Skip to main content

Research Repository

Advanced Search

Adaptive swarm optimisation assisted surrogate model for pipeline leak detection and characterisation.

Adegboye, Mutiu Adesina

Authors

Mutiu Adesina Adegboye



Contributors

Wai-Keung Fung
Supervisor

Abstract

Pipelines are often subject to leakage due to ageing, corrosion and weld defects. It is difficult to avoid pipeline leakage as the sources of leaks are diverse. Various pipeline leakage detection methods, including fibre optic, pressure point analysis and numerical modelling, have been proposed during the last decades. One major issue of these methods is distinguishing the leak signal without giving false alarms. Considering that the data obtained by these traditional methods are digital in nature, the machine learning model has been adopted to improve the accuracy of pipeline leakage detection. However, most of these methods rely on a large training dataset for accurate training models. It is difficult to obtain experimental data for accurate model training. Some of the reasons include the huge cost of an experimental setup for data collection to cover all possible scenarios, poor accessibility to the remote pipeline, and labour-intensive experiments. Moreover, datasets constructed from data acquired in laboratory or field tests are usually imbalanced, as leakage data samples are generated from artificial leaks. Computational fluid dynamics (CFD) offers the benefits of providing detailed and accurate pipeline leakage modelling, which may be difficult to obtain experimentally or with the aid of analytical approach. However, CFD simulation is typically time-consuming and computationally expensive, limiting its pertinence in real-time applications. In order to alleviate the high computational cost of CFD modelling, this study proposed a novel data sampling optimisation algorithm, called Adaptive Particle Swarm Optimisation Assisted Surrogate Model (PSOASM), to systematically select simulation scenarios for simulation in an adaptive and optimised manner. The algorithm was designed to place a new sample in a poorly sampled region or regions in parameter space of parametrised leakage scenarios, which the uniform sampling methods may easily miss. This was achieved using two criteria: population density of the training dataset and model prediction fitness value. The model prediction fitness value was used to enhance the global exploration capability of the surrogate model, while the population density of training data samples is beneficial to the local accuracy of the surrogate model. The proposed PSOASM was compared with four conventional sequential sampling approaches and tested on six commonly used benchmark functions in the literature. Different machine learning algorithms are explored with the developed model. The effect of the initial sample size on surrogate model performance was evaluated. Next, pipeline leakage detection analysis - with much emphasis on a multiphase flow system - was investigated in order to find the flow field parameters that provide pertinent indicators in pipeline leakage detection and characterisation. Plausible leak scenarios which may occur in the field were performed for the gas-liquid pipeline using a three-dimensional RANS CFD model. The perturbation of the pertinent flow field indicators for different leak scenarios is reported, which is expected to help in improving the understanding of multiphase flow behaviour induced by leaks. The results of the simulations were validated against the latest experimental and numerical data reported in the literature. The proposed surrogate model was later applied to pipeline leak detection and characterisation. The CFD modelling results showed that fluid flow parameters are pertinent indicators in pipeline leak detection. It was observed that upstream pipeline pressure could serve as a critical indicator for detecting leakage, even if the leak size is small. In contrast, the downstream flow rate is a dominant leakage indicator if the flow rate monitoring is chosen for leak detection. The results also reveal that when two leaks of different sizes co-occur in a single pipe, detecting the small leak becomes difficult if its size is below 25% of the large leak size. However, in the event of a double leak with equal dimensions, the leak closer to the pipe upstream is easier to detect. The results from all the analyses demonstrate the PSOASM algorithm's superiority over the well-known sequential sampling schemes employed for evaluation. The test results show that the PSOASM algorithm can be applied for pipeline leak detection with limited training datasets and provides a general framework for improving computational efficiency using adaptive surrogate modelling in various real-life applications.

Citation

ADEGBOYE, M.A. 2023. Adaptive swarm optimisation assisted surrogate model for pipeline leak detection and characterisation. Robert Gordon University, PhD thesis. Hosted on OpenAIR [online]. Available from: https://doi.org/10.48526/rgu-wt-2071535

Thesis Type Thesis
Deposit Date Sep 5, 2023
Publicly Available Date Sep 5, 2023
DOI https://doi.org/10.48526/rgu-wt-2071535
Keywords Remote sensing; Pipeline leakage detection; Computational fluid dynamics; Data models
Public URL https://rgu-repository.worktribe.com/output/2071535

Files





You might also like



Downloadable Citations