Methods and systems for training a machine learning model with measurement data captured during manufacturing process

Patent metadata

Field	Value
Publication number	US-12462159-B2
Application number	US-202217842041-A
Country	US
Kind code	B2
Filing date	Jun 16, 2022
Priority date	Jun 16, 2022
Publication date	Nov 4, 2025
Grant date	Nov 4, 2025

Abstract

Methods and systems for training a machine learning model with measurement data captured during a manufacturing process. Measurement data regarding a physical characteristic of a plurality of manufactured parts is received as measured by a plurality of sensors at various manufacturing stations. A time-series dynamics machine learning model encodes the measurement data into a latent space having a plurality of nodes. Each node is associated with the measurement data of one of the manufactured parts and at one of the manufacturing stations. A batch of the measurement data can be built, the batch include a first node and a first plurality of nodes immediately connected to the first node via first edges, and measured in time earlier than the first node. A prediction machine learning model can predict measurements of a first of the manufactured parts based on the latent space of the batch of nodes.

First claim

What is claimed is:

1 . A computer-implemented method of training a machine learning model with measurement data captured during a manufacturing process, the method comprising:
receiving measurement data regarding a physical characteristic of a plurality of manufactured parts as measured by a plurality of sensors at various manufacturing stations; via a time-series dynamics machine learning model, encoding the measurement data into a latent space having a plurality of nodes, each node associated with the measurement data of one of the manufactured parts as measured at one of the manufacturing stations; via a prediction machine learning model, determining a predicted measurement of a first of the manufactured parts at a first of the manufacturing stations based on the latent space of at least some of the measurement data not including the measurement data corresponding to the first manufactured part at the first manufacturing station; via the machine learning model, comparing the prediction measurement of the first manufactured part to the measurement data of the first manufactured part at the first manufacturing station; based on a difference between the prediction measurements and the actual measurement data, updating parameters of the machine learning model until convergence; and based upon the convergence, outputting a trained machine learning model with the updated parameters.

2 . The computer-implemented method of claim 1 , further comprising:
batching the measurement data for processing by the prediction machine learning model, wherein the batching includes building a batch of nodes that are immediately connected to a first node via an edge and timestamped prior to the first node.

3 . The computer-implemented method of claim 2 , wherein the step of predicting is performed on the batch of nodes.

4 . The computer-implemented method of claim 2 , wherein the batching further includes selecting additional nodes for the batch, wherein the additional nodes include all nodes associated with the first manufactured part timestamped prior to the first node.

5 . The computer-implemented method of claim 4 , wherein the additional nodes do not include all nodes associated with the first manufacturing station timestamped prior to the first node.

6 . The computer-implemented method of claim 1 , wherein the measurement data is multimodal measurement data, and the predicted measurement is a multimodal predicted measurement.

7 . The computer-implemented method of claim 1 , further comprising:
utilizing an embedding neural network to embed the measurement data into an array configured for execution by the time-series dynamics machine learning model.

8 . A system of training a machine learning model with measurement data captured during a manufacturing process, the system comprising:
a plurality of sensors located at a plurality of manufacturing stations, each sensor configured to produce measurement data indicating a physical characteristic of a plurality of manufactured parts passing through a respective one of the manufacturing stations, at least one processor programmed to: execute a time-series dynamics machine learning model to encode the measurement data into a latent space having a plurality of nodes, each node associated with the measurement data of one of the manufactured parts as measured at one of the manufacturing stations, execute a prediction machine learning model to determine a predicted measurement of a first of the manufactured parts at a first of the manufacturing stations based on the latent space of at least some of the measurement data not including the measurement data corresponding to the first manufactured part at the first manufacturing station, compare the prediction measurement of the first manufactured part to the measurement data of the first manufactured part measured at the first manufacturing station, based on a difference between the prediction measurements and the actual measurement data, updating parameters of the machine learning model until convergence, and based upon the convergence, output a trained machine learning model with the updated parameters.

9 . The system of claim 8 , wherein the at least one processor is further programmed to:
build a batch of the measurement data for processing by the prediction machine learning model, wherein the batch includes selected nodes that are immediately connected to a first node via an edge and timestamped prior to the first node.

10 . The system of claim 9 , wherein the at least one processor is further programmed to execute the predicting machine learning model with the batch of nodes.

11 . The system of claim 9 , wherein the building of the batch further includes selecting additional nodes for the batch, wherein the additional nodes include all nodes associated with the first manufactured part timestamped prior to the first node.

12 . The system of claim 11 , wherein the additional nodes do not include all nodes associated with the first manufacturing station timestamped prior to the first node.

13 . The system of claim 8 , wherein the measurement data is multimodal measurement data, and the predicted measurement is a multimodal predicted measurement.

14 . The system of claim 8 , wherein the at least one processor is further programmed to:
execute an embedding neural network to embed the measurement data into an array configured for execution by the time-series dynamics machine learning model.

15 . A computer-implemented method of training a machine learning model with measurement data captured during a manufacturing process, the method comprising:
receiving measurement data regarding a physical characteristic of a plurality of manufactured parts as measured by a plurality of sensors at various manufacturing stations; via a time-series dynamics machine learning model, encoding the measurement data into a latent space having a plurality of nodes, each node associated with the measurement data of one of the manufactured parts as measured at one of the manufacturing stations; batching the measurement data to build a batch including a first plurality of nodes that are immediately connected to a first node via first edges and measured in time earlier than the first node, and a second plurality of nodes wherein each of the second plurality of nodes are immediately connected to, and measured in time earlier than, a respective one of the first plurality of nodes via second edges; and via a prediction machine learning model, determining a predicted measurement of a first of the manufactured parts at a first of the manufacturing stations based on the latent space of the batch of nodes.

16 . The computer-implemented method of claim 15 , further comprising:
via the machine learning model, comparing the prediction measurement of the first manufactured part to the measurement data of the first manufactured part at the first manufacturing station; based on a difference between the prediction measurements and the actual measurement data, updating parameters of the machine learning model until convergence; and based upon the convergence, outputting a trained machine learning model.

17 . The computer-implemented method of claim 16 , wherein the batching further includes selecting additional nodes for the batch, wherein the additional nodes include all nodes associated with the first manufactured part measured in time prior to the first node.

18 . The computer-implemented method of claim 17 , wherein the additional nodes do not include all nodes a

Assignees

• Bosch Gmbh Robert

Inventors

• Cabrita Condessa Filipe J
• Willmott Devin T
• Batalov Ivan
• Semedo João D
• Lin Wan-Yi
• Kolter Jeremy
• Thompson Jeffrey

Classifications

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G05B2219/32339

  Object oriented modeling, design, analysis, implementation, simulation language · CPC title