The slip surface mechanism of delayed failure of the Brumadinho tailings dam in 2019

The catastrophic failure of the Feijão tailings dam in Brumadinho, Brazil, on January 25, 2019, tragically claimed 270 lives and caused widespread environmental devastation. This event, occurring three years after tailings disposal ended, highlights a critical gap in our understanding of long-term tailings dam stability. While the immediate causes of dam failures, such as increased loading, earthquakes, rainfall, and weak foundations, are relatively well-understood, the delayed failure mechanism observed in Brumadinho requires further investigation. The delayed nature of the failure makes it particularly concerning, as it challenges the common assumption that dams become safer after tailings deposition ceases. This study aims to elucidate the physical mechanism responsible for this delayed failure by focusing on the Feijão dam's specific characteristics and history. Understanding this mechanism is crucial for improving risk assessment methodologies and ensuring the long-term safety of existing and decommissioned tailings dams worldwide. The mining industry's significant economic role in developing countries underscores the need to enhance safety standards and prevent future tragedies. This research contributes to that vital goal by proposing a novel mechanism explaining the delayed failure and identifying key parameters controlling the time to failure.

Previous research on tailings dam failures has focused primarily on immediate triggers such as rapid tailings discharge, earthquakes, and rainfall. Studies have explored the role of various factors in destabilizing dams, including increased loading, weak foundations, and the impact of environmental factors. The failure of the Fundão dam in Mariana, Brazil, which occurred shortly before the Brumadinho disaster, sparked increased scrutiny of tailings dam safety and led to regulatory changes. However, the literature lacks a comprehensive understanding of the delayed failure mechanism observed in Brumadinho, where the dam failed years after the cessation of tailings deposition. Existing models often don't adequately account for the long-term effects of creep and the complex interaction between different tailings layers. This study builds upon existing knowledge by incorporating a more detailed understanding of tailings behavior, including creep deformation and layer segregation, to explain the delayed failure observed in the Brumadinho disaster. This allows for a more holistic and accurate risk assessment than previous studies.

This study employed a large-deformation finite element method with remeshing and interpolation techniques to model the entire lifecycle of the Feijão dam. This included the initial construction phase, ongoing operation and tailings disposal, and the post-closure period leading to the eventual catastrophic failure. The model incorporated detailed information on the dam's geometry, tailings layering (coarse and fine tailings), and phreatic lines at each stage. The numerical analysis used a sophisticated constitutive model for tailings that captures three key aspects of their behavior: (i) brittle structure and strain softening, (ii) creep deformation under sustained loads, and (iii) segregation into coarser and finer layers during deposition. The brittle behavior was modeled using a Mohr-Coulomb constitutive law with a strain-softening frictional angle, reducing from a peak value to a residual value as the material shears. This accounts for the rapid strength loss observed in the tailings under undrained conditions. Creep was modeled using a generalized Maxwell model (Maxwell-Wiechert model), capturing the time-dependent deformation under constant stress. The model also incorporated the layering of tailings, reflecting the presence of interlayered coarse and fine tailings observed in in-situ cone penetration tests (CPTu). The numerical simulation accounted for the complex loading history of the dam, considering both the increasing load during the construction and operation phases and the constant load after closure. The model tracked stress and strain history throughout the dam's lifecycle, focusing on the development and growth of slip surfaces. The transition from progressive slip surface growth to catastrophic failure was simulated using dynamic analysis. The model parameters (e.g., elastic modulus, Poisson's ratio, cohesion, peak and residual friction angles, and creep parameters) were calibrated based on laboratory test data from the expert panel report and in-situ CPTu data. To validate the results, the model's predictions of surface displacement during the year prior to failure were compared to satellite image data. Analytical criteria for the critical length of the slip surface, adapted from Zhang et al. (2015), were used to verify the results of the numerical simulations. The analysis considered the combined effect of multiple slip surfaces and their potential to trigger catastrophic failure.

The numerical modeling revealed a three-stage process leading to the Brumadinho dam failure: 1. **Slip Surface Initiation and Growth during Construction and Operation:** The upstream construction method, with subsequent raisings, led to the formation of interlayered coarse and fine tailings. Initial slip surfaces nucleated within the weaker fine tailings layers during dam construction and operation, primarily due to rapid undrained discharge of tailings. Although local plastic deformation occurred, the dam remained globally stable after completion. 2. **Creep-Driven Slip Surface Growth after Dam Closure:** After tailings disposal ceased, slip surfaces continued to grow due to creep deformation under constant load. This slow, yet unstable, growth was largely unnoticed due to limited surface displacement (around 30 mm over the last year). The model accurately predicted the slow creep-driven deformation observed from satellite imagery in the year preceding the failure. 3. **Reaching the Critical Length for Catastrophic Propagation:** Catastrophic slip surface growth commenced when the combined length of multiple slip surfaces reached a critical value (approximately 250 m for the uppermost slip surfaces). This critical length was verified using an analytical criterion based on the properties of the tailings and the depth of the slip surfaces. The dynamic analysis showed that within seconds of the critical length being reached, the main slip surface propagated to the surface, leading to the catastrophic collapse of the dam. The model successfully replicated the observed deformation patterns at the onset of failure, further supporting the proposed mechanism. The model predicted failure within three years of closure which matches the actual failure time indicating the accuracy of the model. Parametric studies explored the effects of various factors, such as discharge history (drained vs. undrained), creep rate, tailings brittleness, rainfall, and vertical perforations on the time to failure. The results indicated that creep, driven primarily by the slow undrained discharge of tailings during construction, was the dominant factor leading to failure. However, other factors, such as rainfall and construction activities, could have accelerated the process, acting as additional triggers.

The findings demonstrate that the Brumadinho dam failure can be attributed to a delayed mechanism involving creep-driven slip surface growth within weak fine tailings layers. This mechanism was not immediately apparent due to the slow rate of creep deformation, resulting in only small surface displacements prior to failure. This underscores the limitations of relying solely on surface displacement monitoring for risk assessment of brittle tailings dams. The ability of the model to accurately predict the time to failure (three years after closure) highlights the potential of whole-lifecycle numerical modeling using advanced constitutive models for tailings. This approach can capture subtle internal processes that are difficult to detect through traditional monitoring techniques. The study also underscores the importance of considering the construction and operation history of a dam, as the initial formation and subsequent growth of slip surfaces during this phase significantly influence the likelihood and timing of delayed failure. Understanding the complex interplay between material properties (brittleness and creep rate), loading history, and environmental factors is crucial for developing effective risk assessment and mitigation strategies for tailings dams.

This study presents a novel mechanism explaining the delayed failure of the Brumadinho tailings dam, emphasizing the critical role of creep deformation in the post-closure phase. The advanced numerical modeling employed, accurately predicting both the slow pre-failure displacement and the timing of the catastrophic collapse, offers a powerful tool for future risk assessment of tailings dams. The findings highlight the limitations of relying solely on surface displacement monitoring for dams built with brittle tailings. Future research should focus on further refining the constitutive models for tailings, integrating more detailed hydrological data, and exploring the potential synergistic effects of multiple contributing factors. The development of improved risk assessment methodologies and the implementation of more robust monitoring systems are essential to prevent future tailings dam disasters.

While the model accurately predicted the time to failure and reproduced key aspects of the dam's behavior, certain limitations exist. The model's accuracy relies heavily on the availability and accuracy of input parameters, particularly those related to tailings properties. While calibrated using laboratory data, some uncertainties remain in fully capturing the complex, in-situ behavior of the tailings. The model also simplified some aspects of the dam's structure and the environmental conditions, such as rainfall variability and the detailed effect of construction activities. Further research is needed to refine the model and account for these additional complexities. Finally, the model focuses on the mechanism of failure; it does not directly model the post-failure behavior such as the flow of liquefied tailings.