Predictive design space metrics for materials development

Abstract

A system and a method are disclosed for predicting design space quality for materials development and manufacture. In an embodiment, a processor receives input of a material property and a design space. The processor identifies a best data point. For each respective candidate material of the design space, the processor receives, as output from a model, a respective property value. The processor determines respective property values that exceed the property value of the best data point adds them to a subset of candidate materials. The processor determines a PFIC score for candidates in the subset. The processor generates a plurality of curves, each reflecting a respective probability distribution of property values. The processor determines a CMLI score based on the plurality of respective curves. The processor determines that the design space is high quality based on the PFIC and CMLI scores, and outputs a recommendation to proceed.

What is claimed is: 
     
         1 . A method for determining material design space quality, the method comprising:
 receiving input, from a user, of a specified material property and a design space, the design space comprising a plurality of candidate materials;   identifying a best data point, from a set of training data that was used to train a first model, having a highest property value of the set for the specified material property;   for each respective candidate material of the design space:
 inputting the respective candidate material into the first model; 
 receiving, as output from the first model, a respective property value for the specified material property; 
 determining whether the respective property value exceeds the property value of the best data point; 
 responsive to determining that the respective property value exceeds the property value of the data point, adding the respective candidate material to a subset of candidate materials; 
   determining a Predicted Fraction of Improved Candidates (PFIC) score by dividing a total number of candidate materials in the subset of candidate materials by a total number of candidate materials in the design space;   generating, using a second model, a plurality of curves, each respective curve of the plurality of curves reflecting a respective probability distribution of property values for the specified material property that a respective candidate material will yield;   determining a Cumulative Maximum Likelihood of Improvement (CMLI) score that indicates the probability that the design space comprises at least one candidate material with a property value that exceeds the property value of the best data point based on a subset of the plurality of respective curves;   determining whether the design space is of high quality based on both the PFIC score exceeding a PFIC threshold score, and the CMLI score exceeding a CMLI threshold score; and   responsive to determining that the design space is of high quality, outputting, to the user, a recommendation to proceed with the design space.   
     
     
         2 . The method of  claim 1 , wherein determining the CMLI score further comprises:
 receiving user input of an amount n of the plurality of candidate materials from which the CMLI score is to be derived; and   limiting the subset of the plurality of respective curves to include n curves of the plurality of curves that each reflect a highest likelihood, relative to others of the plurality of curves, that their respective candidate materials have a property value that exceeds the property value of the best data point.   
     
     
         3 . The method of  claim 2 , wherein outputting, to the user, a recommendation to proceed with the design space comprises outputting a ranked list of the candidate materials in order of their likelihood of having a property value that exceeds the property value of the best data point. 
     
     
         4 . The method of  claim 3 , further comprising limiting the ranked list to n candidate materials. 
     
     
         5 . The method of  claim 1 , wherein determining the CMLI score further comprises:
 limiting the subset of the plurality of respective curves to include a predefined percentage of the plurality of curves, each curve of the predefined percentage reflecting a highest likelihood, relative to others of the plurality of curves, that their respective candidate materials have a property value that exceeds the property value of the best data point.   
     
     
         6 . The method of  claim 1 , further comprising:
 determining whether the design space is of low quality based on both the PFIC score being below the PFIC threshold score, and the CMLI score being below the CMLI threshold score; and   responsive to determining that the design space is of low quality, outputting, to the user, a recommendation not to proceed with the design space.   
     
     
         7 . The method of  claim 1 , further comprising:
 determining whether the design space is of unknown quality based on either the PFIC score being above the PFIC threshold score and the CMLI score being below the CMLI threshold score, or the PFIC score being below the PFIC threshold score and the CMLI score exceeding the CMLI threshold score; and   responsive to determining that the design space is of unknown quality, outputting, to the user, a recommendation indicating that the design space is of unknown quality.   
     
     
         8 . A non-transitory computer-readable medium comprising memory with instructions encoded thereon for determining material design space quality, the instructions, when executed, causing a processor to execute operations, the instructions comprising instructions to:
 receive input, from a user, of a specified material property and a design space, the design space comprising a plurality of candidate materials;   identify a best data point, from a set of training data that was used to train a first model, having a highest property value of the set for the specified material property;   for each respective candidate material of the design space:
 input the respective candidate material into the first model; 
 receive, as output from the first model, a respective property value for the specified material property; 
 determine whether the respective property value exceeds the property value of the best data point; 
 responsive to determining that the respective property value exceeds the property value of the data point, add the respective candidate material to a subset of candidate materials; 
   determine a Predicted Fraction of Improved Candidates (PFIC) score by dividing a total number of candidate materials in the subset of candidate materials by a total number of candidate materials in the design space;   generate, using a second model, a plurality of curves, each respective curve of the plurality of curves reflecting a respective probability distribution of property values for the specified material property that a respective candidate material will yield;   determine a Cumulative Maximum Likelihood of Improvement (CMLI) score that indicates the probability that the design space comprises at least one candidate material with a property value that exceeds the property value of the best data point based on a subset of the plurality of respective curves;   determine whether the design space is of high quality based on both the PFIC score exceeding a PFIC threshold score, and the CMLI score exceeding a CMLI threshold score; and   responsive to determining that the design space is of high quality, output, to the user, a recommendation to proceed with the design space.   
     
     
         9 . The non-transitory computer-readable medium of  claim 8 , wherein the instructions to determine the CMLI score further comprise instructions to:
 receive user input of an amount n of the plurality of candidate materials from which the CMLI score is to be derived; and   limit the subset of the plurality of respective curves to include n curves of the plurality of curves that each reflect a highest likelihood, relative to others of the plurality of curves, that their respective candidate materials have a property value that exceeds the property value of the best data point.   
     
     
         10 . The non-transitory computer-readable medium of  claim 9 , wherein the instructions to output, to the user, a recommendation to proceed with the design space comprise instructions to output a ranked list of the candidate materials in order of their likelihood of having a property value that exceeds the property value of the best data point. 
     
     
         11 . The non-transitory computer-readable medium of  claim 10 , further comprising limiting the ranked list to n candidate materials. 
     
     
         12 . The non-transitory computer-readable medium of  claim 8 , wherein the instructions to determine the CMLI score further comprise instructions to:
 limit the subset of the plurality of respective curves to include a predefined percentage of the plurality of curves, each curve of the predefined percentage reflecting a highest likelihood, relative to others of the plurality of curves, that their respective candidate materials have a property value that exceeds the property value of the best data point.   
     
     
         13 . The non-transitory computer-readable medium of  claim 8 , the instructions further comprising instructions to:
 determine whether the design space is of low quality based on both the PFIC score being below the PFIC threshold score, and the CMLI score being below the CMLI threshold score; and   responsive to determining that the design space is of low quality, output, to the user, a recommendation not to proceed with the design space.   
     
     
         14 . The non-transitory computer-readable medium of  claim 8 , the instructions further comprising instructions to:
 determine whether the design space is of unknown quality based on either the PFIC score being above the PFIC threshold score and the CMLI score being below the CMLI threshold score, or the PFIC score being below the PFIC threshold score and the CMLI score exceeding the CMLI threshold score; and   responsive to determining that the design space is of unknown quality, output, to the user, a recommendation indicating that the design space is of unknown quality.   
     
     
         15 . A system for determining material design space quality, the system comprising:
 a processor; and   a non-transitory computer-readable medium comprising memory with instructions encoded thereon, the processor configured, when executing the instructions, to:   receive input, by a user, of a specified material property and a design space, the design space comprising a plurality of candidate materials;   identify a best data point, from a set of training data that was used to train a first model, having a highest property value of the set for the specified material property;   for each respective candidate material of the design space:
 input the respective candidate material into the first model; 
 receive, as output from the first model, a respective property value for the specified material property; 
 determine whether the respective property value exceeds the property value of the best data point; 
 responsive to determining that the respective property value exceeds the property value of the data point, add the respective candidate material to a subset of candidate materials; 
   determine a Predicted Fraction of Improved Candidates (PFIC) score by dividing a total number of candidate materials in the subset of candidate materials by a total number of candidate materials in the design space;   generate, using a second model, a plurality of curves, each respective curve of the plurality of curves reflecting a respective probability distribution of property values for the specified material property that a respective candidate material will yield;   determine a Cumulative Maximum Likelihood of Improvement (CMLI) score that indicates the probability that the design space comprises at least one candidate material with a property value that exceeds the property value of the best data point based on a subset of the plurality of respective curves;   determine whether the design space is of high quality based on both the PFIC score exceeding a PFIC threshold score, and the CMLI score exceeding a CMLI threshold score; and   responsive to determining that the design space is of high quality, output, to the user, a recommendation to proceed with the design space.   
     
     
         16 . The system of  claim 15 , wherein the processor is further configured, when determining the CMLI score, to:
 receive user input of an amount n of the plurality of candidate materials from which the CMLI score is to be derived; and   limit the subset of the plurality of respective curves to include n curves of the plurality of curves that each reflect a highest likelihood, relative to others of the plurality of curves, that their respective candidate materials have a property value that exceeds the property value of the best data point.   
     
     
         17 . The system of  claim 16 , wherein the processor is further configured, when outputting, to the user, a recommendation to proceed with the design space, to output a ranked list of the candidate materials in order of their likelihood of having a property value that exceeds the property value of the best data point. 
     
     
         18 . The system of  claim 17 , wherein the processor is further configured to limit the ranked list to n candidate materials. 
     
     
         19 . The system of  claim 15 , wherein the processor is further configured, when determining the CMLI score, to:
 limit the subset of the plurality of respective curves to include a predefined percentage of the plurality of curves, each curve of the predefined percentage reflecting a highest likelihood, relative to others of the plurality of curves, that their respective candidate materials have a property value that exceeds the property value of the best data point.   
     
     
         20 . The system of  claim 15 , wherein the processor is further configured to:
 determine whether the design space is of low quality based on both the PFIC score being below the PFIC threshold score, and the CMLI score being below the CMLI threshold score; and   responsive to determining that the design space is of low quality, output, to the user, a recommendation not to proceed with the design space.