Automated generation of a machine learning model from computational simulation data
Systems and methods for automatically training a machine learning model are described herein. An example method includes performing a set of computational simulations; and assembling a data set associated with the set of computational simulations. The data set includes data associated with at least one simulation result for at least one computational simulation in the set of computational simulations. The method also includes training a machine learning model with the data set. At least one feature and at least one target for the machine learning model are part of the data set.
1 . A method for automatically training a machine learning model, comprising:
(a) receiving a simulation input from a user, wherein the simulation input comprises a model representing a physical system, a set of material properties associated with the model, and an applied condition that describes a condition configured to be applied to the model during a simulation;
(b) determining, for the applied condition, a parameter range for each of two or more variables associated with the applied condition, including a first variable and a second variable, wherein the parameter range describes a plurality of parameter values that are potential values for the two or more variables associated with the applied condition;
(c) automatically performing a set of computational simulations based on a plurality of (c) combinations of parameterized variable values for each of the two or more variables to produce a set of simulation results, wherein the set of simulation results comprises, for each result, a resulting condition that corresponds to the applied condition when simulated with that combination of parameterized variable values, wherein the resulting condition comprises one or more resulting variables;
(d) assembling a data set based on the set of simulation results, wherein the data set describes, for each of the set of computational simulations, input values of the two or more variables associated with the applied condition for that simulation, and output values associated with the resulting condition for that simulation;
(e) receiving, from a user, a specification that designates two or more features and a target from the data set, wherein each of the target and the two or more features are distinct, and wherein the target and the two or more features are selected from the two or more variables associated with the applied condition and the one or more resulting variables associated with the resulting condition; and
(d) training a machine learning model with the data set according to the specification, wherein the machine learning model when trained is configured to predict a value of the target based on the value of the two or more features.
2 . The method of claim 1 , further comprising, when automatically performing the set of computational simulations based on the plurality of combinations of parameterized variable values:
(a) after performing one or more of the set of computational simulations, training and testing the machine learning model to determine whether a stop criteria has been reached based on an analysis of the predicted value of the target and a configured accuracy threshold associated with the stop criteria; and
(b) where the stop criteria has not been reached, continue performing the set of computational simulations, and training and testing the machine learning model, until the stop criteria has been reached.