Browsing by Author "Pu, Ziqiang"
Now showing 1 - 9 of 9
Results Per Page
Sort Options
- Exploiting gan as an oversampling method for imbalanced data augmentation with application to the fault diagnosis of an industrial robotPublication . Pu, Ziqiang; Oliveira, José Valente de; Li, ChuanO diagnóstico inteligente de falhas baseado em aprendizagem máquina geralmente requer um conjunto de dados balanceados para produzir um desempenho aceitável. No entanto, a obtenção de dados quando o equipamento industrial funciona com falhas é uma tarefa desafiante, resultando frequentemente num desequilíbrio entre dados obtidos em condições nominais e com falhas. As técnicas de aumento de dados são das abordagens mais promissoras para mitigar este problema. Redes adversárias generativas (GAN) são um tipo de modelo generativo que consiste de um módulo gerador e de um discriminador. Por meio de aprendizagem adversária entre estes módulos, o gerador otimizado pode produzir padrões sintéticos que podem ser usados para amumento de dados. Investigamos se asGANpodem ser usadas como uma ferramenta de sobre amostra- -gem para compensar um conjunto de dados desequilibrado em uma tarefa de diagnóstico de falhas num manipulador robótico industrial. Realizaram-se uma série de experiências para validar a viabilidade desta abordagem. A abordagem é comparada com seis cenários, incluindo o método clássico de sobre amostragem SMOTE. Os resultados mostram que a GAN supera todos os cenários comparados. Para mitigar dois problemas reconhecidos no treino das GAN, ou seja, instabilidade de treino e colapso de modo, é proposto o seguinte. Propomos uma generalização da GAN de erro quadrado médio (MSE GAN) da Wasserstein GAN com penalidade de gradiente (WGAN-GP), referida como VGAN (GAN baseado numa matriz V) para mitigar a instabilidade de treino. Além disso, propomos um novo critério para rastrear o modelo mais adequado durante o treino. Experiências com o MNIST e no conjunto de dados do manipulador robótico industrial mostram que o VGAN proposto supera outros modelos competitivos. A rede adversária generativa com consistência de ciclo (CycleGAN) visa lidar com o colapso de modo, uma condição em que o gerador produz pouca ou nenhuma variabilidade. Investigamos a distância fatiada de Wasserstein (SWD) na CycleGAN. O SWD é avaliado tanto no CycleGAN incondicional quanto no CycleGAN condicional com e sem mecanismos de compressão e excitação. Mais uma vez, dois conjuntos de dados são avaliados, ou seja, o MNIST e o conjunto de dados do manipulador robótico industrial. Os resultados mostram que o SWD tem menor custo computacional e supera o CycleGAN convencional.
- Exploiting generative adversarial networks as an oversampling method for fault diagnosis of an industrial robotic manipulatorPublication . Pu, Ziqiang; Cabrera, Diego; Sánchez, René-Vinicio; Cerrada, Mariela; Li, Chuan; Valente de Oliveira, JoséData-driven machine learning techniques play an important role in fault diagnosis, safety, and maintenance of the industrial robotic manipulator. However, these methods require data that, more often that not, are hard to obtain, especially data collected from fault condition states and, without enough and appropriated (balanced) data, no acceptable performance should be expected. Generative adversarial networks (GAN) are receiving a significant interest, especially in the image analysis field due to their outstanding generative capabilities. This paper investigates whether or not GAN can be used as an oversampling tool to compensate for an unbalanced data set in an industrial manipulator fault diagnosis task. A comprehensive empirical analysis is performed taking into account six different scenarios for mitigating the unbalanced data, including classical under and oversampling (SMOTE) methods. In all of these, a wavelet packet transform is used for feature generation while a random forest is used for fault classification. Aspects such as loss functions, learning curves, random input distributions, data shuffling, and initial conditions were also considered. A non-parametric statistical test of hypotheses reveals that all GAN based fault-diagnosis outperforms both under and oversampling classical methods while, within GAN based methods, an average accuracy difference as high as 1.68% can be achieved.
- Exploiting graph neural network with one-shot learning for fault diagnosis of rotating machineryPublication . Shuai Yang; Xu Chen; Yu Wang; Yun Bai; Pu, ZiqiangInsufficient training data often leads to overfitting, posing a significant challenge in diagnosing faults in mechanical devices, particularly rotating machinery. To address this issue, this paper introduces a novel approach employing a graph neural network (GNN) with one-shot learning for fault diagnosis in rotating machinery. Firstly, the Short-Time Fourier Transform (STFT) is utilized for data preprocessing to convert the one-dimensional data into two-dimensional pictures. Subsequently, Feature extraction a convolutional neural network (CNN) is utilized to perform feature extraction. By introducing the adjacency matrix to explore the spatial information within data, a graph neural network (GNN) method is proposed to achieve the fault classification of rotating machinery with small sample. The method utilizes GNN to process structural information between, transferring the distance metric from Euclidean space to non-Euclidean space. Classification accuracy is thereby improved based on information processing in non-Euclidean space.Experiments were implemented on two datasets to verify the proposed method, including an open dataset of the rolling bearing and an experimental rig of the rotate vector (RV) reducer in an industrial robot. Siamese Net, Matching Net, and sparse auto-encoder with random forest (SAE + RF) wereemployed as the comparisons to further prove the effectiveness of the proposed method. Results indicate that the proposed method outperforms all the comparative methods in both rotating machineries.
- Fault diagnosis for wind turbine gearboxes by using deep enhanced fusion networkPublication . Pu, Ziqiang; Li, Chuan; Zhang, Shaohui; Bai, YunThe gearbox will directly affect the safety and reliability of the wind turbine, whose failure leads to low processing accuracy and certain economic losses. To address this issue, a deep enhanced fusion network (DEFN) is proposed for the fault diagnosis of the wind turbine gearbox with the experimental vibration data. In the proposed DEFN, three sparse autoencoders are first applied to extract deep features of three-axial vibration signals, respectively. Second, a feature enhancement mapping is developed to minimize the intraclass distance of the deep features in the three-axial vibration. Finally, the fused three-axis features are put into an echo state network for fault classification. The results of the experiment carried out in a wind turbine show that the proposed DEFN has a good fault diagnosis accuracy compared with other peer models.
- Generative adversarial one-shot diagnosis of transmission faults for industrial robotsPublication . Pu, Ziqiang; Cabrera, Diego; Bai, Yun; Li, ChuanTransmission systems of industrial robots are prone to get failures due to harsh operating environments. Fault diagnosis is of great significance for realizing safe operations for industrial robots. However, it is difficult to obtain faulty data in real applications. To migrate this issue, a generative adversarial one-shot diagnosis (GAOSD) approach is proposed to diagnose robot transmission faults with only one sample per faulty pattern. Signals representing kinematical characteristics were acquired by an attitude sensor. A bidirectional generative adversarial network (Bi-GAN) was then trained using healthy signals. Inspired by way of human thinking, the trained encoder in Bi-GAN was taken out to perform information abstraction for all signals. Finally, the abstracted signals were sent to a random forest for the one-shot diagnosis. The performance of the present technique was evaluated on an industrial robot experimental setup. Experimental results show that the proposed GAOSD has promising performance on the fault diagnosis of robot transmission systems.
- A one-class generative adversarial detection framework for multifunctional fault diagnosesPublication . Pu, Ziqiang; Cabrera, Diego; Bai, Yun; Li, ChuanIn this article, fault diagnosis is of great significance for system health maintenance. For real applications, diagnosis accuracy suffers from unbalanced data patterns, where normal data are usually abundant than anomaly ones, leading to tremendous diagnosis obstacles. Therefore, it is challenging to use only normal data for fault diagnosis under this imbalanced condition. In addition, a single fault diagnosis model can only conduct one fault diagnosis task in most of cases. Accordingly, a one-class generative adversarial detection (OCGAD) framework based on semisupervised learning is proposed to learn one-class latent knowledge for dealing with multiple semisupervised fault diagnosis tasks, i.e., fault detection using only normal knowledge learning, novelty detection from unknown conditional data, and fault classification with unlabeled data. A bi-directional generative adversarial network (Bi-GAN) is first trained with only normal data. A one-class support vector machine is then established using features exacted by Bi-GAN from signals acquired from an attitude sensor for multifunctional fault detection. The presented OCGAD model is validated using an industrial robot with experiments of three fault detection tasks. The results demonstrate that the present model has good performance for dealing with multiple semisupervised diagnosis problems.
- Sliced Wasserstein cycle consistency generative adversarial networks for fault data augmentation of an industrial robotPublication . Pu, Ziqiang; Cabrera, Diego; Li, Chuan; Valente de Oliveira, JOSÉWe investigate the role of the loss function in cycle consistency generative adversarial networks (CycleGANs). Namely, the sliced Wasserstein distance is proposed for this type of generative model. Both the unconditional and the conditional CycleGANs with and without squeeze-and-excitation mechanisms are considered. Two data sets are used in the evaluation of the models, i.e., the well-known MNIST and a real-world in-house data set acquired for an industrial robot fault diagnosis. A comprehensive set of experiments show that, for both the unconditional and the conditional cases, sliced Wasserstein distance outperforms classic Wasserstein distance in CycleGANs. For the robot faulty data augmentation a model compatibility of 99.73% (conditional case) and 99.21% (unconditional case) were observed. In some cases, the improvement in convergence efficiency was higher than 2 (two) orders of magnitude.
- A super resolution method based on generative adversarial networks with quantum feature enhancement: Application to aerial agricultural imagesPublication . El amraoui, Khalid; Pu, Ziqiang; Koutti, Lahcen; Masmoudi, Lhoussaine; Valente de Oliveira, JOSÉSuper-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.
- VGAN: generalizing MSE GAN and WGAN-GP for robot fault diagnosisPublication . Pu, Ziqiang; Cabrera, Diego; Li, Chuan; Valente de Oliveira, JOSÉGenerative 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.