AI and optical sensing could make laser welding self-correcting

A research review published June 3, 2024, outlines how cameras, spectrometers and other optical sensors, paired with AI, could spot laser-weld defects in real time and help correct them before parts leave the line. The framework points to faster, less wasteful manufacturing across industries such as automotive and aerospace.

Why it matters: - Real-time monitoring could catch laser-weld defects before finished parts are scrapped, reworked or shipped. - The approach could reduce destructive testing, scrap, downtime and repair costs in precision manufacturing. - Industries that rely on flawless joints, including automotive, aerospace, shipbuilding, bridge construction and additive manufacturing, could benefit.

What happened: - Researchers from the Guangdong Provincial Welding Engineering Technology Research Center at Guangdong University of Technology published a review on June 3, 2024, in Advances in Manufacturing. - The paper surveys how optical sensors and AI can monitor laser welds as they form and support adaptive control. - The review is identified by DOI 10.1007/s40436-024-00493-1.

The details: - Laser welding uses a concentrated, high-energy beam to make fast, narrow and deep welds. - The process can be disrupted by unstable keyholes, molten-metal flow, plasma shielding and spatter. - Those disruptions can lead to porosity, cracking, underfill, humps and incomplete fusion. - Conventional inspection usually happens after welding, when defects are costly to fix. - Acoustic monitoring can be limited by factory noise and vibration. - A single optical sensor often captures only part of the weld process. - The review links defects to the physics inside the weld, including keyhole instability, thermal contraction, alloy behavior and excessive energy or melt flow. - Radiation-based monitoring systems use light emitted or reflected during welding. - Pyrometers and infrared cameras track temperature and cooling. - Photodiodes measure rapid intensity changes. - Spectrometers analyze plasma emissions. - Vision systems track the molten pool, keyhole and plume. - Active-light systems add an external source to probe the weld. - X-ray imaging can show internal flow and pore formation. - Optical coherence tomography, also called inline coherent imaging, measures weld depth with micrometer-scale resolution. - Magneto-optical imaging can help with seam tracking in extremely narrow joints. - The review compares cost, sampling speed, spatial resolution and factory readiness across these approaches. - Multisensor fusion is presented as a key strategy because each method has blind spots. - AI models such as support vector machines and convolutional neural networks can combine signals, extract features, classify penetration states and help control laser power, speed, focus or wire feed. - The review also examines machine learning and deep learning for signal processing, defect prediction, classification and closed-loop control. - The source URL for the review is the full paper. - Funding came from the National Natural Science Foundation of China, the Guangdong Provincial Natural Science Foundation of China and the Guangzhou Municipal Special Fund Project for Scientific and Technological Innovation and Development.

Between the lines: - The main shift is from post-weld inspection to inline decision-making while the weld is still forming. - The review argues that no single sensor can explain the full event because temperature, light, plasma and geometry each reveal different parts of the process. - AI is most useful when it turns those signals into timely control actions, not just a quality label. - The biggest hurdle is not only accuracy but also interpretability, speed and reliability outside lab conditions. - Standardized datasets and tighter integration of physics-based models with data-driven learning will likely determine how fast the field moves into production.

What's next: - Future systems will need to be accurate, interpretable and fast enough to correct defects before they become permanent. - Researchers will need better labeled datasets, more robust multisensor setups and methods that work across changing factory conditions. - If those pieces come together, laser welding could move toward self-monitoring and self-adjusting production lines.

The bottom line: - Optical sensing plus AI could turn laser welding into a process that detects its own mistakes and corrects them in real time.