Influence Diagram Model Research Articles

Attribute statistical process control under nonconstant process deterioration

AbstractA statistical process control (SPC) model is introduced that incorporates sample data on the number of defectives and allows the probability that an assignable cause of variation in each time interval of a finite production process to be nonconstant. A Limited Memory Influence Diagram (LIMID) model is implemented to facilitate the decision in each time interval on whether or not to investigate the potential presence of the assignable cause of variation and restore the process to working condition. The method determines control limits that minimize the costs of inspection, repair, sampling, and operating under the assignable cause of variation. Sample size and sampling intervals can also be adjusted to further reduce the costs of maintaining quality control. This method expands on the capabilities of models that assume that the conditional probability of an assignable cause occurring remains constant throughout the production horizon.

A risk-based maintenance decision model for subsea pipeline considering pitting corrosion growth

Pitting corrosion growth of subsea pipeline can cause pipeline leak, resulting in the huge economic losses. This study presents a risk-based model to implement the optimal maintenance decision-making of subsea pipelines subject to pitting corrosion. Firstly, a pitting corrosion growth prediction model is built using DeWaard 95 model and dynamic Bayesian network (DBN). Subsequently, the probability of failure, i.e., PoF and consequence of failure, i.e., CoF for pipeline are estimated to obtain the risk profile of corroded pipeline. The maintenance cycle of pipeline is estimated based on total utility function and the acceptable risk. Eventually, a Bayesian influence diagram (BID) model and expected utility theory (EEU) are implemented to determine maintenance decision of corroded pipelines. The methodology is illustrated by a case study, which indicates that it can be a useful tool for maintenance decisions for subsea pipeline subject to pitting corrosion growth.

Intermittent sampling for statistical process control with the number of defectives

An intermittent sampling model for statistical process control (SPC) is introduced to monitor the quality of output from a production process where the number of defective units in a sample is measured in selected time periods. A Limited Memory Influence Diagram (LIMID) model is implemented to determine the time periods in which to collect samples and the decision strategy to minimize total costs of quality control. In periods where samples are collected, the observed defectives determine whether the process is stopped to investigate and repair an assignable cause of variation. Based on sample results, the process can alternatively be allowed to run without interruption until the next predetermined sampling interval, or the results may suggest collecting a sample again in the next period. The model only requires the user to know the result of the current sample to make a decision, in contrast to Bayesian methods that require calculations based on all prior samples and a history of actions. Despite the limited and intermittent nature of the information, the model provides lower quality costs than existing methods for a wide range of production time horizons.

A probabilistic framework for risk management and emergency decision-making of marine oil spill accidents

Offshore oil spills may pose a severe threat to marine ecological environment. In this paper, a new methodology based on Bayesian network (BN) and Influence Diagram (ID) is developed for risk management and emergency decision-making of marine oil spill accidents. The methodology integrates risk management before accident and emergency response after accident, which can balance risk and cost, and render an optimal decision-making. Marine oil spill scenarios including root causations, intermediate and consequent events are identified and modeled using BN considering the dependencies and multi-state of incident process nodes. The probabilities of offshore oil spill incident and the resulting ecological disasters are estimated. The prevention and mitigation measures marine oil spill incidents are identified and added to BN for developing Bayesian ID model, in which the cost and utility of implementing each safety measure are considered. Bayesian ID model can estimate the cost and utility of all safety strategies, and the optimal risk management and emergency strategy are determined by balancing the cost and utility. A case study of marine oil spill accident due to subsea oil pipeline leak is used to illustrate the methodology. It is observed that the methodology can efficiently support the decision-making of oil and gas sector in risk management and emergency response of offshore oil spill accidents.

Expected Value of Partial Perfect Information in Hybrid Models Using Dynamic Discretization

In decision theory models, expected value of partial perfect information (EVPPI) is an important analysis technique that is used to identify the value of acquiring further information on individual variables. EVPPI can be used to prioritize the parts of a model that should be improved or identify the parts where acquiring additional data or expert knowledge is most beneficial. Calculating EVPPI of continuous variables is challenging, and several sampling and approximation techniques have been proposed. This paper proposes a novel approach for calculating EVPPI in hybrid influence diagram (HID) models (these are influence diagrams (IDs) containing both discrete and continuous nodes). The proposed approach transforms the HID into a hybrid Bayesian network and makes use of the dynamic discretization and the junction tree algorithms to calculate the EVPPI. This is an approximate solution (no feasible exact solution is possible generally for HIDs) but we demonstrate it accurately calculates the EVPPI values. Moreover, unlike the previously proposed simulation-based EVPPI methods, our approach eliminates the requirement of manually determining the sample size and assessing convergence. Hence, it can be used by decision-makers who do not have deep understanding of programming languages and sampling techniques. We compare our approach to the previously proposed techniques based on two case studies.

A decision analysis approach for risk management of near-earth objects

Risk management of near-Earth objects (NEOs; e.g., asteroids and comets) that can potentially impact Earth is an important issue that took on added urgency with the Chelyabinsk event of February 2013. Thousands of NEOs large enough to cause substantial damage are known to exist, although only a small fraction of these have the potential to impact Earth in the next few centuries. The probability and location of a NEO impact are subject to complex physics and great uncertainty, and consequences can range from minimal to devastating, depending upon the size of the NEO and location of impact. Deflecting a potential NEO impactor would be complex and expensive, and inter-agency and international cooperation would be necessary. Such deflection campaigns may be risky in themselves, and mission failure may result in unintended consequences.The benefits, risks, and costs of different potential NEO risk management strategies have not been compared in a systematic fashion. We present a decision analysis framework addressing this hazard. Decision analysis is the science of informing difficult decisions. It is inherently multi-disciplinary, especially with regard to managing catastrophic risks. Note that risk analysis clarifies the nature and magnitude of risks, whereas decision analysis guides rational risk management. Decision analysis can be used to inform strategic, policy, or resource allocation decisions.First, a problem is defined, including the decision situation and context. Second, objectives are defined, based upon what the different decision-makers and stakeholders (i.e., participants in the decision) value as important. Third, quantitative measures or scales for the objectives are determined. Fourth, alternative choices or strategies are defined. Fifth, the problem is then quantitatively modeled, including probabilistic risk analysis, and the alternatives are ranked in terms of how well they satisfy the objectives. Sixth, sensitivity analyses are performed in order to examine the impact of uncertainties. Finally, the need for further analysis, data collection, or refinement is determined.The first steps of defining the problem and the objectives are critical to constructing an informative decision analysis. Such steps must be undertaken with participation from experts, decision-makers, and stakeholders (defined here as "decision participants"). The basic problem here can be framed as: “What is the best strategy to manage risk associated with NEOs?” Some high-level objectives might be to minimize: mortality and injuries, damage to critical infrastructure (e.g., power, communications and food distribution), ecosystem damage, property damage, ungrounded media and public speculation, resources expended, and overall cost. Another valuable objective would be to maximize inter-agency/government coordination.Some of these objectives (e.g., “minimize mortality”) are readily quantified (e.g., deaths and injuries averted). Others are less so (e.g., “maximize inter-agency/government coordination”), but these can be scaled. Objectives may be inversely related: e.g., a strategy that minimizes mortality may cost more. They are also unlikely to be weighted equally. Defining objectives and assessing their relative weight and interactions requires early engagement with decision participants.High-level decisions include whether to deflect a NEO, when to deflect, what is the best alternative for deflection/destruction, and disaster management strategies if an impact occurs. Important influences include, for example: NEO characteristics (orbital characteristics, diameter, mass, spin and composition), impact probability and location, interval between discovery and projected impact date, interval between discovery and deflection target date, costs of information collection, costs and technological feasibility of deflection alternatives, risks of deflection campaigns, requirements for inter-agency and international cooperation, and timing of informing the public.The analytical aspects of decision analysis center on estimation of the expected value (i.e. utility) of different alternatives. The expected value of an alternative is a function of the probability-weighted consequences, estimated using Bayesian calculations in a decision tree or influence diagram model. The result is a set of expected-value estimates for all alternatives evaluated that enables a ranking; the higher the expected value, the more preferred the alternative. A common way to include resource limitations is by framing the decision analysis in the context of economics (e.g., cost-effectiveness analysis).An important aspect of decision analysis in the NEO risk management case is the ability, known as sensitivity analysis, to examine the effect of parameter uncertainty upon decisions. The simplest way to evaluate uncertainty associated with the information used in a decision analysis is to adjust the input values one at a time (or simultaneously) to examine how the results change. Monte Carlo simulations can be used to adjust the inputs over ranges or distributions of values; statistical means then are used to determine the most influential variables. These techniques yield a measure known as the expected value of imperfect information. This value is highly informative, because it allows the decision-maker with imperfect information to evaluate the impact of using experiments, tests, or data collection (e.g. Earth-based observations, space-based remote sensing, etc.) to refine judgments; and indeed to estimate how much should be spent to reduce uncertainty.

Computing rank dependent utility in graphical models for sequential decision problems

This paper is devoted to automated sequential decision in AI. More precisely, we focus here on the Rank Dependent Utility (RDU) model. This model is able to encompass rational decision behaviors that the Expected Utility model cannot accommodate. However, the non-linearity of RDU makes it difficult to compute an RDU-optimal strategy in sequential decision problems. This has considerably slowed the use of RDU in operational contexts. In this paper, we are interested in providing new algorithmic solutions to compute an RDU-optimal strategy in graphical models. Specifically, we present algorithms for solving decision tree models and influence diagram models of sequential decision problems. For decision tree models, we propose a mixed integer programming formulation that is valid for a subclass of RDU models (corresponding to risk seeking behaviors). This formulation reduces to a linear program when mixed strategies are considered. In the general case (i.e., when there is no particular assumption on the parameters of RDU), we propose a branch and bound procedure to compute an RDU-optimal strategy among the pure ones. After highlighting the difficulties induced by the use of RDU in influence diagram models, we show how this latter procedure can be extended to optimize RDU in an influence diagram. Finally, we provide empirical evaluations of all the presented algorithms.

Efficiency of influence diagram models with continuous decision variables

A measure of efficiency for influence diagram models with continuous decision variables is presented in order to evaluate whether the additional computational complexity required by a more accurate model is justified. The efficiency measure is a multi-objective utility function that considers both the accuracy and complexity of the ID model. Accuracy is determined as the mean squared error between influence diagram decision rules and an analytical solution. Complexity is assessed by tracking the run time required to obtain the solution. The resulting efficiency score considers the preferences of an individual decision maker for accuracy and complexity. Three influence diagram models are compared using the efficiency measurement, and an iterative solution procedure is introduced to improve model performance.

Influence Diagrams for Capacity Planning and Pricing under Uncertainty

ABSTRACT: Influence diagrams are presented as a complement to analytical methods for solving capacity-planning and pricing problems under uncertainty. An influence diagram model that permits continuous decision variables is used to determine optimal decision rules for a single product monopolist making capacity and price selections once per period. Solutions under different scenarios based on the timing of actual demand observations and the ability to adjust initial capacity are compared to previous solutions in the literature. The results demonstrate that optimal decision rules for capacity and price determined using influence diagrams closely approximate those found using analytical approaches. The influence diagram is used to determine the effect of changing demand function parameters on the optimal solutions. Additionally, the influence diagram model can be adapted to solve capacity-planning problems that do not meet the assumptions required to obtain analytical solutions or heuristics in previous studies.

A failure analysis and remnant life assessment of boiler evaporator tubes in two 250 MW boilers

It has been established from (a) the tube thermal flux calculations, (b) the failure surface fractographic features, (c) the data sheets on evaporator tube temperatures from both Mitsubishi and Seimens, (d) the reported data from the literature and (e) to a certain extent, the tube strain gauge data that the recent waterwall tube failures in two 250 MW boilers were the result of high stresses being active at certain boiler locations. At start-up these stresses were estimated from fractographic features to be around 370–420 MPa in magnitude, which resided between the tube materials yield and ultimate tensile strengths. The defect extension process in the present waterwall tube failures was shown to be fatigue crack growth, which is the most common failure mode in waterwall boiler tubes. By adopting two separate approaches, viz., an “upper bound” ASME XI design line and an influence diagram model approach which was designed specifically for waterwall tubes, detailed remnant life values could be estimated for a number of different situations. In the case of new replacement tubes, both approaches estimated a remnant life of about 9 years. Note that the failure probability of the upper bound ASME XI predictions was demonstrated to be around 10 −5. Also the influence diagram model approach was extremely helpful in identifying the probable root cause of corrosion fatigue failures in waterwall boiler tubes. With respect to the remnant life estimates two separate cases were considered, viz., (1) tubes with defect populations as recorded during the recent NDT study and (2) new replacement evaporator tubes which contained zero defects.

An Influence Diagram Extension of the Unified Partial Method for Common Cause Failures

Modelling Common Cause Failures (CCFs) is an essential part of risk analyses, especially for systems such as nuclear power plants, which are required to have high reliability. The Unified Partial Method (UPM) is the main approach of the UK for modelling CCFs. This paper presents an Influence Diagram model for CCFs which extends UPM and represents uncertainty on system performance. This allows more detailed modelling of CCFs in terms of root causes and coupling factors, creates a context for using information in the industry database, and captures the non-linearity in the way system defences influence reliability. A structured expert elicitation process is used to construct the Influence Diagram model and to identify the non-linear structure of the domain, using an example of Emergency Diesel Generators (EDGs) from nuclear power plants. Insights and experiences from the elicitation process are described.

Software Review of Analytica 1.2

Analytica is an easy-to-learn, easy-to-use modeling tool that allows modelers to represent what they know through influence diagrams. These diagrams show which model quantities are derived from which others and indicate by shape and color the roles that different nodes play in the model, e.g., decision variables, chance variables, outcome variables, deterministic functions, or abstractions of sub-models. A wide variety of built-in probability distributions allow uncertainties about input values to be painlessly specified and propagated through the model via a fast, professional Monte-Carlo simulation engine. Resulting uncertainties and sensitivities about any quantity in the model can be viewed with admirable ease and flexibility by selecting among probability density, cumulative distribution, confidence band, sensitivity analysis, and other displays. Analytica features clever hierarchical model management and navigation features that serious model-builders will appreciate and that novice modelers will learn from as they are led to develop well-structured, well-documented models. Simple continuous (compartmental-flow) and Markov chain dynamic simulation models can be built by paying some detailed attention to arrays and indices, although Analytica does not support true discrete-event simulation. Within its chosen domain—uncertainty propagation through influence diagram models—Analytica is by far the easiest and best tool that we have seen.

UNCERTAINTIES OF ASPHALT LAYER THICKNESS DETERMINATION IN FLEXIBLE PAVEMENTS-INFLUENCE DIAGRAM APPROACH

The thickness of pavement layers is an important parameter used in Pavement Management Systems (PMS). Thickness data are used for pavement condition assessment, performance predictions, selection of maintenance strategies and rehabilitation treatments, basic quality assessment, and as input to overlay thickness design. Pavement thickness is usually determined from direct testing such core samples, nondestructive testing such as radar, or historical records such as pavements network database. This paper proposes the use of Bayesian Influence Diagrams as a tool in providing a probabilistic model for thickness determination procedure in flexible pavements. The Bayesian Influence Diagram Model is presented as a framework for addressing uncertainties involved in capturing quantitative and qualitative information in the asphalt layer thickness determination procedures. The model is also used to perform value of information analysis in the determination of pavement layer thickness. The Influence Diagram representation facilitates the assessment of coherent prior distributions and makes it easier for knowledge engineers and other decision makers to express and understand more general kinds of dependency and independency assumptions.

Neural network-based decision class analysis for building topological-level influence diagram

In order to reduce the burden of modeling decision problems, the concept of decision class analysis (DCA) was proposed. DCA treats a set of decision problems having some degree of similarity as a single unit. This paper presents a scheme within which a neural network is used to implement DCA. An influence diagram model is employed to represent the decision problem, since the diagram is a good tool for knowledge representation of complex decision problems under uncertainty. DCA under consideration is viewed as a classification problem where a set of input–output data pairs is given. We thus utilize a feed-forward neural network with a supervised learning procedure so as to develop DCA and then to generate an influence diagram in the topological level. This paper also presents the results of the neural net simulation with an example of a class of decision problems.

A Methodology for Evaluating Military Systems in a Counterproliferation Role

This paper illustrates a methodology to evaluate how dissimilar military systems support the accomplishment of the United States' counterproliferation objectives. The key questions in evaluating counterproliferation systems are identified. By using decision analysis, an influence diagram model is developed that represents military activities in the counterproliferation process. A value model is developed that enables systems to be evaluated against common criteria. An analysis of intelligence, defensive, and offensive counterproliferation systems suggests that intelligence system improvements may provide the greatest potential to meet the United States' counterproliferation objectives. Sensitivity analysis is conducted to determine which factors in the model are most important. To demonstrate the model, nine systems from the Air Force Vulcan's Forge 1995 wargame are evaluated. This paper illustrates the value of decision analysis, and influence diagrams in particular, by involving decision makers and subject matter experts in structuring complex problems for analysis.

Modeling a class of decision problems using artificial neural networks

This paper presents an artificial neural network to build a decision model, together with a discussion about implementation of decision class analysis. In contrast to evaluating or analyzing decision problems, there has been little research to build decision models such as the influence diagram. In practice, generating an influence diagram requires much time and effort. Furthermore, the resulting model can be generally applicable to only a specific decision problem. In order to reduce the burden of modeling decision problems, the concept of decision class analysis (DCA) is proposed. DCA treats a set of decision problems having some degree of similarity as a single unit. This paper presents a scheme within which a neural network is used to implement DCA, i.e. to model similar decision problems within the same class. An influence diagram model is used to represent the decision problem. It is a good tool for knowledge representation of complex decision problems under uncertainty. After the influence diagram is briefly described and the concept of DCA is introduced, we propose a method for developing influence diagrams using a feedforward neural net. We also present the results of neural net simulation with an example of a class of decision problems.

Optimization of decision networks in structured task environments

This paper considers the problem of determining the optimal distributed decision strategy for a team of decision-makers (DMs) arranged in an arbitrary acyclic organizational structure and controlling a complex structured process. Each DM has access to uncertain and partial information about the task environment and can control only a portion of it. We present an influence diagram model of the joint task-organization system and formulate the optimal team strategy in terms of a set of coupled hypothesis testing tasks at DMs. Specifically, we show that the scope of a local decision task is determined by the interaction of the task structure (what can be measured) and of the information access structure of the organization (who can measure what); while the control structure (who can influence what event) has an impact on the locally perceived costs associated with the decision options available. Theoretical results are illustrated via a numerical example, and connections to existing decision models are discussed.

R&D Decision Advisor: An interactive approach to normative decision system model construction

This paper describes the architecture of R&D Decision Advisor ™, a commercial intelligent decision system for evaluating corporate research and development projects and portfolios. In analyzing projects, R & D Decision Advisor interactively guides a user in constructing an influence diagram model for an individual research project. The system's interactive approach can be clearly explained from a blackboard perspective. The system guides the user to choose among general project features but offers flexibility to capture unique project details. Related research generalizes the underlying architecture and addresses resource-constrained decision situations.

Decision-Analytic Methodology for Cost-Benefit Evaluation of Diagnostic Testers

Abstract A modular influence diagram model is proposed as a decision-analytic framework for reasoning about diagnostic testing in the manufacture of mechanical products. The influence diagram ties product design, manufacturing, and testing decisions to field quality, costs and risks. The decision-analytic theory of “expected value of information” is used to evaluate the cost-benefit of alternate testing systems. The structure of the model highlights research directions for engineering economics in evaluating cost-benefit tradeoffs in the product cycle from design, to manufacturing, marketing and field service. An implementation in the IDES (Influence Diagram Based Expert System) illustrates the potential of applying such a planning model to real-time diagnostic decisions in the manufacture of mechanical components for high-speed printers.

Uncertainty Methods in Expert Systems

The paper introduces the main measures of uncertainty and discusses some of the problems associated with conventional uncertainty propagation methods, such as those based on independent probabilities and the methods used with certainty factors, belief, and possibility functions. Some examples show the importance of the associated errors. Alternative methods, such as log‐linear regression and causal networks or influence diagram models, are discussed. Finally, their structural and parametric learning possibilities are analyzed.