Temperature Prediction Using Multivariate Time Series Deep Learning in the Lining of an Electric Arc Furnace for...

Area of Science:

• Metallurgical Engineering
• Data Science
• Artificial Intelligence

Background:

• Sensor data from extreme industrial environments, like smelting processes, offers valuable insights into structural behavior.
• Electric arc furnaces (EAFs) in ferronickel production are heavily instrumented, generating vast amounts of operational data.
• Deep learning approaches present an opportunity to analyze this data for improved process understanding and operational optimization.

Purpose of the Study:

• To develop and apply a deep learning model for predicting temperatures within a 75 MW electric arc furnace used for ferronickel production.
• To enhance data quality through cleaning and then utilize a multivariate time series model for accurate temperature forecasting.

Main Methods:

• A two-step methodology involving data cleaning to remove redundant and atypical data.
• Implementation of a sequential deep learning model comprising a gated recurrent unit (GRU) layer and a dense layer.
• Utilizing multivariate time series data from 16 radially distributed thermocouples for training and validation.

Main Results:

• The data cleaning process improved the overall quality of sensor readings.
• The developed GRU + Dense deep learning model achieved a high accuracy in temperature prediction.
• An average root mean square error (RMSE) of 1.19 °C was recorded on the test set for predicting furnace lining temperatures.

Conclusions:

• Deep learning models, specifically GRU-based architectures, are effective for temperature prediction in industrial smelting furnaces.
• Accurate temperature prediction can lead to better process control and operational efficiency in ferronickel production.
• This research contributes to the ongoing development of data analytics and AI applications in extreme industrial environments.