Structural Integrity Intelligence for Commercial Vessels

Executive Summary

This document defines Vesselspect as a structural assessment workflow for commercial vessels. It covers inspection planning, structured data capture, sensor-aware fusion of visual evidence and ultrasonic thickness measurements (UTM), and expert-reviewed decision support for ballast tanks, cargo holds, and similar hard-to-access steel spaces. Remote and robotic capture can reduce exposure and improve repeatability, but the core value is the survey-zone model that turns fragmented inspection inputs into a persistent condition record. Where drawings, structural references, or 3D context are available, the workflow links imagery, UTM, repair history, and operational context to the same structural member. The result is condition maps, high-risk-zone prioritization, reporting support, and draft repair scopes with traceable reasoning. Final survey decisions and class sign-off remain with certified surveyors and the relevant classification process.

Initial deployment targets vessels subject to close-up survey and thickness-measurement regimes under class rules, starting with ballast tanks and cargo holds where visual evidence and thickness history are most often fragmented. The workflow is designed to operate with remote inspection providers, robotic platforms, approved UTM suppliers, shipyards, and operator records rather than assuming a single hardware stack or exclusive data channel. Access to structural drawings, survey records, and historical thickness logs materially improves the outputs, but the system can also start from survey-zone references and build a stronger longitudinal record over time. This is the unifying framework: combine long-term visual and thickness measurements, identify high-risk zones, validate against historical and real-vessel or industrially relevant data, and support class-aligned reporting without claiming automatic approval.

The founding team consists of three domain specialists whose expertise maps directly onto the product requirements for combined visual-and-measurement assessment, expert-guided corrosion interpretation, and class-compliant reporting. A mathematician with active work in optimal control, degradation modeling, and fleet-wide Bayesian inference contributes route optimization for limited robot endurance, corrosion progression models, and statistical prediction of when zones will reach class limits. A physicist contributes physically grounded computer vision and photometric/spectral surface models for corrosion and coating assessment, plus measurement uncertainty quantification for UTM fusion. An aerospace-turned-maritime engineer with CAD, robotics, and structural experience contributes the integration layer that maps robotic data to class structural zones, interprets findings against IACS rules and fatigue-critical details, and ensures outputs are surveyor-ready. These are not generic computer-vision or robotics roles: each requires sustained domain expertise that shapes the product from acquisition through class-compliant decision support.

The business model combines software subscriptions, project-based integration, and inspection-support services. Distribution can run through classification-society digital teams, approved service suppliers, shipyards, and operator technical organizations rather than relying on direct hardware sales. The product is positioned as the inspection-to-decision layer that complements remote inspection hardware and strengthens existing survey workflows.

Phase 1 begins with a bounded pilot on one vessel segment, using remote or robotic visual capture plus targeted UTM confirmation under full surveyor oversight. Additional modalities such as thermal, hyperspectral, or local geometry remain secondary, sensor-aware extensions rather than the core promise. The architecture is vessel-type extensible and designed to work across class societies, operators, shipyards, and approved service providers wherever equivalent structural references and inspection records are available.

Problem and Market Opportunity

The inspection problem

Commercial vessel structural surveys in ballast tanks and cargo holds remain overwhelmingly manual and hazardous. During special surveys, intermediate surveys, and close-up inspections required by class rules (IACS UR-Z10), certified surveyors must enter confined spaces, often on scaffolding or staging, to perform visual examination and ultrasonic thickness measurements. Each survey is treated as a fresh exercise: the surveyor documents what is visible and measures at selected spots, but that record is rarely registered to precise structural geometry, rarely compared systematically against prior surveys of the same vessel, and almost never aggregated across sister vessels or the class fleet for trend analysis.

This process has four critical weaknesses:

1. Dangerous and access-constrained. Ballast tanks and cargo holds are confined, poorly lit, often corroded, and may contain residual hazardous atmospheres. Human entry carries well-documented risks; the practical goal is to reduce unnecessary entry and focus close-up access on the zones that genuinely require human attention.

2. Fragmented data streams. Visual inspection and UTM are performed as parallel or sequential activities. Visual imagery (even with AI defect detection) identifies surface corrosion, coating breakdown, and cracks but provides no depth information. Ultrasonic readings give quantitative remaining thickness but lack spatial context and correlation with visual patterns. The result is incomplete evaluations, missed degradation mechanisms, and reports that require hours of manual reconciliation.

3. Non-repeatable and non-persistent. Two surveyors examining the same tank on the same vessel will frame, light, and document findings differently. There is rarely a persistent, zone-based condition record that accumulates visual + thickness history across multiple surveys and vessels.

4. No class-level or fleet learning. Each vessel is surveyed in isolation. Patterns that would be visible across hundreds of sister ships — zone-specific corrosion rates by operating profile, coating performance by cargo type, repeat problem details at manholes or hopper knuckles — remain invisible because data never leaves individual survey reports and is never geometrically registered to a common structural model.

Current remote and robotic inspection tools still inherit this fragmentation: they can deliver imagery and limited UTM, but they often stop at capture and basic annotation. The missing piece is the unified methodology, expert-infused models, historical-data fusion, and class-aligned outputs that turn raw inspection data into actionable structural integrity intelligence.

Why now

The hardware for remote and robotic inspection of confined vessel spaces has matured rapidly. Recent research and commercial developments demonstrate GPS-denied aerial robots and crawlers that can operate in ballast tanks and cargo holds, capturing high-resolution imagery and deploying UT probes under remote supervision. This commoditization of capture shifts the value from the robot itself to what happens after the data is collected: precise structural registration, multi-modal fusion (imaging + UTM), degradation modeling, expert knowledge integration, and class-rule-mapped reporting. The competitive frontier is no longer “can a robot enter a tank” but “what does the combined visual + quantitative dataset tell you that a traditional survey cannot — and how does it stay within class acceptance?”

Market opportunity

The addressable market is every commercial vessel that undergoes class-mandated close-up surveys and thickness measurements: effectively the global fleet of oil tankers, chemical tankers, bulk carriers, and other vessels subject to IACS UR-Z10. The initial target is vessels already engaged with forward-leaning classification societies and operators participating in digitalization pilots, where the value of persistent digital records and predictive maintenance is most immediately recognized.

The per-vessel value of a structural integrity intelligence subscription comes from three sources: reduced unnecessary human entry time and risk during surveys, consistent and traceable documentation that reduces rework and manual reconciliation, and better prioritization of zones that need close review, confirmation, or repair planning. Any change to survey scope or interval remains subject to the attending surveyor and relevant classification-society approval. As the platform accumulates inspection history across vessels and survey cycles, the dataset becomes increasingly valuable for operators, shipyards, service providers, and classification stakeholders trying to understand recurring degradation behaviour.

The product architecture is not tied to a single vessel type, class society, or capture vendor. Any collaboration that provides appropriate structural references, survey procedures, and historical thickness records can enable the same platform on additional vessel segments. The initial program is an entry point; the long-term market is structural integrity intelligence across commercial shipping.

Product: Inspection-to-Decision Structural Assessment Workflow

The product has three layers. The first organizes inspection delivery and data capture. The second builds structural understanding and longitudinal records from that data. The third converts that understanding into class-aligned maintenance and survey decisions. Each layer adds value the one below it cannot provide alone, and together they deliver a unified methodology that merges visual and ultrasonic datasets with operational expertise and long-term records.

Layer 1: Inspection Delivery and Structured Data Capture

Vesselspect covers the inspection workflow from planning through structured data capture. Evidence may come from aerial robots, crawlers, approved remote-inspection providers, or manual survey teams, but it is organized around repeatable survey zones rather than ad hoc files, photos, and spreadsheets. The objective of Phase 1 is not to introduce a new hardware category; it is to ensure that visual evidence, thickness measurements, and metadata are captured in a form that can be fused reliably later.

The capture stack combines route or coverage planning, image-quality checks, zone registration, and confirmation logic. High-resolution imagery and video provide the baseline visual record. Certified UTM remains the quantitative reference for absolute thickness. Optional supporting signals such as thermal, PEC, hyperspectral, or local geometry can be used where they add value, but each modality is treated according to what it can and cannot reliably support. That sensor-aware stance is essential in confined maritime spaces, where lighting, coating condition, residues, access geometry, and endurance limits all affect evidence quality.

Phase 1 therefore focuses on repeatable capture, coverage gaps, and traceability: blur and low-contrast detection, missed-zone flags, recapture prompts, and precise registration of measurements to the intended structural member. The result is a structured inspection record that supports later corrosion analysis rather than a one-off media archive.

The Phase-1 inspection package covers the core requirements of a close-up survey plus thickness-measurement scoping:

• Repeatable full-surface or critical-zone image capture with structural registration
• Digital reporting mapped to survey-relevant structural zones and applicable survey concepts
• Coating condition, corrosion, crack, and deformation screening
• Targeted UTM on ranked hotspots, with manual confirmation where required
• Optional thermal, PEC, spectral, or local geometry cues as secondary evidence where validated

These sensing modes are staged rather than assumed all at once. The baseline visual + UTM workflow comes first; extra sensors are added only where they improve decisions and can be validated against the operational workflow.

Layer 2: Structural Integrity Software

Repeatable capture is necessary but not sufficient. The software layer converts recurring inspections into a persistent structural record that improves as more evidence is registered to the same zones over time.

The first software component is the survey-zone fusion layer. Rather than treating each survey as an isolated event, the product aligns visual findings, certified thickness readings, repair history, and operational context to the same structural members; flags gaps or conflicting evidence; and tracks progression across repeated surveys. Near-term outputs include:

• Repeat change maps and corrosion progression flags
• Thickness maps color-coded against applicable class-society criteria and vessel-specific baselines
• Early corrosion-risk indicators based on recurrent coating failure at known moisture-trap geometries (manholes, brackets, lap joints)
• Trend reviews that show whether a zone appears stable, spreading, or accelerating and when it may require closer attention

Deep learning detection and corrosion-type classification form the computational backbone of the visual screening layer. Semantic segmentation architectures (U-Net, DeepLab variants) and object-detection models (Mask R-CNN, YOLO variants) delineate corrosion patches, coating breakdown, and structural anomalies at pixel level from robotic imagery. Spatiotemporal models that exploit attention across consecutive frames are particularly useful for confined-space drone footage, where sequential overlap provides signal that single-image classifiers miss under the variable lighting and specular reflections typical of ballast tank environments. The closest vessel-specific spatiotemporal reference we found, Azad et al., studies underwater hull video rather than dry ballast tanks, so we use it only as an adjacent reference for sequential ship-inspection video analysis. ¹¹ The detection layer classifies findings by corrosion mechanism — since type governs both progression rate and structural significance — across the principal categories present in vessel tank structures: general wastage (broad uniform thinning, predictable from class allowance tables and UTM trends); pitting corrosion (localized deep attack at weld toes, ballast tank corners, and coating-breakdown points, assessed against IACS UR-Z10 allowable pit density and depth criteria ²); grooving corrosion (along welds and stiffener-to-plate intersections, associated with residual stress); crevice corrosion (at lap joints, bracket slots, and structural connections under differential aeration); galvanic attack (near sacrificial anodes and dissimilar-metal contacts); and microbiologically influenced corrosion (MIC, in partially flooded tank areas where anaerobic sulphate-reducing bacteria accelerate local wastage, typically at rates significantly exceeding purely electrochemical corrosion). Each category maps to a different degradation model in the fusion layer and a different urgency threshold in the repair specification engine. The detection models are further modulated by operational context — trade route, cargo type (acidic bulk cargoes, ore, grain, or chemical), ballast-water source and cycling frequency, and vessel age — all of which the platform ingests from operational records to condition corrosion rate priors for each structural zone.

These are survey-planning and maintenance cues for certified surveyors, not autonomous repair decisions. The mathematical foundation — corrosion degradation models calibrated to real vessel operating profiles, fleet-wide Bayesian inference that treats the entire class as a statistical prior, statistically grounded zone-risk scores — is where the founding team’s applied-mathematics expertise directly shapes the product.

This quantitative analysis is complemented by systematic codification of surveyors’ and naval architects’ tacit knowledge. Through structured interviews with veteran class surveyors and shipyard repair specialists, the platform captures subtle visual cues for early subsurface corrosion, pattern recognition heuristics for high-risk details, and judgment calls on when a finding warrants immediate engineering referral versus continued monitoring. With an aging surveyor workforce, embedding this hard-won knowledge in a searchable database converts individual experience into a durable class and operator resource. The result sharpens the platform’s risk models and triage logic with judgment that data alone cannot provide.

The second software component is a zone-based condition model with 3D context where available. It combines registered multi-modal imagery, structural references, local geometry from reconstruction or prior scans, historical UTM logs where available, and expert annotations into a living record. Its purpose is historical comparison, progression forecasting, and survey-scope optimization. Over time it becomes the place where surveyors review recurring hotspots, compare pre- and post-repair condition, and see not only whether a finding is isolated, stable, or progressing but also what evidence is missing and what follow-up is justified.

Layer 3: Inspection-to-Decision Engine

The third layer transforms analysis into operational and survey action. Instead of outputting a list of detected defects and raw thickness readings, the product produces a priority-ranked worklist where each suspect zone carries a recommended next step: visual review by surveyor (remote or on-site), repeat capture at closer standoff or different lighting, targeted UTM or PEC confirmation, manual dry-film-thickness or pit-depth gauging, or engineering referral for repair assessment. Structural context from drawings and survey-zone references, applicable thickness criteria, and progression forecasts from vessel and fleet baselines jointly determine urgency - a zone whose thickness loss trajectory suggests it may breach criteria before the next special survey is flagged differently from one that available history shows as stable.

A detection and fusion layer grounded in corrosion science and surface physics combines image findings with geometric context, UTM values, and structural significance. A coating blister near a fatigue-critical bracket in a ballast tank carries different operational weight than the same blister on a non-structural longitudinal in a cargo hold. Structural references, applicable survey concepts, and expert rules make that distinction computable and auditable. The most useful Phase-1 output is not an isolated bounding box or thickness value but a localized review package that links each suspect region to its structural reference (frame, stringer, plating strake), measurement confidence, prior survey history, relevant rule context, and recommended next action.

Repair specification generation translates confirmed structural findings into draft engineering documents that surveyors and shipyard repair teams can review and refine. For zones where thickness has been confirmed below the applicable class-society criteria for that structural member, the system can assemble a structured renewal draft: suspected renewal extent referenced to frame bays and strake boundaries; surface-preparation and coating notes drawn from the vessel's maintenance context; and post-repair checks needed to confirm the intervention was executed and documented. For cracks or deformations, the draft flags the defect category and highlights where engineering review or a structural fitness-for-service assessment is required before repair scope can be finalized. These are structured starting points for human review, not autonomous class decisions. By keeping the evidence, structural zone, and proposed next steps together, the output reduces manual cross-referencing and makes maintenance guidance easier to review.

AI in this product is tuned for triage, hotspot prioritization, and surveyor decision support — not autonomous class sign-off. Review literature and current class practice support AI-assisted detection and data fusion, but they also show weak generalization across coating types, repair histories, variable tank conditions, and different vessel operating profiles. The mitigation is disciplined ground-truth collection during the pilot (paired robotic + manual surveyor findings on the same zones), operator- and class-specific fine-tuning, first-principles priors from corrosion physics and class rules that reduce dependence on purely data-driven pattern matching, and explicit human review thresholds for every finding category. Final survey decisions and class recommendations remain with certified surveyors.

Inspection Boundaries and Confirmatory Metrology

A coherent product requires keeping modalities separate because they use different evidence, different instruments, different approval paths, and different class acceptance criteria:

1. Visual and multispectral coating/corrosion screening: image-based, the core Phase-1 product. Qualitative assessment of coating condition, corrosion extent, cracks, and deformation.

2. Thermal subsurface triage: radiometric infrared for delamination, disbonds, and moisture detection behind coatings or in lap joints; optional in Phase 1 as a modular payload.

3. Quantitative thickness measurement: targeted UTM on zones ranked by visual + thermal triage, converting visual flags into precise remaining-thickness values against class minima; pulsed eddy current (PEC) as a drone-deployable pre-screening modality that quantifies relative wall loss — enabling rapid triage of high-wastage locations through coatings and without couplant — with class-accepted absolute thickness values then confirmed by conventional UTM at PEC-flagged hotspots. ⁷ Phase-1 uses manual or semi-automated probe deployment with full surveyor oversight; Phase 2 moves to integrated robotic deployment where validation permits.

4. Spectral coating-chemistry assessment: hyperspectral imaging (350–1000 nm, covering VNIR spectral range across both Cubert Ultris X20 Plus and Resonon Pika L class payloads) for objective differentiation of coating composition, degradation state, and early chemical indicators of substrate corrosion beneath intact coatings; Phase-2 bounded pilot on high-value zones to build ground-truth for AI training. ^{8 12}

5. Local geometric assessment: 3D reconstruction or structured-light for deformation, buckling, or dent mapping on flagged structure; optional workflow.

Public and project evidence supports full-zone visual screening now, passive thermal as a mature optional augmentation, and targeted UTM as the essential quantitative step already required by class rules. The innovation is not inventing new sensors but fusing them natively inside the class structural model with expert and historical context, so the output becomes a combined qualitative-and-quantitative assessment rather than a stack of disconnected inspection artifacts.

The practical workflow follows naturally:

1. Inspection workflow captures visual evidence and, where used, thermal or other supporting cues for the selected tank or hold.
2. AI triage (visual + thermal + structural zone importance) ranks zones for quantitative follow-up.
3. Robot or surveyor deploys UTM probe at highest-priority locations; readings are automatically registered to the structural model.
4. Fusion engine updates the digital twin, applies degradation models, and generates the ranked worklist with class references and forecasts.
5. Surveyor reviews the package remotely or on-site, requests any needed confirmation, and issues the report through the normal class process.

Competitive Landscape

The market around remote vessel inspection and structural assessment is still emerging, driven by safety, labor constraints, and the growing availability of digital inspection data. Relevant players and their positioning relative to Vesselspect are as follows.

Research projects and robotics developers in this space focus on hardware, navigation in GNSS-denied environments, and baseline visual defect detection. Their strength is domain-specific engineering for ballast tanks and cargo holds. Their limitation, and the gap this proposal addresses, is the absence of a production-grade assessment workflow that fuses visual and UTM data with historical records, expert knowledge, and predictive models. Vesselspect is positioned as the complementary inspection-to-decision layer that turns captured inspection data into structural assessment outputs.

Commercial remote-inspection vendors (e.g., those offering drone or crawler systems for tanks) typically provide visual mapping, basic AI annotation, and optional UT payload integration. Most still treat visual and thickness data as parallel outputs rather than a fused, zone-registered, longitudinal record. They may not provide the expert-informed reasoning and decision-support layer that sits between raw multi-modal data and survey action. Vesselspect does not compete primarily on robot hardware; it competes on the workflow intelligence that connects capture, fusion, and reporting.

Classification societies’ own digital initiatives are increasingly exploring remote inspection technologies (RIT) and AI-assisted tools. Societies already accept certain remote-inspection evidence under controlled conditions, provided human oversight and validation are maintained. Collaboration with societies, approved providers, and shipyards can make the workflow easier to deploy, but Vesselspect is framed as an enabler that strengthens existing survey workflows rather than depending on one exclusive channel.

Traditional MRO / shipyard survey teams remain the baseline. Manual close-up surveys with staging, cherry-pickers, and handheld UT gauges are still the dominant method. The competitive position is not that manual survey is obsolete — it is that manual survey cannot economically produce the repeatable, zone-registered, multi-year, class-fleet-comparable structural integrity record that enables the predictive and scope-optimization value described in this document.

This product does not compete on robotic platforms. It competes on the unified assessment intelligence layer itself — a layer that requires both class data access and the domain expertise to reason about corrosion dynamics, structural significance, and predictive maintenance within class rules.

Competitive Advantage: Data Access, Domain Expertise, and Workflow Integration

The product requires two things simultaneously: access to relevant engineering and survey data, and the domain expertise to convert that data into structurally and regulatorily meaningful integrity intelligence. Neither alone produces the same product.

What class, shipyard, and operator data access enables

Collaboration with classification societies, approved service suppliers, shipyards, and operators can provide four categories of data that no third-party robotic vendor can reliably obtain in isolation:

1. Vessel structural models and class drawings that define actual tank geometry, plating boundaries, stiffener locations, and critical details — enabling routes and measurement sequences derived from real engineering rather than photogrammetric reconstruction alone.

2. Class rule references and survey procedures (IACS UR-Z10, UR-Z7, vessel-specific thickness-measurement guidance, repair criteria) that map every structural zone to its relevant inspection requirements, thickness criteria, and acceptance standards.

3. Historical thickness databases where available that provide the longitudinal baseline per zone and per vessel class - the essential input for degradation modeling and time-to-limit forecasting.

4. Fleet-wide anonymized survey records that, in aggregate, reveal statistical patterns invisible to any single operator or surveyor: which zones corrode fastest under which cargo/operating profiles, which coating systems perform best, which structural details are repeat problem areas across sister ships.

What domain expertise enables

The founding team brings three disciplines that are necessary to exploit class data and that generic robotics or computer-vision engineering does not cover.

Applied mathematics contributes corrosion degradation models calibrated to real vessel data, optimal route planning with formal coverage and endurance guarantees for confined spaces with limited robot flight time, and fleet-wide Bayesian inference that uses the entire class population as a statistical prior for individual-vessel risk assessment.

Physics contributes models of how light and infrared interact with coated steel surfaces under tank lighting and humidity conditions, spectral signatures for coating condition, and rigorous uncertainty quantification for fused visual + UTM measurements — reducing dependence on large labeled datasets and improving generalization across coating ages, cargo residues, and lighting environments.

Maritime structural engineering (adapted from aerospace CAD and robotics experience) contributes the domain knowledge to interpret which findings matter: understanding load paths in ballast tank and cargo hold structure, fatigue concentrations at brackets and hopper knuckles, the relationship between coating breakdown and underlying corrosion rate, and the exact mapping of robotic data to class survey task cards and repair recommendations. Credible engagement with class surveyors, superintendents, and shipyard repair teams depends on this domain fluency.

Why both are necessary

Data without the right expertise is just files. A society, shipyard, or operator could license an existing inspection vendor’s software and supply it with structural models, but without physically grounded corrosion and coating models, degradation mathematics, and class-rule-aware risk reasoning, the result would still be fragmented. Conversely, the team’s expertise without real engineering and survey data produces algorithms with limited regulatory grounding. The product requires both.

Business Model and Revenue

Subscription tiers

The primary revenue model is software-and-services per vessel per year, offered in three tiers that reflect increasing depth of analysis and workflow integration.

Base tier: Inspection planning support, structured capture integration, registered multi-modal imagery and UTM, available structural context, and digital reporting mapped to survey zones and applicable survey requirements. This tier replaces fragmented manual photography and ad-hoc thickness logs with a repeatable, indexed, traceable record. Its value is measured in reduced unnecessary human entry, faster survey documentation turnaround, and stronger evidence traceability.

Professional tier: Adds the structural integrity fusion layer: time-series comparison across surveys, corrosion progression maps, thickness deviation trends, vessel-class statistical baselines, and expert-codified risk scoring. This tier enables superintendents and class to make survey-scope decisions based on how integrity is actually evolving, not just what a single snapshot shows.

Enterprise tier: Adds the inspection-to-decision engine with priority-ranked worklists, forecasted intervention windows, API integration with class survey planning systems and operator maintenance management platforms, and full fleet-level analytics across the operator’s population or other approved datasets. This tier turns the product from a reporting tool into a predictive survey-optimization and integrity-management system.

Channel and integration model

Commercial delivery can combine direct subscription sales with channel agreements and integration partnerships. Potential channels include classification-society digital teams, approved UTM or remote-inspection providers, shipyards, and operator technical departments. The common model is that Vesselspect provides the fusion, decision-support, and reporting workflow while counterparties provide access to inspection events, structural context, or field execution. Commercial structure will vary by channel rather than depend on a single revenue-sharing template.

Expansion revenue

Two additional revenue streams emerge as the platform matures. First, a per-event or per-zone confirmatory-metrology module that adds targeted UTM density, hyperspectral chemistry assessment, or 3D deformation mapping to specific survey events, priced as a supplement to the base subscription. Second, anonymized and aggregated structural-integrity intelligence - corrosion rates by zone type, coating performance by cargo profile, effectiveness of repair strategies - offered to engineering, integrity-management, and rule-development teams subject to data-rights and confidentiality terms.

Go-to-Market and Customer Acquisition

Phase 1: Prove value in one lead deployment

The first deployment is a bounded pilot with one vessel segment and one lead inspection environment, ideally involving a shipyard, operator, approved service provider, or class-linked stakeholder. The goal is not immediate revenue. It is operational proof: demonstrating that the workflow produces repeatable coverage and reliable measurement registration inside real ballast tanks and cargo holds; that the fused digital record is useful and trusted by surveyors; that the AI triage and prioritization reduce rather than add workload; and that the structural risk scoring produces outputs worth reviewing. The pilot runs with full manual or surveyor confirmation on every flagged zone and every UTM reading. Nothing is accepted solely on the basis of automated analysis.

During the pilot, the product team collects paired data — robotic findings alongside traditional surveyor findings on the same zones — to build the ground-truth dataset that validates detection performance, UTM placement accuracy, fusion quality, and false-positive/false-negative rates. This dataset is the foundation for every subsequent claim about accuracy, reliability, and class acceptance.

Phase 2: Expand through partner channels

Once the pilot demonstrates measurable value - reduced unnecessary human entry, improved documentation quality and traceability, surveyor acceptance of the fused digital record, and preliminary evidence that prioritized zones align with actual degradation - the product can expand through operator, shipyard, service-provider, and class-linked digital channels. It can be bundled with existing inspection workflows or sold as a standalone subscription, depending on who owns the inspection event and the data relationship.

At this stage, the platform begins accumulating data across multiple vessels and operators. The vessel-class statistical baselines that differentiate the professional and enterprise tiers become increasingly substantive as more survey events contribute to the dataset.

Phase 3: Expand across vessel types and partner ecosystems

With a stable product on one vessel segment and one lead deployment environment, expansion follows two axes. First, the platform extends to additional vessel types and survey scopes such as intermediate surveys, damage surveys, and post-repair verification. Second, the same architecture - structural model access, rule integration, multi-modal fusion, and fleet intelligence - can be deployed through additional operators, shipyards, approved providers, and class-linked channels without requiring fundamental changes to the product architecture.

Team

The founding team consists of three domain specialists whose backgrounds directly address the technical and regulatory requirements of vessel integrity assessment and the broader market.

The applied mathematician has active research and practical experience in optimal control, route optimization under endurance constraints, degradation modeling, and hierarchical statistical inference. For this product, the specific contributions are mission planning that guarantees coverage of critical zones within robot flight-time limits, corrosion progression models that forecast thickness loss trajectories from sparse UTM + visual data, and fleet-wide Bayesian methods that use class-level statistics as a prior for individual-vessel risk. These tasks require mathematical reasoning about spatial coverage, time-series dynamics, and uncertainty propagation — none of which generic robotics software provides.

The physicist brings expertise in physically grounded computer vision, photometric and spectral surface modeling, and measurement uncertainty quantification. For vessel integrity, this translates to models of how RGB, thermal, and hyperspectral signatures relate to coating condition and early corrosion under real tank lighting and humidity; rigorous fusion of visual and UTM data with calibrated confidence bounds; and reduction of dependence on massive labeled datasets through first-principles priors. Sensor characterization, calibration, and traceability — essential for any metrology-adjacent product that must satisfy class requirements — also fall within this role.

The maritime structural engineer (background in aerospace CAD, robotics, and launch-vehicle structural dynamics) brings the domain knowledge to connect robotic data to class engineering reality: mapping image and UTM coordinates to frame stations, strakes, and stiffener positions; interpreting findings against load paths, fatigue-critical details, and IACS repair criteria; and ensuring that every output is formatted and referenced in a way that surveyors and class can directly use. Credible technical engagement with class surveyors, shipyard superintendents, and regulatory stakeholders depends on this fluency.

Two near-term hires close operational gaps: an MRO / shipyard domain expert who understands confined-space survey procedures, staging logistics, coating repair workflows, and class survey acceptance criteria from direct experience; and a regulatory and class liaison engineer experienced with IACS recommendation development, RIT approval pathways, and EASA/IMO-adjacent processes for digital survey tools.

Roadmap

The roadmap separates three phases because each adds sensing depth, regulatory burden, and operational complexity while building on the same unified fusion architecture.

Phase 1 execution gates (must hold before broader deployment)

• Repeatable coverage and image/UTM quality sufficient for class surveyor review
• Reliable missed-zone detection, recapture logic, and probe placement accuracy (cm-level relative to class model)
• Acceptable false-positive rate for visual + thermal triage, validated on paired robotic + manual data
• Digital reports correctly mapped to class structural zones and IACS rule references
• Surveyor acceptance that the ranked worklist and forecasts add value rather than workload
• Alignment with lead survey stakeholders and class or RO reviewers on pilot procedure and evidence requirements

What success at Phase 1 means

If Phase 1 works, the company holds more than an inspection workflow. It holds a growing, zone-registered dataset of visual + UTM findings, a persistent structural integrity model per tank and vessel, and a predictive engine whose accuracy and value improve as more vessels and survey cycles are recorded. The accumulated multi-modal, zone-aligned, longitudinal data - neither the capture system nor the software alone - is what makes the product difficult to replicate and increasingly useful over time. It is exactly the kind of multi-year visual and UTM history, combined with expert-guided machine learning, that this product is designed to use.

Phase 2 and beyond

Phase 2 deepens the quantitative core: moving from targeted manual UTM confirmation to more integrated deployment where measurement uncertainty is fully characterized and validated for the intended workflow. The hyperspectral pilot collects ground-truth coating chemistry data on high-value zones (leading edges of hopper tanks, moisture-prone lap joints, manhole surrounds) to train the visual AI layer to infer coating condition and early corrosion risk from RGB + thermal alone where spectral signatures correlate. Active thermography (brief controlled heating) is evaluated for improved sensitivity to shallow delamination in coated structure. All additions are bounded pilots with explicit pass/fail criteria; nothing is productized before validation against manual reference methods and class surveyor acceptance.

Regulatory Strategy and Risk

Regulatory approach

The regulatory position follows directly from the product’s inspection boundaries and its emphasis on human expert oversight. The product is framed as a documented survey-support and data-fusion workflow, used within existing survey and reporting processes, with explicit surveyor review and confirmation wherever decisions affect class or repair scope. It does not claim autonomous survey release or replacement of surveyor judgment. It positions itself as the tool that makes inspection data more usable by providing the missing unified assessment layer.

Public precedent supports this positioning. IACS has issued recommendations on Remote Inspection Technologies (RIT), and class practice shows that remote-inspection evidence can be used when appropriate safeguards, data quality, and human oversight are in place. Specific acceptance depends on the vessel, society, survey scope, and service-provider arrangements. The pilot therefore operates as decision support with paired evidence and confirmation, building the basis for broader acceptance rather than assuming it.

Engagement with class and RO stakeholders, shipyards, operators, and approved measurement providers is central to the regulatory strategy. Their review of evidence requirements, reporting formats, and data traceability helps ensure the workflow can slot into existing survey processes with minimal friction.

Risk matrix

References

1. EU and commercial vessel-inspection robotics programmes (2026). Public project and product materials on confined-space robotic inspection capabilities.
2. IACS (2023). Unified Requirement UR Z10 – Hull Surveys of Oil Tankers, Chemical Tankers, Bulk Carriers, etc. International Association of Classification Societies.
3. IMO (2024). Remote inspection technologies guidance for surveys of ships and mobile offshore units.
4. DNV (2025). Rules for Classification – Ships, Part 7: Surveys, Chapter 1: Survey Requirements. DNV AS.
5. ABS (2025). Rules for Survey After Construction – Part 1: General. American Bureau of Shipping.
6. Lloyd’s Register (2025). Rules and Regulations for the Classification of Ships – Survey Requirements. Lloyd’s Register Group.
7. Voliro (2025). Drone-Enabled Pulsed Eddy Current (PEC) Sensor for Wall Thickness Measurement. voliro.com
8. Cubert (2025). Ultris X20 Plus — Hyperspectral Camera for UAV/UGV Applications, 350–1000 nm, 164 spectral bands. cubert-hyperspectral.com
9. DJI Enterprise (2025). Zenmuse H30 Series — Multi-Sensor Payload with Radiometric Thermal Imaging. dji.com/zenmuse-h30-series
10. Polanyi, M. (1966). The Tacit Dimension. University of Chicago Press.
11. Azad, M.A., Mohammed, A., Waszak, M., Elvesæter, B., & Ludvigsen, M. (2023). "Multi-label Video Classification for Underwater Ship Inspection." arXiv:2305.17338. Presented at OCEANS 2023 Limerick, IEEE. Used here as an adjacent reference for spatiotemporal ship-inspection video analysis, not as direct ballast-tank evidence. arxiv.org/abs/2305.17338
12. Resonon (2025). Pika L — Lightweight VNIR Hyperspectral Camera, 400–1000 nm, 281 spectral channels. resonon.com/Pika-L