Enhancing consistency in piping and instrumentation diagrams using DistilBERT and smart PID systems

F.S. Gómez-Vega, O. Acuña, Andrea C. Camargo, Jeison D. Jimenez, Sara M. Galeano, Isabella E. Franco, Laura L. Lozano, Jenifer Vásquez, Edwin Puertas

Abstract: Highlights•DistilBERT Application for Inconsistency Detection: The study successfully trained a DistilBERT model to detect inconsistencies in Smart P&ID databases with remarkable results, achieving an F1 score of 99% and a loss of just 0.04%. The model was specifically tuned for the intelligent engineering design process, aiming to minimize human error in reviewing P&IDs by automating inconsistency detection.•Dataset Construction and Views: The research involved building a dataset by tracking changes within the Smart P&ID system. Views were constructed to focus on unresolved inconsistencies and relationships between engineering elements, which provided the necessary labeled data for model training. This data structure allowed the model to identify and resolve inconsistencies in engineering designs.•Scalability and Performance of DistilBERT: The model was able to handle large datasets efficiently, with training experiments demonstrating rapid convergence and high performance. The research highlights the model’s potential for scalability, suggesting that it can generalize well to complex datasets beyond the initial scope of this project, which points to future applications in other industrial contexts.