Prediction of Cover-Subsidence Sinkhole Volume Using Fibre Bragg Grating Strain Sensor Data.

Area of Science:

• Geotechnical Engineering
• Geohazards
• Sensor Technology

Background:

• Sinkholes pose significant risks to infrastructure and human safety, particularly in karstifiable terrains.
• Accurate monitoring and volume estimation of cover-subsidence (C-S) sinkholes are crucial for risk mitigation.

Purpose of the Study:

• To investigate the feasibility of using fibre Bragg grating (FBG) sensor strain data and machine learning (ML) for determining C-S sinkhole volume.
• To compare the performance of different ML algorithms in predicting sinkhole volume.

Main Methods:

• Experimental formation of a C-S sinkhole using a test rig.
• Collection of strain data using FBG sensors during sinkhole development.
• Application and comparison of Weighted Least Squares (WLS) regression, Support Vector Regression (SVR), and eXtreme Gradient Boosting (XGBoost) for volume regression.

Main Results:

• ML algorithms assuming normality were found inappropriate for C-S sinkhole phase classification and volume regression with FBG strain data.
• WLS regression yielded the lowest R2 values.
• SVR showed improved performance over WLS.
• XGBoost achieved the highest R2 values (1.00 for training, 0.97 for testing) and lowest root mean squared errors.

Conclusions:

• eXtreme Gradient Boosting is a highly effective method for determining C-S sinkhole volume using FBG sensor strain data.
• FBG sensors combined with advanced ML techniques offer a promising approach for sinkhole monitoring and assessment.