Investigation on the contamination process of trichloroethylene based on artificial neural network inverse tomography

In response to the issue of insufficient monitoring accuracy regarding the three-dimensional infiltration process of dense non-aqueous phase liquids (DNAPL) in saturated sandy soils. This investigation, the migration process of DNAPL contaminants in saturated sand was simulated using a three-dimensional sand tank experiment. Simultaneously, electrical resistivity imaging (ERT) was employed for dynamic monitoring, and an artificial neural network (ANN) model was utilized for the inversion of electrical resistivity data. This approach systematically obtained the spatial distribution characteristics of contaminants under both ERT and ANN methods. A numerical simulation model was established as an idealized case to compare the resistivity method with the ANN inversion method, thereby verifying the migration rules of DNAPL contaminants. The results indicated that: (1) Contaminate transport: Both ERT imaging and ANN inversion had clearly depicted DNAPL spatial infiltration images. Spatial moment analysis revealed that the numerical simulation results were generally consistent with the previous two methods in terms of horizontal and vertical penetration trends. Based on the numerical model, it was evident that the error of the ANN inversion method was smaller than that of the resistivity method when calculating the centroid error, demonstrated that ANN provides more accurate monitoring of contaminant distribution. (2) Model accuracy verification: The mean squared error (MSE) of ANN inversion for profiles at y = 0.01 m and y = 0.09 m is 0.0256 and 0.0238, respectively. Compared with the measured ERT data, this confirms that the ANN method enhances the inversion accuracy of the DNAPL infiltration process, providing reliable technical support for real-time monitoring and dynamic prediction of contaminants in three-dimensional fields.