Oil & Gas Meters Processing with Artificial Intelligence and Computer Vision
⚡ Pain Points We Tackled
The client, a Canadian Oil & Gas service company, needed to automate reading data from production meters: graphs, printed stickers/barcodes, and handwritten operator notes for extracted oil & gas resources. The existing process was manual, error-prone, and inconsistent across equipment types.
Our Approach
Azati’s ML and computer-vision team developed a pipeline to ‘unfold’ circular meter prints into rectangular images, detect and separate multiple colored curves, identify printed stickers/barcodes, and apply handwriting recognition for operator notes. We trained neural networks to handle diverse equipment formats and data quality, integrated services in cloud infrastructure, and iterated rapidly on pilot prototypes.
Applied Methods and Practices
- Round Disk Image Transformation: Unwrapped circular graphs into rectangular format to enable coordinate-based curve detection.
- Color-Based Curve Segmentation: Highlighted multiple overlapping curves on graphs, overcoming background interference for accurate reading.
- Neural-Network Format Classification: Routed input data from diverse partner equipment to the appropriate processing pipeline.
- Handwriting Recognition: Detected and interpreted handwritten dates and numbers using bounding-box region search and trained OCR/Tesseract networks.
- Agile Iterative Development: Pilot prototyping followed by iterative integration and deployment to the client’s cloud environment.
Solution Features
- Automated Meter Data Processing: Ingests scanned meter outputs and returns structured digital data (curve values, printed labels, handwritten notes).
- Cloud-Based Deployment: Client infrastructure and deployed in the cloud for scalability and partner equipment diversity.
- High-Accuracy Modules: Barcode and sticker processing (90% accuracy), curve reading (>80% accuracy), handwritten note recognition (30–70+% depending on input quality).
- Customizable Neural Network Processing: Adaptable to various data formats and partner-specific equipment configurations.