are now filled with sensors that collect huge amounts of data. This data can be used to train Artificial Intelligence (AI) systems that help predict problems before they happen and keep machines running smoothly. One powerful area of AI for manufacturing is time series classification, which looks at data over time to detect patterns, forecast future events, and classify machine behavior—such as identifying early signs of equipment failure. However, using real factory data to train these AI models is still challenging, especially when the system needs to work in real time. In this study, we introduce a fully tested, closed-loop framework that shows how to build, evaluate, and deploy time-series AI models for real manufacturing environments. We also explain how to select and extract the right features from the data so the AI model performs reliably. To demonstrate this, we use a brand-new dataset collected from a robotic assembly line and walk through the complete process from raw data to real-time classification on the production floor.
Modern manufacturing is changing fast thanks to Industry 4.0, where factories are now filled with sensors that collect huge amounts of data. This data can be used to train Artificial Intelligence (AI) systems that help predict problems before they happen and keep machines running smoothly. One powerful area of AI for manufacturing is time series classification, which looks at data over time to detect patterns, forecast future events, and classify machine behavior—such as identifying early signs of equipment failure. However, using real factory data to train these AI models is still challenging, especially when the system needs to work in real time. In this study, we introduce a fully tested, closed-loop framework that shows how to build, evaluate, and deploy time-series AI models for real manufacturing environments. We also explain how to select and extract the right features from the data so the AI model performs reliably. To demonstrate this, we use a brand-new dataset collected from a robotic assembly line and walk through the complete process from raw data to real-time classification on the production floor.
Digital Twin-Enabled Robot Collision Detection Using Time Series Forecasting
Digital Twin-Enabled Robot Collision Detection Using Time Series Forecasting
One powerful AI tool used in this space is Time-Series Forecasting (TSF), which analyzes past data to predict what will happen next—such as when a machine might fail or when two robotic arms might collide. Another major technology is the Digital Twin (DT), a virtual replica of a real machine or system that can simulate behavior and test decisions in real time. While both TSF and Digital Twins are valuable on their own, combining them into one closed-loop system—where predictions trigger real-time responses—remains a challenge in manufacturing. This research proposes a new framework that connects Time-Series Forecasting with a Digital Twin to detect potential collisions between machines before they occur. We demonstrate this system using a robotic assembly line and walk through how the AI model is trained, tested, and deployed. The framework is designed to be flexible, so it can be applied to different types of manufacturing systems—not just robotics—helping factories become safer, smarter, and more efficient.
One powerful AI tool used in this space is Time-Series Forecasting (TSF), which analyzes past data to predict what will happen next—such as when a machine might fail or when two robotic arms might collide. Another major technology is the Digital Twin (DT), a virtual replica of a real machine or system that can simulate behavior and test decisions in real time. While both TSF and Digital Twins are valuable on their own, combining them into one closed-loop system—where predictions trigger real-time responses—remains a challenge in manufacturing. This research proposes a new framework that connects Time-Series Forecasting with a Digital Twin to detect potential collisions between machines before they occur. We demonstrate this system using a robotic assembly line and walk through how the AI model is trained, tested, and deployed. The framework is designed to be flexible, so it can be applied to different types of manufacturing systems—not just robotics—helping factories become safer, smarter, and more efficient.
Semantic Web and Knowledge Graphs
Semantic Web and Knowledge Graphs
Collaborators: AI Institute at University of South Carolina, Siemens R&D Department, Bosch Center for Artificial Intelligence
My research advances the application of Semantic Web and KG technologies in industrial environments. I developed pipelines to transform raw manufacturing data into RDF-based triples in real time, enabling contextualized reasoning directly on edge devices. Crucially, I investigated how KGs enhance explainability in machine learning (ML) systems, bridging the gap between ML outputs and engineering decision-making. By developing causal KGs, systems can provide human-understandable justifications for predictions, supporting intelligent decision-making processes. In collaboration with Siemens, I co-authored a patent for a federated information system that integrates data across PLC, SCADA, MES, and ERP layers, demonstrating how KG principles can solve interoperability challenges in real industrial settings.
Utilizing Knowledge Graphs for Increased Explainability in Manufacturing
Utilizing Knowledge Graphs for Increased Explainability in Manufacturing
Industry 4.0 is transforming manufacturing by connecting machines, data, and automation in powerful new ways. However, as factories rely more on artificial intelligence and machine learning, a major challenge has emerged: these systems often make decisions that are difficult for humans to understand. In many cases, the AI works like a “black box,” giving answers without explaining why. This research introduces a new framework that makes AI decisions in manufacturing more transparent and easier to interpret. Using a Knowledge Graph, a structured way of linking information, our system explains how a machine learning model reaches its decisions. We applied this approach to a robotic assembly line, where a classifier (DrCIF) was trained to detect defective products with 99% accuracy. Instead of just flagging a product as defective, the system also provides a clear explanation of what caused the problem and why the model made that prediction. This helps technicians diagnose issues faster and builds trust in AI-driven manufacturing systems.
Industry 4.0 is transforming manufacturing by connecting machines, data, and automation in powerful new ways. However, as factories rely more on artificial intelligence and machine learning, a major challenge has emerged: these systems often make decisions that are difficult for humans to understand. In many cases, the AI works like a “black box,” giving answers without explaining why. This research introduces a new framework that makes AI decisions in manufacturing more transparent and easier to interpret. Using a Knowledge Graph, a structured way of linking information, our system explains how a machine learning model reaches its decisions. We applied this approach to a robotic assembly line, where a classifier (DrCIF) was trained to detect defective products with 99% accuracy. Instead of just flagging a product as defective, the system also provides a clear explanation of what caused the problem and why the model made that prediction. This helps technicians diagnose issues faster and builds trust in AI-driven manufacturing systems.
Manufacturing is moving beyond traditional automation toward fully autonomous systems, ie, factories that can make smart decisions, adjust to disruptions, and customize production without human intervention. To reach this level of agility, production lines must be able to understand machine conditions, interpret sensor and IoT data, analyze inspection results, and reallocate resources in real time. However, most current factories still rely on basic, machine-level decision-making and do not take full advantage of the data they generate. This work takes a step toward true manufacturing autonomy by applying Semantic Web technologies to manage and reason over factory data. We demonstrate this through a full use case where a production line continues operating—even when equipment fails—by using structured knowledge and data integration to make system-level decisions. The results show how Semantic Web methods can unify information from different machines, sensors, and processes, enabling a factory to operate intelligently and independently. This approach highlights a promising path toward autonomous manufacturing that is only beginning to be explored at the intersection of AI, data semantics, and industrial systems.
Manufacturing is moving beyond traditional automation toward fully autonomous systems, ie, factories that can make smart decisions, adjust to disruptions, and customize production without human intervention. To reach this level of agility, production lines must be able to understand machine conditions, interpret sensor and IoT data, analyze inspection results, and reallocate resources in real time. However, most current factories still rely on basic, machine-level decision-making and do not take full advantage of the data they generate. This work takes a step toward true manufacturing autonomy by applying Semantic Web technologies to manage and reason over factory data. We demonstrate this through a full use case where a production line continues operating—even when equipment fails—by using structured knowledge and data integration to make system-level decisions. The results show how Semantic Web methods can unify information from different machines, sensors, and processes, enabling a factory to operate intelligently and independently. This approach highlights a promising path toward autonomous manufacturing that is only beginning to be explored at the intersection of AI, data semantics, and industrial systems.
Page updated
Google Sites
Report abuse