01/30/2023
By Danielle Fretwell

The Francis College of Engineering, Department of Chemical Engineering, invites you to attend a doctoral dissertation defense by Lohith Annadevula on “Development Of Advanced Statistical & Computational Models For Safeguards Inspection Planning And Assessment Of Inspection Effectiveness”.

Candidate Name: Lohith Vamsi Mohan Reddy Annadevula
Degree: Doctoral
Defense Date: Thursday, Feb 9, 2023
Time: noon to 1:30 p.m.
Location: SOU250 or via Zoom

Committee:

  • Advisor Sukesh Aghara, Associate Dean of Graduate Studies and Research, Professor, Chemical Engineering, University of Massachusetts Lowell
  • Valmor F. de Almeida, Associate Professor, Chemical Engineering, University of Massachusetts Lowell
  • Stephen T. Lam, Assistant Professor, Chemical Engineering, University of Massachusetts Lowell
  • Kenneth Jarman, Applied Scientist, Pacific Northwest National Laboratory

Abstract:

Advanced statistical techniques have been proposed for the development of models aimed at solving some of the nuclear inspection and safeguards problems of the International Atomic Energy Agency (IAEA). Two important IAEA inspection problems are the computation of defect detection probability (DP) specific to stratum-based inspection and the overall minimum aggregate detection probability (Min ADP) for a facility or statewide inspection. Traditionally, the stratum level defect detection probability is computed using deterministic models’ case- by- case scenario; these models will not work for any other cases apart from the ones they are built for, and no single universal model exists in the literature. A Monte-Carlo (stochastic) method using pseudo-random generators is developed in this work, which is universally applicable to any scenario and can compute stratum-level DP with a user-defined standard error. Two universal deterministic methods based on conditional tree, compositions and multivariate hypergeometric distributions are developed here to compute stratum-level DPs with the same accuracy as case-specific deterministic models of the literature. The usage of machine learning (ML) algorithms has been proposed, and if succeeded, allows for quick prediction of results with acceptable error. From the ML perspective, the stratum-based inspection problem has been identified as a multi-variate regression problem and can be treated with multi-variate regression models. Computation of facility-wide or state-wide minimum aggregate detection probability using the stratum detection probability curves comes under the class of optimization problem. The development of a greedy algorithm (GA) from heuristics, Pareto frontier algorithms from dynamic programming, and greedy ML algorithm from reinforced ML & meta-heuristics has been proposed along with their possible theoretical proofs to solve the minimum ADP optimization problem. A potential extension of the stratum-level stochastic model to the facility/state level has been proposed and taken as one of the thesis objectives apart from the three mentioned before.

All interested students and faculty members are invited to attend the defense.