Blast Furnace Temperature Control Using Image-Based Learning Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing molten iron temperature control methods in blast furnaces face challenges in maintaining accuracy, particularly when disturbances such as changes in unloading speed or iron content occur, and when operation conditions lack historical data, leading to degraded prediction and control accuracy.

Innovation Solution

A learning model generation method using machine learning to convert historical data into image data, which is then processed by a convolutional neural network to determine operational quantities for controlling molten iron temperature, allowing for accurate control and guidance in blast furnace operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a physical model is used for molten iron temperature control, then control accuracy is improved under normal conditions, but prediction accuracy degrades when disturbances occur that are difficult to consider in the physical model

Engineering Contradiction:
Improvemolten iron temperature prediction accuracyVSAvoidadaptability to disturbances
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent combines a physical model with a machine learning model to create a hybrid control system. The physical model provides control accuracy under normal conditions, while the machine learning model compensates for disturbances that the physical model cannot predict, such as changes in unloading speed or iron content in raw materials. This merging allows the system to maintain high prediction accuracy while adapting to various disturbances.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The machine learning model acts as an intermediary that bridges the gap between the physical model and actual process disturbances. It processes historical data and operational parameters to predict temperature deviations caused by disturbances, allowing the physical model to be adjusted accordingly and maintaining overall prediction accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If past operation data is used for temperature control, then prediction accuracy is ensured for similar operation cases, but accuracy degrades when operation conditions lack historical data

Engineering Contradiction:
Improvemolten iron temperature prediction accuracyVSAvoidhandling of new operation conditions
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary actions by collecting and storing historical operation data in advance. This historical data is then used by the machine learning model to predict temperatures for new operation conditions, allowing the system to prepare for and accurately handle previously unseen scenarios without requiring exact historical matches.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The machine learning model uses parameter changes from historical data to adapt to new operation conditions. By learning from patterns in past operations and adjusting its predictions based on current operational parameters, the model can maintain accuracy even when encountering new conditions that differ from historical records.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If machine learning with image data is used, then control accuracy is improved for varying conditions, but device complexity increases

Engineering Contradiction:
Improvetemperature control accuracyVSAvoidlearning model generation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent converts historical operational data into image data representations, creating a visual copy of the process history. This allows the machine learning model to leverage image processing capabilities and patterns recognition to accurately predict temperature variations under varying conditions, while the conversion process structures the data in a manageable format that reduces the complexity of direct numerical analysis.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4002023B1Learning model generation method, learning model generation device, method for controlling molten iron temperature in blast furnace, method for guiding molten iron temperature control in blast furnace, and molten iron manufacturing method
Publication Date: 2024.09.25 JFE STEEL CORP
  • EP4002023B1 patent drawingFigure 1~2
  • EP4002023B1 patent drawingFigure 3
  • EP4002023B1 patent drawingFigure 4

AI summary

A learning model generation method according to the present invention includes the steps of machine learning a behavioral model with which an operator determines an operational quantity of a process or equipment, where input data is image data obtained by imaging history data of one or more observable quantities indicating an operation state of a process or equipment, and output data is an operational quantity of a process or equipment determined by the operator based on the history data, and outputting the machine-learned behavioral model.