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

On 25 January 2019, the upstream-constructed tailings dam at the Córrego do Feijão iron ore mine in Brumadinho, Brazil, failed catastrophically, killing 270 people and releasing 9.7 Mm³ of tailings. The event followed the 2015 Fundão dam failure and prompted regulatory changes and plans to decommission upstream dams. Despite cessation of tailings deposition in 2016 and modern monitoring, the Brumadinho dam failed without clear precursors, and its delayed nature has kept the cause under debate. An expert panel suggested creep and rainfall-induced suction loss; another investigation pointed to drilling-induced weakening. This study addresses the research question of how a catastrophic failure could occur years after closure, proposing and testing a mechanism in which slip surfaces initiate in weak fine-tailings interlayers during construction and progressively grow, then continue to grow post-closure under constant load due to creep until reaching a critical length that triggers unstable propagation and rapid collapse.

The paper situates the Brumadinho failure within a broader context of frequent tailings dam disasters (five to six catastrophic cases annually post-2000) and recognized triggers including rapid loading, seismicity, rainfall, and weak foundations. Prior analyses emphasize upstream construction vulnerabilities, segregation of coarse and fine tailings during deposition, static liquefaction in brittle, contractive tailings, and challenges in stability assessment of upstream dams. Competing post-event hypotheses for Brumadinho include creep with rainfall effects versus localized structural weakening due to vertical perforations. The study builds on fracture and progressive failure mechanics applied to soils and slip surface growth theory, extending them to layered tailings with rate-dependent and brittle behavior.

The authors performed lifecycle numerical modelling of the Feijão dam using a large-deformation finite element approach with an arbitrary Lagrangian–Eulerian style remeshing and interpolation workflow. Static analyses simulated construction, operation with raisings, and post-closure evolution; dynamic analysis captured the onset and first seconds of catastrophic failure. Lagrangian finite strain elasto-plastic increments were computed in Abaqus, with Python-based remeshing and field interpolation. Geometry, layering of coarse and fine tailings beaches for each of the ten raises (1976–2016), and phreatic lines were represented. Stress–strain history was tracked through construction and post-closure. Material modeling used an effective stress Mohr–Coulomb law with cohesion and friction angle reduction (strain-controlled softening from peak to residual) to capture brittleness in drained processes, and a total stress framework for rapid undrained events with peak undrained shear strength proportional to vertical effective stress, softening to a residual value with accumulated inelastic strain. Time-dependent behavior was modeled via a generalized Maxwell (Maxwell–Wiechert) viscoelastic model (Prony series) to represent creep; creep strain contributes to the inelastic strain that drives softening. Parameters were calibrated against laboratory triaxial tests (including two sustained-load creep tests) reported by the expert panel, fitting Prony series with six Maxwell elements (instantaneous shear modulus about 19.2 MPa, long-term about 8.1 MPa, RMS error <1%). Coarse tailings parameters included representative elastic moduli and Mohr–Coulomb strength (e.g., peak friction ~36°, residual ~33°, small cohesion, residual strain at softening onset ~4–6%), fine tailings had lower shear strength (peak friction ~33°) and higher brittleness/sensitivity; undrained parameters adopted typical values for normally consolidated tailings (Poisson’s ratio ~0.495, higher Young’s modulus in total stress). The foundation soil was modeled as stable Mohr–Coulomb material without creep. The modelling incorporated realistic discharge histories, including periods of locally high discharge rates causing undrained loading during operation, followed by consolidation and strength gain. Post-closure, creep under near-constant effective stresses was simulated; at slip surface tips, brittle failures and local undrained growth were considered the unfavourable case. An analytical criterion for onset of catastrophic slip surface growth in a cut planar slope was also applied to validate critical slip surface lengths, using average plane strain stiffness of the overlying coarse beaches, plastic displacement to reach residual strength, slip surface depth, and shear stress ratios at tip and along the surface. Satellite-derived surface displacement rates (2018–2019) were compared against model predictions to corroborate creep-induced deformations prior to failure.

- Mechanism: Failure was precipitated by slip surface initiation within weak fine-tailings interlayers beneath upstream raises during construction, with progressive growth under increasing load, followed by continued post-closure growth driven purely by creep under constant load. When combined slip surfaces reached a critical length, unstable propagation ensued, causing rapid collapse. - Evidence of layering and weakness: CPTu data and construction geometry show interlayered coarse and fine tailings beneath several raises, especially after the fourth raise with a setback, creating loci for subhorizontal slip surface propagation. - Progressive to catastrophic transition: Numerical modelling indicates multiple localized slip surfaces formed by the 7th–10th raises. Post-closure creep caused further initiation and growth. The top two merging slip surfaces at ~60 m depth reached a combined length of ~250 m at failure onset; analytical criterion estimates a critical length ~200 m for that depth, consistent with catastrophic triggering at 250 m. The deeper set at ~70 m depth had a critical length ~210 m vs. combined length ~140 m and did not trigger failure first. - Time to failure: The main slip surface lengthened by ~130 m in the first year post-closure, then by ~50 m over the next two years (stable growth), followed by rapid propagation just before failure (about 32–36 months after closure), matching the January 2019 event (~3 years post-closure). - Rates and deformations: Post-closure growth of the slip surface tip occurred via discrete rapid events, with modeled rates up to ~0.4 m/day; surface displacements remained small (~30 mm in the final year), consistent with satellite observations showing limited precursors. Modeled annual surface displacement vectors and magnitudes matched satellite estimates near the crest and slope. - Dynamic collapse: Within ~0.5 s of unstable growth, the main slip surface daylighted above the starter dam with a bulge matching video evidence; within ~5–11 s, propagation extended upslope beneath the pond (~300 m behind crest), initiating retrogressive failure and liquefaction/runout of saturated tailings. - Discharge history effect: Including realistic undrained loading episodes during operation produced the observed deep, layered slip surface system and eventual failure; assuming fully drained discharge led to different, curved failure zones near the starter dam and did not collapse, though post-closure creep-driven growth in fine layers still occurred but with later failure timing. - Creep sensitivity: Using creep parameters corresponding to lower K0 (0.4) from lab tests would have led to collapse as early as January 2017 (~6 months post-closure). Thus, creep characteristics strongly affect time to failure. - Brittleness sensitivity: Reducing the strain required to reach residual strength to ~4% would lead to failure within ~2 months after closure; increasing it to ~30% may prevent failure entirely. Higher iron content increases brittleness and pore pressure generation, promoting static liquefaction. - Rainfall and suction: While loss of matric suction in near-surface tailings during rainy seasons has limited effect on deep slip surface growth, it can accelerate shallow shear banding and contribute to liquefaction once motion begins. Other site disturbances (e.g., vertical perforations) could have acted as accelerants but were not required by the proposed mechanism. - Monitoring implication: Small pre-failure surface displacements in brittle, layered tailings mean that conventional deformation monitoring may not provide clear warnings; lifecycle numerical modelling with creep and softening is necessary for risk assessment.

The findings resolve the paradox of a catastrophic failure occurring years after closure by demonstrating that creep-driven growth of pre-existing slip surfaces in weak fine-tailings layers can continue under constant load until a critical length is reached, triggering unstable propagation. This mechanism reconciles the limited observable surface deformations with the severe internal weakening and explains why monitoring did not signal imminent collapse. It clarifies the dominant role of construction and discharge history in preconditioning weak layers, highlights the importance of layered stratigraphy from tailings segregation, and quantifies the transition threshold (critical length) using both numerical and analytical approaches. The results underscore that external triggers such as intense rainfall or drilling are not necessary for failure, though they may hasten it; thus, decommissioned upstream dams with brittle, rate-dependent tailings remain at risk in the absence of obvious external loads. Practically, the work implies that robust risk assessment should: model the full dam lifecycle with rate effects and softening; evaluate current slip surface lengths versus analytical critical thresholds; and consider parameter uncertainties (creep rates, brittleness) that strongly influence time to failure. The demonstrated match to satellite displacement data and event kinematics lends credibility to the proposed mechanism and offers a pathway for proactive assessment and mitigation.

The study proposes and validates a physical mechanism for the Brumadinho failure: delayed, creep-driven growth of slip surfaces along fine-tailings layers formed during upstream construction, culminating in catastrophic propagation once a critical length is exceeded. Lifecycle large-deformation FE modelling, calibrated to laboratory creep and strength data and corroborated by satellite displacement and event video, reproduced the observed timing (~3 years post-closure) and initial failure kinematics. An analytical critical-length criterion independently supports the numerically inferred trigger. Sensitivity analyses show that discharge history, creep characteristics, and tailings brittleness control the time to failure; rainfall and other perturbations can accelerate but are not necessary. The main contributions are a mechanistic explanation for delayed failure in decommissioned upstream dams, a validated modelling framework for risk assessment, and a practical criterion to gauge proximity to catastrophic transition. Future research should: refine site-specific parameter characterization of creep and brittleness; extend modelling to 3D and incorporate evolving phreatic conditions and consolidation; integrate remote sensing-based displacement and pore pressure monitoring into predictive workflows; evaluate mitigation strategies (e.g., drainage improvements, buttressing) via lifecycle models; and couple detailed dynamic failure initiation models with efficient runout simulators for consequence assessment.

- Modelling focused on a representative 2D cross-section; three-dimensional effects and spatial variability of layering and properties may alter local stability margins. - For conservative assessment, the highest post-operation phreatic surface was maintained in post-closure modelling; temporal evolution of groundwater and full long-term consolidation effects were simplified, though noted as minor for deep slip growth. - The assumption of undrained conditions for slip surface tip growth represents an unfavourable scenario; actual drainage conditions may vary spatially and temporally. - Material parameters (creep Prony series, softening laws, sensitivity) were calibrated from limited laboratory tests; parameter uncertainty strongly influences predicted time to failure. - Foundation soils were assumed stable with no creep; interactions with foundation heterogeneity were not explored. - External triggers (rainfall-induced suction loss, drilling) were not required by the mechanism but were not explicitly modelled in coupled hydro-mechanical transients beyond sensitivity considerations.