Implementation of a Deep Learning Estimation Model for Identifying Induced Failure Hazard Areas in Construction Equipment

Overturning accidents involving heavy construction vehicles may occur due to soft ground conditions caused by subsurface water accumulation or sinkholes. Although such accidents may appear unexpected during visual inspection, subsurface failure often precedes these events. Therefore, real-time assessment of ground stability is essential for improving operational safety. Electrical Resistivity Tomography (ERT) is a widely used geophysical method for analyzing subsurface conditions through deep ground investigation. However, ERT data interpretation typically involves complex nonlinear inversion processes that are computationally intensive and highly dependent on initial parameter estimates. In this study, an inference-based deep learning approach is proposed to efficiently identify areas susceptible to overturning accidents. The method converts data obtained using a Wenner array configuration into a high-resolution dipole–dipole configuration, thereby eliminating the need for traditional inversion procedures. To evaluate the performance of the proposed framework, three deep learning models—Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM)—were trained and tested using regression-based loss functions. Among these models, the MLP demonstrated the best performance, achieving an average R² value of 0.9935 ± 0.0014 and a root mean square error (RMSE) of 3.44%. The proposed approach leverages the inherent stability of the Wenner array and the high spatial resolution of the dipole–dipole configuration. By employing forward inference, the method significantly reduces the computational complexity associated with conventional inversion techniques. As a result, the efficiency and practical applicability of ERT-based ground inspection are substantially improved. Overall, the proposed framework provides an effective tool for identifying hazardous ground conditions and assessing the susceptibility of construction sites to overturning accidents involving heavy vehicles.