Liquid Level Measurement And Error Compensation Methods Based On Acoustic Resonance Principle

Based on the principle of low frequency audible acoustic resonance,the liquid level measurement method has the acoustic diffraction characteristic which can overcome the parasitic reflection problem of ultrasonic measurement method.However,due to the interference of measuring environment noise,improper installation of measuring device,improper setting of parameters and other reasons,there are defects in the sequence of resonance points obtained by acoustic resonance method,resulting in errors in the measurement results.Aiming at this problem,based on the deep analysis for the defect form of resonance waveform,the compensation method of resonance defect and the corresponding liquid level measurement method are studied from two different angles.First,the resonance sequence liquid level conversion formula is used to calculate the liquid level height,and the error inference model based on the Maximum Likelihood Evidence Reasoning(MAKER)rule is given to correct the defect waveform to realize the compensation of measurement error.Second,after a large number of acoustic resonance data are generated in the long-term operation of the liquid level monitoring system,a deep learning measurement model is established to deduce the liquid level height directly from the original resonance data.The main research work is as follows:(1)Defect compensation and liquid level estimation method of acoustic resonance sequence based on MAKER rule.At different liquid level heights,the resonance frequency sequences with defects are collected and their local features are extracted.Based on the sample set composed of these local features,the initial reference evidence matrices(IREM)and joint reference evidence matrices(JREM)are created.The positions and number of false/missing resonance points are estimated based on MAKER rule evidence fusion reasoning.A complete set of resonance frequency sequence is obtained through defect compensation strategy,and the liquid level estimation is calculated at the conversion point.(2)Frequency domain feature extraction and liquid level estimation method based on STFT and CNN.After obtaining a large number of resonance waveforms at different liquid level heights,a data-driven Convolutional Neural Network(CNN)model is used to estimate liquid level heights.The short time Fourier transform(STFT)is used to transform the resonance wave into a spectrum image to obtain the resonance time-frequency characteristic matrix.The CNN model including convolution layer,pooling layer and nonlinear activation layer is constructed.The input is time frequency characteristic matrix,and the output is liquid level estimation.The training set data is used to optimize the model to further estimate the accuracy of the model.(3)Acoustic resonance liquid level estimation method based on LSTM.Considering the simplicity of the design of the liquid level estimation algorithm and the adequacy of the resonance wave feature,an end-to-end liquid level estimation method based on the Long Short-Term Memory(LSTM)is presented.LSTM network is selected to construct the liquid level estimation model.The input is the pre-processed acoustic resonance liquid level data,and the output is the liquid level estimation value.Finally,at different liquid level heights,the liquid level estimation and measurement effects of the above three new methods are compared and analyzed in detail,and the performance advantages and application scope of each method are explained.