Robot PCB Through-Hole Insertion Using RL and Admittance Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for inserting electronic components into printed circuit boards using industrial robots are inefficient due to complex geometries and susceptibility to damage, requiring manual parameter tuning and high precision sensors, and lack precision in alignment.
Innovation Solution
A method utilizing reinforcement learning (RL) algorithms, specifically the Soft Actor-Critic (SAC) algorithm, integrated with admittance control and convolutional neural networks (CNNs), to guide an industrial robot in inserting electronic components into a printed circuit board by merging images from multiple perspectives and using admittance controllers for precise positioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional hybrid force-position control with preprogrammed trajectory is used, then the robot can perform insertion tasks, but the method lacks precision in alignment and requires manual parameter tuning
Solution Approach 1:
The system uses reinforcement learning to enable the robot to automatically learn and optimize insertion parameters through trial and error, eliminating the need for manual parameter tuning. The robot autonomously improves its insertion performance by receiving feedback from sensors and adjusting its own control parameters.
Solution Approach 2:
The system implements a closed-loop feedback mechanism where sensors detect the insertion state and provide real-time information to the reinforcement learning algorithm, which then adjusts the robot's trajectory and force parameters dynamically to achieve precise alignment.
2Reliability
If impedance or admittance controllers with programmed trajectories are used, then the robot can maintain constant downforce, but the method requires high precision sensors and well-made robotized production machine
Solution Approach 1:
The reinforcement learning algorithm dynamically adjusts control parameters including downforce, trajectory, and insertion speed based on real-time sensor feedback and learned patterns, allowing the system to maintain reliable insertion without requiring extremely high precision sensors or rigid machine structures.
Solution Approach 2:
The system transitions from static preprogrammed trajectories to dynamic adaptive trajectories that are continuously adjusted based on sensor feedback and reinforcement learning, enabling the robot to adapt to variations in PCB and component geometry while maintaining constant downforce control.
3Manufacturing precision
If pure vision system with high-speed machine vision is used, then alignment can be achieved, but the method does not provide force control for safe assembly
Solution Approach 1:
The system merges vision-based alignment with force-controlled insertion by integrating camera data with impedance/admittance control and reinforcement learning, achieving both precise alignment and safe assembly by coordinating visual information with force feedback.
Solution Approach 2:
The reinforcement learning algorithm acts as an intermediary that translates vision system alignment data into appropriate force control parameters, coordinating the visual alignment information with the force-controlled insertion process to prevent component damage.
4Productivity
If conventional methods with preprogrammed spiral path are used, then insertion can be performed, but the method is inefficient for complex geometries and susceptible to damage
Solution Approach 1:
The system replaces static preprogrammed spiral paths with dynamic adaptive trajectories generated by reinforcement learning, allowing the robot to optimize its insertion path in real-time based on detected geometry and component characteristics, improving efficiency while preventing damage.
Data Source
AI summary
A method of inserting an electronic components in through-hole technology, THT, into a printed circuit board, PCB by an industrial robot, based on reinforcement learning, includes grabbing, by means of a tool with universal fingers mounted to the end-effector of the industrial robot, the electronic component to be inserted into the PCB; moving the tool to a starting position being in close proximity to a final position of the electronic component; acquiring at least one image showing the tool, the electronic component and the PCB; calculating, on a basis of the at least one image, at least one movement instruction for the industrial robot; adjusting position of the tool on a basis of the at least one movement instruction, and repeating the steps until the electronic component is in the final position.