Convolutional neural network evaluation of additive manufacturing images, and additive manufacturing system based thereon

A method of training an artificial intelligence module for closed loop control of an additive manufacturing machine operable to perform additive manufacturing processes to build parts, the method comprising:

• collecting a set of additive manufacturing build parameter configuration files corresponding to a plurality of parts;
• collecting sequential time-based parameter data in-process by the additive manufacturing machine;
• capturing build layer images in-process;
• generating build layer image classification data by a convolutional neural network configured to evaluate the captured build layer images;
• capturing images of a built part post-process;
• generating post-process image classification data by at least one other convolutional neural network configured to evaluate the captured built part images;
• inputting to the artificial intelligence module the set of additive manufacturing build parameter configuration files;
• inputting to the artificial intelligence module the sequential time-based parameter data;
• inputting to the artificial intelligence module the build layer image classification data;
• inputting to the artificial intelligence module the post-process image classification data;
• wherein the at least one other convolutional neural network includes a convolutional neural network configured to evaluate two-dimensional images of sectioned parts and another convolutional neural network configured to evaluate three-dimensional images of parts;
• and
• evaluating the additive manufacturing build parameter configuration files, the sequential time-based parameter data, the build layer image classification data, and the post-process image classification data by means of the artificial intelligence module;
• and
• wherein the artificial intelligence module is trained to be incorporated in a closed loop control structure to adjust the initial set of build parameters in-process.

An additive manufacturing system for building a part layer-by-layer according to an additive manufacturing build process, the additive manufacturing system comprising:

• an additive manufacturing machine including a powder bed and an energy source, wherein a beam of energy from the energy source is scanned relative to a layer of powder in the powder bed to build each layer of the part by fusion;
• a build parameter configuration file storing an initial set of build parameters for building the part in the additive manufacturing machine, wherein the initial set of build parameters is based at least in part on a geometric model of the part;
• a closed-loop control structure for adjusting the initial set of build parameters in-process, the closed loop control structure including a slow control loop having a trained artificial intelligence module and a fast control loop having a state machine;
• wherein the slow control loop has a controller gain update period on the order of whole seconds, and the fast control loop has a controller gain update period on the order of microseconds;
• a build layer image sensor arranged to acquire layer images of the part layers in-process;
• and
• wherein the initial set of build parameters, a time-based sequence of adjusted build parameters corresponding to the build process, and the layer images are transmitted as inputs to the trained artificial intelligence module, and
• wherein melt pool data are collected in-process by the additive manufacturing machine, inputted to the artificial intelligence module, and evaluated by means of the artificial intelligence module.

A method of training an artificial intelligence module for closed loop control of an additive manufacturing machine operable to perform additive manufacturing processes to build parts, the method comprising:

• collecting a set of additive manufacturing build parameter configuration files corresponding to a plurality of parts;
• collecting sequential time-based parameter data in-process by the additive manufacturing machine;
• capturing build layer images in-process;
• generating build layer image classification data by a convolutional neural network configured to evaluate the captured build layer images;
• capturing images of a built part post-process;
• generating post-process image classification data by at least one other convolutional neural network configured to evaluate the captured built part images;
• inputting to the artificial intelligence module the set of additive manufacturing build parameter configuration files;
• inputting to the artificial intelligence module the sequential time-based parameter data;
• inputting to the artificial intelligence module the build layer image classification data;
• inputting to the artificial intelligence module the post-process image classification data;
• evaluating the additive manufacturing build parameter configuration files, the sequential time-based parameter data, the build layer image classification data, and the post-process image classification data by means of the artificial intelligence module;
• wherein the artificial intelligence module is trained to be incorporated in a closed loop control structure to adjust the initial set of build parameters in-process, wherein the closed loop control structure includes a slow control loop having a trained artificial intelligence module and a fast control loop having a state machine;
• and
• collecting melt pool data in-process by the additive manufacturing machine;
• inputting to the artificial intelligence module the melt pool data;
• and
• evaluating the melt pool data by means of the artificial intelligence module;
• and
• wherein the slow control loop has a controller gain update period on the order of whole seconds, and the fast control loop has a controller gain update period on the order of microseconds.