Stuyd Of Data-driven Convolutional Neural Network For Tomographic Reconstruction Of Electical Capacitance

The gas-solid fluidized bed is widely used in chemical,electrical,and metallurgical industries.It's important to measure the flow parameters of gas-solid fluidized bed for its flow mechanism research,system control and optimization.Electrical capacitance tomography(ECT)can reconstruct the medium distribution over the crosssection of the pipe by measuring the capacitance vector between the electrode arrays outside the pipe.It is gradually applied for the measurement of gas-solid fluidized bed parameters due to the advantages of real-time,being noncontact,low-cost,high reliability and safety.In this thesis,a data-driven ECT image reconstruction algorithm is presented based on convolutional neural networks.A measurement system is further developed and used to study the particle flow characteristics on a circulating turbulent fluidized bed(CTFB).Firstly,considering the soft field and ill-conditioned problems of ECT reconstruction,a data-driven ECT image reconstruction algorithm based on convolution neural network is proposed.A data set with typical and random flow patterns is established by numerical simulations.The convolution neural network model is studied and trained with the established data set to reconstruct the particle distribution over the cross-section of the pipe.Furthermore,the influence of the model parameters on image reconstruction is analyzed to optimize the model.Compared with the traditional algorithm,the data-driven ECT image reconstruction algorithm has the higher correlation coefficient more than 0.97 for the typical and random flow patterns.The relative image error is less than 4% at the signal-tonoise ratio of 20 d B.The consuming time for frame-by-frame imaging is only 10 ms.Then,an ECT system based on the convolutional neural network model is developed.It consists of hardware and software.The hardware part is composed of four parts: a capacitance sensor array,an electrode selection switch,a capacitance measurement circuit,a data acquisition and control system,which is mainly used for capacitance signal conversion,data acquisition and transmission.The software includes a communication and control module,an image reconstruction unit and an image display module,which can realize the capacitance signals sampling,reconstruction of particle distribution,display and storage of reconstructed results.Finally,the developed ECT system is employed to study the particle flow behavior in the CTFB.Static experiment results show that the correlation coefficients of the ECT system based on convolutional neural network algorithm are higher than 0.85 for the typical flow patterns: stratified flow,annular flow,single-core flow and dualcore flow.The experimental results of the CTFB show that the particle concentration in the central area is lower than that in the edge.The average concentration of particles decreases with increasing the air flow rate and decreasing the initial feeding height within the flow pipe.This is in good agreement with the particle distribution in a typical fluidized bed,which further proves the effectiveness of the ECT system.