Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-09-0266
Machine Learning Based Optimal Acceleration/Deceleration Design for a 3-UPU Type Parallel Kinematic Mechanism
1Department of Power Mechanical Engineering, National Tsing Hua University, Hsinchu City, Taiwan
2Department of Mechanical Engineering, Ming Chi University of Technology, New Taipei City, Taiwan
ABSTRACT
Since large acceleration and deceleration could undermine the machining accuracy and cause machinery damage, an interpolator that plans the motion of the system and a controller that maneuvers the motion are included in a CNC system. Many studies have been focusing on the effects of motion profile on machine stability, but few consider the physical limitation of the machine. In this study, a machine learning based interpolator that plans the motion profile of a pre-determined path of the end effector is proposed for a 3-UPU type Parallel Kinematic Mechanism (PKM). First, a trapezoidal motion profile is created to provide the velocity-ramps at two ends. A kinetostatic model that considers the external loadings exerted at the end-effector, the equivalent inertia forces of the components, the specifications of motors and ball screws, as well as the gravitational effect, are employed to verify the maximum loadings of the joints and define the maximum allowable acceleration of the end-effector. The resulting maximum accelerations in the task space were mapped into the joint space to obtain the allowable jerk constraints of the individual length-varying links in the following optimization algorithm. Then, the interpolator labels the positions on the pre-determined path whose corresponding jerk of the length-varying links exceeds the user-defined limits. A Gaussian-shaped deceleration profile is added to each labeled position that serves as the basis of the path profile. Accordingly, a machine learning model that optimizes the motion profiles by Adam algorithm with PyTorch is developed. The objective of the loss function is to search for the shortest processing time within the user-defined limit for the jerk of the length-varying limbs by adjusting the parameters of the Gaussian profiles. Three illustrative examples in different paths and with jerk limits between 1500mm/s3 to 5500mm/s3 were studied to verify the proposed interpolator of a 3-UPU PKM. In conclusion, the proposed machine learning based interpolator that accounts for the constraints and the motions of both the end-effector and the actuators provides a possible precursor for the development of a cyber-physical system.
Keywords: PKM, 3-UPU, Machine learning, Acceleration/deceleration design, Interpolator
1Department of Power Mechanical Engineering, National Tsing Hua University, Hsinchu City, Taiwan
2Department of Mechanical Engineering, Ming Chi University of Technology, New Taipei City, Taiwan
ABSTRACT
Since large acceleration and deceleration could undermine the machining accuracy and cause machinery damage, an interpolator that plans the motion of the system and a controller that maneuvers the motion are included in a CNC system. Many studies have been focusing on the effects of motion profile on machine stability, but few consider the physical limitation of the machine. In this study, a machine learning based interpolator that plans the motion profile of a pre-determined path of the end effector is proposed for a 3-UPU type Parallel Kinematic Mechanism (PKM). First, a trapezoidal motion profile is created to provide the velocity-ramps at two ends. A kinetostatic model that considers the external loadings exerted at the end-effector, the equivalent inertia forces of the components, the specifications of motors and ball screws, as well as the gravitational effect, are employed to verify the maximum loadings of the joints and define the maximum allowable acceleration of the end-effector. The resulting maximum accelerations in the task space were mapped into the joint space to obtain the allowable jerk constraints of the individual length-varying links in the following optimization algorithm. Then, the interpolator labels the positions on the pre-determined path whose corresponding jerk of the length-varying links exceeds the user-defined limits. A Gaussian-shaped deceleration profile is added to each labeled position that serves as the basis of the path profile. Accordingly, a machine learning model that optimizes the motion profiles by Adam algorithm with PyTorch is developed. The objective of the loss function is to search for the shortest processing time within the user-defined limit for the jerk of the length-varying limbs by adjusting the parameters of the Gaussian profiles. Three illustrative examples in different paths and with jerk limits between 1500mm/s3 to 5500mm/s3 were studied to verify the proposed interpolator of a 3-UPU PKM. In conclusion, the proposed machine learning based interpolator that accounts for the constraints and the motions of both the end-effector and the actuators provides a possible precursor for the development of a cyber-physical system.
Keywords: PKM, 3-UPU, Machine learning, Acceleration/deceleration design, Interpolator
