Real-time monitoring is essential in industrial applications such as tool wear detection and temperature compensation, where maintaining process stability and ensuring product quality are critical. To address the challenges caused by heterogeneous data sources and device types, this paper proposes a real-time edge computing system based on the Intel Edge Insights for Industrial (EII) platform. The system utilizes multiple AI Neural Compute Sticks (NCS) to enable accelerated and parallel inference for multiple tasks. The system integrates two AI models in separate containers. One model performs tool wear detection using a U-Net segmentation network, and the other conducts temperature compensation through a one-dimensional convolutional neural network. Each container operates on an independent NCS to support efficient and isolated task execution. Data communication is managed by the EII Message Bus, stored in InfluxDB, and visualized using Grafana, enabling complete data flow from acquisition to real-time presentation. Experimental results demonstrate that the system can concurrently handle multiple monitoring tasks with low latency and reliable performance. The proposed architecture enhances operational efficiency, supports intelligent automation, and provides a scalable solution for smart manufacturing environments.

AI-driven tool wear monitoring models playa crucial role in the manufacturing sector by accurately forecasting and identifying tool degradation. This capability enables the reduction of downtime and ensures the maintenance of cutting quality. Nevertheless, these models encounter obstacles in noisy manufacturing settings, where environmental variables may disrupt sensor data, impacting the precision of wear predictions. As a result, this article presents a novel UNet-based noise reduction model designed to eliminate diverse environmental noise from cutting sounds. This model is trained using hybrid signals in both the time-domain and frequency domain as part of the loss function. It enables the model to capture both temporal and spectral characteristics of the data, allowing for a more comprehensive representation of the signal's behavior. Experiments show that the Signal-to-Noise Ratio (SNR) can be effectively increased by over 3dB compared to the baseline across various cutting workpieces. Additionally, the proposed method exhibits superior robustness against various types and levels of noise. The results demonstrate that the quality of the cutting sound can be enhanced by over 7dB and 4dB respectively, following the application of the noise reduction technique.

Automated monitoring of tool wear is crucial for maintaining product quality. Furthermore, implementing AI techniques for real-time tool monitoring involves not only developing models but also managing their versions, avoiding the issue of models becoming less accurate as the properties of the machinery change over time. Consequently, this study develops a tool wear prediction system integrated with an artificial intelligent (AI) model management platform. First, this system uses various machine learning models to extract diverse signal features from sensor fusion, thereby boosting the accuracy of tool wear prediction. Secondly, the AI Models Management Platform comprises the C# programming language, Neural Networks Processing Unit (NPU) board, and Docker on both user and server sides, enhancing industrial processes and enabling real-time analysis of sensor data. According to these results, the Ensemble Learning method within the machine learning model demonstrates superior performance, yielding an average root mean squared error (RMSE) of$\mathbf{0.000185}\ mm^{2}$. Additionally, AI model management platform efficiently handle various model versions and streamline data training processes, empowering users to select suitable models and thereby enhancing system robustness.

Self-supervised learning on point clouds has gained a lot of attention recently, since it addresses the label-efficiency and domain-gap problems on point cloud tasks. In this paper, we propose a novel self-supervised framework to learn informative features from partial point clouds. We leverage partial point clouds scanned by LiDAR that contain both content and pose attributes, and we show that disentangling such two factors from partial point clouds enhances feature learning. To this end, our framework consists of three main parts: 1) a completion network to capture holistic semantics of point clouds; 2) a pose regression network to understand the viewing angle where partial data is scanned from; 3) a partial reconstruction network to encourage the model to learn content and pose features. To demonstrate the robustness of the learnt feature representations, we conduct several downstream tasks including classification, part segmentation, and registration, with comparisons against state-of-the-art methods. Our method not only outperforms existing self-supervised methods, but also shows a better generalizability across synthetic and real-world datasets.

In this work, we aim to address the 3D scene stylization problem - generating stylized images of the scene at arbitrary novel view angles. A straightforward solution is to combine existing novel view synthesis and image/video style transfer approaches, which often leads to blurry results or inconsistent appearance. Inspired by the high-quality results of the neural radiance fields (NeRF) method, we propose a joint framework to directly render novel views with the desired style. Our framework consists of two components: an implicit representation of the 3D scene with the neural radiance fields model, and a hypernetwork to transfer the style information into the scene representation. To alleviate the training difficulties and memory burden, we propose a two-stage training procedure and a patch sub-sampling approach to optimize the style and content losses with the neural radiance fields model. After optimization, our model is able to render consistent novel views at arbitrary view angles with arbitrary style. Both quantitative evaluation and human subject study have demonstrated that the proposed method generates faithful stylization results with consistent appearance across different views.