Date of Award

Fall 2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Materials and Infrastructure Systems

First Advisor

David Iseley

Abstract

With the development of the natural gas industry, the demand for pipeline construction has also increased. In the context of advocating green construction, horizontal directional drilling (HDD), as one of the most widely utilized trenchless methods for pipeline installation, has received extensive attention in industry and academia in recent years. The safety of natural gas pipeline is very important in the process of construction and operation. It is necessary to conduct in-depth study on the safety of the pipeline installed by HDD method.

In this dissertation, motivated by the following considerations, two aspects of HDD are studied. First, through the literature review, one issue that has not received much attention so far is the presence of stress problem during the operation condition. Thus, two chapters (Chapters 3 and 4) in this dissertation are related to the pipe stress problem during the operation. Regarding this problem, two cases are considered according to the fluidity of drilling fluid. The more dangerous situation is determined by comparing the pipeline stress in the two working conditions. The stress of pipeline installed by HDD method and open-cut method is compared, and it indicates that the stress of pipeline installed by HDD method is lower. Moreover, through the analysis of influence factors and stress sensitivity, the influence degree of different parameters on pipeline stress is obtained.

Secondly, literature review indicates that the accurate prediction of pullback force in HDD construction is of great significance to construction safety and construction success. However, the accuracy of current analytical methods is not high. In the context of machine learning and big data, three new hybrid data-driven models are proposed in this dissertation (Chapter 5) for near real-time pullback force prediction, including radial basis function neural network with complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN-RBFNN), and support vector machine using whale optimization algorithm with CEEMDAN (CEEMDAN-WOA-SVM), and a hybrid model combines random forest (RF) and CEEMDAN. Three novel models have been verified in two projects across the Yangtze River in China. It is found that the prediction accuracy is dramatically improved compared with the original analytical models (or empirical models). In addition, through the feasibility analysis, the great potential of machine learning model in near real-time prediction is proved.

At the end of this dissertation, in addition to summarizing the main conclusions obtained, three future research directions are also pointed out: (1) stress analysis of pipelines installed by HDD in more complex situations; (2) stress analysis of pipeline during HDD construction; (3) database establishment in HDD engineering.

Included in

Engineering Commons

Share

COinS