[Paper] A Blockchain-based Traceability System for AI-Driven Engine Blade Inspection

Source: arXiv - 2603.08288v1

Overview

The paper introduces BladeChain, a blockchain‑enabled traceability platform that records every inspection event of aircraft engine blades across manufacturers, airlines, maintenance, repair & overhaul (MRO) providers, and regulators. By marrying Hyperledger Fabric with off‑chain storage (IPFS) and AI‑driven defect detection, BladeChain creates an immutable audit trail that can be trusted by all parties in the aerospace supply chain.

Key Contributions

• Multi‑stakeholder Hyperledger Fabric network that enforces a shared, tamper‑evident ledger among OEMs, airlines, MROs, and regulators.
• State‑machine driven lifecycle management: blade status transitions (e.g., “in service”, “inspection due”, “under repair”) are encoded in chaincode, automatically triggering inspections when flight‑hour, cycle, or calendar thresholds are hit.
• AI model provenance tracking: each inspection record logs the exact AI model name and version used for defect detection, enabling regulators to verify how a defect was identified.
• Hybrid on‑chain/off‑chain storage: large inspection artifacts (images, sensor logs) are stored in IPFS; their SHA‑256 digests are anchored on the ledger, guaranteeing integrity without bloating the blockchain.
• Pluggable detection module: organizations can swap or upgrade AI models without changing the underlying ledger or workflow logic.
• Prototype evaluation: demonstrates 100 % lifecycle completion for up to 100 blades, sustaining ~26 transactions per minute, and detecting tampering within ~17 ms.

Methodology

1. System Architecture – BladeChain is built on a permissioned Hyperledger Fabric network with four organizations (OEM, Airline, MRO, Regulator). Each organization runs peers that host the shared ledger and execute chaincode.
2. State Machine & Chaincode – Blade status is modeled as a finite‑state machine (e.g., NEW → IN_SERVICE → INSPECTION_DUE → UNDER_REPAIR → RETIRED). Chaincode enforces valid transitions and emits events when a blade crosses a predefined usage threshold.
3. Inspection Scheduling – Thresholds (flight hours, cycles, calendar days) are configurable per blade type. When a threshold is exceeded, the chaincode automatically creates an “inspection task” entry.
4. AI‑Driven Inspection – An external AI service consumes sensor data or high‑resolution images, runs a defect‑detection model, and returns a result together with the model identifier and version.
5. Off‑chain Artifact Management – Raw inspection data (e.g., images, logs) are uploaded to IPFS. The resulting content identifier (CID) is hashed (SHA‑256) and stored on‑chain alongside the AI provenance metadata.
6. Evaluation Setup – A synthetic workload simulating up to 100 blades was generated. Transactions (status updates, inspection creation, result recording) were measured for throughput, latency, and consensus overhead against a baseline centralized SQL implementation.

Results & Findings

Key takeaways:

• The added consensus cost of Fabric is modest (≈ 30 % throughput reduction) but provides strong tamper‑evidence and multi‑party trust.
• Off‑chain IPFS storage keeps ledger size manageable while still guaranteeing integrity via hash anchoring.
• Automatic inspection triggering eliminates manual scheduling errors, a common source of missed or delayed maintenance in current practice.

Practical Implications

• Regulatory compliance made auditable – Regulators can verify not only what defect was reported but also how it was detected, satisfying emerging AI‑explainability requirements.
• Reduced downtime – Automated, threshold‑driven inspection scheduling ensures blades are inspected exactly when needed, preventing unexpected failures and costly unscheduled removals.
• Inter‑organizational trust without a central authority – By using a permissioned blockchain, OEMs, airlines, and MROs can share data securely without exposing proprietary information to a single database administrator.
• Future‑proof AI upgrades – Since the AI model is referenced by name and version in the ledger, organizations can adopt newer detection algorithms (e.g., higher‑resolution CNNs) without redesigning the workflow or re‑negotiating contracts.
• Scalable to other aerospace components – The state‑machine + off‑chain pattern can be replicated for turbine disks, landing‑gear assemblies, or even non‑aerospace high‑value assets (e.g., wind‑turbine blades).

Limitations & Future Work

• Scale – The prototype was evaluated on up to 100 blades; real‑world fleets involve thousands of components, requiring performance testing at larger scale and possibly sharding or channel partitioning.
• Network latency – While tamper detection is fast, the end‑to‑end latency (including IPFS upload and AI inference) was not deeply profiled; optimizing edge‑computing placement could be necessary for time‑critical inspections.
• Governance model – The paper assumes a static four‑party consortium; future work should explore dynamic membership, revocation, and incentive mechanisms for broader industry adoption.
• AI model verification – Storing only the model name/version does not guarantee the model’s integrity; integrating model hash anchoring or zero‑knowledge proof of inference could strengthen trust.

BladeChain demonstrates how blockchain can move beyond cryptocurrency to solve concrete, safety‑critical traceability challenges in aerospace. For developers building next‑generation maintenance platforms, the paper offers a concrete reference architecture that blends permissioned ledgers, decentralized storage, and AI provenance—components that are increasingly familiar in modern cloud‑native stacks.

Authors

• Mahmoud Hafez
• Eman Ouda
• Mohammed A. Mohammed Eltoum
• Khaled Salah
• Yusra Abdulrahman

Paper Information

• arXiv ID: 2603.08288v1
• Categories: cs.CR, cs.AI, cs.DC
• Published: March 9, 2026
• PDF: Download PDF