Next-Level Welding via AI and Smart Manufacturing

Two OEMs—John Deere and Taiwan-based electric-bus manufacturer Tron-E—leverage artificial intelligence and smart-manufacturing techniques to optimize weld productivity and quality.

In 2020, Deere & Company launched its Smart Industrial operating model, “to accelerate its success in the integration of smart-technology innovation with its legacy of manufacturing,” the firm promoted at the time. “The new Deere Smart Industrial strategy was created to unlock new value for customers, while simultaneously ‘revolutionizing’ the agriculture and construction industries through the rapid introduction of new technologies.”

An example of the model surfaced in 2021, when Deere collaborated with Intel to pilot an artificial-intelligence (AI) solution utilizing computer vision to automatically spot and correct defects in its automated-welding processes. Together the companies developed a neural-network-based inference engine that would log defects in real time and automatically stop the welding process, accelerating defect identification and allowing Deere operators to address issues immediately.

John Deere uses robotic gas-metal-arc welding of mild to high-strength steels throughout its 52 factories globally. In these factories, hundreds of robotic arms consume millions of weld-wire pounds annually. With this R&D project with Intel, it sought to specifically target weld porosity, using a machine-vision solutions based on locating a camera directly on the welding gun, just 12 to 14 in. from the weld. Using Intel’s edge software running on an Intel Core processor-based industrial PC and optimized with Intel’s OpenVino toolkit, the solution examined the streaming video frames and, when it detected defects, it would stop the welding robot.   

While the piloted solution never found its way into production, the opportunity to work with Intel to leverage AI to address weld porosity was a chance to bring together two of John Deere’s core values—innovation and quality. For an update from Deere, MetalForming spoke with Will Chemin, tech lead for welding and fabrication in the firm’s Construction & Forestry Division.

“John Deere has used smart technology for a long time in our industrial robotic-welding cells,” Chemin says. Among the features he notes:

• Touch sense via weld wire or laser to detect any shift in the weld joint; part presence in the robotic-welding fixture; and part identification, where after a series of touch senses, the control will select the appropriate robotic-welding program, eliminating the opportunity for human error.
• Through-the-arc seam tracking, sensing the welding current to maintain a constant contact tip-to-work distance and enabling the robot control to alter the programmed path of the robot to adjust for variations in the weld joint due to part dimensional and fitup tolerances.

“While these applications are not entirely new to the industry, we have made developments in how we use the information,” Chamin says. “As an example, using touch sensing (via laser or weld wire) for part ID, we have automated the reporting process in our MRP system (SAP), so operators no longer need to manually report their part production. By doing this we also eliminate the opportunity for operators to scan incorrect parts, and we can trigger when parts are due for inspection, referring back to the control plan, by enabling counters in the John Deere Assembly Assist Tool (JDAAT).”

Note: JDAAT is an internally created AI-powered manufacturing execution system that tracks all of the work instructions an operator may need to ensure timely, properly performed quality checks.

“We’re always looking to advance in welding and fabrication,” Chemin adds, “leveraging new technologies.” 

The Construction and Forestry Division operates globally, with two of its largest facilities in Iowa, Dubuque and Davenport, which combined house some 150 robotic-welding arms. All use touch sensing, split evenly between weld-wire touch and laser. The Dubuque plant manufactures Deere’s lineup of skid-steer loaders and compact track loaders, crawlers and backhoes, while the Davenport facility builds motor graders, articulated dump trucks and loaders.