Predicting Object Deformation via Machine Learning Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current additive manufacturing techniques face challenges in predicting object deformation during the manufacturing process, particularly due to thermal diffusion and manufacturing errors, which existing first-principle-based simulation methods are slow and lack the necessary resolution.
Innovation Solution
A machine learning model, specifically a deep neural network, is employed to predict object deformation by converting point clouds from 3D object models into edge features and convolving them to generate a predicted point cloud, allowing for accurate geometry prediction and deformation compensation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If first-principle-based simulation methods are used to predict object deformation, then prediction capability is provided, but speed is slow and resolution is insufficient
Solution Approach 1:
The patent replaces first-principle-based simulation methods with a machine learning model that has been trained on simulation data. This substitution transitions from computational physics simulations to a trained predictive model, achieving both high resolution and fast prediction speed. The machine learning model learns deformation patterns from extensive simulation training data and can predict deformations for new objects instantly without running full physics simulations.
Solution Approach 2:
The patent performs extensive simulation computations in advance to train the machine learning model. By pre-computing deformation scenarios across diverse object geometries and material properties during the training phase, the system prepares a knowledge base that enables rapid predictions during actual manufacturing planning without requiring real-time simulation execution.
2Manufacturing precision
If iterative compensation is implemented to reduce manufacturing errors, then manufacturing precision is improved, but computation time increases
Solution Approach 1:
The patent replaces iterative physics-based simulation and compensation calculations with a machine learning model that directly predicts deformations and provides compensation adjustments. The model processes object geometry and manufacturing parameters to output deformation predictions and compensation values in a single computational pass, eliminating the need for repeated simulation cycles while maintaining high geometric accuracy.
Data Source
AI summary
Examples of methods for predicting object deformations are described herein. In some examples, a method includes predicting a point cloud. In some examples, the predicted point cloud indicates a predicted object deformation. In some examples, the point cloud may be predicted using a machine learning model and edges determined from an input point cloud.