Scholarlink

<p>Industrial Internet of Things (IIoT) technology and artificial intelligence (AI) are two main pillars of the fourth industrial revolution. Among their applications, predictive maintenance of rotating components has been receiving more attention in smart factories due to their widespread use. In this work, a container-based IIoT infrastructure along with a lightweight yet powerful AI model is proposed to address this topic. The serverless IIoT infrastructure is based on the KubeEdge (KE) platform for the edges and Kubernetes (K8s) in the cloud, which function as a monolithic containerized infrastructure. Predictive maintenance can be achieved by the distributed monitoring of rotating component’s vibration and its analysis at the edges. Specifically, health indicator (HI) criteria are constructed to reveal the degradation trends. For this purpose, the vibration signals are first transformed into RGB images using the continuous wavelet transform (CWT) and then fed into the proposed AI model named LITA (Lightweight IIoT Transformer UNet Autoencoder). LITA is a lightweight and powerful deep learning model based on a modified Transformer UNet model. The training process of LITA models is handled by the containers in the cloud with virtually unlimited computational resources, and the real-time inference of the predictive maintenance application is performed entirely in a distributed processing manner at the edges. To evaluate the performance of the proposed LITA model and compare it with other methods, standard intelligent maintenance systems (IMS) bearing data were used. The evaluation results demonstrate the effectiveness of the proposed method, achieving 6.84% and 3.57% improvements in monotonicity and robustness standard metrics compared to state-of-the-art methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A containerized Edge AI predictive maintenance infrastructure for rotating components using continuous wavelet transform images and Transformer UNet autoencoder for Industry 4.0

  • 期刊论文

  • 发表日期：

    2025年10月7日

查看全文
  • Mohammad Ali Keshavarz,
  • Saeed Sharifian

摘要

Industrial Internet of Things (IIoT) technology and artificial intelligence (AI) are two main pillars of the fourth industrial revolution. Among their applications, predictive maintenance of rotating components has been receiving more attention in smart factories due to their widespread use. In this work, a container-based IIoT infrastructure along with a lightweight yet powerful AI model is proposed to address this topic. The serverless IIoT infrastructure is based on the KubeEdge (KE) platform for the edges and Kubernetes (K8s) in the cloud, which function as a monolithic containerized infrastructure. Predictive maintenance can be achieved by the distributed monitoring of rotating component’s vibration and its analysis at the edges. Specifically, health indicator (HI) criteria are constructed to reveal the degradation trends. For this purpose, the vibration signals are first transformed into RGB images using the continuous wavelet transform (CWT) and then fed into the proposed AI model named LITA (Lightweight IIoT Transformer UNet Autoencoder). LITA is a lightweight and powerful deep learning model based on a modified Transformer UNet model. The training process of LITA models is handled by the containers in the cloud with virtually unlimited computational resources, and the real-time inference of the predictive maintenance application is performed entirely in a distributed processing manner at the edges. To evaluate the performance of the proposed LITA model and compare it with other methods, standard intelligent maintenance systems (IMS) bearing data were used. The evaluation results demonstrate the effectiveness of the proposed method, achieving 6.84% and 3.57% improvements in monotonicity and robustness standard metrics compared to state-of-the-art methods.