Design And Implementation Of Petrochemical P&ID Drawing Symbol Reconstruction System

In the process of project construction and operation in petrochemical enterprises,Piping and Instrumentation Diagram(P&ID)is the basis for detailed engineering design of the factory.Currently,petrochemical engineering companies rely mainly on professional engineers to manually draw P&IDs,and the quality of the drawings depends on the skills of each individual,leading to low drawing efficiency and error-prone,which can affect the design quality and project schedule.In addition,P&ID drawings are delivered to the client in paper form,which can lead to issues such as loss,difficult archiving,and sharing.Therefore,the delivery of digital P&ID drawings has become a new trend in petrochemical enterprises,which can significantly reduce the time spent by engineers manually drawing P&IDs,ensure the quality and efficiency of the drawings meet or exceed project design requirements,and achieve cost savings and efficiency gains.This article mainly focuses on the following aspects:(1)Through cooperation with CNOOC Petrochemical Engineering Co.,Ltd.,this article collects P&ID drawings in PDF format and converts them into PNG format P&ID images.The images are then processed by sliding window to obtain a dataset containing 51 classes of symbol.Due to the limited number of samples in the dataset and the problem of class imbalance,this article proposes a data augmentation algorithm suitable for P&ID images ——Random Splicing.It can selectively augment one or more categories of symbols in P&ID images,solving the problem of class imbalance in the dataset,and reducing the cost and time of manual production and annotation of P&ID images to some extent.(2)To address the problem of small differences and difficulty in distinguishing between symbols,this article introduces the CBAM(Convolutional Block Attention Module,CBAM)attention mechanism and improves its "series" structure to a "parallel" structure,enhancing the information fusion between channel feature maps and spatial feature maps through concatenation and channel shuffle,and using a 1×1 convolution to achieve channel alignment,resulting in the PCSC＿CBAM(Parallel Concat Shuffle Convolution＿CBAM,PCSC＿CBAM)attention mechanism.This is further introduced into the Neck of YOLOv5.Comparative experiments show that the YOLOv5+PCSC＿CBAM model achieves high precision,recall,and mean Average Precision of symbol recognition,reaching 95.6%,95.0%,and 97.2%,respectively,with only an additional 1.97 M parameters.It can effectively recognize symbols in P&ID drawings and reduce the possibility of false negative and false positive.To address the problem of position offset in YOLOv5 prediction boxes,this article fully utilizes the shape features of the symbols and creates a template library containing 24 template images.It applies a shape-based template matching method in Halcon to obtain accurate placement coordinates and rotation angles for the 51 classes of symbol,which can effectively achieve symbol positioning.(3)This article designs and implements a P&ID symbol recognition system and a P&ID symbol reconstruction system,which meet the company’s needs for automatic recognition and reconstruction of symbol drawings.The P&ID symbol recognition system can complete the recognition and positioning of symbol and generate structured data in only about 1 minute.The P&ID symbol reconstruction system can automatically reconstruct symbol using structured data within 15 minutes.Compared with manual drawing of P&IDs,the P&ID symbol reconstruction system saves about 40% of time.