# A Blockchain-based Traceability System for AI-Driven Engine Blade Inspection

Crypto.world

> Aircraft engine blade inspection records today are spread across paper forms, spreadsheets, and organization-specific databases owned by manufacturers, airlines, maintenance organizations, and…

**arXiv**: 2603.08288

**Published**: 2026-03-09

**Authors**: Mahmoud Hafez, Eman Ouda, Mohammed A. Mohammed Eltoum, Khaled Salah, Yusra Abdulrahman

## AI summary

## Overview

**Research area:** Permissioned blockchain (Hyperledger Fabric) applied to aerospace maintenance records, combined with AI-based defect detection for aircraft engine blades.

**Technical level:** Intermediate. The paper assumes familiarity with blockchain concepts such as endorsement policies, chaincode, and consensus, but explains its design decisions in enough detail for a reader with general distributed-systems background.

**Scope:** BladeChain is a four-organization Hyperledger Fabric prototype that records the full life cycle of aircraft engine blades and cryptographically binds AI-generated inspection results, inspector identity, and image artifacts to an immutable ledger.

## What This Paper Is About

Aircraft engine blade inspection records today are spread across paper forms, spreadsheets, and organization-specific databases owned by manufacturers, airlines, maintenance organizations, and regulators, which makes audits slow and tampering hard to detect. At the same time, AI models can now flag blade defects from images, but those detections are trapped inside local quality-assurance tools with no proof of which model, which image, and which inspector produced a given finding. BladeChain's goal is to close that gap by putting blade life cycle events on a shared permissioned blockchain and recording every inspection together with the AI model name and version, image hash, and IPFS identifier, so regulators can audit both *what* defect was found and *how* it was found.

## Key Contributions

1. **BladeChain itself:** the authors propose a blockchain-based traceability system designed specifically for AI-driven engine blade inspection, targeting the fragmented and non-auditable state of current aviation maintenance records.
2. **On-chain AI model provenance:** each inspection record captures the AI model name and version used for defect detection, letting regulators audit not only what defects were found but how they were found.
3. **Automated inspection scheduling:** a chaincode-enforced state machine triggers inspections when configurable flight-hour, cycle, or calendar thresholds are exceeded, removing manual tracking errors.
4. **A working prototype and evaluation:** the system is implemented on Hyperledger Fabric and evaluated on workloads of up to 100 blades, with a centralized SQL baseline included to quantify the consensus overhead and the associated security trade-off.

## Main Findings

- **Life cycle completion:** the prototype demonstrated 100% life cycle completion across evaluated workloads of up to 100 blades.
- **Throughput:** the system sustained a consistent throughput of 26 operations per minute.
- **Consensus cost is measurable:** a centralized SQL baseline was built and used to quantify the overhead introduced by blockchain consensus, framing it explicitly as a security-versus-performance trade-off rather than a cost that is hidden.
- **Fast tamper detection:** security validation confirmed tamper detection within 17 ms through hash verification.
- **Automatic scheduling works through the ledger, not manual process:** the `LogFlightOperation` chaincode function evaluates three thresholds — 500 flight hours, 500 flight cycles, and 180 calendar days — and moves a blade to `Inspection_Due` when any is exceeded, recording which threshold fired. The paper states these values are configurable despite being hardcoded in the current implementation.
- **Identity cannot be spoofed by the client:** inspector and organization fields in inspection records are derived from the caller's MSP certificate via `GetClientMSPID` and `GetClientID`, not from client-supplied input.
- **No prior system combines these features:** Table I in the paper compares BladeChain against Aleshi et al. (2019), Schyga et al. (2019), Ho et al. (2021), AirChain (2022), Hawashin et al. (2024), Kabashkin (2024), and Alqaryuti et al. (2025), and shows BladeChain as the only one marked as combining multi-org endorsement, parts life cycle management, AI integration, off-chain record storage, and integrity proofs.
- **Note on the available content:** the full paper text was truncated partway through Section IV-E, so the detailed performance tables referenced in Section V and the concluding Section VI were not available. Only the results stated in the abstract are reported above; other benchmark figures are not reported in the available content.

## Methodology in Plain English

The authors began by deriving six requirements from four stakeholder classes — Original Equipment Manufacturers, airlines, Maintenance, Repair and Overhaul providers, and regulators — covering multi-organization traceability, multi-party control, automated scheduling, evidence binding, auditability, and practical scalability. They chose Hyperledger Fabric 2.5 rather than a public blockchain because public chains use pseudonymous identities (which conflict with accountable, auditable participants) and because gas fees and consensus latency are unnecessary for maintenance workloads.

The system is built in four layers. Web dashboards for each stakeholder role sit on top. A Node.js API gateway acts as the orchestrator: it receives an uploaded inspection image, sends it to the AI engine, uploads the annotated artifact to IPFS, computes a SHA-256 hash of the image, and submits a transaction to the Fabric network. The Fabric network itself has four organizations, one peer each with CouchDB state databases, and a three-node Raft ordering service. IPFS and the AI engine sit off-chain.

The smart contract, written in Go, encodes a six-state blade life cycle: `Manufactured`, `In_Service`, `Inspection_Due`, `Under_Inspection`, `Under_Repair`, and `Failed_Scrapped`. The transaction endorsement policy requires a logical majority — signatures from at least three of the four organizations — so no single party can create an inspection record alone.

Defect detection is deliberately pluggable and runs entirely off-chain. The prototype uses YOLOv8 trained on BladeSynth, a synthetic dataset of turbine blade images with pixel-level defect annotations. The detector returns the model name and version, defect labels and confidence scores, bounding-box coordinates, and an overall result of PASS, FAIL, or CRITICAL, all of which is written to the ledger alongside the IPFS CID and SHA-256 hash.

For evaluation, the authors built a network of thirteen Docker containers (three orderers, four peers, four CouchDB instances, an IPFS daemon, and the API gateway) and tested functional correctness, performance, and security.

## Why This Matters

**Impact on research:** The paper reframes AI inspection as an auditable process rather than a black-box point solution. By recording model name and version on-chain alongside cryptographic commitments to the exact images reviewed, it establishes a pattern that other safety-critical inspection domains could adopt, and it gives a concrete baseline (Fabric versus a centralized SQL database) for reasoning about consensus overhead in low-throughput, high-assurance settings.

**Real-world applications:**

- **Regulatory audit of maintenance histories:** a regulator can pull a blade's complete history, retrieve the original and annotated images from IPFS, recompute the SHA-256 hashes, and verify that nothing was altered — without trusting the operator's own database.
- **Model recall and re-evaluation:** if a detection model is later found to have systematic blind spots, every historical inspection made with that model version can be identified and flagged for re-review, which the paper notes is impossible in systems that treat AI detection as a black box.
- **Cross-organizational maintenance disputes:** because inspector and organization identity are drawn from MSP certificates and each transaction carries a majority endorsement, attribution cannot be repudiated by a client application.
- **Automated compliance enforcement:** a blade physically cannot remain in `In_Service` past its inspection interval, because the state transition is enforced by chaincode rather than by a manual tracking process.

**Industry relevance:** The paper cites a global MRO market projected to exceed $125 billion by the early 2030s, with engine maintenance accounting for 46% of civil aviation MRO spending. It also notes that regulatory frameworks such as EASA Part-145 mandate Occurrence Reporting Systems, yet many implementations are passive repositories rather than systems capable of enforcing traceability across organizational boundaries — precisely the gap BladeChain targets.

## Future Directions

1. **Governance beyond technical guarantees:** the threat model explicitly places collusion among a majority of endorsers out of scope, noting it requires governance controls. How such controls would work across OEMs, airlines, MROs, and regulators remains an open question.
2. **Adversarial robustness of the detection module:** attacks against the AI model are also declared out of scope as orthogonal ML security research, leaving unanswered how a model that is compromised or subtly degraded would affect the integrity of an otherwise tamper-evident ledger.
3. **Scale beyond the prototype:** evaluation covered workloads of up to 100 blades with throughput of 26 operations per minute. Whether the design holds for fleet-scale deployments, and how the consensus overhead quantified against the SQL baseline scales, is not established in the available content.
4. **From synthetic to operational imagery:** the prototype detector was trained on BladeSynth, a synthetic dataset. Validating detection performance and the full record-binding workflow on real borescope imagery from in-service engines is a natural next step.

## Target Audience

This paper suits blockchain researchers working on enterprise or permissioned-ledger applications, aerospace maintenance and airworthiness engineers interested in digital records and regulatory compliance, MRO and airline IT architects evaluating Hyperledger Fabric for record-keeping, and machine-learning practitioners who deploy inspection models in safety-critical settings and need provenance and accountability around model outputs. Readers new to blockchain will find the architecture sections accessible, but the design rationale and related-work comparison assume some familiarity with endorsement policies and consensus mechanisms.

## Authors’ abstract

Aircraft engine blade maintenance relies on inspection records shared across manufacturers, airlines, maintenance organizations, and regulators. Yet current systems are fragmented, difficult to audit, and vulnerable to tampering. This paper presents BladeChain, a blockchain-based system providing immutable traceability for blade inspections throughout the component life cycle. BladeChain is the first system to integrate multi-stakeholder endorsement, automated inspection…

[The full abstract on arXiv](https://arxiv.org/abs/2603.08288)

[Read the original paper](https://arxiv.org/abs/2603.08288)

---

Canonical: https://crypto.world/paper/a-blockchain-based-traceability-system-for-ai-driven-engine-blade-2603.08288
