Bayesian Phase Diagram Population via Maximum Entropy Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high computational and experimental costs associated with populating phase diagrams for multiple substances make it inefficient to determine conditions for phase separation in mixtures, which is crucial for formulations in industries like liquid formulation.
Innovation Solution
A computer-implemented method using Bayesian optimization and maximum entropy sampling to selectively calculate and interpolate data points in phase diagrams, reducing the number of required calculations and increasing resolution, allowing for faster and more accurate phase diagram generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If rigorous phase separation experiments are performed to populate phase diagrams, then measurement precision and reliability are improved, but computational expense and time consumption increase significantly
Solution Approach 1:
The patent creates a computational copy of the phase diagram using machine learning models trained on limited experimental data. The model generates predicted phase boundaries and properties that replicate the information obtained from exhaustive experiments, thereby reducing the need for time-consuming rigorous experimentation while maintaining measurement precision.
Solution Approach 2:
The patent performs preliminary computational work by training machine learning models on a small subset of experimental data before full phase diagram population. This preliminary action establishes predictive capabilities that can then rapidly generate phase diagram information without requiring complete exhaustive experimentation, significantly reducing time expense.
2Loss of time
If simulations are used instead of experiments to investigate phase diagrams, then time expense and cost are reduced, but computational expense remains significant
Solution Approach 1:
The patent creates a computational copy or surrogate model that replicates the behavior of complex simulation systems. Once trained on limited simulation data, the machine learning model can rapidly predict phase diagram properties without requiring repeated expensive simulations, thereby reducing computational expense while maintaining the benefits of simulation over experimentation.
Solution Approach 2:
The patent changes the computational approach by transitioning from direct physics-based simulations to machine learning predictions. This parameter change in the computational method reduces the computational resources required, as the trained model can make predictions much faster than running full simulations for each query point in the phase diagram.
3Manufacturing precision
If full resolution phase diagrams are populated using traditional methods, then manufacturing precision is improved, but productivity decreases due to computational expense
Solution Approach 1:
The patent uses machine learning models to create a computational copy of the complete high-resolution phase diagram. The model is trained on a sparse set of data points and then generates predictions for all points in the full-resolution phase diagram, achieving manufacturing precision equivalent to traditional methods while dramatically improving productivity through rapid prediction.
Solution Approach 2:
The patent performs preliminary training of the machine learning model on a limited dataset, establishing the predictive framework in advance. This preliminary action enables the model to subsequently generate full-resolution phase diagrams rapidly without requiring repeated expensive calculations for each data point, thereby improving productivity while maintaining precision.
Data Source
AI summary
Method and system are provided for efficiently populating a phase diagram for modeling of multiple substances. The method may include defining an n-way phase diagram with data points each being an n-tuple describing the n substance inputs, wherein the n-way phase diagram is defined at a user-configured resolution. The method may select an initial subset of data points and calculate their contribution to the phase diagram. The method may then generate a Bayesian model based on the initial subset of calculated data points and predicting the resultant phase and an associated uncertainty of all the uncalculated data points in the defined phase diagram. The method may select a sample subset of the data points using maximum entropy sampling and calculating a resultant phase for each of the selected data points, and incorporate the calculated phases into the Bayesian model. Re-modeling the Bayesian model may predict the resultant phase and an associated uncertainty of all the remaining uncalculated data points in the defined phase diagram. Repeating the selecting a sample subset of the data points using maximum entropy sampling and re-modeling is carried out until a defined termination criterion is met.