Comparison of machine learning models for the detection of partial defects in spent nuclear fuel

Rossa, Riccardo; Borella, A.; Giani, Nicola

doi:10.1016/j.anucene.2020.107680

Cited by 14 publications

(4 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The dataset for the study relies on previous work performed at SCK CEN (Rossa et al, 2020(Rossa et al, , 2018. It includes the PDET responses simulated via Monte Carlo N-Particle -MCNP code (Werner, 2018) for 17 × 17 PWR spent nuclear fuel assemblies with and without diversion, see Fig.…”

Section: Pdet Datasetmentioning

confidence: 99%

“…Recent efforts have been conducted to develop methods that can enhance the processing of the measured data and extract more details of the system configuration. For example, machine learning algorithms were used to quantify the percentage of replaced fuel pins in SNF assemblies (Rossa et al, 2020(Rossa et al, , 2018Aldbissi et al, 2022), to predict parameters of SNF assemblies (Mishra et al, 2021), to detect and localize missing radioactive sources within a small grid (Durbin and Lintereur, 2020), to track elemental and isotopic material flows through material balance areas for safeguards (Shoman and Cipiti, 2018). These methods can help to reduce the inspection time and make the identification of diversion patterns more precise, so that the decision process of the inspectors is facilitated.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Identification of Diversions in Spent Pwr Fuel Assemblies by Pdet Signatures Using Artificial Neural Networks (Anns)

Al-dbissi

Rossa²,

Borella³

et al. 2023

Preprint

View full text Add to dashboard Cite

Section: Pdet Datasetmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Identification of Diversions in Spent Pwr Fuel Assemblies by Pdet Signatures Using Artificial Neural Networks (Anns)

Al-dbissi

Rossa²,

Borella³

et al. 2023

Preprint

View full text Add to dashboard Cite

“…A fourth partial defect scenario was also simulated where the rods were removed without replacement, leaving a water gap in the fuel assembly. The partial defect pattern denoted as case 1 in figure 2 is the same as that used in [12]. The substitution is symmetric and concentrated to the edges of the assembly, making the case one of the more difficult ones for the DCVD to detect.…”

Section: Jinst 15 P07009mentioning

confidence: 99%

Partial defect detection using the DCVD and a segmented Region-Of-Interest

Branger¹,

Grape²,

Jansson³

2020

J. Inst.

View full text Add to dashboard Cite

The Digital Cherenkov Viewing Device (DCVD) is a safeguards instrument available to international nuclear safeguards inspectors. It is frequently used to verify fuel on the gross defect level, and approved for partial defect verification, i.e. to assess that parts of a fuel assembly have not been diverted. The current limit for partial defect verification with the DCVD is on the 50% level. In the verification process, an analysis methodology is used where the inspector places a Region-Of-Interest (ROI) around the fuel assembly and assesses the total Cherenkov light intensity within this region. The intensity is then compared to a predicted value, and deviations from the predicted value are used to flag fuel assemblies for further investigations.In this work, we investigate a slightly different analysis approach, where the ROI is split into two or three segments to more accurately capture changes in light intensity in different regions of the captured image. The purpose is to increase the sensitivity of the DCVD to partial defects below the 50% level. Based on simulations of a Pressurised Water Reactor 17x17 fuel assembly, we conclude that a partial defect on the 30% level decreases the Cherenkov light intensity by at least 15% using one single ROI, by at least 20% using a ROI with two segments, and by at least 22% using a ROI with three segments. The analysis approach using two or three ROI segments instead of one thus appears to be more sensitive to partial defects, and can enable more accurate detection of partial defects on the 50% level as well as partial defect detection below the 50% level.Validation of the approach using a limited set of measurement data of intact fuel assemblies supports that detection of light intensity reductions by 20% and 22% is possible, while ensuring that the false positive rate is kept sufficiently low. However, an optimization of ROI segment splits as well as a more extended validation of the approach is required before the method can be considered reliable and applicable to all fuel assemblies that the DCVD can verify today.

show abstract

“…A number of such fuel libraries already exist for light water reactors (LWRs) [ 5 , 6 ]. There have been numerous research works looking into the possibility of using machine learning techniques for verification of spent nuclear fuel (SNF) LWRs [7] , [8] , [9] , [10] , [11] , [12] , [13] , [14] , [15] with reasonable success. However, such work looking into using conventional and possibly new verification signatures for MSRs is severely limited.…”

Section: Introductionmentioning

confidence: 99%

Data library of irradiated fuel salt and off-gas tank composition for a molten salt reactor concept produced with Serpent2 and SOURCES 4C codes

Mishra,

Elter,

Branger

et al. 2024

Data in Brief

View full text Add to dashboard Cite

Comparison of machine learning models for the detection of partial defects in spent nuclear fuel

Cited by 14 publications

References 10 publications

Identification of Diversions in Spent Pwr Fuel Assemblies by Pdet Signatures Using Artificial Neural Networks (Anns)

Identification of Diversions in Spent Pwr Fuel Assemblies by Pdet Signatures Using Artificial Neural Networks (Anns)

Partial defect detection using the DCVD and a segmented Region-Of-Interest

Data library of irradiated fuel salt and off-gas tank composition for a molten salt reactor concept produced with Serpent2 and SOURCES 4C codes

Contact Info

Product

Resources

About