Lab robot system tests steel heat treatments over 10 times faster

Inside the Institute for Materials Research at Tohoku University, a steel specimen is heated in a furnace, then a robot arm carries it to an X-ray instrument and then to an impact tester. No one touches it between steps. Once the heat treatment itself is finished, the handling, measurement and recovery of each sample take about 800 seconds.

Getting a steel to be both stronger and tougher means searching an enormous range of processing conditions. Even with the same alloy, heat treatment can change the metal's microstructure and mechanical properties dramatically. Bulk structural metals have been particularly hard to automate compared with liquids, powders and thin films, because every solid specimen has to be transported, positioned, heated, characterized and mechanically tested, and each step relies on large, specialized equipment.

The team's answer was a set of dedicated specimen-handling tools and a single computer controlling all the instruments. The platform combines an induction heating furnace, an X-ray diffractometer (XRD) and a micro-impact testing machine, with a robotic arm moving specimens between them. The researchers used it to run one- and two-step tempering of a commercial JIS SCM435H low-alloy steel, a grade defined by Japanese industrial standards. Each run produces processing, structure and property data at the same time.

The authors report that the mechanical evaluation reaches accuracy comparable to measurements that researchers perform by hand, while the experimental sequence runs more than 10 times faster. The platform also carries an AI optimization tool, integrated into the NIMS Orchestration System (NIMO), which already adjusts furnace control parameters automatically. The loop that would close the cycle is described as future work: AI analyzes each result, proposes the next heat treatment, the platform runs it and sends the new data back.

For materials teams, the immediate gain is breadth. The authors say the system could substantially expand the heat-treatment conditions researchers can explore, and it generates the data that AI needs to learn from. The team gives no timeline for industrial use, and the closed loop has not yet been demonstrated. For now, the evidence covers one steel grade, tested in one laboratory, and Miyamoto's stated aim is to speed up the autonomous discovery of new processing routes and structural materials.