Real-time adaptive control of additive manufacturing processes using machine learning

• US 10,234,848 B2
• Filed: 05/24/2017
• Issued: 03/19/2019
• Est. Priority Date: 05/24/2017
• Status: Active Grant

• Current Assignee: Relativity Space, Inc.
• Original Assignee: Relativity Space, Inc.
• Inventors: Mehr, Edward, Ellis, Tim, Noone, Jordan
• Primary Examiner(s): Karim, Ziaul
• Assistant Examiner(s): Kabir, Saad M
• Application Number:
  US15/604,473
• Publication Number:
  US 20180341248A1
• Time in Patent Office: 664 Days
• Field of Search: None
• CPC Class Codes: B22F 10/00 Additive manufacturing of workpieces or articles from metallic powder apparatus or devices therefor B22F12/00
• CPC Class Codes: B22F 10/12 by photopolymerisation, e.g. stereolithography [SLA] or digital light processing [DLP]
• CPC Class Codes: B22F 10/18 by mixing binder with metal in filament form, e.g. fused filament fabrication [FFF]
• CPC Class Codes: B22F 10/25 Direct deposition of metal particles, e.g. direct metal deposition [DMD] or laser engineered net shaping [LENS]
• CPC Class Codes: B22F 10/28 Powder bed fusion, e.g. selective laser melting [SLM] or electron beam melting [EBM]
• CPC Class Codes: B22F 10/38 to achieve specific product aspects, e.g. surface smoothness, density, porosity or hollow structures
• CPC Class Codes: B22F 10/80 Data acquisition or data processing
• CPC Class Codes: B22F 10/85 for controlling or regulating additive manufacturing processes
• CPC Class Codes: B22F 12/90 Means for process control, e.g. cameras or sensors
• CPC Class Codes: B33Y 50/02 for controlling or regulating additive manufacturing processes
• CPC Class Codes: G05B 13/048 using a predictor
• CPC Class Codes: G05B 19/4099 Surface or curve machining, making 3D objects, e.g. desktop manufacturing
• CPC Class Codes: G05B 2219/35134 3-D cad-cam
• CPC Class Codes: G05B 2219/45165 Laser machining
• CPC Class Codes: G05B 2219/49011 Machine 2-D slices, build 3-D model, laminated object manufacturing LOM
• CPC Class Codes: G05B 2219/49017 DTM desktop manufacturing, prototyping
• CPC Class Codes: G05B 2219/49018 Laser sintering of powder in layers, selective laser sintering SLS
• CPC Class Codes: G05B 2219/49023 3-D printing, layer of powder, add drops of binder in layer, new powder
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 3/045 : Combinations of networks
• CPC Class Codes: G06N 3/08 : Learning methods
• CPC Class Codes: G06N 3/084 : Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06N 5/01 : Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
• CPC Class Codes: G06N 7/01 : Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06N 7/02 : using fuzzy logic computing arrangements based on biological models G06N3/00; computing arrangements using knowledge-based models G06N5/00
• CPC Class Codes: Y02P 10/25 : Process efficiency

Abstract

Methods for control of post-design free form deposition processes or joining processes are described that utilize machine learning algorithms to improve fabrication outcomes. The machine learning algorithms use real-time object property data from one or more sensors as input, and are trained using training data sets that comprise: i) past process simulation data, past process characterization data, past in-process physical inspection data, or past post-build physical inspection data, for a plurality of objects that comprise at least one object that is different from the object to be fabricated; and ii) training data generated through a repetitive process of randomly choosing values for each of one or more input process control parameters and scoring adjustments to process control parameters as leading to either undesirable or desirable outcomes, the outcomes based respectively on the presence or absence of defects detected in a fabricated object arising from the process control parameter adjustments.

Claims

1. A method for real-time adaptive control of a post-design free form deposition process or a post-design joining process, the method comprising:

a) providing an input design geometry for an object;

b) providing a training data set, wherein the training data set comprises;

(i) past process simulation data, past process characterization data, past in-process physical inspection data, or past post-build physical inspection data, for a plurality of objects that comprise at least one object that is different from the object to be physically fabricated that is provided in step (a); and

(ii) training data generated through a repetitive process of randomly choosing values for each of one or more input process control parameters and scoring adjustments to the input process control parameters as leading to either undesirable or desirable outcomes, the outcomes based respectively on the presence or absence of defects detected in a fabricated object arising from the process control parameter adjustments;

c) providing one or more sensors, wherein the one or more sensors provide real-time data for one or more object properties as the object is being physically fabricated; and

d) providing a processor programmed to;

(i) predict an optimal set of one or more process control parameters for initiating the free form deposition process or joining process, wherein the predicted optimal set of one or more process control parameters are derived using a machine learning algorithm that has been trained using the training data set of step (b);

(ii) remove noise from the object property data provided by the one or more sensors prior to providing it to the machine learning algorithm, wherein the noise is removed using a signal averaging algorithm, Kalman filter algorithm, nonlinear filter algorithm, total variation minimization algorithm, or any combination thereof;

(iii) provide a real-time classification of detected object defects using the machine learning algorithm that has been trained using the training data set of step (b), wherein the real-time data from the one or more sensors is provided as input to the machine learning algorithm, and wherein the real-time classification of detected object defects is output from the machine learning algorithm; and

(iv) provide instructions to perform the post-design free form deposition process or post-design joining process to fabricate the object, wherein the machine learning algorithm adjusts the one or more process control parameters in real-time while physically performing the free form deposition process or the joining process.

2. The method of claim 1, wherein steps (b) through (d) are performed iteratively and process characterization data, in-process inspection data, or post-build inspection data for each iteration is incorporated into the training data set.

3. The method of claim 1, wherein the free form deposition process or joining process is a stereolithography (SLA), digital light processing (DLP), fused deposition modeling (FDM), selective laser sintering (SLS), selective laser melting (SLM), electronic beam melting (EBM), or welding process.

4. The method of claim 1, wherein the machine learning algorithm comprises an artificial neural network algorithm, a Gaussian process regression algorithm, a logistical model tree algorithm, a random forest algorithm, a fuzzy classifier algorithm, a decision tree algorithm, a hierarchical clustering algorithm, a k-means algorithm, a fuzzy clustering algorithm, a deep Boltzmann machine learning algorithm, a deep convolutional neural network algorithm, a deep recurrent neural network, or any combination thereof.

5. The method of claim 1, wherein the method is implemented using either:

(i) a single integrated system comprising a deposition or joining apparatus, a sensor, and a processor;

or (ii) a distributed, modular system comprising a first deposition or joining apparatus, a first sensor, and a first processor, wherein the first deposition or joining apparatus, the first sensor, and the first processor are configured to share training data and real-time process characterization data via a local area network (LAN), an intranet, an extranet, or an internet.

6. The method of claim 1, wherein the training data set further comprises process characterization data, in-process inspection data, or post-build inspection data that is generated by an operator while manually adjusting the input process control parameters.

13. The method of claim 1, wherein the one or more sensors comprise at least one laser interferometer, machine vision system, or sensor that detects electromagnetic radiation that is reflected, scattered, absorbed, transmitted, or emitted by the object.

14. The method of claim 1, wherein the one or more sensors provide data on acoustic energy or mechanical energy that is reflected, scattered, absorbed, transmitted, or emitted by the object.

15. The method of claim 1, wherein the object defects are detected as differences between object property data and a reference data set that are larger than a specified threshold, and are classified using a one-class support vector machine (SVM) or autoencoder algorithm.

16. The method of claim 1, wherein the object defects are detected and classified using an unsupervised one-class support vector machine (SVM), autoencoder, clustering, or nearest neighbor (kNN) machine learning algorithm and a training data set that comprises object property data for defective and defect-free objects.

7. A system for controlling a post-design free form deposition process or a post-design joining process, the system comprising:

a) a first deposition or joining apparatus for physically fabricating an object based on an input design geometry;

b) one or more process characterization sensors, wherein the one or more process characterization sensors provide real-time data for one or more process parameters or object properties; and

c) a processor programmed to;

(i) provide a predicted optimal set of one or more process control parameters for initiating the free form deposition process or joining process using a machine learning algorithm;

(ii) remove noise from the real-time object property data provided by the one or more process characterization sensors prior to providing it to the machine learning algorithm, wherein the noise is removed using a signal averaging algorithm, Kalman filter algorithm, nonlinear filter algorithm, total variation minimization algorithm, or any combination thereof;

(iii) provide a real-time classification of object defects using the machine learning algorithm, wherein the real-time object property data from the one or more process characterization sensors is provided as input to the machine learning algorithm, and wherein the real-time classification of detected object defects is output from the machine learning algorithm; and

(iv) provide instructions to perform the post-design free form deposition process or post-design joining process to fabricate the object, wherein the machine learning algorithm adjusts the one or more process control parameters in real-time while physically performing the free form deposition process or the joining process and wherein the machine learning algorithm has been trained using a training data set that comprises;

i) past process simulation data, past process characterization data, past in-process physical inspection data, or past post-build physical inspection data, for a plurality of objects that comprise at least one object that is different from the object to be physically fabricated that is provided in step (a); and

ii) training data generated through a repetitive process of randomly choosing values for each of one or more input process control parameters and scoring adjustments to the input process control parameters as leading to either undesirable or desirable outcomes, the outcomes based respectively on the presence or absence of defects detected in a fabricated object arising from the process control parameter adjustments.

8. The system of claim 7, wherein the first deposition or joining apparatus, the one or more process characterization sensors, and the processor are configured as:

(i) a single integrated system;

or (ii) as distributed system modules that share training data and real-time process characterization data via a local area network (LAN), an intranet, an extranet, or an internet.

9. The system of claim 7, wherein the one or more process characterization sensors comprise at least one laser interferometer, machine vision system, or sensor that detects electromagnetic radiation that is reflected, scattered, absorbed, transmitted, or emitted by the object.

10. The system of claim 7, wherein the one or more process characterization sensors provide data on acoustic energy or mechanical energy that is reflected, scattered, absorbed, transmitted, or emitted by the object.

11. The system of claim 7, wherein the object defects are detected as differences between object property data and a reference data set that are larger than a specified threshold, and are classified using a one-class support vector machine (SVM) or autoencoder algorithm.

12. The system of claim 7, wherein the object defects are detected and classified using an unsupervised one-class support vector machine (SVM), autoencoder, clustering, or nearest neighbor (kNN) machine learning algorithm and a training data set that comprises object property data for defective and defect-free objects.

17. The system of claim 7, wherein the first deposition or joining apparatus is a stereolithography (SLA) apparatus, digital light processing (DLP) apparatus, fused deposition modeling (FDM) apparatus, selective laser sintering (SLS) apparatus, selective laser melting (SLM) apparatus, electronic beam melting (EBM) apparatus, or welding apparatus.

18. The system of claim 7, wherein the machine learning algorithm comprises an artificial neural network algorithm, a Gaussian process regression algorithm, a logistical model tree algorithm, a random forest algorithm, a fuzzy classifier algorithm, a decision tree algorithm, a hierarchical clustering algorithm, a k-means algorithm, a fuzzy clustering algorithm, a deep Boltzmann machine learning algorithm, a deep convolutional neural network algorithm, a deep recurrent neural network, or any combination thereof.

19. The system of claim 7, wherein the training data set further comprises process characterization data, in-process inspection data, or post-build inspection data that is generated by an operator while manually adjusting the process control parameters.