Enabling AI-powered field operations for pipeline inspections and corrosion detection

Corrosion causes over 64% of damage incidents in crude oil pipelines, but existing manual inspection methods are slow and inconsistent, especially at remote sites with limited connectivity - delayed detection means asset degradation, safety risk, and unplanned downtime. The client needed a system that could process high-resolution image data, detect corrosion with precision, and run entirely within localized environments with no reliance on cloud infrastructure, balancing machine learning compute needs against local resource constraints.

The solution ingests high-resolution TIFF images from pipeline cameras and applies pre-trained ML models to detect corrosion. These detections are rendered as transparent polygon overlays and fused with JPEG visualizations, accessible via an intuitive gallery interface within the AI Workbench.

• Dual-format image ingestion and storage: To support high-fidelity image processing, the system accepts and stores images in both TIFF and JPEG formats. Engineers upload inspection images in TIFF (for analysis) and JPEG (for visualization) formats using a React-based interface. TIFFs are stored in MinIO, while metadata is recorded in PostgreSQL. This setup removes the need for manual file conversion or preprocessing.

• On-site ML inference: A Flask-based service automatically processes TIFF files using pre-trained ML models. Corrosion is detected and marked spatially. All inference runs locally, ensuring consistent performance at remote sites and avoiding dependency on internet access.

• Overlay visualization engine: The system converts model outputs into polygon overlays and blends them with JPEG images. These composite visuals allow engineers to see exact damage locations while preserving image clarity. Overlay data is stored in PostgreSQL to support traceability and further analysis.

• Gallery interface for field review: Engineers review results through a responsive React gallery built with Material UI. It supports zoom, scroll, and filter functions for fast comparison across pipeline sections. Redux handles state management, ensuring smooth interactions even with large image sets.

• Offline-first architecture: All services run on local infrastructure using Docker containers. Nginx manages routing between components. This setup supports full inspection workflows in bandwidth-limited environments and ensures inspection data remains on-prem.

• Traceability and metadata: The PostgreSQL database links each image to its analysis results, inspection status, and timestamps. Engineers can query by pipeline segment or inspection cycle. This audit-friendly structure also supports trend analysis over time.

Zemoso delivered an offline-capable, AI-powered pipeline inspection platform built for the realities of field operations. By combining machine learning, real-time visualization, and a lightweight deployment model, the system enables faster, more reliable corrosion detection and safer, more efficient asset management.