Predictive maintenance anticipates and prevents component failures by analysing operational data for early signs of degradation. Traditional industry-standard models for aircraft systems are often rule-based, missing complex patterns and limiting scalability. This research develops a deep learning (DL) fault detection pipeline for the engine bleed air system of wide-body twin-engine aircraft, leveraging real-world operational sensor and maintenance data. An interpretable feature engineering framework extracts physically informed features, including dual-engine comparisons, to train a gated recurrent unit (GRU) fault detection model for robust temporal modelling of healthy and faulty conditions. Bayesian optimisation is implemented for hyperparameter tuning. However, the scarcity of representative failure data, which is a common issue in aviation, limits the achievable performance and fidelity of DL models. To address this, a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) is employed to augment the dataset with synthetic failure data. A post-processing block labeling technique is introduced to enhance fidelity, and a novel fidelity savings metric translates model predictions into operational savings. GAN-based augmentation enhances recall, precision, and F1-score, and outperforms traditional augmentation. Further case study results show that the WGAN-GP-augmented model delivers four times the operational savings compared to the industry-standard model, 50\% more than the non-augmented GRU, and 31\% more than traditional augmentation. ... Read more

Scarce failure data often causes unreliable results when making predictions concerning Remaining Useful Life (RUL). This study explores the use of deep generative models (DGMs) for augmenting turbofan engine datasets by CMAPSS to improve these RUL predictions. By implementing Conditional Tabular GANs (CTGAN) and Tabular Variational Autoencoders (TVAE), synthetic data is generated and validated using statistical metrics such as Wasserstein distance and Kolmogorov-Smirnov tests. Then, these new datasets are used in several compositions of both real and synthetic data to train regressors and subsequently let them make RUL predictions. The regressors, such as Random Forest Regressors (RFR) and Convolutional Neural Networks (CNN), evaluate performance improvements through RMSE and MAE metrics. Results indicate that adding synthetic data improves prediction robustness, particularly when data is limited. This highlights the potential of DGMs for Prognostics and Health Management (PHM) applications. ... Read more

Detecting change in sensor measurements is essential for maintaining product quality and ensuring efficiency in manufacturing processes. Traditionally, statistical methods such as control charts are used to detect changes by comparing new sensor measurements with historical data. However, in high-dimensional, short-run (HDSR) settings, where there are many sensors and only limited or no historical observations, change detection becomes challenging and sometimes even impossible. HDSR processes are mostly present in specialized industries where errors can be costly, such as: semiconductors, aerospace or shipping. Previous research highlights several control charts to address HDSR processes and also demonstrated that grouping sensors can improve change detection. Finding groups of sensors was done by incorporating expert knowledge or by combining similar sensor data to increase sample size. In this research, a novel procedure for finding groups of sensors is proposed, by using an algorithm that automatically groups sensors based on the maximization of the probability of detection. The procedure and three state-of-the-art alternatives are applied to a case study involving a semiconductor manufacturing process of a new electron optical module. The results reveal that the proposed procedure finds groups of sensors that reflect sensor covariance and process knowledge. Furthermore, it is shown that the probability of detecting persistent mean shifts is improved compared to the three alternative control charts. Specifically, the proposed procedure had faster detection of shifts and also a higher POD for small magnitude shifts. Areas for future research could be the extension of the proposed procedure to a Bayesian framework. ... Read more

The performance of Remaining Useful Life (RUL) prediction models is often limited by data scarcity, especially in safety-critical systems like aircraft engines where failure data is rare. To address this challenge, we propose the Super-SpaceTime GAN, a framework for generating synthetic condition monitoring (CM) data to enhance RUL predictions. The framework incorporates dual-conditioning on operating conditions (OCs) and RUL labels, an autoencoder-based latent space for denoising, and a supervised loss function to align synthetic data with real degradation trajectories. Evaluated on the CMAPSS FD004 dataset, the SuperSpaceTime GAN generates synthetic data that closely mimic real distributions, as verified using JensenShannon Distance, Principal Component Analysis, t-distributed Stochastic Neighbor Embedding, and a novel autoencoder-based health monitoring metric. The framework demonstrates improvement in prognostic performance in limited training data scenarios, with gains persisting even in data-rich settings. These findings highlight the potential of the Super-SpaceTime GAN to improve RUL predictions by addressing data scarcity, making it a valuable tool for Prognostics and Health Management (PHM). ... Read more

Research aimed at improving engine maintenance practices is vital to ensuring the aviation industry's high safety standards. Both preventive and corrective maintenance approaches result in either higher costs or increased failures. Predictive maintenance aims to achieve an optimal balance by using time series and historical failure data to anticipate future problems and plan engine maintenance activities. While extensive research has been done on the application of AI models for predictive maintenance using simulated datasets, limited research has been done using real world engine data. Simulated datasets lack the complexity of real world systems which undermines the applicability of their findings to real world situations. This paper addresses the gap between theory and industry by utilizing eight years of operating and maintenance data from the GEnx-1B, provided by KLM Engine Services. We expand classic residual based methods with a novel approach, sister engine analysis, to improve the performance of an LSTM-based fault detection model. The data is labeled based on maintenance reports, which provide the most accurate link to the underlying degradation state of an engine. These reports reveal a significant bias toward high pressure turbine degradation in the GEnx-1B. Furthermore, our findings indicate that incorporating sister engine analysis enhances the average F1 score of an AI-based fault detection model by 7.7% and that AI methods outperform the current industry standard when provided with the same input data. The use of physically relevant model features and LIME analysis ensures the model's behavior is interpretable. These results are significant as they help build trust in AI-based solutions for predictive maintenance, which is crucial for their broader adoption. ... Read more

Aviation is a major contributor to global warming and so pathways for decarbonization need to be explored. As battery technology advances, electric regional aviation emerges as a viable option. A key challenge, however, is that the aircraft batteries degrade over time, progressively restraining the operational capability of electric aircraft. Current approaches do not account for this effect. To address this gap, this research presents the Battery DegradationAware Electric Fleet Assignment framework that enables the integration of battery ageing into tactical scheduling. It does this by combining a rolling-horizon fleet assignment model with a battery degradation module that predicts and updates each aircraft’s state of health based on its flown missions, with battery replacement scheduled according to designated strategies to ensure continuity of operations. This framework is evaluated with five distinct experiments, tested on the KLM Cityhopper network and complemented by an additional case study for validation. The experiments validate the framework’s operational and degradation dynamics and demonstrate that battery degradation has a significant impact, with depreciation costs amounting to 0.91 €/km flown. Degradation reduces fleet range capability, lowering ASKs by 0.96%, revenue by 0.76%, and increasing total operating costs by 9.1% as battery replacements become necessary. The sensitivity analysis indicates that future battery price scenarios can make or break profitability. Consequently, operational models that ignore degradation effects will overstate the profitability of electric aircraft. ... Read more

Battery state of health (SOH) estimation is one of the three main analytical tasks of a battery management system (BMS), when viewed from engineering maintenance and prognostics perspective. With the current global effort towards more suitable and greener processes, lithium-ion batteries have shown to be an important element in facilitating this transition. One industry where this can be noticed in particular is the transportation sector, where a strong shift towards battery electric vehicles (BEV) can be observed. Within the aviation sector, current research efforts include electrical flight. However, numerous challenges remain, that are typically observable within a safety critical domain such as aerospace. One these challenges includes the determination of uncertainty in battery SOH prediction. This would provide improved transparency on the capabilities and limitation of a model, when used as part of a battery system. Within this report we propose the use of a bidirectional gated recurrent unit (Bi-GRU) with learnable soft attention, to predict battery SOH based on charge measurements. Uncertainty analysis is enabled through the use of simultaneous quantile regression (SQR) and orthonormal certificate (OC), to be able to highlight and distinguish the aleatoric and epistemic uncertainty of the proposed model. We afterwards evaluate the model for point prediction accuracy using standard metrics, and evaluate the produced uncertainty using specialised test cases and calibration metrics. We achieved strong results using the proposed framework on a 2-phase fast charging dataset published by Toyota. ... Read more

Autonomous robotic exploration must balance fast mapgrowth with robustness under uncertainty. Among frontier-based methods, information-theoretic variants (e.g., Shannon/Rényi) are fast and reliable. This thesis integrates Behavioral Entropy (BE)—which models human-like risk perception—into a planner-in-theloop deep reinforcement learning(DQN) framework that learns the BE risk parameter 𝛼 online.
The policy selects 𝛼 from a discrete action set, uses occlusion-aware visibility to compute BE-based information gain, and delegates motion to a classical A* planner; reward shaping couples coverage/entropy reduction with motion cost. Action-usage analysis reveals an adaptive risk schedule—aggressive early, conservative mid-episode, selectively aggressive late—enabling faster cleanup of residual area. Overall, a reinforcement-learning method with risk-attitude tuning yields a robust, planner-compatible explorer. ... Read more

Inherent subjectivity, inefficiencies, and the substantial cost related to human-based visual inspection of high-pressure turbine (HPT) blades has driven research in alternative automated techniques. The
combination of computer vision (CV) and deep learning (DL) provides a compelling alternative. However, as DL models increase in capability, their large parameter spaces require substantial data to
achieve robust training. Therefore, this study explores the use of two Auxiliary Classifier Generative Adversarial Networks (AC-GANs) to augment proprietary datasets for two failure modes: (a) a 3-
channel red-green-blue (RGB) model for obstructed holes, and (b) a 4-channel RGB plus depth (RGBD) model for foreign object damage, with the depth reconstructed using monocular depth estimation. A Differential Evolution Optimizer (DEO) was used for hyperparameter optimization of a ResNet-18 classification target model. In the obstructed hole dataset, GAN-based augmentation showed significantly improved accuracy, recall, F1, and AUC-ROC (p < 0.05), competing with traditional methods using less augmentation. In the foreign object damage dataset, the inclusion of depth information significantly enhanced accuracy, precision, F1, and AUC-ROC performance (p < 0.05). However, the RGBD augmentation mainly resulted in a trade-off between precision and recall, without statistically significant differences (p > 0.05). ... Read more

For the remaining useful life (RUL) prediction of bearings across varying operating conditions, transfer learning models have demonstrated high accuracy. To make use of the maximum amount of bearing degradation information, multi-source domain adaptation models have been developed to enable predictions across operating conditions and bearing types. These models typically make use of vibration data in the time, frequency, and time-frequency domains. However, in practice, some systems only collect data in the frequency domain, such that prediction models developed for these systems lack access to degradation features from the time or time-frequency domains. To address this, this study proposes a multi-source domain adaptation (MSDA) model for bearing RUL prediction across operating conditions and bearing types using only frequency domain data as input data. The model employs a common feature extractor to capture domain invariant features, and domain specific regressors for RUL prediction. Domain adaptation is done by minimising the maximum mean discrepancy between features from the source and target domains, while the domain specific RUL predictions are aligned by minimising the distance between the RUL predictions. Experimental results across three scenarios demonstrate that the MSDA model achieves accurate RUL predictions. Even with only frequency domain data, the RMSE and Score are comparable to those of other advanced algorithms that use both frequency and time domain data. ... Read more

During operation aircraft brakes degrade due to wear. This degradation can be continuously monitored using brake degradation sensors. Using this monitored degradation data the remaining useful life of the brakes can be estimated by means of a prognostic model based on a Gamma probability distribution. Using the remaining useful life estimations a maintenance schedule can be optimized for a fleet of aircraft each fitted with multiple brakes accounting for maintenance constraints. Considered maintenance constraints include the availability of the maintenance hangar and the aircraft flight schedules. Because the problem contains multi-unit systems opportunistic maintenance can be applied, specifically economic dependance between components is considered in the optimization. Using a rolling horizon approach a long-term maintenance schedule can be created and evaluated. This long-term schedule is simulated using Monte Carlo simulation to achieve robust results on which meaningful conclusions can be drawn. To compare the results of the condition-based maintenance strategy a time-based maintenance strategy with fixed replacement intervals is used. It has been shown that the application of a condition-based maintenance strategy which makes use of prognostics can significantly improve the scheduling performance. ... Read more