Presentation Profile
Automated Quality Control of Petroleum Products via Computer Vision
Currently Scheduled: 10/14/2026 - 1:00 PM - 2:00 PM
Room: Exhibit Hall Entrance
Main Author
Raj Shah - Koehler Instrument Company, Inc.
- Aadi Shah - Koehler Instrument Company, Inc.
Abstract:
Quality control (QC) of petroleum products and pipeline corrosion analysis heavily dependent on manual, vision-based inspection and rating. This process is slow, costly, and susceptible to biases. Standard test methods: copper-strip corrosion test (ASTM D130) and petroleum color scales (ASTM D1500 and ASTM D156) require a human to compare samples against a reference by sight, while corrosion assessment is through labor intensive field inspection. Recent advances in computer vision (CV) and deep learning enable automated, object, and reliable image analysis. This review analyzes applications of CV and machine learning (ML) to petroleum QC across two applications: pipeline corrosion inspection and automated reading and verification of physical lab test results. For corrosion inspection, convolutional neural networks (CNNs) use transfer learning to classify defects based on images acquired from optical sensor data at accuracies exceeding 98%. This field has evolved past binary classification to pixel-level localization and severity grading, utilizing drones and edge computing. In laboratory testing, ML models predict Saybolt color, viscosity, flash point, and cloud point using RGB and HSV image data with 90-99% accuracy. In commercial instrumentation, these same camera-algorithm systems are being integrated to grade the ASTM D130 test, eAectively eliminating operator bias. Automated colorimeters standardized the visual D1500 and Saybolt scales to their electronic equivalent (ASTM D6045) and feeds results automatically into laboratory information management systems (LIMS). Current challenges facing these methods include scarcity of the labeled data sets required for training, limited model interpretability, sensitivity to lighting and image quality, and the need for formal validation before these methods can be incorporated into ASTM, ISO, and IP standards. These approaches make QC more objective and eAicient, allowing for deeper integration of computer vision to testing instrumentation.
