Deep-learning-based Character Recognition of Aerospace Alloy Components for Automated Quality Inspection

As aerospace engine manufacturing advances toward intelligent automation, traditional manual component identification methods suffer from low efficiency and high error rates.We present a vision-based automated recognition system that integrates image processing and deep learning algorithms for the identification of aerospace engine components.Nickel-cobalt alloy components (Inconel 718) are used as test samples.The system employs optical character recognition (OCR) and You Only Look Once version 8 (YOLOv8) algorithms for character detection and classification.Image preprocessing includes grayscale conversion, Gaussian blur, edge detection, and morphological operations to enhance image quality.To overcome the limitations of single-frame recognition under varying lighting conditions and complex backgrounds, a multiframe fusion mechanism is implemented to enhance stability and accuracy.Comparative experiments were conducted to evaluate algorithm performance in terms of recognition accuracy, processing speed, and computational efficiency.The system integrates a programmable automation controller (PAC) and positioning sensors for automated operation.Experimental results demonstrate an approximately 96% recognition accuracy with a processing time of 0.35 seconds per component, confirming the effectiveness of this approach for automated quality inspection in aerospace manufacturing environments.Recent studies have led to significant advances in real-time object detection algorithms and integrated detection-recognition systems.Hussain systematically reviewed the real-time capability and extensibility of YOLO variants, (2) leading to subsequent improvements such as DWRB module-SADown module-LASPPF structure (DSL)-YOLO for enhanced small-object and occlusion detection (3) and YOLO-Average SPP-Fast Block, C2f_SimDCNv2, and Space-to-Depth (ADS) for balanced defect detection efficiency. (4) In automatic optical inspection (AOI) applications, deep learning approaches have proven effective even with limited training data. (5)YOLOv8-based systems have been successfully applied to visual anomaly quantification, (6) while multiscale residual attention networks address challenges in tiny character and oblique text recognition. (7)Two-stage detection-optical character recognition (OCR) systems have been developed for various industrial scenarios, including expiry-date reading under varying illumination, (8) embossed tire text recognition, geometric alignment pipelines, (9) and small-character recognition with contrast enhancement and template matching. (10) In automatic number plate recognition (ANPR), YOLOv8-OCR integration has demonstrated high-speed processing capabilities. (11,12)In comparative studies, cloud-based OCR and handwritten text recognition services have been evaluated for optimal tool selection. (13) Methodologically, dual-branch architectures and global feature extraction have been explored to improve surface defect inspection performance. (14,15) Aerospace components manufactured from advanced alloys-including titanium, nickelcobalt-based superalloys, and high-strength steels-pose significant optical inspection challenges.Nickel-cobalt alloy components (Inconel 718) are particularly challenging to inspect owing to intense specular reflection, low contrast, and surface oxidation or coatings, which reduce the signal-to-noise ratio and edge definition. (16)Traceability markings are often small and irregularly oriented, requiring multiscale, pose-invariant recognition. (17)Metal surface character recognition is sensitive to pose and illumination variations, necessitating precise fixturing and process control.While cloud-based OCR offers convenience, industrial deployment demands real-time latency and data governance compliance.Edge computing on programmable logic/ automation controllers (PLC/PAC) reduce latency but require lightweight models owing to computational constraints, and the industrial Internet of Things (IIoT) integration adds communication overhead.Recent research in industrial character recognition has predominantly employed generalpurpose deep learning models or cloud-based OCR services to enhance inspection accuracy.However, these approaches often struggle with the specific optical and sensing challenges of aerospace materials-such as the intense specular reflection and low contrast exhibited by Inconel 718 surfaces-and frequently lack the necessary integration with industrial control systems for real-time production monitoring.Furthermore, existing vision-based inspection systems typically require substantial network bandwidth and computational resources, limiting their deployment feasibility in edge computing environments where immediate feedback and deterministic response times are critical for production line operations. (14-17)In this study, we present an industrial vision system for automated character recognition on Inconel 718 aerospace components using image sensing and deep learning techniques.The system integrates the