Research
Research
Mechanics-consistent, uncertainty-aware computation with data assimilation — and the question of when the information at hand is sufficient for the decision that has to be made.
Research profile
My work sits at the interface of nonlinear computational structural mechanics, inverse problems and reliability theory. The guiding idea is a single framework: a mechanics-consistent, uncertainty-aware computational method with data assimilation, scalable enough to be used on real structures.
One half of that is reliability itself. Classical safety formats assume a well-behaved limit state. Real nonlinear systems do not oblige: the failure domain is curved and often non-convex, several failure modes compete, and a factor applied to a material property no longer corresponds to a defined probability. My 2023 general partial safety factor theory extends the classical format to such systems using a notion of homogeneity, rather than discarding a format that engineers can actually apply.
The other half runs in the opposite direction — from the structure back to the model. Existing bridges, dams and heritage buildings are not designed but assessed, and part of the evidence is measured rather than assumed. That raises questions of identifiability and information: which observations tell us something we did not already know, what stays unidentifiable however long we measure, and how much uncertainty survives once the data are in.
Underneath both is the mechanics: nonlinear finite element formulations, constitutive models, the numerical machinery needed to follow a structure past its peak load. I do not simply apply existing software. I use difficult, discrete-continuous systems from real infrastructure as benchmarks demanding enough to expose where the algorithms fail.
What I want from all of it is unglamorous: scientific insight that survives contact with a real structure and becomes a method someone can actually use — in an engineering office, in an assessment, or in a code. Twenty years of moving between the university, the engineering office and the standardization committee is what makes that possible; it is also the part of the work that is hardest to do from only one of the three.
Earlier in my career the methods were developed mainly on masonry and heritage structures and carried into European standardization. Both continue — but as fields of application, not as the subject.
Fields
01
Nonlinear computational structural mechanics
Finite element and finite–discrete element formulations, constitutive and interface models, stability, damage and collapse. The numerical machinery needed to follow a structure past its peak load — and the computing to do it at the scale of a real one.
02
Structural reliability and safety concepts
Reliability of nonlinear systems, where the failure domain is curved and several modes compete. The general partial safety factor theory that extends the classical format to them, together with quality control, conformity assessment and code calibration.
03
Information, identifiability and inverse problems
What observations actually reveal about a structure: Bayesian inversion, sensor informativeness, spatially distributed parameter fields — and decision identifiability, the question of when the available evidence already settles the decision that has to be made.
04
Monitoring, digital twins and predictive assessment
Coupling measurement with continuously calibrated models, so that the condition of a bridge or a building becomes a living quantity: data assimilation, service-life prediction, and the case for extending service life on evidence rather than on assumption.
05
Scientific machine learning and complexity
Mechanics-informed operator learning and structure-preserving surrogates — fast, mesh-independent solvers that stay physically admissible — and measures of complexity for systems too large to reason about one member at a time.
06
Engineering practice and standardization
Where the methods are tested. Assessment of existing and infrastructure structures, earthquake resilience and retrofitting, and the translation of research into European and German design rules — the part that decides whether a method is real.
Current directions
- Information sufficiency for reliability decisions. Establishing decision identifiability as a mechanics-based property of structural inverse problems: decision-equivalence classes in the space of distributed parameters, information operators that isolate the directions capable of moving a limit state, and certificates that test whether a decision stays stable across everything the observations allow.
- Identifiability of spatially distributed parameters. Bayesian random-field formulations and multiscale identifiability measures that separate what can be resolved from what cannot, under sparse observations and in the presence of model discrepancy.
- Information-driven experimental design. Placing sensors and designing loading programmes to resolve a decision boundary rather than to reconstruct a whole parameter field — fewer sensors, and better answers from them.
- Mechanics-informed operator learning. Embedding equilibrium, material objectivity and thermodynamic consistency into deep operator learning, to obtain mesh-independent surrogate solvers for families of nonlinear boundary-value problems that stay physically admissible.
- Structure-preserving surrogates for reliability. Reduced-order models that preserve energy balance and passivity, so that very small failure probabilities can be estimated without brute-force Monte Carlo and without losing the physics.
This is my own research programme, pursued independently. Several parts are in preparation rather than in print, and I am glad to discuss any of them with people working on the same questions.
Standardization and transfer
Research becomes useful to practice when it reaches the rulebook. I have worked in European and German standardization since 2013, most substantially on the safety format for numerical methods in EN 1990.
European standardization · CEN
- CEN/TC 250/SC 10/WG 3 — Safety formats and FEM applications
Co-author of the new Annex X of EN 1990, which governs design assistance by numerical methods. - CEN/TC 250/WG 7 T1 — EN 1990 — Basis of structural design
Safety of nonlinear structural systems. - CEN/TC 250/SC 6 PT T1 — EN 1996-1-1 — Design of masonry structures
Proposed and carried the accepted safety verification for the buckling of masonry members. - AhG Reliability, CEN/TC 250/SC 10 — Reliability of the Eurocodes
Preparatory work for the background report of prEN 1990.
German standardization · DIN
- DIN NA 005-06-37 AA — Earthquake safety of masonry structures
- DIN NA 005-51 FBR-01 SO — “GruSiBau” — fundamental safety requirements for building structures
Memberships
- JCSS — Joint Committee on Structural Safety
- GAMM — International Association of Applied Mathematics and Mechanics