Enhancing Efficiency in Piping & Instrumentation Diagrams (P&IDs) through AI-Driven Digitization

Optimizing Piping and Instrumentation Diagrams (P&IDs) in design and engineering is often hindered by manual and repetitive tasks, despite common, reusable patterns. This work leverages AI to identify and automate these recurring patterns, converting P&IDs from non-machine- readable formats (e.g., images, PDFs) into graph-based representations. This conversion significantly improves P&ID accessibility, reducing manual handling and boosting efficiency. Existing digitization solutions lack sufficient accuracy, prompting this study’s comparative approach with three models. First, SqueezeNet 1.1 was applied for unit operation classification across 16 classes but showed limitations in generalizing to unseen data. The second model, a Random Forest using pixel data from CSV files, demonstrated high recall and F1 score on train and validation data but lacked robustness on new data. Finally, the Faster R-CNN with a ResNet-50 backbone and Stochastic Gradient Descent (SGD) as the optimizer, trained over 30 epochs with optimized parameters, achieved superior accuracy and generalization, making it most promising for practical applications.