Search

Wind turbines play an integral role in generating clean energy, but regularly suffer from operational inconsistencies and failures leading to unexpected downtimes and significant Operations & Maintenance (O&M) costs. Condition-Based Monitoring (CBM) has been utilised in the past to monitor operational inconsistencies in turbines by applying signal processing techniques to vibration data. The last decade has witnessed growing interest in leveraging Supervisory Control & Acquisition (SCADA) data from turbine sensors towards CBM. Machine Learning (ML) techniques have been utilised to predict incipient faults in turbines and forecast vital operational parameters with high accuracy by leveraging SCADA data and alarm logs. More recently, Deep Learning (DL) methods have outperformed conventional ML techniques, particularly for anomaly prediction. Despite demonstrating immense promise in transitioning to Artificial Intelligence (AI), such models are generally black-boxes that cannot provide rationales behind their predictions, hampering the ability of turbine operators to rely on automated decision making. We aim to help combat this challenge by providing a novel perspective on Explainable AI (XAI) for trustworthy decision support. This thesis revolves around three key strands of XAI - DL, Natural Language Generation (NLG) and Knowledge Graphs (KGs), which are investigated by utilising data from an operational turbine. We leverage DL and NLG to predict incipient faults and alarm events in the turbine in natural language as well as generate human-intelligible O&M strategies to assist engineers in fixing/averting the faults. We also propose specialised DL models which can predict causal relationships in SCADA features as well as quantify the importance of vital parameters leading to failures. The thesis finally culminates with an interactive Question- Answering (QA) system for automated reasoning that leverages multimodal domain-specific information from a KG, facilitating engineers to retrieve O&M strategies with natural language questions. By helping make turbines more reliable, we envisage wider adoption of wind energy sources towards tackling climate change.

This PhD thesis investigates two key challenges in the area of fine-grained emotion detection in textual data. More specifically, this work focuses on (i) the accurate classification of emotion in tweets and (ii) improving the learning of representations from knowledge graphs using graph convolutional neural networks. The first part of this work outlines the task of emotion keyword detection in tweets and introduces a new resource called the EEK dataset. Tweets have previously been categorised as short sequences or sentence-level sentiment analysis, and it could be argued that this should no longer be the case, especially since Twitter increased its allowed character limit. Recurrent Neural Networks have become a well-established method to classify tweets over recent years, but have struggled with accurately classifying longer sequences due to the vanishing and exploding gradient descent problem. A common technique to overcome this problem has been to prune tweets to a shorter sequence length. However, this also meant that often potentially important emotion carrying information, which is often found towards the end of a tweet, was lost (e.g., emojis and hashtags). As such, tweets mostly face also problems with classifying long sequences, similar to other natural language processing tasks. To overcome these challenges, a multi-scale hierarchical recurrent neural network is proposed and benchmarked against other existing methods. The proposed learning model outperforms existing methods on the same task by up to 10.52%. Another key component for the accurate classification of tweets has been the use of language models, where more recent techniques such as BERT and ELMO have achieved great success in a range of different tasks. However, in Sentiment Analysis, a key challenge has always been to use language models that do not only take advantage of the context a word is used in but also the sentiment it carries. Therefore the second part of this work looks at improving representation learning for emotion classification by introducing both linguistic and emotion knowledge to language models. A new linguistically inspired knowledge graph called RELATE is introduced. Then a new language model is trained on a Graph Convolutional Neural Network and compared against several other existing language models, where it is found that the proposed embedding representations achieve competitive results to other LMs, whilst requiring less pre-training time and data. Finally, it is investigated how the proposed methods can be applied to document-level classification tasks. More specifically, this work focuses on the accurate classification of suicide notes and analyses whether sentiment and linguistic features are important for accurate classification.

Domain-Specific Aspect Languages (DSALs) are a valuable tool for separating cross-cutting concerns, particularly within fields with endemic cross-cutting practices. Agent-Based Modelling (ABM) runtime inspection, which cuts across the core concern of model development, serves as a prime example. Despite their usefulness, DSALs face multiple adoption issues: the literature regarding their development and use is incohesive, coupling to a weave target hinders re-use, and available tooling is immature compared to Domain-Specific Languages (DSLs). We believe these issues can be aided by furthering DSL middle-out techniques for DSALs. We first define the background of what a DSAL is and how they may be used, moving onto how we can use DSL techniques to further DSALs. We develop a middle-out semantic model approach for developing domain-level DSALs with transparent aspect orientation using adaptions of DSL techniques. We have implemented the approach for model-specific DSALs for the in-house framework Animaux, and as middleware-specific DSAL for agent messages in the JADE framework, which can be specialised to models using extension DSALs. We give illustrative result cases using our implementations to provide a base of the user development costs and performance of this approach. In conclusion, we believe the adoption of these technologies aids ABM applications and encourage future work in similar fields. This thesis has given a base philosophy toward DSLs, a novel approach for the development of middle-out DSALs and illustrative cases of this approach.