Semi-Supervised Symbol Detection for Piping and Instrumentation Drawings

Background

Piping and instrumentation diagrams (P&IDs) are technical drawings used to operate and maintain process systems. P&IDs are typically used to identify components and quantities needed for ongoing projects, and to understand the layout and operation of the process system with completed projects. The information provided from P&IDs is crucial to preparing bill-of-quantities, placing a purchase order, developing a work schedule, or performing resource allocation. However, typical programs that create P&IDs are time-intensive and high cost, due to the current need for manual annotation or to conduct additional training on machine learning algorithms.

Invention Description

Researchers at Arizona State University have developed a new semi-supervised learning method for efficient symbol detection in piping and instrumentation diagrams (P&IDs). This method uses a two-stage pipeline that combines deep learning for generic symbol detection and a Siamese network for symbol differentiation. This method significantly reduces the need for manual data annotation while improving detection efficiency.

Potential Applications:

• Automated P&ID analysis for engineering & construction
• Machine learning model development with reduced annotation requirements
• Enhanced document analysis tools for technical drawings and schematics

Benefits and Advantages:

• Lower cost – uses self-supervised learning to reduce manual annotation costs
• High accuracy – transfer learning from MS COCO pretrained models
• Optimized resource use – non-linear correlation between annotated data volume and model performance
• Enhanced performance – K-means coreset sampling over random sampling
• Effective symbol differentiation – high Top-1 and Top-5 accuracy rates

Related Publication: Semi-supervised symbol detection for piping and instrumentation drawings