Efficient and accurate trajectory pose compensation of robot-assisted terahertz system for blade evaluation combining D-H model and machine-learning

The performance evaluation of terahertz measurement technology on complex curved surfaces, such as blades, holds significant importance. Industrial robots can be equipped with probes to realise this technology. However, overcoming the limitation of robot encoder modification and addressing the problem of poor pose accuracy in robot-assisted terahertz detection is one of the main challenges affecting the evaluation results . This paper proposes a compensation method combining model and optimisation-based machine learning for robot-assisted terahertz detection. First, robot trajectory planning that meets the requirements for terahertz measurement of blades is described, and a NURBS surface model of the blade and efficiently plan the probe’s orientation is proposed. Second, a rapid and precise method for optimising trajectory pose accuracy, combining the robot’s D-H model with calibration compensation based on Particle Swarm Optimization-Extreme Learning Machine (PSO-ELM), is proposed. Third, prediction and compensation method is simulated and validated against the trajectory planning results. Finally, precision evaluation is conducted based on the calibration outcomes, and the robot assisted terahertz blade detection experiment is carried out. The results show that compared to the other compensation method, this approach improves both precision and efficiency, laying the foundation for detection devices in terahertz measurement research on curved surfaces.