Predictive modeling for forged components
In general, various aspects of the techniques enable predictive modeling for forged components. A computing device comprising a memory and a processor may be configured to perform the techniques. The memory may store a trained machine learning model that associates training features extracted from data representative of a plurality of training forged components to a plurality of training model results. The memory may also store data representative of a target forged component. The processor may perform a geometrical analysis with respect to the data representative of the target forged component to extract target features, and apply the trained machine learning model to the target features to obtain predicted model results for the target forged component. The processor may also output the predicted model results.
1 . A computing device comprising:
a memory; and
one or more processors configured to:
extract training features from data representative of a plurality of surface geometry zones for a plurality of training forged components, wherein each training forged component from the plurality of training forged components is associated with data representative of a respective subset of surface geometry zones from the plurality of surface geometry zones;
for each training forged component:
obtain a plurality of computational fluid dynamics results for the respective subset of geometry zones by modeling application of a quench process to the respective training forged component according to quench parameters, wherein each geometry zone from the respective subset of geometry zones is associated with a respective subset of computational fluid dynamics results from the plurality of computational fluid dynamics results; and
determine, based on the plurality of computational fluid dynamics results, a respective model for identifying a respective set of fitting parameters;
obtain, based on each respective set of fitting parameters, a plurality of training model results representative of one or more mechanical properties;
store, in the memory, a trained machine learning model that associates the training features to the plurality of training model results;
store, in the memory, data representative of a target forged component having a target surface geometry zone that differs from all of the plurality of surface geometry zones for the plurality of training forged components;
perform a geometrical analysis with respect to the data representative of the target forged component to extract target features;
apply the trained machine learning model to the target features to obtain predicted model results for the target forged component, the predicted model results representative of the application of the quench process to the target forged component according to the quench parameters; and
output the predicted model results.
2 . The computing device of claim 1 , wherein the predicted model results identify a quench distortion for the target forged component as a function of location over the target surface geometry zone.
3 . The computing device of claim 2 , wherein the quench distortion includes one or more of a tensile strength, a transient stress, and a microstructure.
4 . The computing device of claim 1 , wherein to obtain the plurality of training model results, the one or more processors are further configured to:
obtain a respective set of heat transfer coefficients as a function of location over the respective subset of surface geometry zones, wherein the respective set of fitting parameters is identified based on the respective set of heat transfer coefficients; and
apply materials and process modeling to each respective set of fitting parameters to obtain the plurality of training model results.
5 . The computing device of claim 1 ,
wherein the training features extracted from the data representative of the plurality of surface geometry zones for the plurality of training forged components include training principal component coefficients representative of each of the plurality of surface geometry zones,
wherein the one or more processors are, when configured to perform the geometrical analysis, configured to perform principal component analysis with respect to the data representative of the target forged component having the target surface geometry zone to obtain target principal component coefficients representative of the target surface geometry zone of the target forged component, and
wherein the one or more processors are configured to, when applying the trained machine learning model, apply the trained machine learning model to the target principal component coefficients representative of the target surface geometry zone of the target forged component to obtain the predicted model results.
6 . The computing device of claim 1 ,
wherein the features extracted from the data representative of the plurality of surface geometry zones for the plurality of training forged components include training feature maps extracted through application of a convolutional neural network to the data representative of the plurality of surface geometry zones for the plurality of training forge components,
wherein the one or more processors are, when configured to perform the geometrical analysis, configured to apply the convolutional neural network to the data representative of the target forged component having the target surface geometry zone to obtain a target feature map representative of the target surface geometry zone of the target forged component, and
wherein the one or more processors are configured to, when applying the trained machine learning model, apply the trained machine learning model to the target feature map representative of the target surface geometry zone of the target forged component to obtain the predicted model results.
7 . The computing device of claim 1 , wherein the one or more processors are further configured to:
obtain model data representative of a training forged component of the plurality of training forged components; and
apply the trained machine learning model to the model data representative of the training forged component of the plurality of forged components to determine whether the quenching process was performed according to the quench parameters or different quench parameters.
8 . The computing device of claim 1 , wherein to obtain the plurality of training model results, the one or more processors are further configured to:
obtain a respective set of heat transfer coefficients as a function of location over the respective subset of surface geometry zones; and
apply a fitting model to each respective set of heat transfer coefficients as a function of location over the respective subset of surface geometry zones in order to obtain the respective set of fitting parameters, wherein the plurality of training model results include all of the sets of fitting parameters.
9 . The computing device of claim 8 , wherein the one or more processors are configured, when applying the trained machine learning model, to:
apply the trained machine learning model to the target features to obtain predicted fitting parameters for the target forged component; and
apply the fitting model to the predicted fitting parameters to obtain the predicted model results as predicted heat transfer coefficients for the target forged component.
10 . The computing device of claim 1 , wherein the data representative of the target forged component includes one or more of an image of the target forged component and a three-dimensional framework of the target forged component.
11 . A method comprising:
extracting training features from data representative of a plurality of surface geometry zones for a plurality of training forged components, wherein each training forged component from the plurality of training forged components is associated with data representative of a respective subset of surface geometry zones from the plurality of surface geometry zones;
for each training forged component:
obtaining a plurality of computational fluid dynamics results for the respective subset of geometry zones by modeling application of a quench process to the respective training forged component according to quench parameters, wherein each geometry zone from the respective subset of geometry zones is associated with a respective subset of computational fluid dynamics results from the plurality of computational fluid dynamics results; and
determining, based on the plurality of computational fluid dynamics results, a respective model for identifying a respective set of fitting parameters;
obtaining, based on each respective set of fitting parameters, a plurality of training model results representative of one or more mechanical properties;
storing, in a memory, a trained machine learning model that associates the training features to the plurality of training model results;
storing, in the memory, data representative of a target forged component having a target surface geometry zone that differs from all of the plurality of surface geometry zones for the plurality of training forged components;
performing a geometrical analysis with respect to the data representative of the target forged component to extract target features;
applying the trained machine learning model to the target features to obtain predicted model results for the target forged component, the predicted model results representative of an application of a quench process to the target forged component according to quench parameters; and
outputting the predicted model results.
12 . The method of claim 11 , wherein the predicted model results identify a quench distortion for the target forged component as a function of location over the target surface geometry zone.
13 . The method of claim 12 , wherein the quench distortion includes one or more of a tensile strength, a transient stress, and a microstructure.
14 . The method of claim 11 , wherein obtaining the plurality of training model results further comprises:
obtaining a respective set of heat transfer coefficients as a function of location over the plurality respective subset of surface geometry zones, wherein the respective set of fitting parameters is identified based on the respective set of heat transfer coefficients; and
applying materials and process modeling to each respective set of fitting parameters to obtain the plurality of training model results.
15 . The method of claim 11 ,
wherein the training features extracted from the data representative of the plurality of surface geometry zones for the plurality of training forged components include training principal component coefficients representative of each of the plurality of surface geometry zones,
wherein performing the geometrical analysis comprises performing principal component analysis with respect to the data representative of the target forged component having the target surface geometry zone to obtain target principal component coefficients representative of the target surface geometry zone of the target forged component, and
wherein applying the trained machine learning model comprises applying the trained machine learning model to the target principal component coefficients representative of the target surface geometry zone of the target forged component to obtain the predicted model results.
16 . The method of claim 11 ,
wherein the features extracted from the data representative of the plurality of surface geometry zones for the plurality of training forged components include training feature maps extracted through application of a convolutional neural network to the data representative of the plurality of surface geometry zones for the plurality of training forge components,
wherein performing the geometrical analysis comprises applying the convolutional neural network to the data representative of the target forged component having the target surface geometry zone to obtain a target feature map representative of the target surface geometry zone of the target forged component, and
wherein applying the trained machine learning model comprises applying the trained machine learning model to the target feature map representative of the target surface geometry zone of the target forged component to obtain the predicted model results.
17 . The method of claim 11 , further comprising:
obtaining model data representative of a training forged component of the plurality of training forged components; and
applying the trained machine learning model to the model data representative of the training forged component of the plurality of forged components to determine whether the quenching process was performed according to the quench parameters or different quench parameters.
18 . The method of claim 11 , wherein obtaining the plurality of training model results further comprises:
obtaining a respective set of heat transfer coefficients as a function of location over the respective subset of surface geometry zones; and
applying a fitting model to each respective set of heat transfer coefficients as a function of location over the respective subset of surface geometry zones in order to obtain the respective set of fitting parameters, wherein the plurality of training model results include all of the sets of fitting parameters.
19 . The method of claim 18 ,
wherein applying the trained machine learning model includes:
applying the trained machine learning model to the target features to obtain predicted fitting parameters for the target forged component; and
applying the fitting model to the predicted fitting parameters to obtain the predicted model results as predicted heat transfer coefficients for the target forged component.
20 . A non-transitory computer-readable storage medium having instructions stored thereon that, when executed, cause one or more processors to:
extract training features from data representative of a plurality of surface geometry zones for a plurality of training forged components, wherein each training forged component from the plurality of training forged components is associated with data representative of a respective subset of surface geometry zones from the plurality of surface geometry zones;
for each training forged component:
obtain a plurality of computational fluid dynamics results for the respective subset of geometry zones by modeling application of a quench process to the respective training forged component according to quench parameters, wherein each geometry zone from the respective subset of geometry zones is associated with a respective subset of computational fluid dynamics results from the plurality of computational fluid dynamics results; and
determine, based on the plurality of computational fluid dynamics results, a respective model for identifying a respective set of fitting parameters;
obtain, based on each respective set of fitting parameters, a plurality of training model results representative of one or more mechanical properties;
store a trained machine learning model that associates the training features to the plurality of training model results;
store data representative of a target forged component having a target surface geometry zone that differs from all of the plurality of surface geometry zones for the plurality of training forged components;
perform a geometrical analysis with respect to the data representative of the target forged component to extract target features;
apply the trained machine learning model to the target features to obtain predicted model results for the target forged component, the predicted model results representative of an application of a quench process to the target forged component according to quench parameters; and
output the predicted model results.