Prediction Uncertainty Research Articles

Uncertainties in exposure predictions arising from point measurements of carbon dioxide in classroom environments.

Predictions of airborne infection risk can be made based on the fraction of rebreathed air inferred from point measurements of carbon dioxide (CO[Formula: see text]). We investigate the extent to which environmental factors, particularly spatial variations due to the ventilation provision, affect the uncertainty in these predictions. Spatial variations are expected to be especially problematic in naturally ventilated spaces, which include the majority of classrooms in the UK. An idealized classroom, broadly representative of the physics of (buoyancy-driven) displacement ventilation, is examined using computational fluid dynamics, with different ventilation configurations. Passive tracers are used to model both the CO[Formula: see text] generated by all 32 occupants and the breath of a single infectious individual (located in nine different regions). The distribution of infected breath is shown to depend strongly on the distance from the release location but is also affected by the pattern of the ventilating flow, including the presence of stagnating regions. However, far-field exposure predictions based on single point measurements of CO[Formula: see text] within the breathing zone are shown to rarely differ from the actual exposure to infected breath by more than a factor of two-we argue this uncertainty is small compared with other uncertainties inherent in modelling airborne infection risk.

Uncertainty-based knowledge distillation for Bayesian deep neural network compression

Deep learning models have been widely employed across various fields. In real-world scenarios, especially safety-critical applications, quantifying uncertainty is as crucial as achieving high accuracy. To address this concern, Bayesian deep neural networks (BDNNs) emerged to estimate two different types of uncertainty: Aleatoric and Epistemic. Nevertheless, implementing a BDNN on resource-constrained devices poses challenges due to the substantial computational and storage costs imposed by approximation inference techniques. Thus, efficient compression methods should be utilized. We propose an uncertainty-based knowledge distillation method to compress BDNNs. Knowledge distillation is a model compression technique that involves transferring knowledge from a complex network, known as the teacher network, to a simpler one, referred to as the student network. Our method incorporates uncertainty into knowledge distillation to address situations where inappropriate teacher supervision undermines compression performance. We utilize the Epistemic uncertainty of teacher predictions to tailor supervision for each sample individually to take into account teacher's limited knowledge. Additionally, we adjust the temperature parameter of the distillation process for each sample based on the Aleatoric uncertainty of the teacher predictions, ensuring that the student receives appropriate supervision even in the presence of ambiguous data. As a result, the proposed method enables the Bayesian student network to be trained under both appropriate supervision of the Bayesian teacher network and ground truth labels. We evaluated our method on the CIFAR-10, CIFAR-100, and RAF-DB datasets, demonstrating notable improvements in accuracy over state-of-the-art knowledge distillation-based methods. Furthermore, the robustness of our approach was assessed through testing weakly trained teacher networks and the analysis of blurred and low-resolution data, which have high uncertainty. Experimental results show that the proposed method outperformed existing methods.

A Novel Strategy Coupling Optimised Sampling with Heterogeneous Ensemble Machine-Learning to Predict Landslide Susceptibility

The accuracy of data-driven landslide susceptibility prediction depends heavily on the quality of non-landslide samples and the selection of machine-learning algorithms. Current methods rely on artificial prior knowledge to obtain negative samples from landslide-free regions or outside the landslide buffer zones randomly and quickly but often ignore the reliability of non-landslide samples, which will pose a serious risk of including potential landslides and lead to erroneous outcomes in training data. Furthermore, diverse machine-learning models exhibit distinct classification capabilities, and applying a single model can readily result in over-fitting of the dataset and introduce potential uncertainties in predictions. To address these problems, taking Chenxi County, a hilly and mountainous area in southern China, as an example, this research proposes a strategy-coupling optimised sampling with heterogeneous ensemble machine learning to enhance the accuracy of landslide susceptibility prediction. Initially, 21 landslide impact factors were derived from five aspects: geology, hydrology, topography, meteorology, human activities, and geographical environment. Then, these factors were screened through a correlation analysis and collinearity diagnosis. Afterwards, an optimised sampling (OS) method was utilised to select negative samples by fusing the reliability of non-landslide samples and certainty factor values on the basis of the environmental similarity and statistical model. Subsequently, the adopted non-landslide samples and historical landslides were combined to create machine-learning datasets. Finally, baseline models (support vector machine, random forest, and back propagation neural network) and the stacking ensemble model were employed to predict susceptibility. The findings indicated that the OS method, considering the reliability of non-landslide samples, achieved higher-quality negative samples than currently widely used sampling methods. The stacking ensemble machine-learning model outperformed those three baseline models. Notably, the accuracy of the hybrid OS–Stacking model is most promising, up to 97.1%. The integrated strategy significantly improves the prediction of landslide susceptibility and makes it reliable and effective for assessing regional geohazard risk.

A geographically weighted neural network model for digital soil mapping of heavy metal copper in coastal cities

Assessing the spatial distribution of heavy metals in soil is essential for mitigating risks to human health and ensuring the sustainable use of soil resources. This study proposed a geographically weighted neural network (GWNN) model, leveraging deep learning and geographically weighted regression (GWR). The model was designed based on the GWR concept to address the spatial autocorrelation of copper (Cu) in soil and incorporated convolutional neural network (CNN) to capture the nonlinear relationships between Cu and environmental covariates. The GWNN model was compared with GWR, extreme gradient boosting (XGBoost), and artificial neural network (ANN) models. XGBoost was employed to select important environmental covariates and spatial autocorrelation of Cu concentrations was assessed using Moran’s I. The model’s performance was evaluated using 10-fold cross-validation, and prediction uncertainty was quantified with 100 bootstrap models. The results indicated that temperature covariates were the most significant predictors of soil Cu concentrations. The R2 values for Cu prediction accuracy were 0.60 for GWNN, 0.53 for ANN, 0.49 for XGBoost, and 0.32 for GWR. The spatial distribution of Cu showed a trend of higher concentrations in the north and lower concentrations in the south, consistent with spatial clusters identified by local Moran’s I. The mean uncertainty of the 90% confidence interval for GWNN was 16.49%, closely aligning with XGBoost (15.44%) and ANN (16.29%) and significantly outperforming the GWR (18.25%). Overall, the GWNN model demonstrated strong predictive accuracy and low uncertainty, offering an improved approach for digital soil mapping applications.

Interpreting and evaluating digital soil mapping prediction uncertainty: A case study using texture from SoilGrids

Soil information is critical for a wide range of land resource and environmental decisions. These decisions will be compromised when the soil information quality is unsatisfactory. Thus, users of soil information need to understand and consider the uncertainty of the available soil information and be able to judge whether it is fit for purpose. The uncertainty information provided with the SoilGrids2.0 product was examined in a case study. We hypothesised that the soil property predictions for the Netherlands (NL) might be less uncertain than those of New Zealand (NZ) because there were more relevant training data for NL than for NZ. The study objectives were to: 1)understand whether the provided uncertainty information is correct for both countries; 2)explore spatial patterns and relationships in the prediction error and uncertainty information using quantitative tools and new graphical analyses; 3)analyse whether these patterns and relations can be explained; and 4)explore how the uncertainty information and insights derived from graphical analyses might assist an end user to determine whether a map is suitable for their purpose. The study focused on soil texture.Independent datasets showed that the SoilGrids 2.0 uncertainty information was too optimistic for sand and too pessimistic for clay for both countries. The graphical analyses confirmed the initial assumption that NL predictions were more accurate than those for NZ, but they also indicated that some locations in NL have high uncertainty. The graphical analyses allowed only a limited identification of the four sources of uncertainty in digital soil maps, but were quite insightful in helping us to better understand the reliability of the information. A set of recommendations was developed for both producers and consumers of digital soil mapping (DSM) products. This includes the provision of a summary map of accuracy classes. We suggest that more research and educational effort is needed to ensure that digital soil maps are used appropriately.

Modeling human activity comprehension at human scale: prediction, segmentation, and categorization.

Humans form sequences of event models-representations of the current situation-to predict how activity will unfold. Multiple mechanisms have been proposed for how the cognitive system determines when to segment the stream of behavior and switch from one active event model to another. Here, we constructed a computational model that learns knowledge about event classes (event schemas), by combining recurrent neural networks for short-term dynamics with Bayesian inference over event classes for event-to-event transitions. This architecture represents event schemas and uses them to construct a series of event models. This architecture was trained on one pass through 18 h of naturalistic human activities. Another 3.5 h of activities were used to test each variant for agreement with human segmentation and categorization. The architecture was able to learn to predict human activity, and it developed segmentation and categorization approaching human-like performance. We then compared two variants of this architecture designed to better emulate human event segmentation: one transitioned when the active event model produced high uncertainty in its prediction and the other transitioned when the active event model produced a large prediction error. The two variants learned to segment and categorize events, and the prediction uncertainty variant provided a somewhat closer match to human segmentation and categorization-despite being given no feedback about segmentation or categorization. These results suggest that event model transitioning based on prediction uncertainty or prediction error can reproduce two important features of human event comprehension.

Taming Overconfident Prediction on Unlabeled Data From Hindsight.

Minimizing prediction uncertainty on unlabeled data is a key factor to achieve good performance in semi-supervised learning (SSL). The prediction uncertainty is typically expressed as the entropy computed by the transformed probabilities in output space. Most existing works distill low-entropy prediction by either accepting the determining class (with the largest probability) as the true label or suppressing subtle predictions (with the smaller probabilities). Unarguably, these distillation strategies are usually heuristic and less informative for model training. From this discernment, this article proposes a dual mechanism, named adaptive sharpening (ADS), which first applies a soft-threshold to adaptively mask out determinate and negligible predictions, and then seamlessly sharpens the informed predictions, distilling certain predictions with the informed ones only. More importantly, we theoretically analyze the traits of ADS by comparing it with various distillation strategies. Numerous experiments verify that ADS significantly improves state-of-the-art SSL methods by making it a plug-in. Our proposed ADS forges a cornerstone for future distillation-based SSL research.

Harnessing uncertainty: A deep mechanistic approach for cautious diagnostic and forecast of Bovine Respiratory Disease

Bovine Respiratory Disease (BRD) is a prevalent infectious disease of respiratory tract in cattle, presenting challenges in accurate diagnosis and forecasting due to the complex interactions of multiple risk factors. Common methods, including mathematical epidemiological models and data-driven approaches such as machine learning models, face limitations such as difficult parameter estimation or the need for data across all potential outcomes, which is challenging given the scarcity and noise in observing BRD processes. In response to these challenges, we introduce a novel approach known as the Bayesian Deep Mechanistic method. This method couples a data-driven model with a mathematical epidemiological model while accounting for uncertainties within the processes. By utilising 265 lung ultrasound videos as sensor data from 163 animals across 9 farms in France, we trained a Bayesian deep learning model to predict the infection status (infected or non-infected) of an entire batch of 12 animals, also providing associated confidence levels. These predictions, coupled with their confidence levels, were used to filter out highly uncertain diagnoses and diffuse uncertainties into the parameterisation of a mathematical epidemiological model to forecast the progression of infected animals. Our findings highlight that considering the confidence levels (or uncertainties) of predictions enhances both the diagnosis and forecasting of BRD. Uncertainty-aware diagnosis reduced errors to 32 %, outperforming traditional automatic diagnosis. Forecast relying on veterinarian diagnoses, considered the most confident, had a 23 % error, whilst forecast taking into account diagnosis uncertainties had a close 27.2 % error. Building upon uncertainty-awareness, our future research could explore integrating multiple types of sensor data, such as video surveillance, audio recordings, and environmental parameters, to provide a comprehensive evaluation of animal health, employing multi-modal methods for processing this diverse data.

Efficient Bayesian updating for deformation prediction of high rock slopes induced by excavation with monitoring data

This study develops an efficient Bayesian updating method with monitoring data for predicting the deformation of high rock slopes induced by excavation. The importance ranking based on random forest is introduced to identify the key rock parameters as random variables in the Bayesian updating. The surrogate models using support vector machine are constructed to approximate the physical numerical models using FLAC3D for evaluating slope deformation. A practical example involving deformation prediction of the excavated left-bank rock slope for the well-known Baihetan hydropower station in southwest China is presented. The results indicate that the developed Bayesian updating method can efficiently and accurately update the posterior distributions of rock parameters and predict the deformation of high rock slopes induced by excavation. Incorporating the monitoring data of displacement into the Bayesian updating can effectively reduce the uncertainty of rock parameters and displacement prediction. As a result, the displacement predictions made by the Bayesian updating are closer to the monitoring data than the prior displacement predictions. In addition, incorporating more monitoring data of displacement from the previous excavation stages produces more accurate displacement predictions for subsequent excavation stages.

Vortex Trapping of Suspended Sand Grains Over Ripples

AbstractCoastal hydrodynamics and morphodynamics integrate the effects of small‐scale fluid‐sediment interactions; yet, these small‐scale processes are not well understood. To investigate sediment trapping by turbulent coherent structures or vortices, the transport of coarse sand over ripples was analyzed in a small‐oscillatory flow tunnel with phase‐separated Particle Image and Tracking Velocimetry. Results from one of the first direct measurements of vortex‐trapped sand grains under oscillatory flows are presented. The vortices mobilized sand grains along the ripple slopes just prior to flow reversal and transported the suspended sediment grains. During several flow cycles, some sand grains were temporarily trapped in the vortex, prescribing semi‐circular trajectories off‐center from the vortex core in quadrants of the vortex that were closest to the ripple slope, as illustrated by Nielsen (1992, https://doi.org/10.1142/1269). Comparisons of the horizontal sediment grain velocity with the horizontal fluid velocity yielded a linear relationship with a slope of 0.87. The vertical grain velocities also varied linearly with the vertical fluid velocity with a slope of approximately 1 and an offset of −0.08 m . The offset is close to the still water settling velocity for coarse sand grains, as hypothesized during vortex trapping. Additionally, estimates of the off‐center distance, between the centers of the semi‐circular sediment paths and vortex cores, compared well with the ratio of the settling velocity to the radian frequency of the vortex yielding a linear regression slope of 0.99. Improved understanding of vortex trapping effects on sediment dynamics may decrease uncertainty in model predictions of large‐scale coastal hydrodynamics and sediment transport.

Parameterization of Vertical Turbulent Transport in the Inner Core of Tropical Cyclones and Its Impact on Storm Intensification. Part I: Sensitivity to Turbulent Mixing Length

Abstract In the inner core of a tropical cyclone, turbulence not only exists in the boundary layer (BL) but can also be generated above the BL by eyewall and rainband clouds. Thus, the treatment of vertical turbulent mixing must go beyond the conventional scope of the BL. The turbulence schemes formulated based on the turbulent kinetic energy (TKE) are attractive as they are applicable to both deep and shallow convection regimes in the tropical cyclone (TC) inner core provided that the TKE production and dissipation can be appropriately determined. However, TKE schemes are not self-closed. They must be closed by an empirically prescribed vertical profile of mixing length. This motivates this study to investigate the sensitivity of the simulated TC intensification to the sloping curvature and asymptotic length scale of mixing length, the two parameters that determine the vertical distribution of a prescribed mixing length. To tackle the problem, both idealized and real-case TC simulations are performed. The results show that the simulated TC intensification is sensitive to the sloping curvature of mixing length but only exhibits marginal sensitivity to the asymptotic length scale. The underlying reasons for such sensitivities are explored analytically based on the Mellor and Yamada level-2 turbulence model and the analyses of azimuthal-mean tangential wind budget. The results highlight the uncertainty and importance of mixing length in the numerical prediction of TCs and suggest that future research should focus on searching for physical constraints on mixing length, particularly in the low- to midtroposphere, using observations and large-eddy simulations. Significance Statement The parametric representation of subgrid-scale turbulent mixing is one of the major sources of uncertainty in numerical predictions of tropical cyclones (TCs). This study investigates how the numerical prediction of TC intensification is affected by the turbulent mixing length, a length scale that is required to close a turbulence scheme formulated based on the turbulent kinetic energy (TKE). The research highlights the uncertainty and importance of mixing length in numerical prediction of TCs and suggests that future research should focus on searching for physical constraints on the mixing length, particularly in the low- to midtroposphere, using observations and large-eddy simulations.

Can the young water fraction reduce predictive uncertainty in water transit time estimations?

Transit time distributions (TTDs) of streamflow are informative descriptors of catchment hydrological functioning and solute transport mechanisms. Conventional methods for estimating TTDs generally require model calibration against extensive tracer data time series, which are often limited to well-studied experimental catchments. We challenge this limitation and propose an alternative approach that uses the young water fraction (Fywobs), an increasingly used water age metric which represents the proportion of streamflow with a transit time younger than 2-3 months, and that can be robustly estimated with sparsely measured tracer data. To this end, we conducted a proof of concept study by modelling TTDs using StorAge Selection (SAS) functions with oxygen isotopes (δ18O) measurements for 23 diverse catchments in Germany. In a Monte-Carlo approach, we computed the (averaged) marginal TTDs of a prior parameter distribution and derived a model-based Fyw (Fywsim). We compared Fywsim with Fywobs, obtained from δ18O measurements, and constrained the prior SAS parameters distribution. Subsequently, we derived a posterior distribution of parameters and resulting model simulations. Our findings showed that using Fywobs to constrain the model effectively reduced parameter equifinality and simulation uncertainty. However, the value of Fywobs on reducing model uncertainty varied across sites, with larger values (Fywobs≥0.10) leading to simulations with a narrower uncertainty band and higher model efficiency, whilst smaller values (Fywobs≤0.05) had limited influence on reducing model output uncertainty. We discussed the potential and limitations of combining SAS functions with Fywobs, and considered broader implications of this approach for enhancing our understanding of catchment functioning and water quality status.

Fouling fault detection and diagnosis in district heating substations: Validation of a hybrid CNN-based PCA model with uncertainty quantification on virtual replica synthesis and real data

This research presents an advanced fouling fault detection (FFD) model tailored to district heating substations, integrating convolutional neural networks, principal component analysis, and uncertainty quantification (UQ). Through careful hyperparameter tuning, the model attained optimal configurations, showcasing accuracy metrics ranging from 96.58 % to 98.19 %. Key findings include the influence of input vectors, particularly the incorporation of overall heat transfer coefficient (UA), which contributed to heightened accuracy and precision. Visualizing predictive uncertainty emphasized the model's adaptability to dynamic conditions, crucial in scenarios involving progressive fault evolution. Generalizability analysis across diverse substations affirmed the model's adaptability, surpassing existing algorithms. Comparative evaluation against recent models underscored the proposed model's superiority, with an accuracy range exceeding 96.58 %. This research introduces UQ as a unique advantage, offering a state-of-the-art solution for FFD in dynamic systems. The model's validation on real data further solidifies its efficacy.

Interpretable Transformer Neural Network Prediction of Diverse Environmental Time Series Using Weather Forecasts

AbstractTransformer neural networks (TNNs) have caused a paradigm shift in deep learning domains like natural language processing, gathering immense interest due to their versatility in other fields such as time series forecasting (TSF). Most current TSF applications of TNNs use only historic observations to predict future events, ignoring information available in weather forecasts to inform better predictions, and with little attention given to the interpretability of the model's use of explanatory inputs. This work explores the potential for TNNs to perform TSF across multiple environmental variables (streamflow, stage, water temperature, and salinity) in two ecologically important regions: the Peace River watershed (Florida) and the northern Gulf of Mexico (Louisiana). The TNN was tested and its prediction uncertainty quantified for each response variable from one‐to fourteen‐day‐ahead forecasts using past observations and spatially distributed weather forecasts. A sensitivity analysis (SA) was performed on the trained TNNs' attention weights to identify the relative influence of each input variable on each response variable across prediction windows. Overall model performance ranged from good to very good (0.78 < NSE < 0.99 for all variables and forecast horizons). Through the SA, we found that the TNN was able to learn the physical patterns behind the data, adapt the use of input variables to each forecast, and increasingly use weather forecast information as prediction windows increased. The TNN's excellent performance and flexibility, along with the intuitive interpretability highlighting the logic behind the models' forecasting decision‐making process, provide evidence for the applicability of this architecture to other TSF variables and locations.

Insights into the prediction uncertainty of machine-learning-based digital soil mapping through a local attribution approach

Abstract. Machine learning (ML) models have become key ingredients for digital soil mapping. To improve the interpretability of their predictions, diagnostic tools such as the widely used local attribution approach known as SHapley Additive exPlanations (SHAP) have been developed. However, the analysis of ML model predictions is only one part of the problem, and there is an interest in obtaining deeper insights into the drivers of the prediction uncertainty as well, i.e. explaining why an ML model is confident given the set of chosen covariate values in addition to why the ML model delivered some particular results. In this study, we show how to apply SHAP to local prediction uncertainty estimates for a case of urban soil pollution – namely, the presence of petroleum hydrocarbons in soil in Toulouse (France), which pose a health risk via vapour intrusion into buildings, direct soil ingestion, and groundwater contamination. Our results show that the drivers of the prediction best estimates are not necessarily the drivers of confidence in these predictions, and we identify those leading to a reduction in uncertainty. Our study suggests that decisions regarding data collection and covariate characterisation as well as communication of the results should be made accordingly.

Hurricane surge and inundation in the Bahamas, part 1: Storm surge model

AbstractA storm surge model has been developed as part of a pilot study for The Commonwealth of the Bahamas. The hydrodynamic model, TELEMAC‐2D, is used to simulate the response of water level to tide and the wind and atmospheric pressure fields of hurricanes and subsequent inundation over land. The model is used by The Bahamas Department of Meteorology to forecast storm surge and flooding over the islands of Grand Bahama and Eleuthera for incoming hurricanes to assist in preparation for and management of hurricane surge events and has been used in a flood risk assessment, reported in a companion paper. The model has been optimised to run quickly while also resolving the bathymetry and topography that affect the development and propagation of storm surge. Wind fields are generated within TELEMAC‐2D based on hurricane warning bulletins. The model has been validated for its representation of water level and against tide gauge measurements during four historical hurricanes: Irene, Sandy, Matthew and Dorian. Factors contributing to uncertainty in forecast predictions are discussed and recommendations are provided to improve the performance in future. The pilot study provides a template for future expansion to cover the other inhabited islands of The Bahamas.

Progress in Research on Terpenoid Biosynthesis and Terpene Synthases of Lauraceae Species

Lauraceae, an important family of Angiospermae, comprises over 2500 species widely distributed in tropical and subtropical evergreen broad-leaved forests. This family is renowned for its rich resource of terpenoids, particularly monoterpenes, sesquiterpenes, and diterpenes. These compounds not only impart specific scents to Lauraceae species but also play crucial roles in plant growth, development, and environmental adaptation. These compounds also possess extensive bioactivities, such as antioxidant, antibacterial, anti-inflammatory, and neuroprotective effects, making them valuable in the fields of perfumery, cosmetics, food, and medicine, and thus holding significant economic value. Recent advancements in high-throughput technologies, especially genomics, transcriptomics, and metabolomics, have significantly advanced our knowledge of the chemical constituents and biosynthetic pathways of terpenoids in Lauraceae species. Such progress has also shed light on the diversity and functionality of the terpene synthases (TPSs) gene family, a key enzyme involved in terpenoid biosynthesis. This paper reviews the latest research findings on the biosynthetic pathways of terpenoids and their key enzyme-encoding gene families in Lauraceae plants. We also analyze the evolutionary patterns of TPS gene family members of four Lauraceae species at the whole-genome level and summarize their mechanisms of action in secondary metabolite synthesis. Furthermore, this paper highlights the current research challenges and proposes prospects, such as the complexity of gene families, the uncertainties in functional predictions, and unclear regulatory mechanisms. Our objective is to provide scientific foundations for the in-depth analysis of terpenoid biosynthesis mechanisms and the development and utilization of natural products in Lauraceae plants.

Probabilistic Air Traffic Complexity Analysis Considering Prediction Uncertainties in Traffic Scenarios

This article presents a methodology for analyzing probabilistic air traffic complexity by integrating prediction uncertainties in convective weather scenarios. With the Performance Review Unit (PRU) model as a base, this method modifies the original framework by incorporating a weather-related complexity indicator. The approach was tested in Austrian airspace using ensemble weather forecasts and historical flight plan data. The results demonstrated that a probabilistic model effectively assesses traffic complexity and captures trends in complexity over time, providing greater reliability in high-complexity sectors. Validation revealed a strong alignment between simulator complexity values and probabilistic complexity, especially in sectors characterized by dense data distributions. In contrast, sectors with more elongated distributions tended to overestimate complexity. Quantitative analysis indicated that the error between the probabilistic mean complexity and the simulator complexity values ranged from 12% to 23%, with higher errors in sectors with lower complexity. This validation confirmed the model’s ability to predict complexity trends, thereby assisting flow manager positions (FMPs) in traffic flow and airspace management. Overall, this study demonstrated that probabilistic complexity assessment provides a deeper understanding of traffic behaviour, facilitating more effective air traffic flow management in uncertain and dynamic conditions.

A deep learning approach for pose error prediction in parallel robots

Parallel robots are increasingly favored in industrial automation due to their high precision, stability, and rigidity in high-speed and heavy-load tasks. However, factors such as manufacturing tolerances, assembly deviations, loads, and environmental conditions can lead to pose inaccuracies, manifesting as both geometric and non-geometric errors. Traditional model-driven approaches effectively predict geometric errors but often neglect non-geometric factors, potentially impacting the overall accuracy. This paper presents a deep learning-based method for comprehensive pose error prediction in parallel robots, addressing both geometric and non-geometric errors. The proposed approach employs an encoder-decoder network that takes target pose inputs and outputs predicted pose errors. The encoder integrates multi-layer convolutional sparse coding (CSC) blocks and efficient channel attention (ECA) modules to extract features from the network inputs. Monte Carlo dropout is utilized to estimate prediction uncertainties, enhancing the reliability of the network predictions. Experimental validation on a Stewart platform demonstrates the high accuracy and robustness of the method under various loading conditions. Comparative analyses with several typical models affirm the superior performance of the proposed network, indicating its potential to improve the operational accuracy and reliability of parallel robots in high-precision applications.

Hail hazard modeling with uncertainty analysis and roof damage estimation of residential buildings in North America

This research presents a statistical approach for hail risk modeling that incorporates the uncertainties of hail model prediction to provide insight into assessing the roof damage of a residential house in hail events. By quantifying the inherent uncertainties in evaluating hailstorm characteristics, this study extends the current existing hail models. The hail data are sourced from the Community Collaborative Rain, Hail and Snow Network (CoCoRaHS) in the U.S. In the modeling process, the largest hail diameter reported in the CoCoRaHS database serves as a primary input variable to estimate the number of observations for the largest hail diameter, hailstorm duration, and hit rate. The assessment of hail risk in this study focuses on the probability of hail damage and resultant repair costs for five types of roofs in North America (unrated roof and impact-resistant roofs with UL 2218 rating classes 1 to 4). The probability of hail damage is calculated as the failure probability by integrating all individual hailstone hits having variable diameters during a hailstorm with fragility curves, which estimate the probability that hailstones will fracture asphalt shingles (allowing water infiltration) or that they dislodge enough granules to cause visible damage requiring replacement for aesthetic reasons. The results reveal that an impact-resistant roof (impact-resistant rating classes 1 to 4) is associated with lower hail risks, with 60 % to 98 % reduction on average compared to unrated roofs. This study provides a comprehensive uncertainty modeling approach for hail hazard and risk, enabling better-informed decision-making and risk management strategies.