关键词: 电磁超声, 液位测量, 二次激励共振法, 卷积神经网络

Abstract:

In high-temperature and high-pressure nuclear power plant cooling water pipelines, traditional liquid level monitoring methods have issues of contact and intrusion. Firstly, a longitudinal wave electromagnetic asonic transducer with a Halbach-like structure is designed. By optimizing the magnetic circuit, the signal strength is enhanced by approximately 11.0 dB. A runway coil design is employed to ensure the divergence of the acoustic beam, facilitating the reduction of echo interference. Secondly, a multi-sensor deployment strategy for liquid level monitoring is proposed. In this scheme, 4 MHz three-cycle thickness measurement pulses are excited on the pipe wall, and the wall resonance frequency is determined from the time interval between echoes. The corresponding fundamental and harmonic resonance waves are then generated according to the wall thickness. Finally, empirical mode decomposition is employed to denoise the echo signals, and normalized envelope features are extracted and fed into a convolutional neural network to classify solid–gas and solid–liquid interfaces across multi-sensor. The range of liquid level height can be effectively determined through the interface judgment results of multi-sensor. Experimental results demonstrate that the proposed method provides stable measurements of inclined and fluctuating liquid levels within metallic pipelines, highlighting its strong potential for industrial applications.

Key words: electromagnetic acoustic transducer (EMAT), water level measurement, secondary excitation resonance method, convolutional neural network (CNN)