Prolonging Robot Lifespan Using Fatigue Balancing with Reinforcement Learning

Over the last quarter century, industrial robots have become an indispensable element in modern manufacturing and assembly. While the costs of implementing these robot systems can be high, over time these expenditures are offset by their longevity and productivity. One of the major costs of existing robotic systems is the maintenance and replacement of individual robots. Currently, process and manufacturing robots are programmed by robotic technicians with predetermined motions based on process requirements and product specifications. These repetitive motions tend to wear out specific joints faster than others, causing total robot failure and precipitating the replacement of the entire robot. This paper leverages machine learning algorithms to modify nonproduction critical robot movements to balance the physical load at each joint, with the goal of extending the useful life of these industrial robots in a manufacturing environment. Experimental results show that Q-Learning balanced the average load on robot joints by roughly 14%.