Robot Liquid Manipulation Simulation Using Learned Surface Motion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Simulating the movement of flexible or liquid objects with high accuracy results in a large computation load and prolonged simulation times.
Innovation Solution
A simulation device utilizing supervised learning to generate a learning model that predicts the movement of representative points on the surface of the operation subject, reducing the computation load by substituting detailed simulations with a more efficient prediction method.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high accuracy simulation of liquid object movement is performed using finite element method or particle method, then simulation precision is improved, but computation load increases and simulation time becomes longer
Solution Approach 1:
The patent performs preliminary high-accuracy simulations to generate training data, then uses this data to train a neural network model. The trained model can subsequently predict liquid object movement with much lower computation load and faster speed, effectively preparing the system in advance to avoid repeated heavy computations.
Solution Approach 2:
The patent creates a virtual copy of the physical simulation process by training a neural network model to replicate the behavior of finite element or particle method simulations. This neural network copy can predict liquid object movement without requiring the computationally intensive original simulation methods, thus reducing both computation load and time.
2Measurement precision
If high accuracy simulation of liquid object movement is performed using finite element method or particle method, then simulation precision is improved, but computation load becomes large
Solution Approach 1:
The patent creates a virtual copy of the physical simulation process by training a neural network model to replicate the behavior of finite element or particle method simulations. This neural network copy can predict liquid object movement without requiring the computationally intensive original simulation methods, thus reducing both computation load and time.
Solution Approach 2:
The patent substitutes the mechanical computation process of finite element methods or particle methods with a neural network-based prediction system. The neural network, once trained, replaces the need for repeated heavy mechanical computations, significantly reducing computation load while maintaining prediction accuracy.
3Manufacturing precision
If detailed simulation of operation subject movement is performed, then manufacturing precision of simulation results is improved, but device complexity increases
Solution Approach 1:
The patent creates a virtual copy of the physical simulation process by training a neural network model to replicate the behavior of finite element or particle method simulations. This neural network copy can predict liquid object movement without requiring the computationally intensive original simulation methods, thus reducing both computation load and time.
Data Source
AI summary
A simulation device that reduces the computation load required for simulation of the movement of an operation subject is provided. The simulation device includes a processor configured to set framing conditions for a model representing the operation subject, set conditions for external force applied to the operation subject, simulate the movement of the operation subject under the framing conditions and the conditions for external force, generate learning data that includes data representing the framing conditions, the conditions for external force, and the movement of representative points located on the surface of the operation subject during simulation of the movement of the operation subject, and by supervised learning using the learning data, generate a learning model that takes the framing conditions, the conditions for external force, and the initial conditions of the representative points as input and outputs data representing the movement of the representative points.