Wind turbine blade maintenance is expensive, dangerous, time-consuming, and prone to misdiagnosis. A potential solution to aid preventative maintenance is using deep learning and drones for inspection and early fault detection. In this research, five base deep learning architectures are investigated for anomaly detection on wind turbine blades, including Xception, Resnet-50, AlexNet, and VGG-19, along with a custom convolutional neural network. For further analysis, transfer learning approaches were also proposed and developed, utilizing these architectures as the feature extraction layers. In order to investigate model performance, a new dataset containing 6000 RGB images was created, making use of indoor and outdoor images of a small wind turbine with healthy and damaged blades. Each model was tuned using different layers, image augmentations, and hyperparameter tuning to achieve optimal performance. The results showed that the proposed Transfer Xception outperformed other architectures by attaining 99.92% accuracy on the test data of this dataset. Furthermore, the performance of the investigated models was compared on a dataset containing faulty and healthy images of large-scale wind turbine blades. In this case, our results indicated that the best-performing model was also the proposed Transfer Xception, which achieved 100% accuracy on the test data. These accuracies show promising results in the adoption of machine learning for wind turbine blade fault identification.

We consider the sparse recovery problem of signals with an unknown clustering pattern in the context of multiple measurement vectors (MMVs) using the compressive sensing (CS) technique. For many MMVs in practice, the solution matrix exhibits some sort of clustered sparsity pattern, or clumpy behavior, along each column, as well as joint sparsity across the columns. In this paper, we propose a new sparse Bayesian learning (SBL) method that incorporates a total variation-like prior as a measure of the overall clustering pattern in the solution. We further incorporate a parameter in this prior to account for the emphasis on the amount of clumpiness in the supports of the solution to improve the recovery performance of sparse signals with an unknown clustering pattern. This parameter does not exist in the other existing algorithms and is learned via our hierarchical SBL algorithm. While the proposed algorithm is constructed for the MMVs, it can also be applied to the single measurement vector (SMV) problems. Simulation results show the effectiveness of our algorithm compared to other algorithms for both SMV and MMVs.

It is well-known that many signals of interest can be well-estimated via just a small number of supports under some specific basis. Here, we consider finding sparse solution for Multiple Measurement Vectors (MMVs) in case of having both jointly sparse and clumpy structure. Most of the previous work for finding such sparse representations are based on greedy and sub-optimal algorithms such as Basis Pursuit (BP), Matching Pursuit (MP), and Orthogonal Matching Pursuit (OMP). In this paper, we first propose a hierarchical Bayesian model to deal with MMVs that have jointly-sparse structure in their solutions. Then, the model is modified to account for clumps of the neighbor supports (block sparsity) in the solution structure, as well. Several examples are considered to illustrate the merit of the proposed hierarchical Bayesian model compared to OMP and a modified version of the OMP algorithm.

Optimization is showcased as an accessible and powerful tool in this attractive and instructive text. Optimization techniques are presented with mathematical motivation, without extensive proofs, and with illustrative Julia code. This text is appropriate for an undergraduate or other introductory course on optimization. It is also suitable for a practicing engineer who wants a broad first taste of the field, along with access to working code. Compared with similar texts, this book generally includes a broader set of topics, including uncertainty propagation, multiobjective optimization, multidisciplinary optimization, Gaussian process surrogate models, and expression optimization. The authors provide the building blocks of optimization algorithms rather than presenting a deep treatment of optimization theory. This extensive use of Julia is the most noticeable innovation of the book. Functions and code are used to illustrate techniques as diverse as automatic differentiation, particle swarm optimization, and Bayesian Monte Carlo methods. The code is presented entirely in the text of the book and duplicated online in Jupyter notebooks. This extensive use of Julia is the most noticeable innovation of the book. Functions and code are used to illustrate techniques as diverse as automatic differentiation, particle swarm optimization, and Bayesian Monte Carlo methods.

Long-wave infrared (LWIR) hyperspectral imagers (HSI) image a scene by collecting high-resolution spectra at each pixel. The data are similar to a camera image but have a large number of narrow spectral bands rather than the familiar broad three bands of red-green-blue in a traditional digital camera. Materials in LWIR are emissive (rather than reflective) and have unique spectra that can be used for material detection and identification. The measured spectra are a convolution of the material spectra (emissivity), the black body temperature (Planck curve), other interacting environmental spectral sources, and measurement error. One approach to material identification is temperature-emissivity separation (TES), which separates or deconvolves the material spectra from the temperature curve. To accomplish this task, we develop a unique flexible model which combines the mathematical model of the physical processes within a Bayesian nonparametric framework. In addition to offering interpretable estimates of model parameters, this model is able to identify material emissivity spectra and cluster pixels into appropriate material groups. We demonstrate our method using both a synthetic and measured dataset. The online supplementary material contains an appendix of the details of the sampling algorithm.