Reverse Engineering Piping and Instrumentation Drawings: A Deep Learning Approach

Data visualization is an easy way to communicate knowledge ever since human evolution. There are many visualization techniques including different types of plots, graphs, charts, and many more. Despite its usefulness, there is one common problem with visualization. The static image that generates the visualization is separated from the raw data. When a research study requires raw data it's hard to reverse engineer and take the raw data back. There are very limited tools available to reverse engineer and extract raw data from the above-mentioned static image. However, it's challenging to analyze all the data visualization techniques at a single glance. So, this research is focused on data extraction and summarization from Piping and Instrumental Drawings(P&ID). The dataset contains symbols taken from 2432 instances of piping and instrumental drawings. The goal of this study is to improve the accuracy of the prebuilt Convolutional Nural Network(CNN) model with hyperparameter optimization techniques and create a web application where users can extract and summarize text and symbols data from any piping and instrumentation drawing. As future directions, the applicability of other visualization techniques is also discussed at the conclusion.