Associate Laboratory of Energy, Transports and Aeronautics

Funder

Organizational Unit

Publications

A super resolution method based on generative adversarial networks with quantum feature enhancement: Application to aerial agricultural images

Publication . El amraoui, Khalid; Pu, Ziqiang; Koutti, Lahcen; Masmoudi, Lhoussaine; Valente de Oliveira, JOSÉ

Super-resolution aims to enhance the quality of a low-resolution image to create a high-resolution one. Remarkable advances are witnessed in this field using machine learning techniques. This paper presents a superresolution method based on generative adversarial networks (GAN) with quantum feature enhancement. The proposed framework uses a feature enhancement layer inspired by the quantum superposition principle. The layer was added to the state -of -the art super-resolution GAN (SRGAN) original model to enhance its performance. The model was trained and evaluated using two publicly available high-resolution aerial images datasets taken by an unmanned aerial vehicle. A set of statistically significant experiments are reported to show its performance. The structural similarity index metric (SSIM), t-distributed stochastic neighbor embedding (t -SNE) and peak signal-to-noise ratio (PSNR) are adopted to evaluate the performance of this proposal against SRGAN model. Results show that this proposal outperforms SRGAN in term of image reconstruction quality by 8% in similarity.

2024Journal article

Restricted

VGAN: generalizing MSE GAN and WGAN-GP for robot fault diagnosis

Publication . Pu, Ziqiang; Cabrera, Diego; Li, Chuan; Valente de Oliveira, JOSÉ

Generative adversarial networks (GANs) have shown their potential for data generation. However, this type of generative model often suffers from oscillating training processes and mode collapse, among other issues. To mitigate these, this work proposes a generalization of both mean square error (mse) GAN and Wasserstein GAN (WGAN) with gradient penalty, referred to as VGAN. Within the framework of conditional WGAN with gradient penalty, VGAN resorts to the Vapnik V-matrix-based criterion that generalizes mse. Also, a novel early stopping-like strategy is proposed that keeps track during training of the most suitable model. A comprehensive set of experiments on a fault-diagnosis task for an industrial robot where the generative model is used as a data augmentation tool for dealing with imbalance datasets is presented. The statistical analysis of the results shows that the proposed model outperforms nine other models, including vanilla GAN, conditional WGAN with and without conventional regularization, and synthetic minority oversampling technique, a classic data augmentation technique.

2022Journal article

Restricted

Study on thermal comfort by using an atmospheric pressure dependent predicted mean vote index

Publication . Ruivo, Celestino; Gameiro da Silva, Manuel; Broday, Evandro Eduardo

The thermal environment index Predicted Mean Vote is a descriptor largely applied to evaluate the comfort sensation of people in moderate environments. Software tools based on the Fanger's method have been created and used, which application is limited for the sea level atmospheric pressure. A procedure for estimating the index of thermal comfort of individuals in environments at air pressure different from barometric pressure 0 m of altitude was recently introduced. In present study, indoor air state associated with neutral thermal comfort conditions of individuals in seating activity are predicted for atmospheric pressure values in the range between 65.0 kPa and 101.3 kPa. It was observed that neutral temperature diminishes with the decrease of barometric pressure. Dependencies of index PMV on the activity level, the clothing insulation, relative, air velocity and on relative humidity for two atmospheric pressure values are investigated. It is concluded that the impact of pressure of the environment on the index of thermal comfort must be also considered when sizing an air-conditioning system.

2021Journal article

Restricted

Free vibrations analysis of composite and hybrid axisymmetric shells

Publication . Simões Moita, José Mateus; Araujo, Aurelio L.; Franco Correia, Victor; Mota Soares, Cristóvão M.

The free vibration of laminated composite (C) and hybrid axisymmetric shell structures, consisting of a composite laminated material sandwiched between two functionally graded material laminas (F1/C/F2), is analysed in the present work. The numerical solutions are obtained by expanding the variables in Fourier series in the circumferential direction and using conical frustum finite elements in the meridional direction. The implemented finite element is a simple conical frustum with two nodal circles, with ten degrees of freedom per nodal circle. This model requires only a reduced number of finite elements to model the geometry of axisymmetric structures, the integration procedures use one Gauss point, and the through the thickness properties variation in FGM laminas is modelled by a small number of virtual layers, resulting a very high computational efficiency. The inhouse developed code presents very good solutions when compared with results obtained by alternative available models.

2022-04Journal article

Restricted

Optimized design of neural networks for a river water level prediction system

Publication . Lineros, Miriam López; Luna, Antonio Madueño; Ferreira, Pedro M.; Ruano, Antonio

In this paper, a Multi-Objective Genetic Algorithm (MOGA) framework for the design of Artificial Neural Network (ANN) models is used to design 1-step-ahead prediction models of river water levels. The design procedure is a near-automatic method that, given the data at hand, can partition it into datasets and is able to determine a near-optimal model with the right topology and inputs, offering a good performance on unseen data, i.e., data not used for model design. An example using more than 11 years of water level data (593,178 samples) of the Carrión river collected at Villoldo gauge station shows that the MOGA framework can obtain low-complex models with excellent performance on unseen data, achieving an RMSE of 2.5 × 10−3 , which compares favorably with results obtained by alternative design.

2021-10Journal article

Open Access