Method and apparatus for additive manufacturing of a workpiece using statistical learning model

1 . A method for additive manufacture of a workpiece, the method comprising:

obtaining a dataset that defines the workpiece in a plurality of workpiece layers arranged one on top of another;

producing a respective material layer having a defined material surface formed from a particulate material;

recording at least one image of the respective material layer;

feeding the at least one image to a previously trained statistical learning model in order to determine at least one defect vector representing a plurality of individual defect probabilities, wherein each individual defect probability from the plurality of individual defect probabilities represents an individual indicator as to whether or not a defined layer defect from a plurality of possible layer defects is present in the respective material layer;

in response to the plurality of individual defect probabilities indicating that no defined layer defect from the plurality of possible layer defects is present in the respective material layer, selectively solidifying the particulate material on the defined surface using a structuring tool based on the dataset in order to thereby produce a defined workpiece layer from the plurality of workpiece layers;

in response to the plurality of individual defect probabilities indicating that at least one defined layer defect from the plurality of possible layer defects is present in the respective material layer, reworking or reproducing the respective material layer, and repeating the recording and the feeding in order to determine the at least one defect vector again; and

repeating the producing, the recording, the feeding, and the selectively solidifying such that further defined workpiece layers from the plurality of workpiece layers are produced one on top of another, wherein the respective material layers each are inspected using the previously trained statistical learning model.

2 . The method of claim 1 wherein:

the respective material layer is illuminated from a plurality of mutually different directions;

a plurality of images of the respective material layer is recorded;

each image from the plurality of images shows the respective material layer with a different illumination direction; and

the at least one defect vector is determined using the plurality of images.

3 . The method of claim 2 wherein the plurality of images together is fed to the previously trained statistical learning model as input data.

4 . The method of claim 1 wherein an elevation map of the respective material layer is determined using the at least one image, and wherein the elevation map is fed to the previously trained statistical learning model as input data.

5 . The method of claim 1 wherein:

the at least one defect vector includes a plurality of defect vectors;

each defect vector from the plurality of defect vectors represents a plurality of individual defect probabilities; and

each defect vector from the plurality of defect vectors represents individual defect probabilities in relation to a selected pixel region in the at least one image.

6 . The method of claim 5 wherein each defect vector from the plurality of defect vectors represents the individual defect probabilities in relation to another individual pixel in the at least one image.

7 . The method of claim 5 wherein at least one of morphological or dimensional properties of a defined defect in the material layer are determined using the plurality of defect vectors.

8 . The method of claim 1 wherein the previously trained statistical learning model comprises a convolutional neural network.

9 . The method of claim 8 wherein the convolutional neural network has an encoder/decoder architecture.

10 . The method of claim 1 wherein:

the at least one image of the respective material layer is normalized based on a reference image; and

the reference image shows a homogeneous, diffusely reflective surface or a defect-free material layer.

11 . The method of claim 1 wherein:

the at least one defect vector is stored, together with a timestamp that identifies the respective material layer, as a historical defect vector, and

the further defined workpiece layers are produced based on the historical defect vector.

12 . The method of claim 1 wherein:

the at least one image of the respective material layer is stored, together with a timestamp that identifies the respective material layer, as at least one historical image; and

the further defined workpiece layers are produced based on the at least one historical image.

13 . The method of claim 1 wherein:

individual properties of the respective material layer are stored together with a respective timestamp that identifies the respective material layer, and

the workpiece is authorized for use based on the individual properties of each respective material layer.

14 . A method for training a statistical learning model, the method comprising:

obtaining a plurality of training images that show a plurality of particulate material layers, wherein the plurality of particulate material layers shown in the plurality of training images contain a plurality of possible layer defects;

marking the plurality of possible layer defects in the plurality of training images in order to obtain a plurality of annotated versions of the plurality of training images;

defining a stop criterion based on the plurality of annotated versions of the plurality of training images;

repeatedly feeding the plurality of training images to the statistical learning model as input data and repeatedly obtaining output data from the statistical learning model in response to the plurality of training images, wherein:

the output data includes at least one defect vector representing a plurality of individual defect probabilities, and

each individual defect probability from the plurality of individual defect probabilities represents an individual indicator as to whether or not a defined layer defect from the plurality of possible layer defects is present in a respective particulate material layer; and

iteratively optimizing the statistical learning model using the plurality of training images and the stop criterion.

15 . The method of claim 14 wherein the plurality of training images are recorded while manufacturing of a defined workpiece.

16 . The method of claim 15 wherein the stop criterion is defined based on the defined workpiece.

17 . The method of claim 15 wherein the stop criterion is at least one of a defect criterion or a time criterion.

18 . A method for additive manufacture of a workpiece, the method comprising:

obtaining a dataset that defines the workpiece in a plurality of workpiece layers arranged one on top of another,

producing a material layer having a defined surface formed from a particulate material,

recording at least one image of the material layer and inspecting the material layer based on the at least one image in order to determine individual properties of the material layer,

selectively solidifying the particulate material on the defined surface using a structuring tool based on the dataset and based on the individual properties of the material layer, such that a defined workpiece layer from the plurality of workpiece layers is produced from the material layer, and

repeating the producing, the recording, and the selectively solidifying such that further defined workpiece layers from the plurality of workpiece layers are produced, wherein:

the at least one image of the material layer is inspected using a previously trained statistical learning model,

the previously trained statistical learning model is used to determine at least one defect vector that represents a plurality of individual defect probabilities,

each individual defect probability from the plurality of individual defect probabilities is an individual indicator as to whether or not a defined layer defect from a plurality of possible layer defects is present in the material layer, and

the selectively solidifying is carried out based on the at least one defect vector.

19 . An apparatus for additive manufacture of a workpiece, the apparatus comprising:

a memory configured to store a dataset defining the workpiece in a plurality of workpiece layers arranged one on top of another;

a production platform;

a layer-forming tool;

a structuring tool;

a camera oriented toward the production platform; and

an evaluation and control unit configured to:

produce a material layer having a defined surface from a particulate material on the production platform using the layer-forming tool,

record at least one image of the material layer using the camera,

feed the at least one image to a previously trained statistical learning model in order to determine at least one defect vector representing a plurality of individual defect probabilities, wherein each individual defect probability from the plurality of individual defect probabilities represents an individual indicator as to whether or not a defined layer defect from a plurality of possible layer defects is present in the respective material layer,

in response to the plurality of individual defect probabilities indicating that no defined layer defect from the plurality of possible layer defects is present in the respective material layer, selectively solidify the particulate material on the defined surface using the structuring tool such that a defined workpiece layer from the plurality of workpiece layers is produced from the material layer, and

in response to the plurality of individual defect probabilities indicating that at least one defined layer defect from the plurality of possible layer defects is present in the respective material layer, rework or reproduce the respective material layer.