Search

Excavation autonomy is an area of industrial interest as there is a significant skills shortage for excavator operators, yet the industry is facing higher budget constraints than ever before. In particular, the excavation task of trenching is heavily impacted as it is difficult to perform but is also integral for most construction projects. Traditionally, autonomy in excavation is often rule-based or control theory-based, which is difficult to scale and apply to different scenarios such as different ground conditions or trench design, common in construction. Reinforcement learning has been applied with successes in similar fields, like robotics, demonstrating its capability in handling non-linearity and complex environments. This thesis addresses the deployment reinforcement learning to a trench excavation task, to provide a potential operator-like behaviour. Twin-Delayed Deep Deterministic Policy Gradients (TD3) were identified as the most appropriate for the research, due to their superior performance on robotics tasks. To deploy TD3 to trenching, a three-part strategy was used that focused on developing the environment, state, and reward function that TD3 would be involved with. First, the overall system architecture and environment for a reinforcement learning-based autonomy system was designed, which utilised the Common Data Environment to store and deploy a trained algorithm and the excavation plans. This was designed alongside driver data analysis to provide additional context to the challenges of trenching. Next, the selection of optimal machine sensors was studied, including the comparison of feature selection methods against existing sensor arrays. These were applied to a neural network trained on driver data, and the assumption was this would be like the policy of a trained TD3 algorithm. This was done to determine what impact the state has on the predictive performance of the policy whilst avoiding the inclusion of a reward function, allowing several machine tasks to use the same approach. Finally, TD3 was deployed to an excavator, focused on developing the reward function for trenching. This was done by breaking the trenching task into smaller tasks for developing the distance and mass rewards, before unifying them into one reward function. By focusing on individual elements of a reward function, the impact can be understood more effectively. TD3 was then deployed to perform a trench excavation task, where it was able to dig within the trench region and dump soil into a hopper. The novel findings of this research were 1) Exploratory analysis of driver data during trenching 2) An autonomy roadmap for excavation 3) An architecture of an RL-based autonomy system for excavation that utilises the driver, simulation, and Common Data Environment 4) The use of feature selection methods to determine machine sensor inputs for the policy network of TD3, without using the reward function 5) A new methodology for developing a reward function 6) The deployment of TD3 to perform a trench excavation task.

his study investigated the integration of long-term monitoring into the structural engineering design process to improve the design and operation of civil structures. A survey of civil and structural engineering professionals, conducted as part of this research, identified the cost and complexity of in-situ monitoring as key barriers to their implementation in practice. Therefore, the research focused on the use of ambient vibration monitoring as it is offers a low cost and unobtrusive method for instrumenting new and existing structures. The research was structured around the stages of analysing ambient vibration data using operational modal analysis (OMA), defined in this study as: i) pre-selection of analysis parameters, ii) pre-processing of the data, iii) estimation of the modal parameters, iv) identification of modes of vibration within the modal estimates, and v) using modal parameter estimates as a basis for understanding and quantifying in-service structural behaviour. A method was developed for automating the selecting of the model order, the number of modes of vibrations assumed to be identifiable within the measured dynamic response. This method allowed the modal estimates from different structures, monitoring periods or analysis parameters to be compared, and removed part of the subjectivity identified within current OMA methods. Pre-processing of ambient acceleration responses through filtering was identified as a source of bias within OMA modal estimates. It was shown that this biasing was a result of filtering artefacts within the processed data. Two methods were proposed for removing or reducing the bias of modal estimates induced by filtering artefacts, based on exclusion of sections of the response corrupted by the artefacts or fitting of the artefacts as part of the modal analysis. A new OMA technique, the short-time random decrement technique (ST-RDT) was developed on the basis of the survey of industry perceptions of long-term monitoring and limitations of existing structural monitoring techniques identified within the literature. Key advantages of the ST-RDT are that it allows the uncertainty of modal estimates and any changes in modal behaviour to be quantified through subsampling theory. The ST-RDT has been extensively validated with numerical, experimental and real-world case studies including multi-storey timber buildings and the world's first 3D printed steel bridge. Modal estimates produced using the ST-RDT were used as a basis for developing an automated method of identifying modes of vibration using a probabilistic mixture model. Identification of modes of vibration within OMA estimates was previously a specialized skill. The procedure accounts for the inherent noise associated with ambient vibration monitoring and allows the uncertainty within the modal estimates associated with each mode of vibration to be quantified. Methods of identifying, isolating and quantifying weak non-linear modal behaviour, changes in dynamic behaviour associated with changes in the distributions of mass or stiffness within a structure have been developed based on the fundamental equations of structural dynamics. These methods allow changes in dynamic behaviour associated with thermally-induced changes in stiffness or changes in static loading to be incorporated within the automated identification of modes of vibration. These methods also allow ambient vibration monitoring to be used for estimating structural parameters usually measured by more complex, expensive or delicate sensors. Examples of this include estimating the change in elastic modulus of simple structures with temperature or estimating the location and magnitude of static loads applied to a structure in-service. The methods developed in this study are applicable to a wide range of structural monitoring technologies, are accessible to non-specialist audiences and may be adapted for the monitoring of any civil structure.

Improving healthcare systems mandates a shift towards personalised and preventive medicine, especially in the management of cardiovascular diseases, the world leading cause of death. The anatomy and function of the heart can be accurately and non-invasively assessed using cardiac magnetic resonance. Nevertheless, it is not yet fully understood how their interplay modulates disease outcomes. In this regard, this Thesis (1) reviews the crucial role of computational modelling towards precision cardiology, by building the 'digital twin' of a patient; (2) develops a fully automated pipeline to construct 3D computational models of human left ventricle from 2D cardiac magnetic images, using state-of-art deep learning approaches; and (3) proposes an enhanced structural and functional analysis to disentangle how adverse outcomes are modulated in cardiovascular disease. This is showcased in acute myocardial infarction, the deadliest cardiovascular disease not only on first manifestation, but on reoccurrence and secondary adverse outputs. In this scenario, the approach introduced in this Thesis is shown to provide additional prognostic information towards a better post-infarction management. Moreover, it identifies 3D shape and contraction patterns that are related to adverse outcomes, building a more profound understanding of remodelling after infarction. Furthermore, the automation of the analysis removes the burden of manual annotations and intra- and inter-observer variability, as confounders of prognosis. All this argues for the clinical impact of this Thesis. Ultimately, the Thesis evidences the potential of 3D computational models for cardiovascular disease management, towards personalised and preventive care.