Integration of machine learning and building information modelling for predictive railway maintenance

Sresakoolchai, Jessada ORCID: 0000-0003-3692-6422 (2023). Integration of machine learning and building information modelling for predictive railway maintenance. University of Birmingham. Ph.D.

Preview

Sresakoolchai2023PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (12MB) | Preview

Abstract

The aims of this thesis are to integrate machine learning or artificial intelligence (AI) and building information modelling (BIM) for predictive railway maintenance. The study process starts by collecting data. Data sources in this thesis consist of two main sources which are field data and numerical data. For field data, it is mainly from MRS Logística S.A. which is a railway authority in Brazil. Examples of field data are track design, track measurement, defect inspection report, maintenance report, and maintenance manual. For numerical data, data are numerically generated using simulations such as multi-body simulations and finite element models. To ensure that numerical data are reliable and realistic, every simulation and finite element model is validated using relevant field data. If differences between numerical data and field data are in the acceptable range, it demonstrates that numerical data can be used as representatives of field data. An advantage of using numerical data is it can create data diversity and variation which is required for machine learning model development. Then, data are prepared and processed to make them in forms that can be used to train machine learning models. Different machine learning models will be used in this thesis. The techniques can be grouped into three types which are supervised learning, unsupervised learning, and reinforcement learning. The application of these machine learning models is to develop predictive maintenance in the railway system such as defect detection, defect severity evaluation, deterioration prediction, and maintenance plan preparation. To fulfil the aim of the thesis, BIM models are developed and integrated with machine learning which can be conducted using different techniques. Results from the study show that the developed machine learning models can fulfil their purposes with high performance and the developed BIM models can be fully integrated with machine learning as a data management platform. Contributions of the study are the developed approach can improve the maintenance efficiency and support predictive maintenance in the railway, machine learning models can support decision making and have high performance according to their functions, and the railway system can improve reliability, availability, maintainability, safety, including passenger comfort when applies the developed approach proposed in this thesis.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):