When Alejo Sfriso, corporate consultant at SRK Consulting Argentina, stepped up to the podium at the Life of Mine | Mine Waste and Tailings 2025 conference in Brisbane, his message was as direct as it was disruptive: it’s time to leave deterministic factor-of-safety thinking behind. “Factor of safety doesn’t exist anymore in concrete, steel or bridge engineering,” Alejo declared. “It’s still holding on in geotechnical engineering for dams, but not for much longer.”
Together with the paper co-authors, Alejo has helped develop a standardised, performance-based framework for assessing the vulnerability of upstream-raised tailings dams to flow liquefaction. Their paper outlines a major leap forward - one that abandons trigger-based thinking in favour of a structured, reproducible process grounded in deformation modelling, statistical probability, and modern risk-informed design principles.
The paper Standardising Vulnerability Assessments of Tailings Dams - Advancing Beyond Trigger Analyses was co-authored by Alejo, Mauro G. Sottile, and Arcesio Lizcano of SRK Consulting. The work draws from a multi-year collaboration across SRK’s international numerical modelling team and reflects deep field and technical experience in tailings dam engineering.
Critical state thinking meets computational modelling
The philosophical heart of Alejo’s argument rests on one fundamental insight from critical state soil mechanics: all tailings storage facilities (TSFs), no matter how idiosyncratic, share a common stress path through a space defined by critical and instability lines. “We no longer need to treat each dam like a medieval cathedral, built without standardisation,” he said. “We know the critical state line exists, and that allows us to test vulnerability in a standardised way.”
The team’s approach hinges on applying three standardised deteriorating actions:
- A load applied at the dam crest
- A toe contraction (simulating deformation of the structure or foundation)
- A rise in the phreatic surface (representing drainage state changes)
Each of these actions independently pushes the dam’s stress state toward instability. If a design holds up under these three perturbations, Alejo explained, “we cannot find any other trigger - credible or not - that will cause failure for that section.” The framework eliminates the need to invent dam-specific triggers, simplifying and unifying how vulnerability is assessed.
From trigger analysis to vulnerability surfaces
But the breakthrough came when the team began combining these actions - modelling not just singular triggers, but their compound effects. “Franco Boni once challenged me,” Alejo recalled. “‘What happens if two moderate triggers combine - do they bring failure where each one alone wouldn’t?’” That question prompted the development of what the team now calls the “vulnerability surface.”
The vulnerability surface is a three-dimensional boundary that maps combinations of crest load, toe contraction, and phreatic rise to system failure. It is defined numerically through Radial Basis Function (RBF) interpolation - a machine learning technique that converts discrete simulation results into a smooth, continuous surface.
“It’s the only 3D concept I can still visualise,” Alejo joked. “But it can be multi-dimensional too - math doesn’t care.”
The critical shift is that the vulnerability surface is no longer binary. Rather than separating ‘safe’ from ‘unsafe,’ it is embedded in a fuzzy transition zone using a Gaussian margin of uncertainty. “The boundary is blurred,” said Alejo. “It’s no longer a line. It’s a field of probabilities.”
Risk-informed design for modern tailings governance
The framework directly integrates with Eurocode 7 (Second Generation), which introduces a hierarchy for defining soil parameters based on uncertainty: measured, derived, and representative values. Alejo stressed that each parameter must now be treated probabilistically, and Eurocode 7 provides a procedure for assigning these distributions consistently.
To compute the probability of failure (PoF), the framework uses Monte Carlo simulations drawing from probabilistic inputs:
- Crest load distributions based on site operations
- Toe contraction distributions based on design tolerances
- Phreatic surface distributions based on drainage or climate data
Each triplet of values is tested against the vulnerability surface. “If the point lies below, it’s okay. If above, it’s not. And that’s done 100,000 times,” Alejo explained. “Computers love this stuff.”
In a simulated example using a representative dam cross-section, the team validated their method with over 650 modelled scenarios. The annual PoF was approximately 50 percent - consistent with the high sensitivity of the cross-section to toe deformation. “We used a vulnerable design to stress-test our own method,” said Alejo. “And we found it robust, smooth, and computationally efficient.”
A tool to empower TARPs and tailings governance
One of the most immediately useful applications of the framework is its potential to support Trigger Action Response Plans (TARPs). Unlike conventional TARPs, which rely on fixed thresholds, this approach can account for compound stress interactions - a critical improvement for modern risk governance.
“If your water table rises by two metres, the load that would cause failure might drop from 120 kPa to just 20 kPa,” Alejo said. “That’s actionable intelligence.” It means operators can better calibrate their TARPs, alert thresholds, and intervention strategies to how risk evolves across multiple domains.
Moreover, real-time sensor data from monitoring systems can be mapped onto the interpolated vulnerability surface, enabling early warning systems that are both automated and context-aware.
First presented in The Rock Wrangler article.
