Manufacturing process control with deep learning-based predictive model for hot metal temperature of blast furnace

Patent metadata

Field	Value
Publication number	US-2020172989-A1
Application number	US-202016787670-A
Country	US
Kind code	A1
Filing date	Feb 11, 2020
Priority date	Sep 27, 2017
Publication date	Jun 4, 2020
Grant date	—

Abstract

A blast furnace control system may include a hardware processor that generates a deep learning based predictive model for forecasting hot metal temperature, where the actual measured HMT data is only available sparsely, and for example, measured at irregular interval of time. HMT data points may be imputed by interpolating the HMT measurement data. HMT gradients are computed and a model is generated to learn a relationship between state variables and the HTM gradients. HMT may be forecasted for a time point, in which no measured HMT data is available. The forecasted HMT may be transmitted to a controller coupled to a blast furnace, to trigger a control action to control a manufacturing process occurring in the blast furnace.

First claim

What is claimed is:

1 . A method of controlling a manufacturing process in a blast furnace, the method executed by at least one hardware processor, the method comprising:
receiving manufacturing process data associated with a blast furnace, the manufacturing process data including at least state variables and control variables used in operating the blast furnace, the state variables including at least a hot metal temperature (HMT) and other state variables, wherein the manufacturing process data includes at least a plurality of measured HMT at different time points, of a product produced in the blast furnace; generating imputed HMT by interpolating the measured HMT; generating HMT gradients based on at least the imputed HMT; defining a causal relationship between at least one of the other state variables and the HMT gradients, the relationship generated as a neural network model; training the neural network model using as training data, a weighted combination of the imputed HMT up to a last known measured HMT and predicted HMT up to the last known measured HMT, the trained neural network model trained to predict a current point in time value for the HMT, in which no measured HMT for the current point in time is available, wherein the trained neural network model predicts the HMT corresponding to a time period starting from the time of the last measure HMT data point for a number of time periods until the number of time periods advances to the current point in time and uses the predicted HMT corresponding to each of the number of time periods to predict the current point in time value for the HMT.

2 . The method of claim 1 , further including:
transmitting the current point in time value for the HMT to a controller to trigger a control action to control a manufacturing process occurring in the blast furnace, the control action including at least selectively opening a conduit to the blast furnace to control content amount of input to the blast furnace.

3 . The method of claim 1 , wherein the neural network model comprises at least a long short-term memory network.

4 . The method of claim 1 , wherein the manufacturing process data is stored as a time series data.

5 . The method of claim 1 , further comprising autonomously retraining the neural network model responsive to receiving a new measured HMT, using the new measured HMT as the last known measured HMT.

6 . The method of claim 1 , wherein the manufacturing process includes at least a continuous blast furnace operation.

7 . The method of claim 1 , wherein the plurality of measured HMT at different time points includes at least a plurality of measured HMT measured at irregular time intervals.

8 . A computer program product for controlling a manufacturing process in a blast furnace, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, wherein the computer readable storage medium is not a transitory signal per se, the program instructions readable by a device to cause the device to perform a method comprising:
receiving manufacturing process data associated with a blast furnace, the manufacturing process data including at least state variables and control variables used in operating the blast furnace, the state variables including at least a hot metal temperature (HMT) and other state variables, wherein the manufacturing process data includes at least a plurality of measured HMT at different time points, of a product produced in the blast furnace; generating imputed HMT by interpolating the measured HMT; generating HMT gradients based on at least the imputed HMT; defining a causal relationship between at least one of the other state variables and the HMT gradients, the relationship generated as a neural network model; training the neural network model using as training data, a weighted combination of the imputed HMT up to a last known measured HMT and predicted HMT up to the last known measured HMT, the trained neural network model trained to predict a current point in time value for the HMT, in which no measured HMT for the current point in time is available, wherein the trained neural network model predicts the HMT corresponding to a time period starting from the time of the last measure HMT data point for a number of time periods until the number of time periods advances to the current point in time and uses the predicted HMT corresponding to each of the number of time periods to predict the current point in time value for the HMT.

9 . The computer program product of claim 8 , wherein the device is caused to further perform:
transmitting the current point in time value for the HMT to a controller to trigger a control action to control a manufacturing process occurring in the blast furnace, the control action including at least selectively opening a conduit to the blast furnace to control content amount of input to the blast furnace.

10 . The computer program product of claim 8 , wherein the neural network model comprises at least a long short-term memory network.

11 . The computer program product of claim 8 , wherein the manufacturing process data is stored as a time series data.

12 . The computer program product of claim 8 , wherein the device is caused to further perform:
autonomously retraining the neural network model responsive to receiving a new measured HMT, using the new measured HMT as the last known measured HMT.

13 . The computer program product of claim 8 , wherein the manufacturing process includes at least a continuous blast furnace operation.

14 . The computer program product of claim 8 , wherein the plurality of measured HMT at different time points includes at least a plurality of measured HMT measured at irregular time intervals.

15 . A system comprising:
a hardware processor; a memory device coupled with the hardware processor; the hardware processor configured to at least:
receive manufacturing process data associated with a blast furnace, the manufacturing process data including at least state variables and control variables used in operating the blast furnace, the state variables including at least a hot metal temperature (HMT) and other state variables, wherein the manufacturing process data includes at least a plurality of measured HMT at different time points, of a product produced in the blast furnace;
generate imputed HMT by interpolating the measured HMT;
generate HMT gradients based on at least the imputed HMT;
define a causal relationship between at least one of the other state variables and the HMT gradients, the relationship generated as a neural network model;
train the neural network model using as training data, a weighted combination of the imputed HMT up to a last known measured HMT and predicted HMT up to the last known measured HMT,
the trained neural network model trained to predict a current point in time value for the HMT, in which no measured HMT for the current point in time is available, wherein the trained neural network model predicts the HMT corresponding to a time period starting from the time of the last measure HMT data point for a number of time periods until the number of time periods advances to the current point in time and uses the predicted HMT corresponding to each of the number of time periods to predict the current point in time value for the HMT.

16 . The system of claim 15 , wherein the hardware processor is further configured to transmit the current point in time value for the HMT to a controller to trigger a control action to control a manufacturing process oc

Assignees

• IBM

Inventors

• Lee Young Min
• Yeo Kyong Min

Classifications

• G01B21/085

  using thermal means · CPC title

• G01K3/04

  in respect of time · CPC title

• G01K17/20

  across a radiating surface, combined with ascertainment of the heat-transmission coefficient · CPC title

• C21B7/06 Primary

  Linings for furnaces · CPC title

• G01K7/42 Primary

  Circuits effecting compensation of thermal inertia; Circuits for predicting the stationary value of a temperature · CPC title