Study of the possibility of determining mass flow of water from neutron activation measurements with flow simulations and neural networks

Mass flow of water in a pipe can be measured in a non-intrusive way by the pulsed neutron activation (PNA) technique. From such measurements mass flow can be estimated by various techniques of rime averaging performed on the time-resolved detector signal(s). However, time averaging methods have a fe w percent systematic error which in addition is not a constant but varies w ith flow and measurement parameters. Achieving a precision better than 1% f rom PNA measurements is a hitherto unsolved task. In this paper a methodolo gy is suggested to solve this task and is tested by simulation methods. The method is based on the we of artificial neural networks to determine mass flow rate from the time resolved detector signal. To achieve this, the netw ork needs to be trained on a large number of real detector data. it is sugg ested that these data should be obtained by advanced numerical simulation o f the PNA measurement. In this paper we use a simplified simulation model f or a feasibility study of the methodology. It is shown that a neural networ k is capable to determine the mass flow rate with a precision of about 0.5 %.