Leveraging AI, TRUMPF is enhancing the overarching system of lasers, optics, sensor technology, and software, reducing production time and the volume of scrap and rework. Its EasyModel AI and VisionLine Detect solutions only need user training rather than extensive machine programming expertise. This streamlines the welding process by improving image processing and minimizing the impact of external interference such as dirt, scratches or poor lighting.
The application of artificial intelligence (AI) in welding isn’t altogether new. A 2019 study demonstrated its promise to revolutionize traditional approaches to welding. Harnessing the power of AI, manufacturers address longstanding challenges such as poorly controlled welding parameters and weld geometry. These technological innovations could help to minimize weld quality problems, enhancing efficiency and productivity.
Managing Welding Parameters and Variables
AI enables an intricate analysis of numerous process parameters and joint geometry parameters that affect welding operations. These include arc voltage, arc current, welding speed, base material thickness, joint profile, joint grooves and gap, weld bead height, and weld penetration. The nonlinear attributes of welding make these relationships difficult to establish and control. This complexity is even more evident when the processes are adapted to robotics, considering the vast range of data provided by sensors and monitoring devices.
However, despite this increased data availability, challenges persist due to difficulties in establishing accurate relationships between welding parameters and variables. These challenges include arc instability, joint position errors, distortion, weld undercut, porosity, irregularities in the weld bead, lack of weld penetration, lack of fusion, heat affected zone (HAZ) softening, microstructure deterioration, and crack susceptibilities. Addressing these requires a coordinated system of real-time sensors, monitoring devices, and AI methods capable of efficient adjustment, monitoring, prediction, and control of welding parameters.
To tackle these challenges, researchers have turned to nonlinear methods with AI capabilities such as the Taguchi method, response surface method (RSM), artificial neural network (ANN), genetic algorithm (GA), fuzzy logic systems, adaptive neuro-fuzzy inference systems (ANFIS), and particle swarm optimization (PSO). These methods provide a framework for identifying relationships between input parameters and output variables, thereby enabling better control of welding parameters to improve weld quality.
For instance, artificial neural networks have been used to predict welding parameters for weld bead height in butt joints, with more accuracy than regression models. Similarly, other studies have successfully applied these AI models to predict and control weld quality in laser transmission welding of thermoplastics.
Current developments in AI for welding robot control have proven effective in addressing challenges in motion planning, path optimization, and trajectory reach enhancement. AI approaches have shortened the time of path planning and improved welding efficiency, preventing damage to robots and equipment in real-time.
Harnessing AI Advances
"TRUMPF [demonstrated] two new AI-based solutions for laser welding applications at the 'Laser World of Photonics' fair in Munich,” says Florian Kiefer, Head of Services Product Management Laser Technology at TRUMPF. “EasyModel AI enables the customer to create an algorithm without programming knowledge. Meanwhile, the AI Filter for our image processing system VisionLine Detect allows the customer to apply the algorithm in their production.”
According to Kiefer, what sets TRUMPF apart is focusing on more than just the laser device itself, instead developing comprehensive solutions. The commercial adoption process is relatively easy for companies already using TRUMPF lasers and can upgrade their systems to use AI solutions without any further hardware investments, he said.
"Industries which heavily rely on laser welding will have the benefit of an even more robust process combined with the possibility of scaling the process faster,” says Kiefer. “With the usage of AI, we can capture experts’ knowledge in an algorithm and therefore transfer it quicker."
Kiefer also pointed out that TRUMPF’s AI solutions have benefits for companies of all sizes. However, he acknowledged that high volume production with a variation in part quality, fixture variation or pollution and variation in lighting/reflection is the best application. He added that while the addition of AI solutions may lead to a marginal increase in process time—a few 10s of milliseconds—the advantages clearly outweigh the extra time.
Another company, Amada Weld Tech is driving advancements in AI and machine learning applications for resistance and laser welding. Their key focus area is weld process monitoring, which employs AI/ML to examine the success of a manufacturing process by collecting and analyzing physical signals from the process.
One significant advancement lies in the high-resolution data acquisition process. To achieve precise predictions about weld quality, the company emphasizes the collection of microsecond-level data, stressing that more data corresponds to more accurate results. This high-resolution data collection facilitates better sorting of good welds from bad ones by AI/ML algorithms.
Amada has also leveraged networking to take data collection from local to global, while investing in network security measures. By connecting process monitors via ethernet, the company eases the transfer of information and allows process engineers to collect and analyze data across multiple factories worldwide. This approach rapidly enriches the weld repository and provides more refined judgments of the welding process.
Finally, the collected data is deployed in AI and ML algorithms. This offers insights on process efficiency, equipment performance, production rate, defects, and anomalies. By applying AI/ML to this data, Amada can find new correlations and features, thereby expanding the capabilities of the process engineer and operator.
Intelligent Adaptation
Current methods of robot operations present significant limitations—inefficient parameter identification processes and the lack of an intelligent response mechanism to adapt and learn from new situations. T Systems is working on surpassing these limitations. In collaboration with car manufacturers, they developed and trialed an AI architecture that not only identifies the quality of a weld seam but also evaluates it to assess the laser welding process of the robot.
This AI architecture enables the laser welding robot to auto-identify parameters, providing an efficiency unmatched by prior methods. It also equips the robot with a situational understanding of various scenarios. The concept enables collective learning among robots as well as continuous learning during operation. This significant expansion of the robot's experiential understanding of weld seam guidance can be systematically integrated and documented.
The fusion of intelligence allows laser welding robots to operate more efficiently, with greater safety, cost-effectiveness, and flexibility. The automotive partnerships helped demonstrate that the fusion of deep learning and reinforcement learning significantly enhances the functionality of laser welding robots, improving the laser welding process. Such capabilities are crucial to meeting the safety and efficiency needs of manufacturers in automotive and other sectors.