The equipment on the offshore platform will be subjected to all kinds of harm from the sea, such as saltwater fog, high humidity, mechanical shock, etc. Equipment failure will result in a high-cost shutdown. The old way of periodic maintenance may be too frequent or too infrequent. A Three-Tier Edge-Cloud Collaborative Architecture for Online Condition Monitoring and Smart Fault Diagnosis of Offshore Platform Equipment is Proposed in this paper. The three layers of the system are: the terminal sensing layer for vibration signal collection; the edge computing layer, which uses a light-weight Lite-1D-CNN model for anomaly detection based on knowledge distillation; and the cloud analysis layer, which employs a ten-class fault classification model of multi-scale residual networks (MS-ResNet). An adaptive data transmission method based on the confidence of anomaly detection at the edge and task scheduling between edge and cloud nodes has also been introduced to reduce transmitted data by 58.3%. Based on the experiments in the bearing dataset at Case Western Reserve University, the edge model achieved an accuracy of 94.6% for binary classification and a latency of 6.8ms; the cloud model reached 97.1% accuracy for ten-class classification. The accuracy and weighted end-to-end latency of the collaborative edge-cloud model have been 96.2% and 44ms, respectively. The experiments on the bearing dataset at Paderborn University have also been verified for fault diagnosis of offshore platform equipment and achieved an accuracy of 86.5% in a different field.
