Reliability of Ageing & Degrading Structures

Reliability is not stationary

For an ageing structure, today's margin is not the same as tomorrow's. Capacity can reduce through corrosion, fatigue cracking, creep, wear, embrittlement, environmental attack or degradation of joints and coatings. Loads can also change as duty cycles, occupancy, process conditions or operational procedures evolve. A lifecycle reliability model therefore treats resistance and demand as time-dependent quantities. The engineering objective may be to predict Pf at future dates, remaining useful life, the probability of reaching a maintenance threshold, or the inspection interval required to maintain a target reliability. This perspective is more informative than repeatedly performing unrelated deterministic checks at isolated ages.

Model the degradation mechanism, not just an age factor

Age itself is not a damage mechanism. The model should represent the physical process that changes capacity: wall-thickness loss from corrosion, crack extension from fatigue, creep strain or rupture damage, section loss from erosion, stiffness reduction or another measurable state. A stochastic degradation model may include a random initiation time, random growth rate and environmental covariates. Where several mechanisms act together, avoid simply summing arbitrary degradation percentages. Define how each mechanism changes the structural resistance model and whether the mechanisms are independent, correlated or sequential.

Time-varying resistance and demand

A useful generic limit state is g(t)=R(t)-S(t), where both resistance R and load effect S may be stochastic processes. Resistance can be represented directly or through underlying state variables such as remaining thickness, crack size or material strength. Demand may have ordinary operational variability plus rare events. The reliability at a future time is the probability that the limit state remains safe according to the chosen definition. If failure is irreversible, first-passage probability is normally more relevant than the probability of being unsafe at one isolated instant.

g(t) = R(t) - S(t)
R_sys(t) = P[T_f > t]

Data sources and degradation uncertainty

Degradation data can come from coupons, fleet histories, thickness surveys, crack measurements, operating records and mechanistic models. The main challenge is separating population variability from limited knowledge. A wide distribution inferred from sparse inspection data can reflect epistemic uncertainty rather than true unit-to-unit scatter. Hierarchical or Bayesian models can be useful when data exist across several components or fleets because they allow information sharing without pretending all assets are identical. Environmental and duty variables should be retained where they explain degradation; otherwise the model may extrapolate poorly when operating conditions change.

Inspection and condition updating

Inspection converts an ageing reliability problem from open-loop prediction to condition-based assessment. Measured wall thickness, crack size, corrosion depth or vibration signature updates the uncertain degradation state. Measurement error and probability of detection must be included. A no-find NDT result is evidence, but not proof of zero damage. Repeated inspections can progressively narrow uncertainty and reveal whether the actual degradation rate differs from the prior model. The updated posterior state should then be propagated to the next decision point. This provides a rational basis for inspection intervals and for extending life when evidence supports it.

Maintenance, repair and intervention models

Maintenance changes the future reliability trajectory and should be represented explicitly. A coating renewal may reset corrosion rate but not restore lost thickness; a repair may restore geometry with its own workmanship uncertainty; a component replacement may effectively reset the age but not the surrounding system. Reliability-centred maintenance studies compare intervention options using risk, cost and availability, not simply expected life. For safety-critical applications, the decision criteria will normally include minimum reliability or deterministic code requirements that remain mandatory even when a probabilistic lifecycle model is used.

Competing modes and system ageing

Ageing structures often have several failure modes: corrosion can reduce static strength and simultaneously accelerate fatigue; creep can interact with fatigue; degradation of supports can redistribute load to other components. System reliability should preserve these modes rather than collapsing them too early. Common environmental exposure creates dependence between components, so assuming independent degradation may exaggerate the benefit of redundancy. Scenario modelling, fault trees, system simulation or multi-state models can be appropriate depending on the architecture and data.

Verification and lifecycle decisions

Verify the deterministic resistance model, degradation law, numerical time stepping and limiting cases before relying on lifecycle probabilities. Compare predicted degradation with historical inspections and back-cast known assets where possible. Report Pf or reliability versus time, state-variable quantiles, dominant sensitivities and the effect of planned inspections or repairs. State clearly whether uncertainty grows because of inherent future variability or because knowledge is poor. The strongest lifecycle assessment links each probability result to an actionable decision: inspect, repair, reduce load, collect data, or accept continued operation to a defined date.

Do not model “age” as a generic percentage knock-down when the underlying degradation mechanism can be represented directly. Mechanism-based state variables make inspection data and maintenance actions meaningful.

Engineering judgement — governing sensitivities

For Reliability of Ageing & Degrading Structures, the most useful review question is not simply whether the solver has produced a plausible contour or scalar result, but whether the model preserves how stochastic load, resistance degradation, inspection information and time correlation combine. Treating yearly states as independent can overstate the rate at which risk accumulates. This is where apparently small modelling choices can change the engineering conclusion. The analyst should identify the variables that can move the governing response, separate physical uncertainty from deliberate conservatism, and show that the selected modelling fidelity is proportionate to the decision being supported. Where the response is close to an acceptance boundary, sensitivity cases should bracket credible changes rather than apply arbitrary percentage perturbations.

Key takeaways

• Ageing reliability requires time-dependent resistance and demand, not repeated unrelated snapshots.
• Inspection and maintenance should update or change the degradation state explicitly.
• Competing degradation modes and common environmental exposure can make component reliabilities strongly dependent.