A 5G-Enabled Edge–Cloud Collaborative Framework for Anomaly Detection in Industrial Compressor PLC Data

Industrial compressor systems are essential in manufacturing, and their failure can cause substantial economic losses and safety risks. Traditional anomaly detection methods based on centralized cloud processing often suffer from high latency and limited real-time capability. Although some recent edge-based solutions perform local processing, they do not always achieve a satisfactory balance between detection accuracy and inference latency, which is crucial in industrial applications. This paper proposes a 5G-enabled edge–cloud anomaly detection framework for industrial compressor Programmable Logic Controller (PLC) data. The framework deploys lightweight machine learning models on edge nodes for real-time inference, while using the cloud for model training and updating. The proposed Long Short-Term Memory Autoencoder (LSTM-AE) is optimized for edge deployment, achieving a mean F1-score of 0.843 ± 0.013 with an inference latency of 94 ms. Compared with traditional cloud-based methods, the proposed system reduces latency and improves the timeliness of anomaly detection. Compared with other edge-based approaches, such as Anomaly Transformer, which provides higher accuracy but 312 ms latency, and the Gated Recurrent Unit (GRU) autoencoder, which provides lower latency of 87 ms but lower accuracy, the proposed method achieves a more favorable balance between detection performance and inference efficiency. By assigning training to the cloud and keeping inference at the edge, the framework also improves economic and energy efficiency compared with cloud-only solutions. These results demonstrate the potential of integrating 5G-enabled edge computing with machine learning for real-time, scalable, and cost-effective anomaly detection in industrial environments.