Enhancing robotic manipulator performance through analyzing vibration, identifying deep-learning-based modal parameters, and estimating frequency response functions

The present study’s objective is to develop a computationally effective prediction model that can be applied to offline optimization algorithms to determine the best robot configuration and increase milling productivity. To improve the machining robot’s performance, this work proposes a deep learning-based method which can anticipate the modal parameters that characterize the robots’ vibration behavior in different configurations. Given the posture-dependent nature of the vibrations produced by the robot, the present study begins with an elaborate data-gathering process employing high-resolution accelerometers and hammer tests, encompassing a wide range of robot postures and trajectories. A large number of hammer tests are performed to provide an accurate, realistic, and relevant dataset for deep learning (DL). With the use of a feedforward neural network, our model can accurately forecast critical modal parameters, which include natural frequencies, damping ratios, and modal stiffnesses. This suggested model is almost perfect for being implemented in optimization algorithms since it only takes around 90 ms to predict multiple frequency modes, which requires FRFs to be recalculated in each iteration. It is of significance in improving the efficiency of the optimization process. It makes it very much possible to make real-time or near real-time adjustments during or after planning robotic milling operations. These findings emphasize the effectiveness of incorporating machine learning into robotic system analysis and significantly advance the optimization of robot manipulators.