IIT Madras develops AI framework for real time gearbox fault detection

Chennai: Indian Institute of Technology Madras (IIT-M) researchers have developed a new artificial intelligence (AI) framework to detect faults in gearboxes used in cars and other vehicles, pumps, and heavy industrial machines. The system works in real time even with vibration sensors used to detect faults in a gearbox placed in non-ideal locations, avoiding the need for costly design changes.

The IIT-M team created a method that uses reinforcement learning (RL), a branch of AI where a computer program learns through trial and error. In factories, engineers normally use vibration sensors to monitor gearboxes. However, placing these sensors in the right locations is often difficult because of heat, wiring, space limits, or exposure to fluids, making fault detection unreliable.

The IIT-M framework solves this by using a ‘multi-sensor fusion approach'. Data from several sensors, even when placed in non-ideal positions, is combined and filtered using an ‘adaptive filtering technique'. This helps isolate fault signals from background noise. The RL agent then selects the best parameters for accurate analysis.

"In the lab, things are simpler as the person performing experiments usually knows where the fault is and can place sensors immediately next to it. However, in real industrial settings, the field engineers do not always know precisely where the issue is, and if they do, placing sensors close enough is not always possible. That is where the proposed framework comes in," said Shahis Hashim, researcher at engineering asset management (EAM) group, department of mechanical engineering, IIT-M. "While it is currently tested on gearboxes, the proposed approach can be applied across many industrial systems," said EAM group researcher Sitesh Kumar Mishra.

The method was tested on a gearbox in the laboratory and was able to assess faults quickly and reliably. A gearbox controls the speed and torque of machines by transferring power from one component to another, making it essential in vehicles and industrial equipment. Monitoring gearboxes helps detect faults early, preventing breakdowns, costly repairs, and safety risks. "Our algorithm fuses data from non-ideal sensor locations, enhancing fault detection—an urgent need we address," said professor Piyush Shakya, who leads the EAM group at IIT-M.