Applications Of Neural Networks To The Measurement Of Multi-Phase Flow

Multi-phase flow processes exit widely in industry filed, the parameter measurement of these processes is very important for resource exploitation or process control. Petrolic mixtures in petroleum pipelines form oil/gas two- phase flow, oil/water two-phase flow or oil/gas/water three-phase flow. Visualization of phase distribution, flow regime identification and phase concentration measurement are maj or research directions of the measurement in petroleum pipeline systems. While powder flowrate is a key parameter of the measurement in pneumatically conveyed powder processes. This dissertation focuses on the application of neural networks to the measurement of multi-phase flow in petroleum pipelines and pneumatically conveyed powder processes. Following are the main works and contribution of this dissertation: 1. An overall introduction of process tomography based on various sensing techniques was presented. Electrical capacitance tomography (ECT), including sensing technique and image reconstruction algorithms were especially reviewed. Furthermore, an introduction of neural networks and their application to soft-sensor modeling was presented. 2. Oil/gas two-phase flow in pipelines being the case study, an image reconstruction model based on BP neural networks was proposed for visualization of the phase distribution within cross-section of pipeline. With static experiments on the 12-electrode ECT experiment equipment for multi- phase flows in Zhejiang University and dynamic experiments on the experiment equipment in the experimental station of Daqing Oil Field, tests for the reconstruction model were carried out. Experimental results showed that the quality of images reconstructed met the expected accuracy requirement: in the static experimental tests the highest fidelity of reconstructed images was 100%, while the lowest fidelity of images was 90.8%, and the average fidelity of images was 98.5%. And in the dynamic experimental tests the reconstructed images accorded well with the images from digital camera monitoring the process. The speed of image reconstruction is about 53 frames/s on the Pentiuml300 compatible computer. 3. For imaging oil/water two-phase flow with noticeable difference between permittivities in pipelines, an image reconstruction model based on BP neural networks was proposed. The results from simulation based on the finite element method showed that the highest fidelity of reconstructed images was 100%, while the lowest fidelity of images was 88.8%, and the average fidelity of images was 95.5%. The speed of image reconstruction was about 52 frames/s on the Pentium/300 compatible computer. 4. To explore the image reconstruction method for oil/gas/water three- phase flow in pipelines, three reconstruction models based on BP network, Elman feedback network and grouping BP network were proposed respectively. Elman feedback network was applied to ECT image reconstruction algorithms. Moreover grouping BP network was used to solve the problem of time-space consuming during the network training. Preliminary results from simulation showed that for the BP network model the highest fidelity of reconstructed images was 100%, the lowest fidelity of images was 95.2%, and the reconstruction speed was about 52 frames/s on the Pentium/300 compatible computer. The Elman network model and grouping BP network improved the quality of reconstructed images, under the same tests the lowest fidelity of reconstructed images f...