Why Vesselinspect

What robotic inspection delivers today

Several projects and vendors are developing autonomous robotic systems for vessel inspection, including aerial drones and crawlers for ballast tanks, cargo holds, and confined spaces. Initiatives like the AUTOASSESS project aim to replace human surveyors in hazardous environments with fully autonomous inspection systems, reducing inspection time dramatically (from 15 days to 3 days in some estimates) while improving safety and data consistency. ¹

Current systems focus on visual imaging with AI for defect detection (corrosion, coating breakdown, cracks) and basic 3D mapping. Some incorporate ultrasonic thickness measurement (UTM) tools, but these are often separate workflows: visual data identifies surface issues, while quantitative thickness readings are taken at limited spots chosen manually or by simple rules. The output is typically a set of images, defect annotations, and isolated thickness readings — valuable, but fragmented.

This is important progress toward removing humans from dangerous confined spaces. But the data streams remain separated. Visual imagery locates surface defects and corrosion patterns but lacks penetration depth. Ultrasonic measurements provide wall-thickness data but without rich spatial context or correlation to visual degradation. Prioritization of measurement points is limited by robot flight/operation time, leading to incomplete assessments, missed trends, and reports that require extensive manual reconciliation by class surveyors.

What Vesselinspect adds

Vesselinspect starts from the same autonomous robotic image and measurement acquisition but adds a unified structural integrity interpretation layer. Three capabilities distinguish it from fragmented visual-plus-measurement approaches.

Classification-supported structural mapping, developed in collaboration with IACS member societies and shipyards, aligns robotic imagery and UTM data with vessel CAD-based structural models and class drawings. Each finding — whether a visual corrosion patch or a thickness reading — is placed in its exact structural zone (e.g., ballast tank plating between frames 12–15, longitudinal stiffener, or cargo hold hopper plate) rather than appearing as an isolated coordinate or image tag. Risk scoring then combines visual severity, thickness deviation from class minima, and the zone’s engineering importance (load path, fatigue history, corrosion-prone geometry).

Vessel-class-wide learning exploits the consistency of structural designs across sister vessels and vessel classes. By aggregating findings from multiple vessels of the same type (or operator fleet), the platform builds degradation baselines for every structural zone. When a new finding appears — say accelerated pitting near a ballast tank manhole — Vesselinspect compares it against how similar findings in the same zone have progressed across dozens of other vessels to forecast stability or growth toward class limits. Systematic codification of experienced surveyors’ and naval architects’ tacit knowledge sharpens these forecasts.

Modular multi-modal sensing fuses qualitative imaging with quantitative measurements from the start. The Phase 1 payload combines high-resolution RGB/ multispectral imaging for visual corrosion and coating assessment with optional radiometric thermal for subsurface moisture or delamination detection. Phase 2 integrates targeted ultrasonic thickness measurement (or pulsed eddy current) on AI-ranked hotspots, plus hyperspectral for objective coating chemistry. All modalities feed the same structural model and historical record, producing unified condition maps rather than parallel reports.

Together, these capabilities convert robotic inspection data into class-ready, forward-looking maintenance intelligence. Operators and class societies gain objective, auditable records for surveys, precise hotspot prioritization that maximizes limited robot time, trend-based predictions of when zones will require intervention, and data-driven coating maintenance programs.

Side-by-side comparison

Classification-supported structural mapping

Vesselinspect’s collaboration with classification societies (DNV, ABS, Lloyd’s Register and other IACS members) and shipyards integrates detailed vessel structural models, class drawings, and survey procedures directly into the inspection workflow. Every robotic-captured image and UTM reading is aligned with the vessel’s internal architecture, placing each finding in its exact structural zone — ballast tank bottom plating, side shell longitudinal, cargo hold frame bracket, or deck girder — rather than only on a visible surface mesh.

Once located within its zone, the system applies risk scoring that combines three inputs: visual severity of the defect (corrosion area, crack length, coating breakdown), quantitative deviation from class minimum thickness (per IACS UR-Z10 and vessel-specific rules), and the engineering importance of the zone (primary load path, high-stress detail, historical fatigue hotspot). A minor coating blister in a low-load fairing receives different priority than the same finding on a ballast tank boundary plating near a critical stiffener. The resulting report links each finding to the relevant class rule references, allowable damage criteria, repair recommendations, and survey task mappings specific to that zone — eliminating hours of manual cross-referencing traditionally required by surveyors and superintendents.

Existing robotic inspection approaches position findings on 3D meshes derived from photogrammetry or LiDAR, providing good traceability for visual surveys. They do not, however, incorporate the deeper class-approved structural dataset, zone-importance weighting, or direct mapping to IACS UR-Z10 thickness measurement requirements that Vesselinspect derives from society and shipyard engineering data.

Because Vesselinspect maintains a consistent structural model across an entire vessel class (or operator’s sister-ship fleet), the same framework applies unchanged to every vessel of the same type. This consistency is the prerequisite for vessel-class-wide learning.

Vessel-class-wide learning and predictive intelligence

Because Vesselinspect maintains a single structural model across every vessel of a given class or fleet segment, the platform can aggregate data beyond individual inspections.

Each inspection adds time-stamped visual findings and UTM readings to a longitudinal record for every structural zone. Over repeated robotic surveys on multiple vessels, the system calculates zone-specific corrosion and degradation rates, identifies repeat problem areas (for example, accelerated pitting at ballast tank manholes or coating failure at cargo hold hopper knuckles), and quantifies progression trends. Bayesian methods continuously refine risk predictions by combining fresh robotic data with the accumulated fleet baseline: a thickness loss rate that took 24 months to approach class limits on 25 sister vessels sets a concrete expectation for the current one. The consistent class structural model ensures data from one vessel is directly comparable to every other of the same type, producing reliable class-level statistics without manual re-alignment.

This data-driven analysis is complemented by systematic collection of surveyors’ and naval architects’ tacit knowledge. Through structured interviews, Vesselinspect codifies the expertise that experienced professionals apply but rarely document: subtle visual cues for early corrosion under intact coatings, pattern recognition for moisture-trap geometries, and judgment calls on when a finding warrants close-up survey or immediate repair. With an aging surveyor workforce, embedding this knowledge in a searchable database before retirement turns individual experience into a durable fleet resource.

The practical outcomes are direct. Class surveyors and superintendents receive prioritized work lists that concentrate limited robot operation time and human review on the highest-risk zones, with progression forecasts showing which findings need attention before the next special survey and which can be monitored. Coating maintenance decisions draw on objective, fleet-wide wear data that supports life-cycle cost and fuel-efficiency calculations (clean hulls, but here focused on internal tank integrity). For class surveys, lease transitions, and sale/purchase, the platform produces auditable, class-aligned condition reports covering each zone’s history and projected trajectory.

Current robotic inspection initiatives already improve safety and basic documentation for confined-space surveys. Vesselinspect extends this by embedding class structural models, multi-modal fusion (imaging + UTM), and fleet data aggregation into the analysis layer. The result is a shift from reacting to visible defects and sparse thickness readings on individual vessels to anticipating class-level integrity issues before they affect vessel availability, safety, or residual value — precisely the unified qualitative-and-quantitative assessment demanded by AUTOASSESS Challenge 5.

Beyond visible light and isolated measurements

Current vessel inspection approaches — whether manual or early robotic — treat visual imaging and ultrasonic thickness measurement as largely separate activities. Visual AI detects surface corrosion and coating anomalies; UTM is performed at pre-defined or randomly selected spots, often without tight coupling to the visual record. Defects beneath intact coatings (early corrosion, blistering, moisture in lap joints or behind frames) remain invisible until they produce surface symptoms or are caught by chance thickness readings.

Vesselinspect’s modular payload architecture addresses this from day one by design for fused multi-modal data.

Phase 1: Visual + thermal triage for confined spaces. High-resolution RGB imaging (core) is augmented by optional radiometric thermal cameras on the same robotic platform. In the stable temperature environment of a ballast tank or cargo hold, delamination, coating disbonds, and moisture pockets produce measurable thermal signatures under passive conditions. The AI triage layer fuses RGB corrosion mapping with thermal contrast: a visual anomaly that also shows subsurface thermal signature receives higher priority for targeted UTM follow-up. Lightweight drone- or crawler-ready thermal payloads add negligible operational time.

Phase 2: Confirmatory quantitative metrology and spectral assessment. For zones flagged by visual + thermal triage, Phase 2 introduces targeted ultrasonic thickness measurement (or adapted pulsed eddy current) to convert visual flags into precise remaining-thickness values against class minima. Snapshot hyperspectral imaging (400–1000 nm) captures spectral signatures distinguishing coating composition, UV/chemical degradation, and early substrate corrosion indicators beneath intact paint — providing objective chemistry data to train the visual AI layer. Both enter as bounded, high-value-zone pilots that build ground-truth datasets before wider deployment.

Thermal, ultrasonic, and spectral sensing are not yet standard in commercial vessel robotic inspection, though research platforms are exploring combinations. By designing for native multi-modal fusion tied to class structural models, Vesselinspect delivers the unified assessment framework that Challenge 5 explicitly requires: merging visual and ultrasonic datasets with expert and historical knowledge to produce actionable, class-compliant structural integrity intelligence.

References

1. AUTOASSESS Project (2026). Open Call #2 Guide for Applicants – Challenge 5: Combined Qualitative and Quantitative Assessment of Structural Integrity Using Imaging and Measurements. autoassess.eu
2. IACS (2023). UR-Z10 – Hull Surveys of Oil Tankers, Chemical Tankers, Bulk Carriers, etc. (Rev. 34). International Association of Classification Societies.
3. DNV (2025). Rules for Classification of Ships – Part 7: Surveys and Maintenance. DNV AS.
4. ABS (2025). Rules for Survey After Construction. American Bureau of Shipping.
5. IMO (2024). MSC.1/Circ.1649 – Guidelines for the Use of Remote Inspection Technologies (RIT) in Surveys.
6. Voliro (2025). Drone-Enabled Pulsed Eddy Current (PEC) Sensor for Wall Thickness Measurement. voliro.com
7. Cubert (2025). Ultris X20 Plus — Hyperspectral Camera for UAV/UGV Mapping. cubert-hyperspectral.com
8. DJI (2025). Zenmuse H30 Series — Multi-Sensor Payload with Thermal Imaging. dji.com/zenmuse-h30-series
9. Polanyi, M. (1966). The Tacit Dimension. University of Chicago Press.