Year
1986
Abstract
There is an increasing demand for effective methods of nuclear materials accountancy. In order to keep accounts for any material balance area, it is necessary to measure, or estimate, the inputs and outputs during the assessment period, and the initial and final inventories. Inputs, outputs, and some components of the inventory usually pose no problem since the material is contained in drums and cylinders and other containers which can be weighed and sampled. The difficulty lies in the in-process inventory, e.g. material contained in reactors and kilns. The only direct way of measuring in-process inventory would be to shut down the plant, remove the material inside, and measure it; this operation is in some cases very expensive and could be even unfeasible. Although in-process inventories cannot be measured directly, a plant manager usually has a fairly good idea of how much material the plant contains: his knowledge of the plant enables him to infer the inventory, at least roughly, from the observable data. A possible solution to the problem is, therefore, to create an expert system which incorporates the knowledge on which such estimates are based. An experimental system of this kind has been developed by JRC-Ispra in collaboration with British Nuclear Fuels Ltd. (BNFL). The program, which is realized in the expert system shell SAVOIR, performs an inventory evaluation for the UF6 production plant at BNFL's Springfields Works, by using both measured values and inferences based on engineering expertise. The performance of the system is under extensive evaluation. This paper will present a detailed description of the knowledge base and some preliminary results on its performance.