Knowledge Embedded Lightweight Vision Transformer for Machine Condition Monitoring

•A prior knowledge embedding method is proposed for the machine condition monitoring of milling operation.•The prior knowledge of machining is fused with vibration signal and wear features before input to the designed intelligent model.•A lightweight visual transformer neural network, named Lsep-ViT, is designed to efficiently learn the fused features of machined surface texture.•The agility and latency of the feature extraction model are greatly enhanced with no more than 30 ms inference latency.