Development of adaptive systems for monitoring deformations of soil bases and foundations using real-time sensors and machine data analysis methods

Abstract

The article is devoted to the development of adaptive systems for monitoring deformations of soil bases and foundations, aimed at transitioning from discrete observations to continuous data streams on the stress-strain state of the soil base–foundation–structure system under conditions of dense urban development and complex soils exhibiting rheological properties. The need for algorithms capable of simultaneously suppressing anthropogenic and temperature-induced noise while generating settlement forecasts for early anomaly detection before deformations exceed the elastic stage is substantiated. The architecture of a sensor network is examined, comprising hundreds of measurement channels with cross-verification of readings (string strain gauges, inclinometers, hydrostatic levels) and parallel deployment of fiber-optic Bragg grating sensors to minimize drifts; the practical value of dynamic polling frequency, adjustable from hourly intervals to 10 Hz during vibrational and loading impacts, is demonstrated, enabling the accumulation of datasets on the order of tens of millions of measurements and rendering advanced time series analysis methods feasible. Quantitative examples compare signal filtering approaches: simple smoothing provides limited noise reduction, whereas the adaptive Kalman filter and wavelet transform achieve variance suppression by orders of magnitude, albeit with computational complexity differences critical for real-time edge devices. Correlation analysis of influencing factors reveals the dominance of static loading alongside notable effects from groundwater levels and vibrations, characterized by nonlinear threshold effects and consolidation time lags. Predictive modeling results over a 30-day horizon are presented, with recurrent networks and hybrid ARIMA+LSTM models attaining high accuracy (MAE ≈ 0.4 mm, R² ≈ 0.98) when supplied with preprocessed data; the systemic interdependence of data preprocessing and model training stages is underscored. Implementation barriers are highlighted, including the absence of standardized data protocols and regulatory verification of algorithmic forecasts, alongside prospects for digital twins and cloud-based aggregation of anonymized observations to scale predictive structural safety.