A blockchain and CNN-based platform for automated concrete crack detection and defect management
Olfa Loukil, Amira Talha, Nesrine Affes, Atef DaoudPurpose
This article proposes an automated platform that integrates AI-based crack detection with blockchain to enhance construction sustainability by proactively monitoring reinforced concrete buildings for structural pathologies. Its core goal is to protect human lives by establishing a transparent and traceable framework for compliance inspections and stakeholder accountability, particularly relevant for prevalent structural issues in the Tunisian context. The objective is not to “store everything on-chain,” but to create verifiable, time-stamped and non-repudiable evidence for each critical step of defect management.
Design/methodology/approach
The system employs convolutional neural networks (CNNs) for automated crack detection from photos, evaluating various models. For better scalability, operational data is stored off-chain (Firebase), while an Ethereum smart contract records tamper-evident evidence on-chain by only storing compact metadata (report ID, stakeholder address/role and timestamp) and cryptographic hashes of the offline report. MetaMask is used to authenticate users and sign transactions, ensuring stakeholder accountability.
Findings
In the context of crack detection, the combined CNN with Batch Normalization and Dropout achieved the best crack detection performance with a validation ROC AUC of 0.830 and a test ROC AUC of 0.789, demonstrating the most robust and balanced detection capability. The results demonstrate that each inspection is recorded as an immutable audit record, providing traceability and non-repudiation. Performance evaluation shows an average confirmation latency of approximately 2.5 seconds and a maximum throughput of about 25 transactions per second, with a transaction success rate between 98% and 100%. The system records only report metadata and cryptographic hashes on-chain, resulting in a stable operational cost of approximately 85,000 gas per inspection. These results indicate that the proposed approach is operationally feasible for real construction supervision workflows.
Research limitations/implications
The blockchain ensures data integrity after recording but cannot guarantee the correctness of the original inspection input. The approach also depends on network configuration and regulatory adoption in construction practice.
Originality/value
This work's originality is the unique creation of an automated, end-to-end platform by combining AI pathology recognition (CNNs) and blockchain data traceability (Ethereum). This offers an immutable and transparent solution for compliance and accountability in reinforced concrete rehabilitation, establishing a new standard for structural safety by addressing a critical, context-specific problem.