Abstract:In the industrial production process, ultrasonic flowmeter plays an important role with the advantages of non-contact measurement and suitable for various fluid media. Aiming at the problems of poor anti-interference ability and low detection accuracy of ultrasonic flow detection, a four-channel non-full tube ultrasonic flowmeter combined with pattern recognition is proposed. The system uses a high-performance chip with floating point operation for FFT calculation and calculates the integrated velocity of four channels, and identifies the liquid level of the non-full pipe with the edge calculation chip, and then improves the identification stability through the liquid level correction model. In the liquid level recognition model, the feature extraction module and the spatial attention mechanism module are used to extract the effective features, and the random forest classification module is used to classify the liquid level. The experimental results show that the DC-SAM algorithm can converge faster than other models, and the accuracy can reach 96. 6%. In the flow experiment, the system can achieve 96. 5% accuracy and good linearity compared with the calibrated flowmeter. The system can accurately identify liquid level and non-full pipe flow, and meet the stability requirements of detection while maintaining high measurement accuracy, which proves the feasibility of operation and deployment of edge calculation in ultrasonic flow detection.