With the advantages of low latency, low power consumption, and high parallelism, optical neural networks (ONN) offer a promising solution for time-sensitive and resource-limited artificial intelligence applications. However, the performance of the ONN model is often diminished by the gap between the ideal simulated system and the actual physical system. To bridge the gap, this work conducts extensive experiments to investigate systematic errors in the optical physical system within the context of image classification tasks. Through our investigation, two quantifiable errors—light source instability and exposure time mismatches—significantly impact the prediction performance of ONN. To address these systematic errors, a physics-constrained ONN learning framework is constructed, including a well designed loss function to mitigate the effect of light fluctuations, a CCD adjustment strategy to alleviate the effects of exposure time mismatches and a ’physics-prior based’ error compensation network to manage other systematic errors, ensuring consistent light intensity across experimental results and simulations. In our experiments, the proposed method achieved a test classification accuracy of 96.5% on the MNIST dataset, a substantial improvement over the 61.6% achieved with the original ONN. For the more challenging QuickDraw16 and Fashion MNIST datasets, experimental accuracy improved from 63.0% to 85.7% and from 56.2% to 77.5%, respectively. Moreover, the comparison results further demonstrate the effectiveness of the proposed physics-constrained ONN learning framework over state-of-the-art ONN approaches. This lays the groundwork for more robust and precise optical computing applications.

In the mobile industrial human-machine interaction (HMI), to establish the data connection, the engineer has to manually select the target machine from a long list, which may lead to wrong connection and waste of time. We observe that the engineer should face to the machine during the interaction to ensure that the machine works accurately, and this characteristic makes the proximity estimation algorithm suitable to simplify the data connection. However, due to the densely deployed machines, the existing algorithms cannot provide sufficient accuracy with limited latency. In this paper, we implement a testbed to evaluate the performance in the mobile industrial HMI. Based on the experimental results, we propose the definition of received signal strength indicator (RSSI) difference and then use it to design the face-to-machine proximity estimation (FaceME) algorithm. The experimental results prove that FaceME can provide guaranteed estimation accuracy and low-time complexity.

To establish data connection in industrial wireless networks, the engineer has to enter the address of the target machine or manually select it from a list. Since the machines are densely deployed in the plant, the manual data connection becomes a non-trivial issue which may lead to connection error and waste of time. The existing technologies, such as QR code scanning and proximity estimation, simplify the process of data connection to the closest node. However, they can not be used to connect with any node located in line-of-sight of the engineer, and thus limits their application. In this paper, we propose a scheme called LightCon which controls the display module (LCD screen or LED light) of nodes to display different visible symbols such as colors and numbers, thus the engineer can select the corresponding symbol on the mobile device to establish connection. Moreover, symbol assignment algorithms are proposed to reduce the complexity of symbol selection when the number of nodes are greater than the number of symbols. The theoretical and experimental results prove that LightCon reduces the complexity of line-of-sight connection with guaranteed accuracy.