Intelligent Signal Processing

Title: An unsupervised bearing fault diagnosis based on deep subdomain adaptation under noise and variable load condition

Abstract:

Deep learning-based approaches for diagnosing bearing faults have attracted considerable attention in the last years. However, in real-world applications, these methods face challenges. For proper training of these models, a considerable amount of labeled data are necessary, and due to limitations in industry, obtaining this amount of data may not be possible. Because of load variations, the distribution of training and test data may vary, which reduces the accuracy of the trained model for various working conditions. Furthermore, noise has a significant impact on bearing fault diagnosis performance in real-world industrial applications. This study introduced the deep subdomain adaptation convolutional neural network (DSACNN) method to overcome these challenges in real scenarios. The local maximum mean discrepancy (LMMD) method reduces the difference between each class distribution in the source and target domains. We validated our proposed method by CWRU bearing dataset under various loads and noise with different SNRs. The results show that DSACNN outperforms other comparative methods in anti-noise performance and reduction of domain distribution discrepancies.

Title: An Intelligent Gearbox Fault Diagnosis under Different Operating Conditions using Adversarial Domain Adaptation

Abstract:

Effective gearbox diagnostic procedures can assist in rotary machinery’s reliable and safe operation. On the other hand, the constant change in working conditions and the lack of labeled data have made fault diagnosis difficult. Changes in working conditions that cause discrepancies in data distribution and a lack of labeled data dramatically reduce fault diagnosis accuracy in deep learning algorithms. Unsupervised domain adaptation (UDA) has been utilized to overcome these challenges in various applications in recent years. This paper presents a novel method based on the CNN model and hybrid domain adaptation called deep coral adversarial network (DCAN) to solve these issues. The CNN model is used to extract features, and the distribution discrepancy between domains is decreased by applying two modules of domain adversarial learning and deep coral adaptation. The proposed method’s performance was evaluated using the SEU gearbox dataset. The results demonstrate the proposed method’s proper performance in diagnosing gearbox faults under different operation conditions. Index Terms—Gearbox fault diagnosis, adversarial domain adaptation, coral, domain adaptation

Title: Synthetic to Real Framework based on Convolutional Multi-Head Attention and Hybrid Domain Alignment

Abstrack:

Domain adaptation (DA) has obtained remarkable results in unsupervised fault diagnosis methods. However, its performance is highly related to two significant factors: Firstly, proposed DA methods should alleviate the global and local distribution gap to precisely matching all distributions in the source and target domain. Most distance-based DA methods assume global DA or concentrate only on local aligning distributions on the country. Secondly, the generalization of most proposed unsupervised fault diagnosis methods relies on labeled faulty data collected from sensors. Contrarily, collected data in real-world scenarios are mostly unlabeled, which considerably declines the model’s generalization. We proposed a synthetic to the real framework to overcome two significant challenges. A convolution multi-head attention network based on hybrid multi-layer domain alignment (CMHA-HMLDA) is conducted concurrently to align global and local distributions. It also alleviates the gap between real and synthetic data more accurately to maintain a robust data-driven model for bearing fault diagnosis. Furthermore, the proposed method is reliable in real scenarios because of transferring employed knowledge of labeled synthetic data into unlabeled real data. To show the supervisory of our proposed method in diagnosing unlabeled real health states, we validated it with a synthetic dataset made from the benchmark bearing Case Western Reserve University (CWRU) dataset. We compared it with recently published unsupervised fault diagnosis methods. Consequently, we achieved state-of-art results that show our proposed method is capable of realizing unlabeled real bearing faults from synthetic data, and it is practical in realworld scenarios

Title: Intelligent Fault Diagnosis of Rolling Bearing Based on Deep Transfer Learning Using Time-Frequency Representation

Abstrack:

With the expansion of deep learning (DL) and machine learning (ML) methods, fault diagnosis based on data-driven models has recently become controversial. However, due to the lack of sufficient labeled data in fault diagnosis, the depth of proposed DL models is less than other models in computer vision areas, which decreases the generalization and accuracy of models. Deep transfer convolutional neural network (DTCNN) with powerful feature extracting is used to tackle this dilemma. In this study, DenseNet201, ResNet152V2 and, MobileNetV2 are chosen as DTCNN models for feature extraction. Firstly, vibration signals are converted into time-frequency RGB images by continuous wavelet transform (CWT). Then, the high-level features of images are extracted by the DTCNN models. Finally, different types of bearing faults are classified by DL and ML classifiers. The experiment is validatedn on the famous Case Western Reserve University (CWRU) bearingn data set. The result demonstrates that the proposed DTCNN models achieve the best accuracy rate in the classification task and are faster to learn than many other existing DL and ML models