Robust artificial neural network for reliability and sensitivity analyses of complex non-linear systems.

Artificial Neural Networks (ANNs) are commonly used in place of expensive models to reduce the computational burden required for uncertainty quantification, reliability and sensitivity analyses. ANN with selected architecture is trained with the back-propagation algorithm from few data representatives of the input/output relationship of the underlying model of interest. However, different performing ANNs might be obtained with the same training data as a result of the random initialization of the weight parameters in each of the network, leading to an uncertainty in selecting the best performing ANN. On the other hand, using cross-validation to select the best performing ANN based on the ANN with the highest R2 value can lead to biassing in the prediction. This is as a result of the fact that the use of R2 cannot determine if the prediction made by ANN is biased. Additionally, R2 does not indicate if a model is adequate, as it is possible to have a low R2 for a good model and a high R2 for a bad model. Hence, in this paper, we propose an approach to improve the robustness of a prediction made by ANN. The approach is based on a systematic combination of identical trained ANNs, by coupling the Bayesian framework and model averaging. Additionally, the uncertainties of the robust prediction derived from the approach are quantified in terms of confidence intervals. To demonstrate the applicability of the proposed approach, two synthetic numerical examples are presented. Finally, the proposed approach is used to perform a reliability and sensitivity analyses on a process simulation model of a UK nuclear effluent treatment plant developed by National Nuclear Laboratory (NNL) and treated in this study as a black-box employing a set of training data as a test case. This model has been extensively validated against plant and experimental data and used to support the UK effluent discharge strategy.

An innovative way of thinking nuclear waste management - Neutron physics of a reactor directly operating on SNF.

A solution for the nuclear waste problem is the key challenge for an extensive use of nuclear reactors as a major carbon free, sustainable, and applied highly reliable energy source. Partitioning and Transmutation (P&T) promises a solution for improved waste management. Current strategies rely on systems designed in the 60's for the massive production of plutonium. We propose an innovative strategic development plan based on invention and innovation described with the concept of developments in s-curves identifying the current boundary conditions, and the evolvable objectives. This leads to the ultimate, universal vision for energy production characterized by minimal use of resources and production of waste, while being economically affordable and safe, secure and reliable in operation. This vision is transformed into a mission for a disruptive development of the future nuclear energy system operated by burning of existing spent nuclear fuel (SNF) without prior reprocessing. This highly innovative approach fulfils the sustainability goals and creates new options for P&T. A proof on the feasibility from neutronic point of view is given demonstrating sufficient breeding of fissile material from the inserted SNF. The system does neither require new resources nor produce additional waste, thus it provides a highly sustainable option for a future nuclear system fulfilling the requests of P&T as side effect. In addition, this nuclear system provides enhanced resistance against misuse of Pu and a significantly reduced fuel cycle. However, the new system requires a demand driven rethinking of the separation process to be efficient.

New parametrization method for dissipative particle dynamics.

We introduce an improved method of parametrizing the Groot-Warren version of dissipative particle dynamics (DPD) by exploiting a correspondence between DPD and Scatchard-Hildebrand regular solution theory. The new parametrization scheme widens the realm of applicability of DPD by first removing the restriction of equal repulsive interactions between like beads, and second, by relating all conservative interactions between beads directly to cohesive energy densities. We establish the correspondence by deriving an expression for the Helmoltz free energy of mixing, obtaining a heat of mixing which is exactly the same form as that for a regular mixture (quadratic in the volume fraction) and an entropy of mixing which reduces to the ideal entropy of mixing for equal molar volumes. We equate the conservative interaction parameters in the DPD force law to the cohesive energy densities of the pure fluids, providing an alternative method of calculating the self-interaction parameters as well as a route to the cross interaction parameter. We validate the new parametrization by modeling the binary system SnI(4)SiCl(4), which displays liquid-liquid coexistence below an upper critical solution temperature around 140 degrees C. A series of DPD simulations were conducted at a set of temperatures ranging from 0 degrees C to above the experimental upper critical solution temperature using conservative parameters based on extrapolated experimental data. These simulations can be regarded as being equivalent to a quench from a high temperature to a lower one at constant volume. Our simulations recover the expected phase behavior ranging from solid-liquid coexistence to liquid-liquid coexistence and eventually leading to a homogeneous single phase system. The results yield a binodal curve in close agreement with the one predicted using regular solution theory, but, significantly, in closer agreement with actual solubility measurements.