Tailings dams represent one of the most environmentally sensitive infrastructures in the mining industry. To address the need for continuous and accurate monitoring, this paper presents an adaptive forecasting framework that combines Internet of Things (IoT) technologies with machine learning (ML) models to detect early signs of structural and ecological risks.

The proposed system architecture is modular and scalable and enables the automated training, selection, and deployment of predictive models for multivariate sensor data. Each sensor data flow is independently analyzed by using a configurable set of algorithms (including linear, convolutional, recurrent, and residual models).

The framework is deployed via containers with a CI/CD pipeline and includes real-time visualization through Grafana dashboards. A use case involving tiltmeters and piezometers in an operational tailing dam shows the system’s high predictive accuracy, with mean relative errors below 4% across all variables (in fact, many of them have a mean relative error below 1%).

These results highlight the potential of the proposed solution to improve structural and environmental safety in mining operations.

This work has been partially supported by the SEC4TD project, funded by EIT with Project Agreement Number 21123, and for UPC authors by the Spanish Ministry of Science and Innovation under grant PID2024-156150OB-I00, as well as by the Catalan Government under contract 2021 641 SGR 00326.