🤖 AI Summary
This work addresses the longstanding disconnect between defect localization and structured reporting in industrial inspection, which has traditionally relied on manual intervention. The authors propose a decoupled three-stage pipeline: the Eyes module leverages YOLOv8-x-obb for high-precision oriented defect detection; the Bridge module maps detection outputs to structured prompts via parameter-free spatial encoding; and the Brain module employs a 4-bit quantized Qwen-2.5-1.5B model, fine-tuned with QLoRA and retrieval-augmented fine-tuning (RAFT), to generate standardized JSON reports. Evaluated on a small-scale synthetic dataset, the approach significantly outperforms generic end-to-end large models, achieving a BLEU-4 score of 0.41, a hallucination rate of only 4%, an expert rating of 8.6/10, and an inference speed of 47 tokens per second on a single T4 GPU—surpassing a 671B-parameter API baseline in overall performance.
📝 Abstract
Automated industrial inspection requires both precise defect localization and structured maintenance report generation; in current practice these tasks are handled separately, with linguistic interpretation left to human experts. This paper describes a decoupled, edge-deployable pipeline for wind turbine blade inspection built from three components that each handle a distinct sub-task. The Eyes a YOLO26-x-obb oriented bounding-box detector localizes defects at dataset-native resolution. The Bridge a deterministic, parameter-free encoding module maps each detected bounding box to grid-referenced spatial tokens embedded in a structured prompt. The Brain a 4-bit quantized Qwen-2.5-1.5B model adapted with Quantized Low-Rank Adaptation (QLoRA) on 947 synthetically generated maintenance reports generates a structured JSON report from that prompt. Retrieval-Augmented Fine-Tuning (RAFT) further grounds each recommendation in indexed maintenance procedures. Five ablation experiments, scored by BLEU-4, ROUGE-L, Hallucination Rate (HR), and an LLM-as-a-Judge rubric, compare the pipeline against a monolithic vision-language model (VLM) baseline and against partial configurations in which one component is removed. The complete system achieves BLEU-4 0.41, HR=4%, and Expert Score = 8.6/10 compared with 0.07, 65%, and 3.3/10 for the zero-shot VLM baseline. The QLoRA-adapted 1.5B model generates higher-quality reports than a 671B-parameter generalist API model given identical detection evidence, at 47 tokens per second on a single T4-class GPU. The results show that purpose-built decoupled architecture with a small domain-specific training corpus outperforms a generalist end-to-end model on this structured generation task.