Passive acoustic leak detection in energy conversion systems of sodium fast reactors

• Main Author: Riber Marklund, Anders
• Format: Doctoral Thesis
• Language: English
• Published: KTH, Reaktorteknologi 2016
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-199415; http://nbn-resolving.de/urn:isbn:978-91-7729-239-5

Reaching the goals of Generation IV nuclear power is challenging. However, no less than six reactor concepts have been identified as capable of fulfilling the demands. Among these, the Sodium Fast Reactor (SFR), probably represents the most mature technology as about 20 SFR plants have been operated to this day.   One design-specific issue of the SFR is the risk of leak and sodium-water reaction inside a steam generator. Standard monitoring is based on hydrogen detection, resulting in high sensitivity but slow response. The alternative of acoustic leak detection methods has been studied since the 1970s since they are able to respond much faster. Demonstrating low false alarm rate while detecting the fairly weak and possibly unknown acoustic signals of leaks has however proven to be difficult.   Today, the CEA performs R&amp;D, notably within the scope of the ASTRID project, with the aim of eliminating the sodium-water reaction risk. This is achieved by a Brayton cycle, using a nitrogen turbine and compact sodium-nitrogen heat exchangers. In case of a leak in this system, the low solubility of nitrogen in sodium and the high pressure in the tertiary circuit would increase the secondary pressure, locally deteriorate performance and possibly result in harmful hydrodynamic effects. Together with the risks of a potential gas leak over to the reactor, this motivates the use of leak detection also for this design.   This thesis concerns passive acoustic leak detection, primarily for a SFR sodium-nitrogen heat exchanger, arguing that this method is suitable based on experiments, numerical simulations and studies on algorithms. The word “passive” here refers to a system that does not send out any signals, but rather records the noise of the plant and detects leaks as changes in this signal. The thesis covers experiments on normal operation and leak-simulating setups as well as machine-learning based detection methods intended to be of interest also for change detection in general. === <p>QC 20170109</p>