Improving Stockline Detection of Radar Sensor Array Systems in Blast Furnaces Using a Novel Encoder-Decoder...

Area of Science:

• Metallurgical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate stockline measurement is crucial for optimized blast furnace (BF) charging operations.
• Harsh BF environments present challenges like noise interference and aberrant measurements for stockline detection.
• Traditional methods for stockline detection are often time-consuming and labor-intensive.

Purpose of the Study:

• To propose a novel deep learning architecture for robust stockline detection in blast furnaces.
• To address challenges of noise suppression and signal distortion in BF stockline measurements.
• To develop a learning-based approach that reduces manual effort compared to traditional methods.

Main Methods:

• A novel encoder-decoder architecture combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) was developed.
• The model suppresses noise, classifies distorted signals, and regresses stockline depth using historical data.
• Experiments were conducted using an eight-radar array system in an actual blast furnace.

Main Results:

• The proposed CNN-LSTM model effectively suppresses noise and classifies distorted signals in stockline measurements.
• LSTM enabled modeling of longer historical data for robust stockline tracking.
• The learning-based approach significantly saved time and effort compared to traditional denoising techniques.

Conclusions:

• The developed encoder-decoder architecture provides an effective solution for stockline detection in challenging blast furnace environments.
• The novel method demonstrates superior performance and efficiency for real-time stockline tracking.
• This approach offers a promising alternative to conventional methods, enhancing blast furnace operational optimization.