کنفرانس بین المللی کنترل ، ابزار دقیق و اتوماسیون

صفحه اصلی / هشتمین کنفرانس بین المللی کنترل ، ابزار دقیق و اتوماسیون

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

نویسندگان :

Mohammadreza Ghorvei¹ Mohammadreza Kavianpour² Mohammad TH Beheshti³ Amin Ramezani⁴

1- Tarbiat Modares University 2- Tarbiat Modares University 3- Tarbiat Modares University 4- Tarbiat Modares University

کلمات کلیدی :

Unsupervised Fault diagnosis, multi-head attention, hybrid domain alignment, synthetic data

چکیده :

Unsupervised Domain adaptation (UDA) has performed outstandingly in unsupervised fault diagnosis. However, its performance is highly related to two significant factors: Firstly, proposed UDA methods should alleviate the global and local distribution gap to match all distributions in the source and target domain precisely. On the country, Most distance-based DA methods assume global domain adaptation or concentrate only on local aligning distributions. 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 adaptation (CMHA-HMLDA) is conducted to simultaneously 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, our proposed method is reliable in real scenarios because it employs labeled synthetic data in the source domain to transfer knowledge into unlabeled real data in the target domain. To show the supervisory of our proposed method in diagnosing unlabeled real health states, we validated it with a synthetic dataset made from bearing benchmark Case Western Reserve University(CWRU) dataset and compared it with recently published UDA methods. Consequently, we achieved the 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 real-world scenarios.