Pattern Recognition Method for Detecting Engineering Errors on Technical Drawings
Paper in proceeding, 2022

Many organizations are looking for how to automate repetitive tasks to reduce manual work and free up resources for innovation. Machine Learning, especially Deep Learning, increases the chance of achieving this goal while working with technical documentation. Highly costly engineering hours can be saved, for example, by empowering the manual check with AI, which helps to reduce the total time for technical documents review. This paper proposes a way to substantially reduce the hours spent by process engineers reviewing PIDs (Piping Instrumentation Diagrams). The developed solution is based on a deep learning model for analyzing complex real-life engineering diagrams to find design errors - patterns that are combinations of high-level objects. Through the research on an extensive collection of PID files provided by McDermott, we prove that our model recognizes patterns representing engineering mistakes with high accuracy. We also describe our experience dealing with class-imbalance problems, labelling, and model architecture selection. The developed model is domain agnostic and can be re-trained on various schematic diagrams within engineering fields and, as well, could be used as an idea for other researchers to see whether similar solutions could be built for different industries.

Artificial Intelligence

engineering drawings

Deep Learning

Piping and Instrumentation Diagrams (P&IDs)

object recognition

Author

Rimman Dzhusupova

McDermott

Richa Banotra

McDermott

Jan Bosch

Chalmers, Computer Science and Engineering (Chalmers), Interaction Design and Software Engineering

Helena Holmström Olsson

Malmö university

2022 IEEE World AI IoT Congress, AIIoT 2022

642-648
9781665484534 (ISBN)

2022 IEEE World AI IoT Congress, AIIoT 2022
Seattle, USA,

Subject Categories

Software Engineering

Embedded Systems

Computer Science

DOI

10.1109/AIIoT54504.2022.9817294

More information

Latest update

8/4/2022 9