Real-time monitoring of electrohydrodynamic spraying modes research based on lightweight YOU ONLY LOOK ONCE-v11–Group Shuffle Attention deep learning model

The evolution of electrohydrodynamic (EHD) spraying modes is affected by a complex interplay of operational parameters, thereby imposing a significant workload on related research. Consequently, developing more efficient and streamlined real-time monitoring methodologies for the recognition of EHD spraying modes becomes indispensable. In this paper, an EHD spraying mode recognition and real-time monitoring method is proposed and verified. Over 24 500 experimental images covering eight EHD modes were collected to construct a dataset for model training after data augmentation. The selection of the YOLOv11 (YOU ONLY LOOK ONCE, version 11) baseline model was acquired via comparing YOLO models from v5 to v13. With the integration of Group Shuffle Attention (GSA) and Spatial-Adaptive Attention, the final YOLOv11–GSA architecture was established. The model achieved a great preference with Precision of 95.2%, Recall of 95.5%, and mAP50 of 97.8% by only a minimal increase in parameter/Floating Point Operation due to the attention mechanism focusing more on regions with distinct features in attention heatmaps. In addition, critical parameter–mode interactions were systematically identified. Specifically, increasing the volume flow rate and enlarging the nozzle–collector distance were confirmed to induce an elevation in transition voltages; in contrast, a smaller nozzle diameter was found to reduce the transition voltage. Additionally, a nozzle tilt angle ranging from 30° to 60° was verified to effectively widen the stability range of the twin-jet mode. This work provides a methodological reference for EHD experiments, with real-time capability enabling directional EHD mode regulation.