Intelligent Coal Mining Automation

The Client

The client is a mining operator specializing in underground coal extraction. Their operations depend on precise excavation of coal layers while avoiding surrounding rock formations. As safety risks and operational complexity increased, they needed a reliable way to automate excavation processes and reduce human exposure to hazardous environments.

The Challenge

Underground coal excavation was performed using heavy machinery controlled manually from ground level, exposing operators to dangerous working conditions. The process carried significant safety risks, including incidents with severe or fatal consequences.

A key geological constraint was the presence of a thin rock layer embedded within the coal seam. This layer, located approximately midway between the bottom and top of the excavation front, needed to be accurately identified to guide the excavation path. While visible to the human eye, detecting it programmatically was challenging due to environmental conditions.

The available visual data came from infrared cameras mounted on excavation machines, capturing images in dusty, low-visibility conditions at regular intervals. These images were often contaminated with previously excavated material, making it difficult to distinguish relevant geological features.

Without an automated detection system, the excavation process could not be safely or consistently optimized, limiting the ability to transition toward autonomous mining operations.

What We Did

Approach

We developed a computer vision solution capable of identifying the thin rock layer within the coal seam using infrared imagery captured in real-world mining conditions. The discovery phase revealed that robust classification between coal, rock, and visual noise (e.g. debris and dust) would be critical for reliable performance.

Given the complexity of the visual environment, we focused on supervised learning with carefully labeled data to ensure the model could generalize across varying conditions. The solution was designed not only to detect the geological feature but also to provide actionable input for automated excavation path planning.

A key design decision was to prioritize robustness to noise and environmental variability over purely theoretical accuracy, ensuring reliable operation in production settings.

Methodology

• Data assessment:
  Collected infrared image data at 10 cm excavation intervals, capturing real- world conditions including dust and debris contamination
• Approach selection:
  Selected a computer vision classification approach tailored to distinguish coal, rock layers, and previously excavated material
• Model development:
  Labelled imaging data (coal vs. rock) and trained a model to detect the thin rock layer within the coal seam
• Validation framework:
  Evaluated model performance on unseen excavation scenarios, focusing on robustness to noise and varying visibility conditions
• Integration:
  Deployed the model into the excavation workflow to guide automated path planning for subsequent mining cycles
• Compliance:
  Aligned with industrial safety and operational requirements for autonomous mining systems

The Outcome

Successful deployment of automated geological feature detection

The computer vision model accurately identified the thin rock layer within the coal seam, enabling reliable guidance for automated excavation.

Secondary outcomes:

• Improved safety by reducing reliance on manual machine control
• Enhanced precision in excavation path planning
• Robust performance in dusty, low-visibility underground conditions
• Production deployment enabling scalable automation of mining operations

Facing a similar challenge?

If you're working with hazardous environments, complex visual data, or industrial automation, we can help design AI systems that improve safety and operational precision.

Book a Technical Consultation