Research On ECT Two Phase Flow Measurement Method Based On Neural Network

Electrical Capacitance Tomography(ECT) is a new kind of PT(Process Tomography), which appeared in the1980s. It developed on the basis of medical CT technology, detection capacitance can be gained through the capacitance sensors which are evenly distributed around the tested pipes, therefore the media distribution of the internal pipes can be reconstructed by the projection information which is embodied in the detection capacitance and the corresponding image algorithm. ECT is the mainstream of the study of PT because of its advantages of non-invasive, simple structure, low cost. ECT can not only analyze the parameter measurement of flow, distributed capacitor affected by flow pattern, but also plays an important in detecting the Two-phase flow parameters measurement and the real-time controlling for system in pneumatic conveying systems. ECT has broad prospects for development based on the above characteristics; therefore, study of ECT has an important and profound significance.Based on a large amount of references, this thesis has applied the theory of feature extraction and the criterion function optimization into the neural network, which makes the ECT satisfies the need of industrial inspection better. Work can be done as follows:1. On account of the problem of low flow identification efficiency of ECT, the thesis has put forward a way to identify the two-phase flow which is combined by rough neural network and feature extraction of ECT. This method has defined the concept of the flow pattern characteristic parameters, then has normalized the independent measuring capacitance signals which are obtained from the ECT. The feature information can be extracted from the feature parameters, then input them to the rough neural network to be trained. Finally inputting the testing samples to the trained rough neural network to complete the flow pattern identification. Simulation experimental results has shown that this method has a higher identification precision compared with the traditional BP neural network, moreover, it has strong ability to resist noise, which provides a new method for the study of the image flow pattern identification for ECT.2.This thesis has proposed Hopfield network model based on the multi-criterion optimization, and confirms the energy function of Hopfield network model on the basis of multi-criterion optimization; By introducing three objective functions to the energy function of Hopfield network, in order to get the optimal solution in the reconstruction of the image. Thus a method of ECT image reconstruction which is combined by the theory of multi-criteria optimization and the Hopfield network theory. The image reconstruction can be achieved better through this method. The simulation experiment show that the Hopfield network based on the multi-criteria optimization is a method of small reconstruction errors and high precision of ECT image reconstruction algorithm, which is superior to the traditional LBP algorithm, landweber iteration algorithm.