UTFacultiesETEventsPhD Defence Lisandro Jimenez Roa | Reliability and Maintenance for Engineering Systems: Fault Trees, Degradation Modelling and Maintenance Optimisation

PhD Defence Lisandro Jimenez Roa | Reliability and Maintenance for Engineering Systems: Fault Trees, Degradation Modelling and Maintenance Optimisation

Reliability and Maintenance for Engineering Systems: Fault Trees, Degradation Modelling and Maintenance Optimisation

The PhD defence of Lisandro Jimenez Roa will take place in the Waaier Building of the University of Twente and can be followed by a live stream.
Live Stream

Lisandro Jimenez Roa is a PhD student in the Department of Formal Methods and Tools. (Co)Promotors are prof.dr. M.I.A. Stoelinga from the Faculty of Electrical Engineering, Mathematics and Computer Science, prof.dr.ir. T. Tinga from the Faculty of Engineering Technology and prof.dr. T.M. Heskes from the Radboud University Nijmegen.

Modern infrastructures, machines, and manufacturing processes require effective management through sustainable policies under constrained resources, focusing crucially on determining "when" and "how" to intervene. The Prognostics and Health Management (PHM) paradigm provides a systematic framework for leveraging data collection and computational models, supporting the management of virtually any engineering component or system. This dissertation divided into three parts delves into three key aspects of PHM: Reliability Modelling, Markov Process-based Prognostics, and Maintenance Optimisation. Data-driven techniques are crucial in these areas, enhancing the automation of model development and deployment.

Part I of the dissertation centres on fault trees, specifically on the automatic inference of Fault Tree (FT) models. Traditionally, these graph-based models are manually constructed through iterative collaboration between system experts and FT modellers, a method prone to human error and potentially resulting in incomplete models. With the increasing availability of data, methodologies that automate this process, discover patterns, and reduce dependency on manual intervention have gained significant relevance. To this end, this dissertation proposes the use of Multi-Objective Evolutionary Algorithms (MOEAs) to automatically infer FT models and casts the optimization as a multi-objective task. This is implemented through the FTMOEA algorithm, which focuses on three optimization metrics, including FT size and accuracy-related metrics. Although FTMOEA consistently produced compact FT structures, it faced scalability issues. To address this, the SymLearn toolchain was developed, which uses a divide-and-conquer approach by identifying modules and symmetries in the failure dataset, thus breaking the inference problem into smaller tasks. Additionally, to improve robustness and scalability, the FTMOEA-CM extension incorporates additional metrics from the confusion matrix.

Part II I of the dissertation focuses on prognostics, specifically the stochastic degradation modelling of sewer mains. Sewer systems are vital to social welfare but present challenges due to their large scale, slow degradation, and limited ability to monitor the entire network. Accurately modelling the degradation profile is essential for targeting inspections and maintenance, improving the reliability and availability of the networks. Various degradation models exist, including physics-based and data-driven approaches, each with its own advantages and limitations. This part addresses the application of data-driven approaches based on Markov chains, commonly used to model stochastic sequences through states and transitions, particularly to model damage severity levels in sewer mains with data collected through Closed Circuit Television cameras. A case study from a Dutch sewer network is presented, starting with Discrete Time Markov Chains (DTMCs) for degradation modelling and examining two Markov chain structures. Due to dataset challenges, such as interval-censored data, more advanced analysis was required. The Turnbull estimator for non-parametric analysis was applied to establish a ground truth, showing that inhomogeneous-time Markov chains more accurately capture non-linear stochastic behaviour. However, the implementation also uncovers issues such as overfitting, reducing the models' predictive power.

Finally, Part III of the dissertation focuses on the maintenance of sewer mains, where obtaining optimal maintenance policies for such large-scale systems is a complex task. Among the various techniques available, Reinforcement Learning (RL) approaches remain largely unexplored for devising strategic maintenance actions in sewer mains. In this part, the sequential decision-making problem is framed using Deep Reinforcement Learning (DRL) for component-level maintenance of sewer mains. This framework considers damage severity levels, tests different degradation model assumptions, and evaluates their impact on maintenance policy. The results show that agent-based policies outperformed heuristics by learning optimal sequences of maintenance actions, providing evidence that DRL-based techniques offer a flexible framework with the potential to improve heuristics and support maintenance decision-making for sewer mains. However, training these models to achieve the desired behaviour remains a challenging task.