Model adaptation and online learning for unstable environments

Patent metadata

Field	Value
Publication number	US-11144842-B2
Application number	US-201615346707-A
Country	US
Kind code	B2
Filing date	Nov 8, 2016
Priority date	Jan 20, 2016
Publication date	Oct 12, 2021
Grant date	Oct 12, 2021

Abstract

Methods, systems, and apparatuses for adapting a predictive model for a manufacturing process. One method includes receiving, with an electronic processor, a plurality of data points for a plurality of manufactured parts and the predictive model. The predictive model outputs a label for a manufactured part provided by the manufacturing process indicating whether the manufactured part is accepted or rejected. The method also includes estimating, with the electronic processor, a drift for each of the plurality of data points and generating, with the electronic processor, an adapted version of the predictive model based on the predictive model and the drift for each of the plurality of data points. In addition, the method includes outputting, with the electronic processor, a label for each of the plurality of manufactured parts using the adapted version of the predictive model.

First claim

What is claimed is:

1. A method of adapting a predictive model for a manufacturing process, the method comprising:
receiving, with an electronic processor, the predictive model, the predictive model outputting a label for a manufactured part provided by the manufacturing process, wherein the label indicates whether the manufactured part is accepted or rejected, the predictive model trained using a first plurality of data points associated with a first plurality of manufactured parts and an actual label for each of the first plurality of manufactured parts;
receiving, with the electronic processor, a second plurality of data points associated with a second plurality of manufactured parts;
determining, with the predictive model, a first predicted label for each of the second plurality of manufactured part;
estimating, with the electronic processor, a drift for each of the second plurality of data points, wherein estimating the drift includes
training a temporary predictive model using the second plurality of data points and the first predicted label for each of the second plurality of manufactured parts,
determining, with the temporary predictive model, a second predicted label for each of the second plurality of manufactured parts, and
comparing the first predicted label and the second predicted label to estimate the drift for each of the second plurality of data points;

generating, with the electronic processor, an adapted version of the predictive model based on the predictive model and the drift for each of the second plurality of data points; and
outputting, with the electronic processor, a third predicted label for each of the second plurality of manufactured parts using the adapted version of the predictive model.

2. The method of claim 1 , wherein determining the first predicted label for each of the second plurality of manufactured parts includes estimating, with the electronic processor, the first predicted label for each of the second plurality of manufactured parts based on the second plurality of data points and the predictive model, wherein the first predicted label for each of second plurality of manufactured parts is indicative of whether each of the second plurality of manufactured parts are within specifications.

3. The method of claim 1 , further comprising:
comparing, with the electronic processor, the adapted version of the predictive model and the temporary predictive model to obtain a difference; and
when the difference is greater than a threshold, updating, with the electronic processor, the adapted version of the predictive model.

4. The method of claim 3 , wherein comparing the adapted version of the predictive model and the temporary predictive model to obtain the difference includes calculating a Kullback-Leibler Divergence between the adapted version of the predictive model and the temporary predictive model.

5. The method of claim 3 , wherein updating the adapted version of the predictive model includes
(a) calculating a gradient of the difference at each of the second plurality of data points;
(b) updating the drift at each of the second plurality of data points based on the gradient at each of the second plurality of data points;
(c) updating the adapted version of the predictive model based on the updated drift for each of the second plurality of data points to obtain an updated adapted version of the predictive model;
(d) generating an updated predicted label for each of the second plurality of manufactured parts based on the second plurality of data points and the updated adapted version of the predictive model;
(e) generating a new temporary predictive model based on the second plurality of data points and the updated predicted label for each of the second plurality of manufactured parts;
(f) comparing the updated adapted version of the predictive model and the new temporary predictive model to obtain an updated difference;
(g) when the updated difference is greater than the threshold, repeating (a) through (f); and
(h) when the updated difference is less than the threshold, applying the updated adapted version of the predictive model.

6. The method of claim 1 , wherein generating the adapted version of the predictive model based on the predictive model and the drift for each of the second plurality of data points includes
(a) setting the adapted version of the predictive model to the predictive model;
(b) generating a new temporary predictive model based on the second plurality of data points;
(c) comparing the adapted version of the predictive model and the new temporary predictive model to determine a difference; and
(d) when the difference is greater than a threshold, updating the drift for each of the second plurality of data points, updating the adapted version of the predictive model based on the updated drift for each of the second plurality of data points, and repeating (b) through (d) using the updated adapted version of the predictive model as the adapted version of the predictive model.

7. A system for adapting a predictive model for a manufacturing process, the system comprising:
a database storing at least one of a first plurality of data points associated with a first plurality of manufactured parts and a second plurality of data point associated with a second plurality of manufactured parts; and
a controller communicably coupled to the database, the controller including an electronic processor, the electronic processor configured to
receive the second plurality of data points;
receive the predictive model, the predictive model outputting a label for a manufactured part provided by the manufacturing process, wherein the label indicates whether the manufactured part is accepted or rejected, the predictive model trained using the first plurality of data points associated with the first plurality of manufactured parts and an actual label for each of the first plurality of manufactured parts;
determine, with the predictive model, a first predicted label for each of the second plurality of manufactured parts;
estimate a drift for each of the second plurality of data points, wherein the drift for each of the second plurality of data points is estimated by
training a temporary predictive model using the second plurality of data points and the first predicted label for each of the second plurality of manufactured parts,
determining, with the temporary predictive model, a second predicted label for each of the second plurality of manufactured parts, and
comparing the first predicted label and the second predicted label to estimate the drift for each of the second plurality of data points;

generate an adapted version of the predictive model based on the predictive model and the drift for each of the second plurality of data points; and
output a third predicted label for each of the second plurality of manufactured parts using the adapted version of the predictive model.

8. The system of claim 7 , wherein the first predicted label for each of the second plurality of manufactured parts indicates whether each of the second plurality of manufactured parts are within specifications.

9. The system of claim 7 , wherein the electronic processor is further configured to
compare the adapted version of the predictive model and the temporary predictive model to obtain a difference; and
when the difference is greater than a threshold, update the adapted version of the predictive model.

10. The system of claim 9 , wherein the electronic processor is configured to compare the adapted version of the predictive model and the temporary predictive model to obtain the difference by calculating a Kullback-Leibler Divergence between the adapted version o

Assignees

• Bosch Gmbh Robert

Inventors

• Das Subhro
• Lade Prasanth
• Srinivasan Soundar
• Ghosh Rumi

Classifications

• G05B13/048

  using a predictor · CPC title

• G05B23/0254

  based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G05B17/02

  electric · CPC title