The nozzle is the key element of the water jet generator for energy conversion. In order to explore the influence of the nozzle diameter on the pressure characteristics of the supercharged pulsed water jet plenum chamber, a supercharged pulsed water jet pressure acquisition system was established, and the equations of motion and theoretical pressurization ratio equations of the supercharged pulsed water jet generator were established. The pressurization chamber pressure acquisition experiments under different nozzle diameters were carried out. The research results show that the pressurized pulsed water jet generator has a critical nozzle diameter of 0.6 mm. When the nozzle diameter is less than the critical diameter, the pressure in the boost chamber is equal to the product of the driving pressure and the boost ratio. As the nozzle changes, there is no significant change in the peak pressure and frequency of the boost chamber. When the nozzle diameter is greater than the critical diameter, there is a non-linear relationship between the boost chamber pressure and the driving pressure. As the nozzle diameter gradually increases, the actual boost ratio gradually decreases, and the peak pressure of the boost chamber further decreases. The nozzle diameter can no longer provide a load for the establishment of fluid pressure in the boost chamber. The results of this research provide a research basis for further controlling the pressure characteristics of the boost pulse water jet.
Open working areas in CNC press brakes require protective systems that can distinguish normal sheet handling from hazardous intrusion while maintaining a short response time. This study developed an embedded vision-based hazardous-region monitoring system that integrates region-constrained image acquisition, edge inference, state decision-making, and GPIO-based electrical-signal output. The hazardous region was defined as the overlap between the tool-motion risk region and the sheet-operation region, and a trained state-recognition model was deployed on a Jetson TX2 through a PyTorch-ONNX-TensorRT conversion workflow. To evaluate the complete protective chain without directly connecting the prototype to the machine emergency-stop circuit, an on-site simulated platform with a CSI camera, Jetson TX2, GPIO wiring, and an LED load was constructed. TensorRT optimization reduced the single-image inference time from 25.48 ms to 10.19 ms. The measured image acquisition, preprocessing/inference, and post-processing/GPIO latencies were 8.1 ms, 16.1 ms, and 4.6 ms, respectively, producing an average end-to-end response time of approximately 29.2 ms. Frame-level tests in four representative videos showed hazardous-state detection rates above 99.0% and false-positive rates of 4.0%-4.2%. These results indicate that the proposed system can convert visual hazardous-state recognition into a measurable external output on an embedded platform. The present validation is limited to simulated-load testing, and functional-safety certification, emergency-stop linkage, braking-distance verification, and fail-safe diagnosis remain necessary before practical safety deployment.