Remote Monitoring of Production Operations using a Smart Gateway Device

近年来，深度学习算法在许多认知应用中的准确率已经超过了人类水平。这一趋势促使研究人员使用深度学习算法来分析从传感器收集的数据，用于许多应用，如健康和自动驾驶。在深度学习算法的设计中，优化的首要焦点是精确度和吞吐量。然而，通过在一些应用中引入深度学习，需要对海量数据进行分析，实时处理是关键；需要新的技术来加速这一过程。 边缘设备由于其低功耗和高效率，在加速深度学习算法方面尤其有希望。使用网关设备应用深度学习模型被称为边缘计算。名称EDGE表示在框架的一个边缘应用模型，这与云计算不同，是在远程服务器上应用深度学习模型。针对云计算中边缘设备在数据管理和时延方面的性能和可靠性较低的问题，引入了FOG架构。 该方案旨在开发一种基于计算机视觉和深度学习的模型，并在网关设备上实现，以加快视频推理的速度和可靠性。将所提出的系统应用于从制造车间记录的视频，以获得实时检测，以进行过程监控和主动的效率改进。为了达到网关设备和深度学习的最佳性能，提出了雾计算来有效地使用这些设备和算法。

In recent years, deep learning algorithms have surpassed human-level accuracy for many cognitive applications. This trend has motivated researchers to use deep learning algorithms to analyze collected data from sensors for many applications, such as health and autonomous driving. In the design of deep learning algorithms, the primary focus of optimization has been accuracy and throughput. However, by introducing deep learning in some applications, the massive amount of data should be analyzed, and real-time processing is critical; new techniques are needed to accelerate the process. Edge devices are especially promising for accelerating deep learning algorithms due to their low power budget and high efficiency. Using gateway devices to apply deep learning models is referred to as edge computing. The name edge indicates applying the model at one edge of a framework, which is different from cloud computing, applying the deep learning model on a remote server. Because of the low capability and reliability of the edge devices on data management and latency on cloud computing, fog architecture is introduced. This proposal aims to develop a computer vision and deep learning-based model to be implemented on a gateway device to accelerate the video inferencing and reliability. The proposed system is applied to the video recorded from manufacturing floors to obtain real-time inspection for process monitoring and proactive efficiency improvement. To achieve the best performance in gateway device and deep learning, fog computing is proposed to use the device and algorithms effectively.