Abstract
In tunnel construction work, dust is generated at almost all activities from rock blasting, rock drilling, rock bolting, transport operations and spraying wet concrete on tunnel walls. Each of these activities had the potential to expose workers to airborne contaminants that, if not adequately controlled, can lead to significant harm, such as lung dysfunction. Therefore, it is essential to monitor dust concentration in the air during tunnel construction work to minimize health damage caused by dust. The Digital Dust Indicator and the Low Volume Sampler are famous devices for the dust concentration monitoring in tunnel construction. However, their common weak point is that they lack instantly because they require post-measurement work to know dust concentration readings. “Low cost”, “portable” and “real-time” are the keywords for sophistication of air dust monitoring devices for tunnel construction.
Introduction
Tunneling operations release significant amounts of respirable dust—particles smaller than four microns that can penetrate deep into the lungs, causing chronic respiratory diseases. Current monitoring solutions like Low Volume Samplers are accurate but slow, requiring manual weighing, while Digital Dust Indicators provide real-time data but are prone to clogging in the harsh, humid environments of a tunnel face. This paper introduces the Video Dust Meter (VDM), a low-cost, real-time, and maintenance-free sensor developed by researchers at Yamaguchi University to address these challenges.
Methodology
The VDM prototype integrates a 12-megapixel camera with a Jetson Nano for edge-AI processing. The system's primary innovation lies in its lighting geometry. Unlike previous models where forward-facing LEDs obscured the background, the VDM employs an inward-facing ring light with a 3mm slit. This configuration concentrates light on particles near the lens, exploiting the Mie scattering effect to make microscopic particles appear as bright, visible circles against a dark background.
To evaluate the system, experiments were conducted in a 27-meter-long plastic greenhouse simulating a tunnel environment. Arizona Test Dust (AC Fine) was used as a standardized simulant. The researchers tested three algorithmic pipelines for particle counting:
- Case A: Traditional computer vision (background subtraction, edge detection, and circle detection).
- Case B: Pure Deep Learning using the YOLO (You Only Look Once) model.
- Case C (Hybrid): A combined approach where background subtraction pre-processes frames before they are fed into a YOLO model.
Results and Discussion
The hybrid approach (Case C) emerged as the most robust method. While pure machine learning (Case B) struggled with low-contrast raw video, the pre-processing step in Case C effectively "cleaned" the data, allowing the YOLO model to ignore noise from vibrations or ambient light changes.
The VDM demonstrated high accuracy, with particle counts showing a strong correlation ($R^2 = 0.9059$ for Case C) compared to the industry-standard LD-5R digital dust indicator. The system successfully processed video at 30 frames per second, providing the real-time feedback necessary for worker safety.
Conclusion
The VDM offers a promising "set-and-forget" solution for air quality monitoring in civil engineering. While the greenhouse trials were successful, future work will focus on field testing in active mountain tunnels to assess durability against extreme humidity and varying rock dust types. Additionally, further calibration is required to convert visual particle counts into standard mass concentration readings (CPM).
Summary updated