Observe the pattern
Structure accident records so recurring exposure and control failures can be seen.
- Fatal-fall pattern analysis
- Transparent descriptive evidence
Fire & process safety researcher
Evidence that becomes engineering action.
I am Dr Abderrahim Zermane. My work connects accident records, causal analysis, explainable machine learning and fire physics so safety decisions remain traceable and honest about uncertainty.
3,321fatal-fall scenarios
128fatal cases reconstructed through fault trees
954hydrogen incidents linked to physics
UCL + UPMresearch and teaching experience
Research programme
One trajectory, not disconnected methods.
The question determines the method. The objective stays constant: preserve the route from evidence to interpretation and keep the boundary visible.
Explain or predict
Fault trees reconstruct pathways; comparative machine learning tests predictive value.
- Fault Tree Analysis
- Seven-model comparison
Translate into physics
Connect incomplete evidence to physical descriptors, propagation and uncertainty.
- Hydrogen reconstruction
- City-scale fire scenarios
EvidenceSources and assumptions remain traceable.
MethodDifferent tools answer different questions.
BoundaryEstimates are not presented as measurements.
DecisionAutomation supports professional judgement.
Flagship publications
Three papers that define the programme.
The established Safety Science contributions remain first; the hydrogen paper marks the transition into dedicated fire-safety research.
01 · Safety Science2023 · first author
Predicting fatal fall accidents using random forest
Seven models were tested on 3,321 scenarios. Random Forest offered the strongest practical balance and factor-level interpretation.
3,321 scenarios7 classifiersFirst author
Read the study summary →
02 · Safety Science2022
Fatal work-at-height accidents using fault trees
128 cases reconstructed as combined pathways rather than a flat list of causes.
Study summary →
03 · Fire Technology2026
Linking Accident Data and Physics-Based Modeling for Hydrogen Fire Safety
954 incidents translated into comparative physical descriptors with uncertainty propagation.
Milestone paper →Projects & research software
What the methods become in practice.
Real collaboration photography and project-specific diagrams replace generic technology imagery.
University College London · 2023–2025Research appointment
Tomorrow’s Cities: urban fire modelling
Physics-based urban fire-spread work within a multi-hazard and fragility framework.
Open the project →
Conceptual model-output viewResearch alpha
Open Urban Fire Framework
A reproducible exposure-to-fire workflow with reduced physics, verification templates and uncertainty-aware outputs.
Open the framework →
Workflow based on the working interfacePrivate working prototype
SEED EIA
158 review findings with source pages, maps, confidence, officer correction and audit history connected.
Open the project →01 / 03
Experience
Research shaped by teaching and safety practice.
Urban fire modelling, engineering education and earlier construction and industrial safety roles form one professional trajectory.
Research Fellow — Physics-based Fire Modelling
University College London
Teaching and Research Assistant
Universiti Putra Malaysia
Construction and industrial safety practice
DMA Construction · Bonatti SPA · Sonatrach
Research notes
Methods explained without hiding their limits.
Short essays connect the publications to the reasoning behind them.
Occupational safety
Why fatal-fall records need three layers of analysis
Description, causal reconstruction and prediction are not interchangeable.
Machine learningWhat “explainable” should mean in safety research
Feature importance guides questions; it does not prove causation.
Fire modellingWhy an urban fire map is not automatically a risk model
Geometry, inference, physics and uncertainty must remain separate.
Collaboration
Let’s make difficult safety evidence usable.
Research collaboration, invited teaching and technical work across fire safety, process safety, hydrogen safety and explainable risk analytics.