Rapid closed-loop control based on machine learning

Patent metadata

Field	Value
Publication number	US-11079745-B2
Application number	US-201514952665-A
Country	US
Kind code	B2
Filing date	Nov 25, 2015
Priority date	Nov 25, 2015
Publication date	Aug 3, 2021
Grant date	Aug 3, 2021

Abstract

A system for rapidly adapting production of a product based on classification of production data using a classifier trained on prior production data is provided. A production control system includes a learning system and an adaptive system. The learning system trains a production classifier to label or classify previously collected production data. The adaptive system receives production data in real time and classifies the production data in real time using the production classifier. If the classification indicates a problem with the manufacturing of the product, the adaptive system controls the manufacturing to rectify the problem by taking some corrective action. The production classifier is trained using bootstrap data and corresponding example data extracted from prior production data. Once the bootstrap data is labeled, the corresponding example data is automatically labeled for use as training data.

First claim

The invention claimed is:

1. A method performed by a computing device of generating a classifier to label data of a target data type, the data representing an activity, the method comprising:
receiving bootstrap data and example data, the bootstrap data and the example data having a correlation in space or time, the bootstrap data and the example data representing the activity, the example data being of the target data type;
receiving from a user a labeling of the bootstrap data to generate labeled bootstrap data;
automatically labeling the example data based on the correlation to generate training data that includes the example data and labels; and
training the classifier based on the training data, the classifier for inputting data of the target data type and outputting a label.

2. The method of claim 1 wherein the activity is additive manufacturing.

3. The method of claim 1 wherein the activity is in a field that is selected from a group consisting of manufacturing, oceanography, meteorology, traffic engineering, container processing, medical imaging, autonomous vehicle operations, and satellite monitoring.

4. The method of claim 1 wherein the classifier is a neural network.

5. The method of claim 1 wherein the classifier is selected from a group consisting of a support vector machine, a decision tree, a k-nearest neighbor algorithm, a linear regression algorithm, a relevance vector machine, and ensembles of classifiers.

6. A method performed by a computing device of generating a classifier to label data of a target data type, the data representing an activity, the method comprising:
receiving bootstrap data and example data, the bootstrap data and the example data having a correlation in space or time, the bootstrap data and the example data representing the activity, the example data being of the target data type;
receiving from a user a labeling of the bootstrap data to generate labeled bootstrap data;
automatically labeling the example data based on the correlation to generate training data; and
training the classifier using a machine learning technique based on the training data
wherein the activity is a manufacturing process and the bootstrap data is snapshot data of the manufacturing process and the example data is video data of the manufacturing process.

7. The method of claim 6 further comprising applying the trained classifier to label newly collected video data during the manufacturing process and providing instructions to alter the manufacturing process in real time.

8. The method of claim 7 wherein a label indicates whether a characteristic of a product is acceptable or not acceptable.

9. The method of claim 8 wherein labels that indicate that the characteristic is not acceptable include a label indicating that manufacturing of the product should be aborted and a label indicating that the characteristic is correctable during the manufacturing process.

10. A method performed by a computing device of generating a neural network classifier to label data of a target data type, the data representing an activity, the method comprising:
receiving bootstrap data and example data, the bootstrap data and the example data having a correlation in space or time, the bootstrap data and the example data representing the activity, the example data being of the target data type;
receiving from a user a labeling of the bootstrap data to generate labeled bootstrap data;
automatically labeling the example data based on the correlation to generate training data; and
training the neural network classifier based on the training data
wherein the activity is a manufacturing process and the bootstrap data is from a first stream of sensor data of the manufacturing process and the example data is from a second stream of sensor data of the manufacturing process.

11. The method of claim 10 further comprising applying the trained neural network classifier to label newly collected example data during the manufacturing process and providing instructions to alter the manufacturing process in real time.

12. The method of claim 11 wherein a label indicates whether a characteristic of a product is acceptable or not acceptable.

13. The method of claim 12 wherein labels that indicate that the characteristic is not acceptable include a label indicating that manufacturing of the product should be aborted and a label indicating that the characteristic is correctable during the manufacturing process.

14. A method performed by a computing device of generating a classifier to label data of a target data type, the method comprising:
receiving seed data having a seed data type;
receiving from a user a labeling of the seed data to generate labeled seed data;
training a bootstrap classifier to label data of the seed data type using the labeled seed data;
receiving bootstrap data and example data, the bootstrap data and example data representing different data streams, the bootstrap data being of the seed data type and the example data being of the target data type;
applying the bootstrap classifier to label the bootstrap data
automatically labeling the example data based on the labeled bootstrap data to generate training data; and
training a production classifier to label data of the target data type using the training data.

15. The method of claim 14 wherein the seed data, the bootstrap data, and the example data are collected during different occurrences of same activity.

16. The method of claim 15 further comprising applying the production classifier to label newly collected video data during a manufacturing process and providing instructions to alter the manufacturing process in real time.

17. The method of claim 16 wherein a label indicates whether a characteristic of a product is acceptable or not acceptable.

18. The method of claim 17 wherein labels that indicate that the characteristic is not acceptable include a label indicating that manufacturing of the product should be abandoned and a label indicating that the characteristic is correctable during the manufacturing process.

19. The method of claim 14 wherein the bootstrap data and the corresponding example data are collected during the same occurrence of an activity.

20. The method of claim 14 wherein the bootstrap data is from a first stream of sensor data of the activity and the example data is from a second stream of sensor data of the activity.

21. The method of claim 14 wherein the seed data, the bootstrap data, and the example data are collected during additive manufacturing.

22. The method of claim 14 wherein the received data relates to a field that is selected from a group consisting of manufacturing, oceanography, meteorology, traffic engineering, container processing, medical imaging, autonomous vehicle operations, and satellite monitoring.

23. The method of claim 14 wherein the classifier is a neural network.

24. A method performed by a computing device of generating a classifier to label data of a target data type, the method comprising:
receiving seed data having a seed data type;
receiving from a user a labeling of the seed data to generate labeled seed data;
training a bootstrap classifier to label data of the seed data type by applying a machine leaning technique with the labeled seed data as input;
receiving bootstrap data and example data, the bootstrap data and example data representing different data streams, the bootstrap data being of the seed

Assignees

• L Livermore Nat Security Llc

Inventors

• Giera Brian

Classifications

• B22F12/90

  Means for process control, e.g. cameras or sensors · CPC title

• B22F10/85

  for controlling or regulating additive manufacturing processes · CPC title

• B22F10/38

  to achieve specific product aspects, e.g. surface smoothness, density, porosity or hollow structures · CPC title

• B22F10/28

  Powder bed fusion, e.g. selective laser melting [SLM] or electron beam melting [EBM] · CPC title

• B29C64/393

  for controlling or regulating additive manufacturing processes · CPC title