Please use this identifier to cite or link to this item:
Title: Machine learning based automated evaluation of piping design
Authors: Tan, Wei Chian
Keywords: DRNTU::Engineering::Industrial engineering::Automation
DRNTU::Engineering::Maritime studies::Maritime science and technology
Issue Date: 2019
Source: Tan, W. C. (2019). Machine learning based automated evaluation of piping design. Doctoral thesis, Nanyang Technological University, Singapore.
Abstract: A methodology to automate the process of evaluating the layout of a piping design in the form of a Piping and Instrumentation Diagram (P&ID) according to a set of rules of the marine and offshore industry, mainly from the International Maritime Organisation and Lloyd's Register, is proposed. The method starts with transforming a P&ID into a vector $\mathbf{x}$ in $R^d$. The transformation is done based on a concept introduced for piping known as Histogram of Connectivity (HoC). The proposed descriptor captures two essential properties of a P&ID: the attributes of each component and the connectivity among the components. Next, a linear Support Vector Machine (SVM) is used to learn a classifier from existing compliant and non-compliant designs. Subsequently, the linear classifier can be used to check if an unseen design complies with the standards. In addition, to enable a follow up on a non-compliant design including corrections or modifications, a method to analyse the reason of non-compliance prediction by the learnt SVM model is introduced. Next, on top of the Histogram of Connectivity developed, the concept of multi-resolution is introduced to look at connectivity beyond components which are connected immediately for cues on non-compliance. This is done by extracting the connectivity information among the components connected via a series of intermediate components. Known as Pyramid of Connectivity Histogram (PoC), the feature vectors extracted are fed into a linear SVM for learning and prediction subsequently. Following the HoC and PoC, a method to learn piping designs of different aspects of international rules in an integrated manner is introduced. Instead of learning designs of different aspects separately, learning using an SVM is done in a common feature space for all aspects. Based on the concept of HoC established previously, the method begins with mapping the mathematical representation of each piping layout into a common feature space for all different aspects or datasets. Next, learning by a linear SVM is performed. Lastly, an end-to-end learning approach based on the latest Convolutional Neural Network (CNN) architectures and transfer learning to perform vision-based analysis of piping designs is introduced. Having a piping design in the form of an image, a framework known as Piping Net (PipNet) is introduced to understand the design and interpret if it complies with applicable engineering regulations. Designs and corresponding labels (compliant or non-compliant) are fed into an existing trained CNN in the form of images for transfer learning, with the subsequently obtained fine-tuned network called PipNet. The developed system has demonstrated outstanding performance on all five challenging datasets of piping designs introduced in this work.
DOI: 10.32657/10220/48128
Schools: School of Mechanical and Aerospace Engineering 
Organisations: Lloyd's Register
Research Centres: Robotics Research Centre 
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:MAE Theses

Files in This Item:
File Description SizeFormat 
wei_chian_tan_thesis_19.pdf9.15 MBAdobe PDFThumbnail

Page view(s) 20

Updated on Jun 23, 2024

Download(s) 5

Updated on Jun 23, 2024

Google ScholarTM




Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.