Systems and methods for predicting part defects during additive manufacturing

A method of detecting weld defects in a build surface of an additive manufacturing system, the method comprising:

• obtaining an image of at least a portion of the build surface;
• subdividing at least a portion of the image corresponding to a location of a part in the build surface into a plurality of regions;
• and
• identifying the presence of weld defects in the build surface based at least in part on light intensities of the plurality of regions.

An additive manufacturing system comprising:

• a build surface;
• one or more laser energy sources;
• an optics assembly configured to direct laser energy from the one or more laser energy sources toward the build surface to melt at least a portion of a layer of material disposed on the build surface to form one or more welds;
• a photosensitive detector configured to image at least a portion of the build surface;
• and
• a processor configured to perform the steps of:
• providing information related to the presence of weld defects in one or more sequential layers of one or more parts to a trained weld-based defect prediction statistical model;
• and
• predicting formation of part defects in the one or more parts using the trained weld-based defect prediction statistical model.

A method for training a weld-based defect prediction statistical model, the method comprising:

• obtaining training data, wherein the training data includes weld defect data and part defect data associated with a plurality of sequential layers of a plurality of separate parts formed with an additive manufacturing system;
• generating a trained weld-based defect prediction statistical model using the training data;
• and
• storing the trained statistical model on non-transitory computer readable memory for subsequent use.

An additive manufacturing system comprising:

• a build surface;
• one or more laser energy sources;
• an optics assembly configured to direct laser energy from the one or more laser energy sources toward the build surface to melt at least a portion of a layer of material disposed on the build surface to form one or more welds;
• a photosensitive detector configured to image at least a portion of the build surface;
• and
• a processor configured to perform the steps of:
• obtaining one or more images of the build surface with the photosensitive detector after fusing the material in one or more sequential layers during formation of one or more parts;
• obtaining one or more process parameters associated with formation of the one or more sequential layers;
• providing the one or more images and the one or more process parameters to a trained multivariate defect prediction statistical model;
• and
• predicting the formation of part defects in the one or more parts using the trained multivariate defect prediction statistical model.

A method of predicting formation of part defects in one or more parts formed with an additive manufacturing system, the method comprising:

• obtaining one or more images of a build surface after fusing material in one or more sequential layers during formation of the one or more parts;
• obtaining one or more process parameters associated with formation of the one or more sequential layers of the one or more parts;
• providing the one or more images and the one or more process parameters to a trained multivariate defect prediction statistical model;
• and
• predicting the formation of part defects in the one or more parts using the trained multivariate defect prediction statistical model.

A method for training a multivariate defect prediction statistical model, the method comprising:

• obtaining training data, wherein the training data includes predicted part defect data determined based at least in part on weld quality for a plurality of parts, process parameter data associated with formation of a plurality of sequential layers of the plurality of parts, and images of a build surface of an additive manufacturing system after fusing material of the plurality of sequential layers for the plurality of parts,
• generating a trained multivariate statistical model using the training data;
• and
• storing the trained multivariate statistical model on non-transitory computer readable memory for subsequent use.