Accurate reconstruction of incomplete sensor fields is critical for state awareness in safety-critical digital twins (DTs). Thermocouples in high-temperature and radiation environments may drift, fail, or become intermittently unavailable, causing partial observability. This study develops a sparse-sensor reconstruction framework for Oregon State University’s High Temperature Test Facility (HTTF) and investigates six methods under intra- and inter-experiment protocols: three conditional generative models (i.e., a diffusion model [DM], variational autoencoder [VAE], and generative adversarial network [GAN]), two deterministic neural-network models (graph neural network-gated recurrent unit [GNN–GRU] and gated recurrent unit [GRU]), and ridge regression. For intra-experiment reconstruction within the PG-29 depressurized conduction cooldown (DCC) test, the three generative models combine graph-based spatial encoding with GRU. For solid-temperature (TS) thermocouples, the graph-temporal VAE achieves the lowest error, with a mean absolute error (MAE) of 11.73 °C, a root mean square error (RMSE) of 20.54 °C, and an R2 of 0.98. The deterministic GNN–GRU and plain GRU obtain RMSE values of 25.26 °C and 35.22 °C, respectively, demonstrating the benefit of spatial modeling. All three generative models reduce RMSE by more than half compared with ridge regression. The graph-temporal DM maintains an R2 of 0.94 when known TS sensors decrease from 91 to 45. The framework also reconstructs fluid-temperature (TF) thermocouples effectively, with the graph-temporal DM achieving an RMSE of 30.09 °C and an R2 of 0.97. For inter-experiment reconstruction, models trained on PG-27 pressurized conduction cooldown (PCC) are evaluated on PG-29 using section-context conditioning. The section-context DM achieves the best transfer performance, with an RMSE of 42.84 °C and an R2 of 0.91. These results demonstrate the value of spatiotemporal generative modeling for missing-sensor reconstruction in safety-critical DT applications.
Tritium transport and accountancy are central challenges for the design, safety, and performance assessment of fusion energy systems, requiring predictive modeling across multiple spatial and temporal scales. This paper presents recent developments in the Tritium Migration Analysis Program, version 8 (TMAP8), an open-source tritium transport modeling tool built on the Multiphysics Object-Oriented Simulation Environment (MOOSE) framework following high software quality assurance standards. TMAP8 enables fidelity-flexible simulations spanning from pore microstructure length scale to component-scale transport and up to system-level fuel cycle modeling within a single application. We introduce new modeling capabilities together with a substantially expanded suite of verification, validation, and benchmarking cases. The verification cases demonstrate accurate numerical solutions, while the validation cases assess predictive capability against experimental measurements. In addition, TMAP8 reproduces and benchmarks a recent tritium fuel cycle model, demonstrating ability over a wide range of scales. All verification and validation cases, input files, and analysis scripts are released as open source, supporting rigorous software quality assurance, transparent benchmarking, workforce development, and community reuse. These developments, in addition to the deployment of a simplified physics-based syntax, further establish TMAP8 as a robust multiscale modeling application for tritium transport analysis in fusion energy systems.
Municipal solid waste (MSW) represents a large, low-cost biomass resource for bioenergy and bioproduct production. However, its heterogeneity and compositional variability limit its direct use as a conversion-ready feedstock. This study evaluated three mechanical pre-screening technologies vibratory screening, trommel screening, and a dynamic disc screen (DDS) for the recovery of the organic fraction of MSW (OFMSW), and removal of plastic contaminants from unprocessed black bag MSW. All systems separated MSW into fines and overs fractions based on particle size and density, with compositional analysis confirming enrichment of biodegradable organics in the fines and preferential concentration of plastics in the overs. Vibratory screening produced an organic fines fraction with 45.6% purity and 63.1% OFMSW recovery, whereas trommel screening achieved higher purity (70.5%) but substantially lower recovery (20.6%) on a dry basis. In contrast, DDS screening produced high-purity fines, with OFMSW purity ranging from 77.8% to 99.0% and OFMSW recovery varying between 25.3% and 99.1% across five batches. Across all DDS batches, plastic decontamination efficiency remained high (85.7–95.5%), demonstrating robust contaminant removal over a range of feed compositions and operating conditions. Extended operation using a larger waste mass (∼1,672 kg) further demonstrated the operational stability of the DDS under the tested conditions. The higher motor-speed condition was associated with lower OFMSW recovery in the fines fraction, suggesting a potential influence of motor speed on DDS separation performance. Overall, DDS-based mechanical pre-screening offers a scalable and effective approach for producing high-purity, low-contaminant OFMSW with the potential for downstream biochemical valorization and sustainable MSW management.
Hybridizing existing hydropower plants with energy storage or other generation can expand the flexibility, reliability services, and economic value provided by existing hydropower assets. However, permitting delays and direct regulatory costs can create barriers. So far in the US, nine hydro-hybrids have received Federal Energy Regulatory Commission (FERC) approval and thirteen total hydro-hybrids have been developed. Analyzing these nine cases along with similar projects shows that regulatory delays have increased recently. While hydro-hybrids have not yet been affected by the worst of the delays, the risk of large delays (>500 days) has increased significantly. The worst delay observed was 762 days, which effectively increased the project cost by up to 28%. FERC could reduce this barrier by lowering regulatory costs and the risk of large delays with a stronger and broader categorical exclusion under the National Environmental Policy Act (NEPA) and by narrowly defining FERC’s jurisdiction over hybrid assets. A broader categorical exclusion for typical hybridization projects would reduce repeated work across projects and streamline the part of the process that carries the most timeline risk. Narrowly defining FERC’s jurisdiction over secondary storage or generation assets would increase certainty about the process and streamline many projects. For example, hybridization projects that do not affect hydropower operations and already go through environmental reviews at the state or local level may not need federal regulation to ensure responsible development and operation. This would significantly reduce the cost and risk of large delays to licensees without negative environmental or social outcomes.
Current model validation methods employ standard statistical analysis (to measure statistical differences between data populations) or machine learning (to identify response surface from data) methods to quantitatively measure the difference between simulated and experimental data sets. Such methods do not explicitly encode the causal or mechanistic structure of the processes that generated the data (e.g., boundary conditions, governing physical relationships) into the comparison metric itself. This can generate the erroneous conclusion that, when two data populations are close enough, the models that have generated them are similar. In addition, when simulated and experimental data differ beyond the acceptance criteria, model calibration techniques are used to tweak simulation model parameters to reduce the gap between simulated and experimental data. This might give the false expectation that a simulation model matches reality. This paper presents an alternative approach that moves away from current data-driven methods for validation toward model-driven methods based on causal inference. The presented methods are designed to identify and quantify the causal relationships between data elements rather than looking at their associations. The outcome of such an analysis when applied to each data set (e.g., simulated and experimental data sets) is a structural causal model that analytically represents such causal associations. Validation is then performed by comparing structural causal models generated by simulated and measured data.