Data-driven predictive modeling in risk assessment: Challenges and directions for proper uncertainty representation.

Data-driven collapse strength modelling for the screen pipes with internal corrosion defect based on finite element analysis and tree-based machine learning

BackgroundOccupational hazards such as solvents and noise in the electronics industry are serious. Although various occupational health risk assessment models have been applied in the electronics industry, they have only been used to assess the risks of individual job positions. Few existing studies have focused on the total risk level of critical risk factors in enterprises.MethodsTen electronics enterprises were selected for this study. Information, air samples and physical factor measurements were collected from the selected enterprises through on-site investigation, and then the data were collated and samples were tested according to the requirements of Chinese standards. The Occupational Health Risk Classification and Assessment Model (referred to as the Classification Model), the Occupational Health Risk Grading and Assessment Model (referred to as the Grading Model), and the Occupational Disease Hazard Evaluation Model were used to assess the risks of the enterprises. The correlations and differences between the three models were analyzed, and the results of the models were validated by the average risk level of all of the hazard factors.ResultsHazards with concentrations exceeding the Chinese occupational exposure limits (OELs) were methylene chloride, 1,2-dichloroethane, and noise. The exposure time of workers ranged from 1 to 11 h per day and the frequency of exposure ranged from 5 to 6 times per week. The risk ratios (RRs) of the Classification Model, the Grading Model and the Occupational Disease Hazard Evaluation Model were 0.70 ± 0.10, 0.34 ± 0.13, and 0.65 ± 0.21, respectively. The RRs for the three risk assessment models were statistically different (P < 0.001), and there were no correlations between them (P > 0.05). The average risk level of all of the hazard factors was 0.38 ± 0.18, which did not differ from the RRs of the Grading Model (P > 0.05).ConclusionsThe hazards of organic solvents and noise in the electronics industry are not negligible. The Grading Model offers a good reflection of the actual risk level of the electronics industry and has strong practicability.

With the rapid development of socioeconomics and the continuous advancement of urbanization, water environment issues in plain river networks have become increasingly prominent. Accurate and reliable water quality (WQ) predictions are a prerequisite for water pollution warning and management. Data-driven modeling offers a promising approach for WQ prediction in plain river networks. However, existing data-driven models suffer from inadequate capture of spatiotemporal (ST) dependencies and misalignment between direct prediction strategy assumptions with actual data characteristics, limiting prediction accuracy. To address these limitations, this study proposes a spatiotemporal graph neural network (ST-GNN) that integrates four core modules. Experiments were performed within the Chengdu Plain river network, with performance comparisons against five baseline models. Results suggest that ST-GNN achieves rapid and accurate WQ prediction for both short-term and long-term, reducing prediction errors (MAE, RMSE, MAPE) by up to 46.62%, 37.68%, and 45.67%, respectively. Findings from the ablation experiments and autocorrelation analysis further confirm the positive contribution of the core modules in capturing ST dependencies and eliminating data autocorrelation. This study establishes a novel data-driven model for WQ prediction in plain river networks, supporting early warning and pollution control while providing insights for water environment research.

The data-driven interwell simulation model (INSIM) has been recognized as an effective tool for history matching and interwell-connectivity characterization of waterflooding reservoirs. INSIM-FT-3D (FT: front tracking) was recently developed to upgrade the applicationdimension of INSIM series data-driven models from two-dimensional (2D) to three-dimensional (3D). However, INSIM-FT-3D cannot accurately infer the dynamic change of well-connectivity and predict well's bottom-hole pressure (BHP). The main purpose of this study intends to expand the capability of INSIM-FT-3D to empower for the assimilation of BHPs, the reliable prediction of water breakthrough and the characterization of dynamic interwell-connectivities. The default setting of well index (WI) in INSIM-FT-3D based on Peaceman's equation does not yield accurate BHP estimates. We derive a WI that can honor the BHPs of a reference model composed of a set of 1D connections. When history matching BHPs of a 3D reservoir, we show that the derived WI is a better initial guess than that obtained from Peaceman's equation. We also develop a flow-path-tracking (FPT) algorithm to calculate the dynamic interwell properties (allocation factors and pore volumes (PVs)). Besides, we discuss the relationship between the INSIM-family methods and the traditional grid-based methods, which indicates that the INSIM-family methods can calculate the transmissibility of the connection between coarse-scale cells in a more accurate manner. As an improvement of INSIM-FT-3D, the newly proposed data-driven model is denoted as INSIM-FPT-3D. To verify the correctness of the derived WI, we present a 1D problem and a T-shaped synthetic reservoir simulation model as the reference models. BHPs and oil production rates are obtained as the observed data by running these two reference models with total injection/production-rate controls. An INSIM-FPT-3D model is created by specifying the transmissibilities and PVs that are the same as those in the reference model. By applying the derived WIs in INSIM-FPT-3D, the resulting BHPs and oil rates obtained agree well with the reference model without further model calibration. Applying INSIM-FPT-3D to a synthetic multi-layered reservoir shows that we obtain a reasonable match of both BHPs and oil rates with INSIM-FPT-3D. Compared with the FrontSim model, the INSIM-FPT-3D model after history matching is shown to match the dynamic PVs from FrontSim reasonably well and can correctly predict the timing of water breakthrough. By allowing for the assimilation of BHP data, we enable INSIM-FPT-3D to history match a green field with limited production history and forecast the timing of water breakthrough. The improved INSIM-FPT-3D leads to more accurate characterization of the interwell connectivities.

Multi-objective optimization of hydrocyclone by combining mechanistic and data-driven models

SummaryRecently, we have developed two computationally efficient data-driven models—interwell numerical simulation model (INSIM) and INSIM with front tracking (INSIM-FT)—for history matching, prediction, characterization, and optimization of waterflooding reservoirs. Then, stemming from the INSIM family, we derived a new data-driven model referred to as INSIM with flow-path tracking (FPT), for more-accurate interwell connectivity calculations, dynamic flow-path tracking, and waterflooding predictions. The model is a connection-based simulation model that is developed on the basis of a two-phase-flow material-balance equation. With the new model, we can characterize a reservoir by history matching the historical well flow-rate data without the detailed petrophysical properties of the reservoir. In INSIM-FPT, we provide an automatic and systematic workflow that incorporates Delaunay triangulation and imaginary wells to construct the model connection map. We apply a modified depth-first searching method to track all influential flow paths between an injector/producer pair for more-accurate calculations of dynamic allocation factors and control pore volumes (PVs). In addition, we provide a method to visualize a saturation field for a history-matched INSIM-FPT model. On the basis of the saturation map, we design a workflow to evaluate possible drilling locations and future performance of infill wells.For application, we create a synthetic reservoir with two different scenarios to demonstrate the reliability of INSIM-FPT. The results show that the dynamic allocation factors and control PVs between injector/producer pairs in the history-matched INSIM-FPT models are consistent with those obtained from the true streamline simulation model. Furthermore, the oil-saturation field generated with INSIM-FPT reasonably matches that obtained with the true model. It shows that the future predictions of infill wells on the basis of history-matched INSIM-FPT models are reasonably accurate but can be improved if more observed data are collected from near the planned infill wells. We also test a large-scale field problem with 65 wells, which shows INSIM-FPT can reasonably match and predict the field data.

Study on risk assessment models for the aggregation of vehicles transporting hazardous chemicals

Data-driven modeling of geometry-adaptive steady heat conduction based on convolutional neural networks

This brief communication reports on the main findings of the LEMTOX workshop, held from 9 to 12 September 2007, at the Helmholtz Centre for Environmental Research (UFZ) in Leipzig, Germany. The workshop brought together a diverse group of stakeholders from academia, regulatory authorities, contract research organizations, and industry, representing Europe, the United States, and Asia, to discuss the role of ecological modeling in risk assessments of pesticides, particularly under the European regulatory framework. The following questions were addressed: What are the potential benefits of using ecological models in pesticide registration and risk assessment? What obstacles prevent ecological modeling from being used routinely in regulatory submissions? What actions are needed to overcome the identified obstacles? What recommendations should be made to ensure good modeling practice in this context? The workshop focused exclusively on population models, and discussion was focused on those categories of population models that link effects on individuals (e.g., survival, growth, reproduction, behavior) to effects on population dynamics. The workshop participants concluded that the overall benefits of ecological modeling are that it could bring more ecology into ecological risk assessment, and it could provide an excellent tool for exploring the importance of, and interactions among, ecological complexities. However, there are a number of challenges that need to be overcome before such models will receive wide acceptance for pesticide risk assessment, despite having been used extensively in other contexts (e.g., conservation biology). The need for guidance on Good Modeling Practice (on model development, analysis, interpretation, evaluation, documentation, and communication), as well as the need for case studies that can be used to explore the added value of ecological models for risk assessment, were identified as top priorities. Assessing recovery potential of exposed nontarget species and clarifying the ecological relevance of standard laboratory test results are two areas for which ecological modeling may be able to provide considerable benefits.

Biologically based dose-response (BBDR) models predict health outcomes (response) resulting from the presence of a toxicant at a biological target (dose). The benefits of BBDR models are many, and research programs are increasingly focusing on mechanistic research to support model development; however, progress has been slow. Impediments to progress include the complexity of dose response modeling, the need for a multidisciplinary team and consistent funding support, and difficulty in identifying and extracting the needed data. Of immediate concern is the lack of transparency of published models to the supporting data and literature, difficulty in accessing model code and simulation conditions sufficient to allow independent replication of results, and absence of well-defined quality criteria. Suggestions are presented to improve the development and use of BBDR models in risk assessment and to address the above limitations. Examples from BBDR models for methylmercury neurotoxicity and 5-fluorouracil embryotoxicity are presented to illustrate the suggestions including what kinds of databases are needed to support model development and transparency, quality assurance for modeling, and how the internet can advance database development and collaboration within the biological modeling community.

RUL (remaining useful life) estimation is one of the main functions of the predictive analytics systems for rotary machines. Data-driven models based on large amounts of multisensory measurements data are usually utilized for this purpose. The use of adjustable bearings, on the one hand, improves a machine’s performance. On the other hand, it requires considering the additional variability in the bearing parameters in order to obtain adequate RUL estimates. The present study proposes a hybrid approach to such prediction models involving the joint use of physics-based models of adjustable bearings and data-driven models for fast on-line prediction of their parameters. The approach provides a rather simple way of considering the variability of the properties caused by the control systems. It has been tested on highly loaded locomotive traction motor axle bearings for consideration and prediction of their wear and RUL. The proposed adjustable design of the bearings includes temperature control, resulting in an increase in their expected service life. The initial study of the system was implemented with a physics-based model using Archard’s law and Reynolds equation and considering load and thermal factors for wear rate calculation. The dataset generated by this model is used to train an ANN for high-speed on-line bearing RUL and wear prediction. The results show good qualitative and quantitative agreement with the statistics of operation of traction motor axle bearings. A number of recommendations for further improving the quality of predicting the parameters of active bearings are also made as a summary of the work.

A new holistic model of risk assessment: Towards an exhaustive conceptualization of risk elements

Modeling the behavior of groundwater levels is necessary to implement sustainable groundwater resource management. Groundwater is a non-linear and complex system, which can be modeled by data-driven models. This study evaluates the performances of data-driven models, support vector machine regression (SVR) and artificial neural network (ANN), for forecasting groundwater levels of confined and unconfined systems at 1-, 2-, and 3-month ahead. This is the first time that confined and unconfined aquifers have been compared using data-driven models. In addition, to identify the optimal input combination, a hybrid gamma test (GT) and genetic algorithm (GA) was used. The coefficient of correlation (R), Mean Absolute Error (MAE), root mean squared error (RMSE), Nash–Sutcliffe efficiency (NSE), and developed discrepancy ratio (DDR) were applied to evaluate the SVR and ANN models. Results showed that the SVR and ANN models were more accurate for the unconfined system than the confined system for forecasts up to 3-month ahead. In both hydrogeological systems for 1-month ahead, the models performed better than for 2- and 3-month ahead forecasts, and the accuracy of the models decreased as the months ahead increased. The SVR model performed better than the ANN model for 1-, 2-, and 3-month ahead groundwater-level forecasting. The SVR model could be successfully used in predicting monthly groundwater in confined and unconfined systems.

Listeria monocytogenes in Retail Delicatessens: An Interagency Risk Assessment—Model and Baseline Results

In model predictive control, fully data-driven prediction models can be used besides common (non-)linear prediction models based on first-principles. Although no process knowledge is required while relying only on sufficient data, they suffer in their extrapolation capability which is shown in the present work for the control of an air separation unit. In order to compensate for the deficits in the extrapolation behavior, a further data source, here a digital twin, is deployed for additional data generation. The plant data set is augmented with the artificially generated data giving rise to a hybrid model in terms of data generation. It is shown that this model can significantly improve the prediction quality in former extrapolation areas of the plant data set. Even conclusions about the uncertainty behavior of the prediction model can be found.