The Problem with Calendar-Based Inspection Cycles
Most offshore wind operators run inspection programmes on fixed annual or biennial schedules. Every foundation in the array gets surveyed on the same calendar cycle regardless of condition history, prior findings, or environmental exposure differences across the site. This is a sensible default for new arrays where no condition history exists, and it satisfies the baseline requirement under most classification society survey schemes (the DNV Offshore Standard for floating wind and BSI PAS 8700 for fixed offshore wind both use period-based survey intervals as the entry-level requirement).
The limitation becomes apparent as arrays mature and condition data accumulates. In a large offshore wind array of 50 to 150 foundations, the structural condition of individual monopiles diverges over time. Foundations in higher wave-loading positions, those with documented corrosion findings from previous surveys, or those with identified CP system concerns occupy a genuinely different risk category than foundations with clean records across multiple survey cycles. Surveying all of them at the same interval is an allocation inefficiency. It also means that some high-risk foundations get surveyed no more frequently than low-risk ones, which is exactly backwards from a risk-based perspective.
The shift from calendar-based to condition-based or risk-based resurvey scheduling is well-established in principle within the integrity management literature, and classification societies including DNV allow operators to move toward this model under so-called Enhanced Survey Programme (ESP) equivalency arrangements, subject to demonstrating that the condition monitoring data quality is sufficient to support risk stratification. The practical barrier has always been the quality of inspection data, particularly how systematically anomaly findings are recorded and compared across survey cycles.
What Structured Anomaly Detection Outputs Enable
The output of an automated anomaly detection process is a structured inventory: each finding has a location identifier, a classification type, a severity indicator, and a reference back to the source frame in the survey footage. When the same foundation is surveyed in successive years and detection is applied consistently, the outputs are comparable in a way that human-described narrative findings often are not.
This comparability is what makes risk-based resurvey scheduling tractable. The questions a condition-based maintenance model needs to answer are: Is the anomaly population at this foundation growing? Are existing findings from prior surveys progressing, stable, or showing improvement? Are there new finding types appearing that were not present in earlier surveys? These questions require structured records with stable classification vocabularies and consistent location referencing, which is precisely what an automated detection pipeline produces.
Consider a simplified but realistic scenario. A typical offshore wind operator has been running annual surveys with Ecodetect processing for two years across a 60-foundation array. At the end of year two, the anomaly inventory contains 120 survey-cycles of structured data. A straightforward analysis of anomaly density per foundation, finding severity distribution, and delta between Year 1 and Year 2 findings produces a ranked list of foundations by condition trajectory. The top decile, perhaps six foundations showing new findings or escalating severity, clearly warrant a 12-month resurvey interval. The bottom quartile, foundations with no findings across two clean surveys, could reasonably extend to 24 or even 36 months under a demonstrated condition-monitoring regime.
How Anomaly Density Translates to Resurvey Risk Scoring
The mechanics of translating anomaly detection outputs into resurvey risk scores depend on the specific risk model an operator is running, and we are not prescribing a particular scoring methodology here. Different classification societies have different frameworks, and individual operators have CMMS configurations that reflect their own engineering judgment. What we can describe is how detection outputs map into common scoring inputs.
Finding count and severity distribution is the most direct input. A foundation with five moderate corrosion findings, one structural crack finding at high confidence, and two CP depletion flags presents a very different risk profile than one with a single low-severity marine growth finding. Most risk scoring frameworks weight findings by type and severity; automated outputs provide both attributes systematically.
Finding progression between surveys is equally important. A corrosion finding that has been recorded at moderate severity for three consecutive surveys without progression is a different maintenance priority than one that advanced from minor to moderate in the last survey interval. Automated detection produces comparable records across surveys; calculating deltas between periods is straightforward when the underlying data is structured consistently.
Location and clustering also matters. Isolated findings at random foundation positions are a lower concern than findings clustering on foundations at the windward edge of the array, which may indicate a common load or environmental exposure pattern requiring broader investigation.
What This Approach Cannot Do on Its Own
Risk-based resurvey scheduling based on anomaly detection outputs is a useful tool, not a complete integrity management solution. A few limits are worth being explicit about.
Detection-based models identify what is visible. Subsurface or internal structural conditions cannot be assessed from ROV visual inspection alone, regardless of how automated the processing is. The classic example is fatigue crack initiation at welds, which may not produce visible surface indications until a crack is well-developed. Condition-based scheduling derived from visual anomaly records should always be considered alongside other integrity indicators, including stress analysis, fatigue life calculations, and non-destructive testing where indicated.
The model also cannot substitute for the engineer's judgment about site-specific factors. A foundation in a documented scour-prone location or one that experienced an impact event (vessel contact, dropped object) has a risk modifier that is not captured by its surface anomaly record. Resurvey scheduling decisions should incorporate these modifiers, which typically require input from the operations and vessel management teams rather than from inspection records alone.
And there is a data accumulation delay. The first year of automated detection produces a baseline. The second year produces the first progression data. A sound risk stratification requires at least two comparable survey cycles, ideally three or more. Operators shifting from calendar-based to condition-based scheduling during this accumulation phase should phase the transition gradually, maintaining conservative default intervals for foundations without a multi-year detection record until that evidence base exists.
Getting Started Without Overhauling Your Programme
The practical path toward condition-based scheduling does not require restructuring an entire inspection programme at once. The most straightforward starting point is to begin accumulating structured detection outputs from existing annual surveys, using the first two to three years of data as evidence building rather than immediately changing any survey intervals. During this period, the anomaly inventory becomes the database that will eventually support the risk scoring model.
At the two-year mark, the first comparison analysis is possible. Many operators use this to identify the clear outliers at both ends of the risk distribution: the few foundations with consistent clean records that are natural candidates for interval extension, and the few with escalating findings that may warrant more frequent monitoring. Making interval adjustments for these clear cases is a defensible first step, provided the operator can demonstrate to their classification surveyor that the detection data quality and consistency supports the decision.
