A Study of Using Pulsed Neutron Activation for Accurate Measurements of Water Flow in Pipes

Sammanfattning: The objectives of this work were 1) to investigate the performance of the pulsed neutron activation (PNA) method for determination of mass flow of water in pipes, and 2) to improve the accuracy such that an error less than 1% become possible with the PNA method. The ultimate purpose is to develop this method into a high-accuracy portable and non-intrusive flow-meter that easily can be used as a calibration tool at a number of field applications, e.g. feed-water flow at nuclear reactors.

The first part of this work was to systematically explore the accuracy of traditional PNA analysis methods. Two experimental series were performed and analysed. In both cases a flow loop was built in which flow measurements were performed by both the PNA method and by some independent calibration method. The first pilot measurements were made on a preliminary loop with moderately stable flow and moderate calibration accuracy. These measurements were made primarily to obtain basic experience with the method, some rough indication of its accuracy and, most important, to gain knowledge of designing a high-quality set-up for high accuracy investigation of the PNA method. Thus, a dedicated loop was designed and built with a high accuracy flow calibration. The results of the measurements performed show that the error of the analysis methods are functions of the flow and measurement parameters such as the flow velocity, the source-detector spacing etc, here called bias functions.

The second part of the work concerns development of methods by which the accuracy of the PNA-based flow measurement can be improved. It was suggested that artificial neural networks (ANNs) be used for high accuracy evaluation of a PNA measurements. It was also suggested that the data required for training of the network be generated by simulations rather than by measurements. To investigate the feasibility of this method, a simulation model was developed and tested. In the model, the neutron and photon transport was simulated by the uncollided flux. The turbulent flow model used was based on eddy diffusion with a superimposed power law velocity profile. This model gives sufficient qualitative agreement with the measurements which makes it suitable for the testing of the ANN-based evaluation method. A network, trained with simulated data was able to reconstruct the reference mass flow, used in the simulation, with an error smaller than for traditional methods. For the case where no external disturbances (noise in the detector signals etc) was assumed, the accuracy was better than 0.5%. Thus the work performed in this thesis suggests that it is possible to develop a PNA-based flow-meter with high accuracy.