Check out some of the Instrumentation and Control group’s recent research highlights below:
Lifecycle Prognostics
We feel that effective prognostic systems should seamlessly predict the remaining useful life (RUL) from beginning of component life to end of component life; we have termed this Lifecycle Prognostics. When a component is first put into service, the only information available may be past failure times of similar components or the expected distribution of failure times derived from reliability analyses of these data (Type I Prognostics). These data provide an estimated life for the average component operating under average usage conditions. As the component operates, it begins to consume its available life at a rate largely influenced by the system and environmental stresses. Information from these recorded stresses can be used to update the expected failure time distribution (Type II Prognostics). The incorporation of stressor information allows for the estimation of the RUL for an average component operating under specific usage conditions. After continued operations, measurable levels of degradation may evolve, which allows for further improvement of the failure time distribution estimate by incorporating these health indicators (Type III Prognostics). Current research typically focuses on developing methods for one or more of the three types of prognostics individually. We have developed a framework for using Bayesian methods to transition between prognostic model types and to update failure time distribution estimates as new information becomes available. Over the last decade, these methods have been developed and integrated into a MATLAB toolbox and successfully applied to several target applications.
Sharp, M., Coble, J., Nam, A., Hines, J., and B. Upadhyaya, “Lifecycle Prognostics: Transitioning Between Information Types”, Proc IMechE Part O: J Risk and Reliability, November, 2014.
Enhanced Risk Monitors with Integrated Equipment Condition Assessment:
Enhanced risk assessment of critical active components will improve asset protection and management, allowing for safe, reliable generation during extending operating cycles and longer reactor lifetimes. Incorporation of dynamic health assessment of key active components in risk monitors will also support operational economics by (1) informing operations and maintenance decisions to optimize maintenance activities, (2) optimizing asset performance, and (3) supporting mission success by ensuring reliable component operation. Recent research has investigated methods to integrate real-time estimates of failure probability and event probability in dynamic probabilistic risk assessment (PRA), including (i) developing a framework for integrating equipment condition assessment into event probability estimation as new conditions arise, (ii) testing the framework with simulated data, (iii) constructing the information flow and statistical updating mechanisms of the predictive risk monitor, (iv) developing prototype predictive risk monitor tools, and (v) performing case studies for selected assets.
Coble J., Coles G., Meyer R., Ramuhalli P., Incorporating equipment condition assessment in risk monitors for advanced small modular reactors, Chemical Engineering Transactions, 33, 913-918, 2013.
Regularization Methods for Improved Monitoring and Diagnostics:
Most data-based predictive modeling techniques have an inherent weakness in that they may give unstable or inconsistent results when the predictor data is highly correlated. Predictive modeling problems of this design are usually under constrained and are termed ill-posed. We have developed regularization methods necessary for producing accurate and consistent prediction results of ill-posed surveillance and diagnostic problems. The applications include plant-wide sensor calibration monitoring and the inferential sensing of nuclear power plant feedwater flow using the following techniques: neural networks, non-linear partial least squares techniques, linear regularization techniques implementing ridge regression, and informational complexity measures.
Gribok, A., J.W. Hines, A. Urmanov and R.E. Uhrig, “Heuristic, Systematic, and Informational Regularization for Process Monitoring”, International Journal of Intelligent Systems, special issue on Intelligent Systems for Process Monitoring, Vol. 17 No. 8, pp. 723-750, Wiley Publishers, 2002.
Prognostic Uncertainty Analysis and Reduction:
Confidence in the accuracy and uncertainty of RUL estimates generated from data driven models has been a focus in reliability analysis across many industries as the use of these types of models becomes more widespread. However, one of the major drawbacks of prognostic modeling is that often there is a lack of past information about the system available to arrive at reasonable RUL estimates with low error and uncertainty values. Typically, this information has been obtained from physical models and simulations of the component or system, but real-world behavior and uncertainty cannot be accurately quantified using only these data. Our research developed a new Bayesian statistical updating method for use during prognostic modeling, specifically a degradation path model. The developed method uses the RUL estimates and statistical values from a Type I model as additional prior information in the degradation path model inputs. The algorithm can also use the operating or stressor conditions of the system as prior information. The research demonstrated that the inclusion of the Type I prior information in the degradation path model can reduce the RUL prediction error.
Nam, A., B.A. Jefferies, J.W. Hines, B.R. Upadhyaya, “A Bayesian Statistical Updating Method: Error Reduction in Remaining Useful Life Estimation”, 2015 American Nuclear Society Nuclear Plant Instrumentation and Control & Human Machine Interface Technologies, Charlotte, NC, 2015.
Sensor Calibration Verification
Each year industry needlessly spends vital resources to calibrate safety critical sensors. The Electric Power Research Institute has shown that only 3-5% of these sensors in the power industry require recalibration. Performing invasive maintenance when it is not required reduces sensor reliability and may damage the instrumentation. We have developed data-based modeling techniques to identify when sensors require recalibration. Because uncertainty techniques have been developed to provide confidence in decisions, these techniques have been implemented in the nuclear industry for safety critical sensors. A three volume set of NUREGs were written for the Nuclear Regulatory Commission, and the techniques have been implemented in other industries as well.
- NUREG/CR-6895, Vol. 1, Hines, J.W., and R. Seibert, “Technical Review of On-Line Monitoring Techniques for Performance Assessment, Volume 1: State-of-the-Art,” U.S. Nuclear Regulatory Commission, Washington, D.C., 2006.
- NUREG/CR-6895, Vol. 2, Hines, J.W.,D. Garvey, R. Seibert, and A. Usynin ., “Technical Review of On-Line Monitoring Techniques for Performance Assessment, Volume 2: Theoretical Issues,” U.S. Nuclear Regulatory Commission, Washington, D.C., 2007.
- NUREG/CR-6895, Vol. 3, Hines, J.W., J. Garvey, D. Garvey, and R. Seibert, “Technical Review of On-Line Monitoring Techniques for Performance Assessment, Volume 3: Limiting Case Studies,” U.S. Nuclear Regulatory Commission, Washington, D.C., 2008
Coupling of Asset Data and Computerized Maintenance Management Systems for Prognostic Data Mining
In most complex industrial settings, large amounts of process data are collected from high value assets and recorded periodically. As parts begin to degrade and components fail, maintenance personnel are responsible for making repairs and recording these repairs in a computerized maintenance management system (CMMS). Unique prognostic methods can be used to increase knowledge about the degradation of the system and provide accurate predictions of RUL before the system reaches an undesirable level of operation. These lifecycle prognostic models require important data that are stored within the CMMS and plant computer. Recent research centers on a concept that uses data mining based on Big Data efforts in order to couple the plant computer data with the CMMS so that prognostic information can be gathered and analyzed. Once the plant computer and CMMS are coupled, important failure data can be mined and used to update or build prognostic models based on component failure times, stressor information, and signal and monitoring residual values. An effective implementation of this concept means that the results can be used as a priori prognostic information in lifecycle prognostic models.
Welz, Z., J.W. Hines, and B.R. Upadhyaya, “Investigation into the Coupling of a Nuclear Plant Computer to a Computerized Maintenance Management System for Prognostic Data Mining. 2015 Prognostic and Health Management Conference.