Hybrid Sintering State Prediction Using Physics and Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current additive manufacturing techniques, such as metal printing, face challenges in controlling the shape of end objects due to porosity and deformation during the sintering process, which can result in significant shape changes and long simulation times for physics-based simulation engines.
Innovation Solution
A hybrid approach combining physics-based simulation engines with machine learning models, specifically deep neural networks, to predict sintering states at larger time increments, reducing simulation time while maintaining accuracy by iteratively refining displacement fields and utilizing multiple machine learning models tailored to different sintering stages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If physics-based simulation engines are used to simulate sintering processes, then accuracy of shape prediction is improved, but simulation time increases significantly
Solution Approach 1:
The system performs preliminary actions by pre-calculating and storing sintering state data at various time increments using physics-based simulation. These pre-computed results are then used to train machine learning models, which can rapidly predict sintering states during actual manufacturing without requiring full physics-based simulation runs, thus reducing simulation time while maintaining accuracy
Solution Approach 2:
The patent creates simplified copies of the complex physics-based simulation by training machine learning models to replicate the simulation outcomes. These ML model copies can predict sintering states instantly without requiring the computationally intensive physics-based calculations, achieving fast predictions with accuracy comparable to the original physics-based simulation
2Volume of moving object
If physics-based simulation engines simulate larger objects, then manufacturing capability is improved, but simulation time increases significantly
Solution Approach 1:
The patent segments the simulation process into two parts: a comprehensive physics-based simulation performed once on representative samples to generate training data, and rapid ML-based predictions for actual large-scale manufacturing. This segmentation allows the system to handle larger objects efficiently by using the pre-trained ML models rather than performing full physics-based simulations on every large object
Solution Approach 2:
The system creates simplified computational copies through ML models that can rapidly simulate large objects without the computational burden of full physics-based simulation. These ML copies enable fast prediction of sintering states for large-scale manufacturing while maintaining accuracy, directly addressing the productivity constraint
3Productivity
If machine learning models are used to predict sintering states, then simulation speed is improved, but accuracy may be compromised without proper training
Solution Approach 1:
The system performs preliminary action by comprehensively training the machine learning model using high-quality physics-based simulation data covering various sintering conditions, temperatures, and time increments. This thorough preliminary training ensures the ML model achieves high prediction accuracy before being deployed for rapid manufacturing simulations
Solution Approach 2:
The patent implements feedback mechanisms where the ML model predictions are continuously refined by comparing them against physics-based simulation results during the training phase. This feedback loop ensures the ML model learns to accurately predict sintering states while maintaining simulation speed, resolving the accuracy-speed trade-off
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This hybrid method significantly reduces simulation time for sintering processes, achieving faster and accurate predictions of object deformation, allowing for more efficient control over the shape of end objects and enabling larger object simulations in a fraction of the time required by traditional methods.
Implementation Method 1
A machine learning model may be trained using training data that includes a simulated input sintering state at a start time and a simulated output sintering state at a target time. The trained machine learning model may then be used to predict a sintering state of an object at a second time based on a simulated sintering state of the object at a first time.
Implementation Method 2
simulate, using a physics engine, a first sintering state of an object at a first time... predict, using the trained machine learning model, a second sintering state of the object at a second time based on the first sintering state
Data Source
AI summary
Examples of methods are described herein. In some examples, a method includes simulating, using a physics simulation engine, a first sintering state of an object at a first time. In some examples, the method includes predicting, using a machine learning model, a second sintering state of the object at a second time based on the first sintering state. In some examples, a prediction increment between the first time and the second time is different from a simulation increment.