Proposed automated data analysis for nuclear waste vitrification at SRS: a case study

Summary form only given. The Savannah River Site (SRS), in operation since the early 1950s, produces plutonium and tritium. About 70 million gallons of radioactive waste have been generated and stored on the site in tanks. The Defense Waste Processing Facility (DWPF) will immobilize this high-level radioactive waste by converting it into a stable borosilicate glass waste form suitable for long term storage. A melter fills two foot diameter, ten feet tall stainless steel canisters with the borosilicate glass. After a canister is filled, the top is sealed by welding a 12.9 cm diameter plug into the canister nozzle using an upset resistance weld. Initially the DWPF will be operated for just under 2 years using simulated (nonradioactive) feed materials. During this time a large amount of data will be generated with respect to the range of operating conditions under which the DWPF welder is expected to operate. The quality of the welds will be determined. One means which appears very attractive as a possible model or tool for assessing weld acceptability is the use of neural networks. The task is akin to pattern recognition