Smart Inspection for Corrosion and Equipment Monitoring Using Robotics with Cloud-Based Computer Vision

The objective of this work is to develop and operationalize a smart inspection system that uses cloud-hosted computer vision models and robotic inputs to detect and classify corrosion, read equipment status, and improve asset integrity at surface facilities. The solution aims to enhance operational efficiency and reduce manual inspection workloads at offshore facilities. Methods, Procedures, Process: The system leverages deep learning models (ResNet for corrosion segmentation and YOLO for equipment detection) trained on historical inspection and open-source datasets. These models are deployed on a cloud-native environment for scalable inference. A robotic or mobile camera captures facility images, which are then processed via a multi-threaded data workflow. Detection outputs, metadata, and inspection results are stored using a standardized schema. A python-based user interface provides end users with interactive visualizations, filtering, and measurement correction capabilities. The complete system supports gauge reading, valve state detection, and corrosion classification with a quantifiable corrosion score. The system successfully detects corrosion and equipment states from visual inputs across a variety of surface facility environments. For corrosion detection, the model generates pixel-level segmentation masks, categorizes corrosion severity, and outputs a corrosion score on a scale of 0-100. For equipment monitoring, the system identifies valve states (open/close) with bounding boxes and confidence scores, while also calculating gauge readings using geometric anglebased interpretation. Multithreaded execution reduced inference latency by ~40%, and cloud integration enabled processing scalability without additional local infrastructure. Cloud-based architecture allows users to upload new images, run inference workflows, and view results directly through a web interface. Data is automatically structured, stored, and linked via metadata, supporting historical traceability and audit. Despite training on limited datasets, the models demonstrated decent generalization across unseen images, with model accuracy expected to improve as more labeled inspection data becomes available. This approach transforms conventional inspection by offering consistent, fast, and intelligent analysis, reducing the need for manual interpretation and helping teams focus on high-risk anomalies. It represents a step forward in integrating computer vision and cloud robotics into mainstream asset integrity workflows This paper introduces a practical and scalable application of AI for surface facility inspection by combining robotics, cloud computing, and computer vision. Unlike conventional methods, it offers real-time feedback, multithreaded processing, and a robust cloud-hosted workflow, making it easier to deploy and scale across operations. It contributes to the growing use of AI in field-level operational advancements and digital asset integrity management