Underground abnormal sensor condition detection based on gas monitoring data and deep learning image feature...

Area of Science:

• Mining Engineering
• Sensor Technology
• Data Science

Background:

• Underground gas sensors are crucial for mining safety but prone to anomalies.
• Manual monitoring of numerous sensors in complex mining environments is infeasible.
• Accurate identification of gas sensor status is vital for preventing accidents.

Purpose of the Study:

• To develop a deep learning feature engineering approach for analyzing underground gas sensor data.
• To establish a method for detecting sensor anomalies by analyzing inter-sensor relationships.
• To improve the reliability of gas monitoring systems in underground mines.

Main Methods:

• Utilized deep learning feature engineering to model relationships between underground gas sensors.
• Transformed time-series gas concentration data into recurrence plots (RPs) for image-based analysis.
• Analyzed spatial and temporal correlations between sensors (T0, T1, T2) based on airflow patterns.

Main Results:

• Demonstrated that sensor relationships can be effectively modeled using position and time.
• Showcased that malfunctioning sensors can be detected by correlated data from other sensors.
• Confirmed high correlations between sensors in specific time series subsections, useful for status checks.

Conclusions:

• The proposed deep learning approach accurately identifies underground gas sensor status.
• Recurrence plots and machine vision methods offer efficient analysis of sensor data.
• This feature-based analysis enables reliable gas concentration monitoring, prediction, and early warning systems in mining.