Measurement Of Suspended Particle Size Distribution Based On Improved Rbf Neural Network

As a typical two-phase flow,suspension is widely distributed in various fields of daily life,such as hydrology,ecological environment,waterway construction,energy and chemical industry.The measurement of its parameters has gradually attracted people's attention,especially the measurement of particle size distribution,more inclined to be intelligent and effective.In this paper,ultrasonic attenuation experiments were performed on suspension samples based on the attenuation effect of ultrasonic waves.The attenuation signal containing information on the particles in the suspension is obtained.By denoising the attenuated signal,an effective signal is obtained and the feature values are extracted.An artificial neural network is used to construct a direct relationship between the characteristic values of the attenuation signal and the particle size distribution.A measurement of the particle size distribution is achieved.The main tasks are:(1)The glass bead samples of the mixed particle size were subjected to drying and weighing,and the samples were sieved in 17 particle size intervals by using the sieving method,and 50 sets training samples and 3 sets test samples belongs to different distribution are configured.(2)Based on the attenuation effect of ultrasonic wave,the ultrasonic attenuation experiment of 50 sets of suspension is carried out on the experimental platform based on the focused ultrasonic transducer,the signals at 670 kHz,750 kHz and 830 kHz are obtained,and the signal is denoised and characterized value extracted.(3)The RBF artificial neural network model is constructed,and the genetic algorithm is used to optimize the model for its limitations.The results show that the optimized RBF neural network has higher learning efficiency and faster convergence speed,and its measurement results are closer to the standard sieving method than the unoptimized RBF.(4)The experimental error is calculated and analyzed,and the Bland-Altman scatter plot is introduced to evaluate the consistency of the two methods.It proves that the proposed method has good reliability.And the relative error average is controlled within 1.3%,the absolute error average is less than 0.75,which indicates that the proposed method has a relatively ideal measurement effect within the experimental range.