During the operation of nuclear power plants, a large amount of low and intermediate level waste (LILW) is generated, which is usually prepared into 200-L and 400-L waste drums. To ensure the safe disposal of these waste drums, they must be analyzed to determine the type and activity of the nuclides contained within them. Non-destructive assay (NDA) has been widely used in the detection of waste drums in nuclear power plants, along with segmented gamma scanning (SGS) and tomographic gamma scanning (TGS). However, the low measurement accuracy of SGS and the long measurement time of TGS limit the practical application of these methods.

This sudy aims to shorten the measurement time while maintaining high measurement accuracy by proposing a new neural network-based method for measuring the activity of waste drum.

When the waste drum was filled with a uniform distribution of medium and rotated at a constant speed during measurement, the point source was equivalent to a ring source. The equivalent ring source in the waste drum possessed an activity equal to the total activity of all sources. The neural network model is established, the count rate of the detector at different positions is used as input, and the radius of the equivalent ring source is used as output. Finally, the total activity of the waste drum is calculated. The simulated measurement is carried out in a 400-L waste drum, the medium is concrete, the radioactive source is Co-60, and 50 groups of single-source and 10 groups of multi-source are generated randomly. Different methods are used to reconstruct the activity of the waste drum.

When there is only one radioactive source in the waste drum, the mean relative error (MRE) of activity reconstruction by the new method is 4.26%, which is much lower than that of SGS (68.15%) and close to that of TGS with 60 grids (3.97%). When there are multiple radioactive sources in the waste drum, the MRE of activity reconstruction by the new method is 24.27%, which is lower than that of SGS (48.02%) and close to that of TGS with 60 grids (28.61%). This new method achieves the equal measurement accuracy of TGS but reduce the measurement time to 1/20 of TGS.

Compared to traditional measurement methods, the new method greatly shortens the measurement time while maintaining high precision, thereby providing technical support for the measurement of LILW.