This paper is the third in a series extending our previous work (anonymous, 2023; anonymous, 2024) on modeling heating system processes. It compares three neural network approaches for process identification: two physics-informed models and a traditional neural network-based model. Raissi et al. (2019) introduced Physics-Informed Neural Networks (PINNs), where physical laws are integrated into the neural network's optimization process. More recently, Natale et al. (2022) proposed Physically Consistent Neural Networks (PCNNs), which incorporate a parallel network structure to enforce consistency with physical laws. Building on the PCNN framework, we developed a novel physics-based neural network combined with a feedforward neural network (PBNN).

CONTEXT

Welding has long been critical in manufacturing, providing strong, durable joints across industries. Traditional inspection methods like x-ray and ultrasonic testing detect defects post-welding, resulting in time, cost, and resource waste. To overcome these limitations, real-time monitoring systems have emerged, integrating current and voltage sensors with AI, machine learning, and big data analysis. Our research aims to transform the welding industry by capturing high-speed data for immediate defect detection and prevention, enhancing both quality and efficiency.

PURPOSE OR GOAL

The goal of this research is to minimize welding defects and increase process efficiency. By integrating sensors, a data acquisition (DAQ) system, and AI/ML models, combined with big data analysis, we have developed a real-time defect identification and prediction system. The system monitors critical welding parameters like current and voltage at high sampling rates (4K-8K samples/second), allowing for immediate detection of anomalies like porosity or distortion. AI models continuously refine predictions by analyzing vast amounts of data, adjusting parameters dynamically to prevent defects. This approach reduces material waste, rework, and production costs, making it ideal for industries seeking high-quality output.

APPROACH

Our system collects data in real-time using sensors connected to a DAQ, which samples at up to 8K rates for precision. Data undergoes pre-analysis, where noise is filtered out, ensuring accuracy for real-time analysis. Advanced algorithms, powered by big data analytics, detect parameter deviations, adjusting settings to prevent defects. AI models, trained on historical data, identify patterns associated with defects, while ML algorithms improve accuracy with every session. A Mini PC handles data storage and processing in real-time, using LabVIEW for monitoring and Power BI for deeper analysis, enabling users to visualize trends and optimize the welding process.

ACTUAL OR ANTICIPATED OUTCOMES

The implementation of real-time monitoring, enhanced by big data analysis, has shown significant reductions in common weld defects such as cracks, porosity, and lack of fusion. High-speed data acquisition (2K, 4K, and 8K sample rates) ensures immediate response to parameter changes, preventing defects. AI and ML models provide predictive insights, identifying irregularities that could lead to defects and improving over time. This approach lowers rework, reduces waste, and boosts overall efficiency. The system is adaptable to future AI advancements and various welding methods, making it scalable for different industrial applications.

CONCLUSIONS/RECOMMENDATIONS/SUMMARY

To further enhance this system, we recommend automating sensor calibration, adding redundancy in data storage, and integrating IoT connectivity for remote monitoring and predictive maintenance. Overall, the real-time use of AI, ML, and big data analysis revolutionizes welding quality control, reducing costs, improving efficiency, and ensuring defect-free welds. As the industry continues to evolve toward automation and data-driven processes, this technology offers a scalable, future-proof solution.

This research presents a comprehensive framework for integrating digital twin technology with autonomous control in a scale-model backhoe, addressing challenges in precision, adaptability, and real-time synchronization. The proposed system combines advanced trajectory planning using the A* algorithm, Ackermann kinematics for vehicle motion, and forward kinematics modeling for a 3DOF robotic manipulator. These components are seamlessly integrated with a digital twin to enable accurate real-time simulation and validation of navigation and control algorithms.

The digital twin acts as a bridge between the physical system and virtual environment, utilizing Firebase for real-time communication and data synchronization. This ensures that sensor data, control commands, and simulations are aligned, enabling adaptability to changing environmental conditions and validating the robustness of the framework. The trajectory planning system demonstrated high accuracy, with error metrics quantified through simulations, while the manipulator model achieved precise positioning, validating its mathematical foundation.

This work contributes to the field of autonomous heavy machinery by presenting a scalable and modular system that aligns with the principles of Industry 5.0, emphasizing sustainability, efficiency, and human-machine collaboration. The proposed methodology not only validates the feasibility of autonomous control for scaled prototypes but also sets the foundation for expanding the framework to full-scale machinery. Future directions include improving adaptability to dynamic environments, integrating IoT and cloud systems for distributed control, and enhancing energy efficiency to align with global sustainability goals.

This paper presents the design and implementation of a full-stack web application for enhancing predictive maintenance and fleet performance monitoring within smart city initiatives. The application combines real-time data collection, predictive analytics, and a user-friendly interface, allowing fleet managers to optimize maintenance schedules and improve operational efficiency. Leveraging 5G technology for high-speed data transmission, the system enhances connectivity and facilitates seamless communication between fleets and infrastructure. The platform delivers actionable insights into fleet performance trends and maintenance requirements. This study addresses challenges in data extraction from diverse fleet types and outlines future integration plans, including IoT devices and virtual reality training modules for emergency response scenarios. The findings highlight the potential of advanced technologies to enhance fleet management practices, contributing to sustainable urban mobility and the advancement of smart cities.

This paper initially explores applying Digital Twin (DT) strategies to advanced technological training for engineering students while integrating the latest devel-opments in Industry 4.0 into research on Solid State Lighting (SSL) technologies. Modules based on AlGaN far-UVC LEDs will serve as the foundation for the development of new products aimed at photonic disinfection and disease preven-tion. The development of these modules requires a preliminary stage of basic re-search, which includes the design and fabrication of far-UVC LED crystals. Thermal simulation inputs are provided by an infrared camera that measures the p-contact temperature of the far-UVC LEDs in Japan. This data is then used to inform the simulation system in Chile, enabling precise thermal modeling. This study emphasizes the importance of individual LED performance, as it underpins the functionality of the final device. The following stage, currently under devel-opment, extends the DT to analyze the complete thermal performance of the final far-UVC LED module, focusing on heat generation and dissipation within the system. This approach supports the development of photonic disinfection pro-cesses, addressing the demand for efficient and reliable far-UVC LED modules for HealthTech applications. 11 mw light output power in a single far-UVC LED has been achieved at Riken, which paw way to the development of 500 mW far-UVC LED module using 50 pieces of LEDs for Healthcare (surgical theater). The ultimate goal is to further improve irradiance distribution, enhance optical ef-ficiency, and extend the LEDs' lifetime through optimized thermal management in healthcare environments.