Robot Controller Noise Subtraction for Pressing Force Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robots struggle to finely control pressing force during operations like deburring and polishing due to operating noise interference, which prevents detection of small forces and accurate control of the tool's contact with the workpiece.
Innovation Solution
A controller system that utilizes a state observation unit, label data acquisition unit, and learning unit to generate a learning model correlating no-load state variables with force sensor data, allowing for estimation and subtraction of noise to accurately control the pressing force during machining operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a force sensor is used to detect pressing force, then pressing force detection is enabled, but operating noise prevents detection of small forces
Solution Approach 1:
The patent extracts and separates the operating noise component from the total force sensor signal. By using a learning model to identify and extract the noise portion based on robot state information, the system can isolate the actual pressing force signal even when it is smaller than the noise level, thereby resolving the detection limitation
Solution Approach 2:
The learning model acts as an intermediary that processes robot state variables and predicts noise components. This intermediary system bridges the gap between raw sensor data and actual pressing force by filtering out noise through learned correlations, enabling accurate detection of small forces
2Object-affected harmful factors
If operating noise is absorbed with spring or rubber, then noise reduction is achieved, but pressing force control is lost
Solution Approach 1:
The patent replaces the mechanical noise absorption approach (springs or rubber) with an information-processing approach using a learning model. Instead of physically damping the noise, the system uses computational methods to identify and subtract noise components from sensor readings, preserving the mechanical coupling necessary for precise force control
Solution Approach 2:
The learning model serves as an intermediary that processes sensor data to separate noise from actual force signals. This allows the mechanical system to remain rigid and responsive for precise control while the computational layer handles noise filtering, combining the advantages of both approaches
3Object-affected harmful factors
If machine learning is used to estimate disturbance, then noise removal is improved, but vibration and tool-specific operating noise are not accounted for
Solution Approach 1:
The patent enhances the learning model to handle multiple types of noise sources universally. By incorporating robot state variables including drive unit rotational and vibrational frequencies, the model can adapt to and remove various noise types (vibration, tool-specific operating noise, gravitational effects) through a single unified approach, making the system versatile across different machining operations
Data Source
AI summary
A controller has a state observation unit to acquire a present state of a robot as a state variable, a label data acquisition unit to acquire, as a label, a detected value of a force sensor attached to an arm and to detect necessary data for control of a pressing force, and a learning unit to generate a learning model indicative of the correlation between the state variable acquired in a no-load state and the label acquired under the state variable acquired in the no-load state and to estimate the detected value of the force sensor. The controller controls the pressing force by using the detected value of the force sensor acquired in the present state of the robot acquired in a loaded state and the detected value of the force sensor estimated by the learning unit based on the present state of the robot acquired in the loaded state.