Creating predictive damage models by transductive transfer learning

1. A method for creating predictive damage models, the method comprising:

receiving an existing first predictive damage model configured to generate at least one of maintenance and servicing recommendations for an existing asset of a first domain;

updating the first predictive damage model to a second predictive damage model, the second predictive damage model applicable to an asset of a second domain, and the second domain having a difference from the first domain;

updating the first predictive damage model including:

identifying latent space between the first domain asset and the second domain asset;

building a regression model from projected source data of the first domain asset;

creating target dependent variables of the second predictive damage model;

applying one of classification and regression techniques to determine a function expressing the target dependent variables;

determining a set of data points from the determined function to develop a second regression model;

applying the second regression model to the set of data points to predict target dependent variables for the second predictive damage model;

evaluating the second predictive damage model using the predicted target dependent variables;

performing a sensitivity study to determine whether a directionality parameter of the second predictive damage model is correct;

based on determining the results of the evaluating step are within an acceptable predetermined range, providing maintenance or servicing recommendations generated by the second predictive damage model to a user platform display; and

based on determining the results of the evaluating step are outside the acceptable predetermined range, repeating the process by rebuilding the regression model to further refine the regression model.

2. The method of claim 1 , the identifying latent space including optimizing a user-specified function by solving a semidefinite programming problem.

3. The method of claim 1 , including using the identified latent space to bridge from the first predictive damage model to a second predictive damage model.

4. The method of claim 1 , including applying principal component analysis to the identified latent space.

5. The method of claim 1 , the rebuilding the regression model including retuning the regression model by comparing residual root mean square errors against known labels of the second domain asset.

6. A non-transitory computer readable medium containing computer-readable instructions stored therein for causing a computer processor to create a predictive damage models comprising:

receiving an existing first predictive damage model configured to generate at least one of maintenance and servicing recommendations for an existing asset of a first domain;

updating the first predictive damage model to a second predictive damage model, the second predictive damage model applicable to an asset of a second domain, and the second domain having a difference from the first domain;

updating the first predictive damage model including:

identifying latent space between the first domain asset and the second domain asset;

building a regression model from projected source data of the first domain asset;

creating target dependent variables of the second predictive damage model;

applying one of classification and regression techniques to determine a function expressing the target dependent variables;

determining a set of data points from the determined function to develop a second regression model;

applying the second regression model to the set of data points to predict target dependent variables for the second predictive damage model;

evaluating the second predictive damage model using the predicted target dependent variables;

performing a sensitivity study to determine whether a directionality parameter of the second predictive damage model is correct; and

if the results of the evaluating step are within an acceptable predetermined range, providing maintenance or servicing recommendations generated by the second predictive damage model to a user platform display;

else repeating the process by rebuilding the regression model to further refine the regression model.

7. The non-transitory computer-readable medium of claim 6 , including instructions to cause the processor to perform the identifying latent space by optimizing a user-specified function by solving a semidefinite programming problem.

8. The non-transitory computer-readable medium of claim 6 , including instructions to cause the processor to perform a step of using the identified latent space to bridge from the first predictive damage model to a second predictive damage model.

9. The non-transitory computer-readable medium of claim 6 , including instructions to cause the processor to perform a step of applying principal component analysis to the identified latent space.

10. The non-transitory computer-readable medium of claim 6 , including instructions to cause the processor to include in rebuilding the regression model a step of retuning the regression model by comparing residual root mean square errors against known labels of the second domain asset.