Electric Resistance Welding (ERW) is a crucial technology in the automotive tubing manufacturing industry. However, setting welding parameters still relies on the operator’s experience, resulting in variability in weld quality and production efficiency. This research develops an Internet of Things integrated predictive model incorporating real-time thermal imaging, Adaptive Statistical Features, and Edge AI on an ESP32 microcontroller. The system captures weld temperature distributions, extracts 12 statistical features (such as contrast, entropy, skewness), and utilizes machine learning for predictive parameter optimization. Experimental results demonstrate that the Artificial Neural Network model achieves 84.4% defect detection accuracy, 6,666 inferences per second, and consumes only 36.87 kB of memory. By reducing human dependence and enabling real-time decision making, this system aligns with Industry 4.0 objectives, enhancing production efficiency and resource utilization in high-frequency ERW. The proposed system provides a cost-effective, scalable solution for industrial applications, fostering intelligent and sustainable manufacturing.