Arresting failure propagation in buildings through collapse isolation

The study addresses the problem that many building collapses are driven by the propagation of local-initial failures caused by extreme or abnormal events (earthquakes, floods, storms, landslides, explosions, impacts, and errors). Contemporary robustness design emphasizes extensive connectivity to prevent collapse initiation by redistributing loads, but this can become counterproductive for large initial failures by allowing failing regions to drag down intact parts. The research proposes and investigates a hierarchy-based collapse isolation strategy that deliberately controls the sequence of failures so that connections fail before critical vertical load-bearing elements (columns), thereby arresting propagation. The aim is to maintain sufficient connectivity for operational conditions and for small initial failures (enabling alternative load paths) while enabling controlled segmentation of the structural system to isolate collapse when large failures occur. The work situates this concept in the context of persistent catastrophic collapses and increasing hazard intensities, highlighting the need for robust, threat-independent solutions.

The paper situates the work within decades of research on structural robustness and progressive collapse following events such as Ronan Point (1968). Existing guidelines (Eurocode EN 1991-1-7, DoD UFC 4-023-03, GSA guidance, ASCE standards) largely prescribe enhanced connectivity (tying forces, full-strength connections) to improve alternative load paths and prevent collapse initiation. Prior studies emphasize prevention of initiation rather than propagation control, and several disasters have occurred following large initial failures despite connectivity measures. The authors reference work on progressive collapse mechanisms, earthquake-induced gravity-load collapses, vulnerability to floods and wind, and landslide impacts, as well as biomimetic inspiration (lizard tail autotomy) for controlled segmentation. They also reference precast connection guidelines (fib) and analytical frameworks (MCFT) and simulation methods (Applied Element Method) used in robustness analysis.

Overview of approach: The hierarchy-based collapse isolation design distinguishes between small initial failures (for which collapse initiation can be prevented via alternative load paths) and large initial failures (which would otherwise trigger at least a partial collapse). The design enforces a failure hierarchy such that beam–column connection failure occurs before column failure, preventing successive column losses and thus arresting propagation. Cost-effective implementation balances reduced connection strength with increased column capacity.

Building designs: Two 15 m × 12 m, two-story (2.6 m floor-to-floor) precast reinforced-concrete building designs were developed: Design H (hierarchy-based) and Design C (conventional). Both used precast columns with corbels, partially precast beams, and hollow-core slabs. Design H used partial-strength beam–column connections and enhanced columns; Design C used full-strength connections per conventional practice and code-compliant columns without explicit hierarchy checks.

Local design details (Design H):

• Top continuity bars (beam ties) dimensioned using UFC 4-023-03 tie force provisions. Required tie strengths implemented with three 25 mm bars for internal beams and three 20 mm bars for perimeter beams, continuous through columns (lap splices/couplers).
• Dowel bars between beams and corbels configured to achieve partial-strength connections that cap force transfer and allow rotations. Design checks based on fib: shear yielding (Von Mises), concrete splitting (strut-and-tie design), and plastic hinge formation in dowels under combined shear/bearing/friction. Final configuration adopted two 20 mm dowel bars per connection to ensure connection failure precedes column failure.
• Columns: Initially designed per codes, then verified with Modified Compression Field Theory (MCFT) using Response 2000 to ensure column capacity envelopes (M-N and M-V) exceed the maximum forces that can be transmitted through the partial-strength connections during collapse. This enforces the hierarchy (connections fail before columns).

Global verification via simulation: High-fidelity collapse simulations used the Applied Element Method (AEM) via Extreme Loading for Structures (ELS) to capture large displacements, fracture, contact, and collisions. The model included columns, beams, corbels, and hollow-core slabs with specialized interfaces (friction 0.6 concrete–concrete; 0.5 at elastomeric bearings). Mesh used ~150 mm elements, totaling 98,611 elements. Concrete and steel constitutive models incorporated tension softening, shear interaction (Mohr–Coulomb), discrete rebar springs with elastic–yield–hardening behavior, and perfect bond to concrete. Ground floor column bases were fully restrained; reflecting boundaries represented debris rebound. Loads were applied as added mass to slabs. Column removal was modeled via element deletion to simulate sudden loss scenarios. Material properties were based on lab tests and standards; key parameters are tabulated in Extended Data Tables 1–2.

Initial failure scenarios and test planning: Simulations evaluated three critical scenarios focusing on edge/corner regions:

• Scenario 1: Corner + adjacent edge column removal; collapse at 11.5 kN/m²; local partial collapse.
• Scenario 2: Two corner columns + intermediate edge column; collapse at 8.5 kN/m²; segmentation along one axis.
• Scenario 3: One corner + two adjacent edge columns; collapse at 7.0 kN/m²; segmentation along both axes; predicted 50:50 intact-to-collapsed area. Scenario 3 was selected for testing.

Full-scale experimental program (Design H): A full-scale specimen matching Design H was built and tested in two phases to examine both small and large initial failures under the same structure:

• Loading protocol: Uniformly distributed load of 11.8 kN/m² was applied only on bays expected to collapse in Phase 2 (more than 8,000 sandbags on two floors), nearly twice the Eurocode accidental situation value (6 kN/m²). Simulations showed partial loading produced similar collapse states to full loading but imposed higher unbalanced moments and larger drifts on the remaining structure, providing a more demanding test.
• Phase 1 (small failure): Quasi-static removal of two penultimate-edge columns (C8 and C11) using custom removable hydraulic jack supports. Goal: demonstrate development of alternative load paths and prevention of collapse initiation.
• Phase 2 (large failure): Dynamic removal of the corner column (C12) between previously removed columns using a three-hinged collapsible steel column with a central hinge unlocked by external actuation, triggering sudden loss.
• Instrumentation: 57 embedded strain gauges, 17 displacement transducers, 5 accelerometers, plus high-resolution cameras and drones. Data used to analyze load redistribution, member responses, drifts, and failure progression.

Comparative baseline (Design C): In simulations, Design C used the same continuity bars but full-strength connections (e.g., two 32 mm dowels) per fib guidance to maximize connectivity; columns designed per codes without explicit hierarchy checks. This served as a benchmark for evaluating hierarchy-based isolation.

Validation: Simulation predictions were compared with experimental measurements (e.g., axial loads in selected columns, drifts at column tops). Good agreement was shown for response evolution following sudden column removal (C12).

• Design H (hierarchy-based) developed stable alternative load paths (ALPs) after small initial failures more severe than those considered by building codes. In Phase 1, quasi-static removal of two penultimate-edge columns (C8, C11) did not initiate collapse under 11.8 kN/m² loading (almost 2× Eurocode accidental). Beams over removal locations carried loads in flexure (deflections too small for catenary action), with increased tensile strains in continuous reinforcement confirming load redistribution; slabs and adjacent columns contributed significantly to ALPs.
• Under a large initial failure (Phase 2: sudden removal of corner column C12 after Phase 1 removals), Design H arrested collapse propagation along a predefined segment border (columns C3–C7–C6–C10), isolating the collapse to bays around C8, C11, C12. Connection failures occurred (rupture of continuity bars, dowel/corbel failures) before any column failure, as intended by the hierarchy, preventing successive column loss and global collapse. Debonding of slab reinforcement along the border completed segmentation.
• Measured dynamic response of the upright segment indicated strong horizontal tugs from collapsing bays: resultant drift at the top of column C9 peaked at about 39.5 mm, then decayed with oscillation to a residual drift (~15.4 mm), consistent with strong but survivable demands comparable to seismic drift limits for rare events. Column C7 experienced a substantial transient increase in compressive strain (peak change approx. −244 με) as a pivot resisting pull-in forces.
• Simulations predicted that both Design H and Design C could prevent collapse initiation for small failures exceeding code scenarios. However, for large initial failures, Design H isolated collapse, whereas Design C’s increased connectivity caused pulling-down effects and total collapse of the building in simulation.
• The full-scale test validated the computational model’s ability to capture key phenomena (axial load redistributions, drifts, failure sequence), supporting the robustness of the design and analysis approach.
• Overall, hierarchy-based collapse isolation provides a low-cost “last line of defence” that does not compromise performance under small failures and can prevent catastrophic propagation following large failures.

The findings directly address the core research question: can collapse propagation after large initial failures be arrested by controlling the hierarchy of failures in a building system? By ensuring that beam–column connections fail before columns, the approach intentionally segments the structure, stopping propagation paths that would otherwise lead to progressive, system-wide collapse. This contrasts with conventional robustness strategies that seek maximal connectivity; while helpful for small failures, those can unintentionally transmit destructive dynamic forces during large failures, enabling intact parts to be dragged into collapse. The experiments confirm that hierarchy-based connections enable ALPs under small failures and trigger controlled disconnection under severe events, preserving a significant portion of the structure for safe evacuation and reducing casualties and losses. The approach reframes robustness design to include propagation control, providing threat-independent resilience suitable for an era of increasing extreme events. The validation through both high-fidelity simulations and a full-scale test lends credibility to adopting this design philosophy in practice.

The study introduces and validates a hierarchy-based collapse isolation strategy that arrests failure propagation by enforcing a failure sequence in which beam–column connections fail before columns. A full-scale precast RC building test and corroborating AEM simulations show that this approach (Design H) prevents collapse initiation after severe small failures and isolates collapse under large failures, in contrast to a conventional full-strength connection design (Design C) that exhibited total collapse in simulation under the same large-failure scenario. This provides a practical, low-cost last line of defence that enhances building resilience without compromising everyday performance or code-required robustness. Future work should extend development, testing, and implementation across different building types, materials, and connection systems, and refine design guidelines and analytical tools to mainstream hierarchy-based collapse isolation in structural robustness practice.

• Scope of validation: The experimental validation is based on a single, purpose-built full-scale precast reinforced-concrete, two-story building with specific geometry, detailing, and loading protocol. Generalization to other structural systems, heights, materials, and connection typologies requires further testing and analysis.
• Loading and scenarios: The test focused on edge/corner column removal scenarios with partially applied gravity loads (though simulations suggest equivalence to full loading for collapse patterns). Other hazard types, interior failures, and different dynamic removal mechanisms were not experimentally tested here.
• Design comparisons: The conventional benchmark (Design C) was validated via simulation rather than experiment; while simulations were validated against the Design H test, direct experimental comparison for Design C was not performed.