Bayesian Belief Network Framework Research Articles

A Bayesian belief network framework for nuclear power plant human reliability analysis accounting for dependencies among performance shaping factors

A challenge to Human Reliability Analysis (HRA) for Nuclear Power Plants (NPPs) lies in the fact that dependencies among Performance Shaping Factors (PSFs) are difficult to deal with due to insufficient knowledge, information and data available. Existing treatment relies heavily on the subjective expert judgment and the dependencies are compromised with the quantities of PSFs, simultaneously, neglects their uncertain interactions. This study proposes a Bayesian Belief Network (BBN) framework for structuring the uncertain dependencies among PSFs and estimate the Human Error Probabilities (HEPs) giving due account to such dependencies. An Exploratory Factor Analysis (EFA) technique is used to analyze human error events and cluster the dependent PSFs into clusters, which serve as the nodes connecting the parent PSF nodes with the child HEP node. Monte Carlo (MC) sampling operationalizes the framework, accounting for the uncertainty that affects PSF clustering and the data filling of conditional probability tables is performed by a Fenton approach.The framework is illustrated by considering 89 human error reports of China NPPs. The results show that the human mental related PSFs complexity, stress/stressor and fitness for duty are highly and steadily dependent, however, the dependencies of the PSFs experience/training and work processes are determined by the specific system situations.

Probabilistic evaluation of CPT-based seismic soil liquefaction potential: towards the integration of interpretive structural modeling and bayesian belief network.

This paper proposes a probabilistic graphical model that integrates interpretive structural modeling (ISM) and Bayesian belief network (BBN) approaches to predict cone penetration test (CPT)-based soil liquefaction potential. In this study, an ISM approach was employed to identify relationships between influence factors, whereas BBN approach was used to describe the quantitative strength of their relationships using conditional and marginal probabilities. The proposed model combines major causes, such as soil, seismic and site conditions, of seismic soil liquefaction at once. To demonstrate the application of the propose framework, the paper elaborates on each phase of the BBN framework, which is then validated with historical empirical data. In context of the rate of successful prediction of liquefaction and non-liquefaction events, the proposed probabilistic graphical model is proven to be more effective, compared to logistic regression, support vector machine, random forest and naive Bayes methods. This research also interprets sensitivity analysis and the most probable explanation of seismic soil liquefaction appertaining to engineering perspective.

Temporal characteristics of an integrated diagnostics risk score before and after heart failure hospitalizations in a large real-world population of patients with implantable devices

Abstract Background Diagnostic parameters measured in implantable cardioverter defibrillators (ICD) and cardiac resynchronization therapy defibrillators (CRTD) have been shown to change before and after heart failure (HF) events with fluid overload. Purpose We investigated the temporal characteristics of an integrated diagnostic risk score before and after HF events in a large real-world cohort of patients with ICD/CRTD devices. Methods We linked a de-identified database of aggregated electronic health record (EHR) data (2007–2017) to a manufacturer's device database with continuous diagnostic monitoring data. Patients with ICD/CRTD implants with intra-thoracic impedance diagnostic feature were included for this study. The previously defined integrated diagnostic risk score was derived by combining daily diagnostic data, including intra-thoracic impedance, night-time heart rate, activity, heart rate variability, and atrial fibrillation (AF) burden, ventricular rate during AF, CRT pacing, ventricular tachycardia episodes and shocks, in a Bayesian Belief Network framework. HF event was defined as an inpatient, emergency department, or observation unit stay in a hospital with primary diagnosis of HF and intravenous diuretic administration. Temporal average of daily risk score across all patients in the 60 days pre and post HF events were compared in HF events with and without readmission within 60 days and with no HF event during pre and post follow-up days. Results A total of 17,886 patients with 1.8±1.2 years of follow-up met inclusion criteria. The average age of patients was 66.6±12.3 years, with 72% being males, and 51% with ICD devices. The average integrated diagnostics risk score in HF events with and without readmission and with no events is shown in Figure. A total of 1174 patients had 1425 HF events with no readmission for HF within 60 days and 282 patients had 295 HF events which were followed by readmission for HF within 60 days. A total of 17,839 patients had no HF events over 86,858 follow-up months. The average daily risk score across all patients was higher on all 60 days pre and post HF event with readmission compared to HF events with no readmission (p<0.001) and both were higher compared to follow-up period with no events (p<0.001). The risk score recovers less often after HF events which are followed by readmission within 60 days compared to HF events with no readmission. Conclusions In a large real-world population of patients with ICD/CRTD devices, the average integrated diagnostics risk score was higher before and after HF events with readmission compared to HF events with no readmission. Re-admissions are more likely in patients with smaller risk score recovery after HF events. Funding Acknowledgement Type of funding source: Private company. Main funding source(s): Medtronic Plc

Assessing relative potential economic impacts of an oil spill on commercial fisheries in the Great Australian Bight using a Bayesian Belief Network framework

Review based experiences of oil-spill induced impacts on fisheries were used to develop a Bayesian Belief Network (BBN) to evaluate the potential economic consequences of a (hypothetical) oil spill in the Great Australian Bight, an area currently being explored by several major international oil companies. The model differs from most previous BBN analyses in that it also considers the relatively short term economic impacts of such a spill, encompassing market and fisher behaviour, as well as impacts on the stocks affected directly by the spill. The results suggest that impacts vary depending on the characteristics of the fishery (i.e. main species, location etc.) and the timing and extent of the oil spill, but market, management and fisher behavioural responses (and their ability to respond) were found to have the greatest impact on the final economic outcomes.

Predicting food web responses to biomanipulation using Bayesian Belief Network: Assessment of accuracy and applicability using in-situ exclosure experiments

Ecological networks are useful for describing the complex trophic interactions within an ecosystem and hold great potential for ecosystem-based management. However, owing to the complexity and limited knowledge on the trophic interactions of natural food webs, it is challenging to make quantitative predictions about ecological community response to management interventions. Here, we use stable isotope mixing models in conjunction with Bayesian Belief Networks (BBN) to develop and examine the trophic interactions for six empirically determined aquatic food webs in tropical reservoirs. Using BBN, we predicted potential trophic cascade outcomes to predator removals, validated the predictions against data observed from in-situ biomanipulation experiments, and identified influential species using sensitivity analyses. Comparisons among various food web modelling frameworks demonstrated the importance of weighted connectance and network-centric approach for quantitative predictions, suggesting that the Bayesian Belief Network framework can play an important role in ecosystem-based management.

A Bayesian belief network framework to predict SOC dynamics of alternative management scenarios

Understanding the key drivers that affect a decline of soil organic carbon (SOC) stock in agricultural areas is of major concern since leading to a decline in service provision from soils and potentially carbon release into the atmosphere. Despite an increasing attention is given to SOC depletion and degradation processes, SOC dynamics are far from being completely understood because they occur in the long term and are the result of a complex interaction between management and pedo-climatic factors. In order to improve our understanding of SOC reduction phenomena in the mineral soils of Veneto region, this study aimed to adopt an innovative probabilistic Bayesian belief network (BBN) framework to model SOC dynamics and identify management scenarios that maximise its accumulation and minimise GHG emissions.Results showed that the constructed BBN framework was able to describe SOC dynamics of the Veneto region, predicting probabilities of general accumulation (11.0%) and depletion (55.0%), similar to those already measured in field studies (15.3% and 50%, respectively). A general enhancement in the SOC content was observed where a minimum soil disturbance was adopted. This outcome suggested that management strategies of conversion from croplands to grasslands, no tillage and conservation agriculture are the most promising management strategies to reverse existing SOC reduction dynamics. Moreover, measures implying SOC stocks were also those providing major benefits in terms of GHGs reduction emissions. Finally, climate change scenarios slightly affected management practice. Advancements in our BBN framework might include more detailed classes at higher resolution as well as any socio-cultural or economic aspect that should improve the evaluation of prediction scenarios.

Predicting pediatric clinic no-shows: a decision analytic framework using elastic net and Bayesian belief network

No-shows are becoming a major problem in primary care facilities, creating additional costs for the facility while adversely affecting the quality of patient care. Accurately predicting no-shows plays an important role in the overbooking strategy. In this study, a hybrid probabilistic prediction framework based on the elastic net (EN) variable-selection methodology integrated with probabilistic Bayesian Belief Network (BBN) is proposed. The study predicts the “no-show probability of the patient(s)” using demographics, socioeconomic status, current appointment information, and appointment attendance history of the patient and the family. The proposed framework is validated using ten years of local pediatric clinic data. It is shown that this EN-based BBN framework is a comparable prediction methodology regarding the best approaches found in the literature. More importantly, this methodology provides novel information on the interrelations of predictors and the conditional probability of predicting “no-shows.” The output of the model can be applied to the appointment scheduling system for a robust overbooking strategy.

A case study of Bayesian belief networks in industrial work process design based on utility expectation and operational performance

PurposeThe primary purpose of this study is to illustrate a practical approach for industrial work process design that, in an integrative manner, captures essential concerns from different parties associated with manufacturing. It aims explicitly to incorporate utility expectation from the perspectives of operational managers, floor workers, and financial planners into the decision making process.Design/methodology/approachThrough a real industrial scenario, the case study illustrates the use of a Bayesian belief network (BBN)‐based expert system and influence diagram in work process design. What‐if analysis is performed. Statistical tests are then used to benchmark and validate the experimental results and actual data.FindingsThe results suggest that the proposed BBN framework is effective in modeling and solving the work design problem. The findings can draw meaningful insights into the adoption and capacity of BBN in the fields of ergonomics, worker health management, and performance improvement.Practical implicationsPractically, the industrial problem is to compare the new stand‐up sewing cells against the traditional sit‐down sewing layout while taking into consideration of ergonomic effect (repetitive motion injury (RMI) likelihood), floor space (SF), yield (%), and cost ($). The study illustrates the use of an expert system and influence diagram to evaluate different alternatives for ergonomic work design in production process.Social implicationsThe results of this study can potentially improve health safety management and worker ergonomics.Originality/valueThe paper is among the few systematic studies that have applied BBN and influence diagram to production ergonomics and worker health management. A methodological framework utilizing these probabilistic reasoning techniques are developed. This new framework can capture essential concerns from different parties in manufacturing.

Using hidden multi-state Markov models with multi-parameter volcanic data to provide empirical evidence for alert level decision-support

For the purposes of eruption forecasting and hazard mitigation, a volcanic crisis may be represented as a staged progression of states of unrest, each with its own timescale and likelihood of transition to other states (or to climactic eruption). If the state conditions can be interpreted physically, e.g., in terms of advancing materials failure, this knowledge could be used directly to inform decisions on alert level setting. A multi-state Markov process provides one simple model for defining states and for estimating rates of switching between states. However, for eruptive processes, such states are not directly observable and must be inferred from latent markers, such as seismic activity, gas output, deformation rates, etc., some of which may be contradictory. Interpretations of uncertain data will be liable to error, so a model is needed which can simultaneously estimate both elements: the transition likelihood of a hidden process and the probabilities of state misclassification. We describe the concept and underlying principles of continuous-time hidden Markov models and demonstrate them in a decision-support context with a preliminary working implementation using MULTIMO data. Where multi-parameter streams of raw, processed or conditioned data of different kinds are available, these can be input in near real-time to appropriate hidden multi-state Markov models, the outputs of each providing their own objective analyses of eruptive state in probabilistic terms. These separate, multiple indicators of state can then be input into a Bayesian Belief Network framework for weighing and combining them as different strands of evidence, together with other observations, data, interpretations and expert judgment.