Reliability modeling and forecasting with two-stage stochastic degradation models: bridge reinforced concrete rebar example
Abstract
Stochastic degradation models offer distinct advantages compared to traditional reliability analyses. Many degradation processes occur in two distinct stages or phases. This is not unusual or unexpected because (a) often the design includes protective coverings or coatings that beneficially delay the onset of a degradation process, (b) some physical or chemical processes naturally do not occur until there has been a time-delay of some type, or (c) after progressing gradually for some time, a degradation process may become more aggressive and fundamentally change.
There have already been several useful two-stage degradation models in the literature including the gamma-gamma and Weiner-Weiner models and others. In this talk, we introduce a new two-stage degradation model called the Weibull-gamma model. The first stage is modeled by a time-to-event distribution, instead of a stochastic process, to more closely resemble the physical behavior of some delayed failure mechanisms, while the second stage is a stochastic process similar to other models. In the new model, both stages are modeled as a function of stress covariates, which can then be used to conduct accelerated testing. The model is demonstrated using an example of a steel rebar used in reinforced concrete to build more reliable bridges. Testing data was collected for three years under accelerated conditions to evaluate the feasibility and reliability of new steel materials exposed to more stressful conditions. While the new model to useful to predict reliability, it is not particularly useful to forecast future degradation. In the final part of the talk, various machine learning forecasting methods are discussed and demonstrated.
Short bio
David Coit is a Professor in the Department of Industrial & Systems Engineering at Rutgers University, Piscataway, NJ, USA. He has also had visiting professor positions at CentraleSupelec Université Paris-Saclay, Paris, France, Tsinghua University, Beijing, China, and others. His current teaching and research involves system reliability modeling and optimization, and energy systems optimization. He has over 140 published journal papers and over 100 peer-reviewed conference papers (h-index 65). He is currently an Associate Editor for IEEE Transactions on Reliability and Journal of Risk and Reliability and for 15 years was a Department Editor for IISE Transactions. He has been a special issue editor for IISE Transactions, Reliability Engineering & System Safety (RESS) and Computers & Industrial Engineering. His research has been funded by USA National Science Foundation (NSF), U.S. Army, U.S. Navy, industry, and power utilities. His NSF grants included a CAREER grant to develop new reliability optimization algorithms considering uncertainty. He was also the recipient of the P. K. McElroy award, Alain O. Plait award and William A. J. Golomski award for best papers and tutorials at the Reliability and Maintainability Symposium (RAMS). Prof. Coit received a BS degree in mechanical engineering from Cornell University, an MBA from Rensselaer Polytechnic Institute (RPI), and MS and PhD in industrial engineering from the University of Pittsburgh. He is a fellow of Institute of Industrial & Systems Engineers (IISE).
Sandwiches are offered at the end of the seminar.
Stochastic degradation models offer distinct advantages compared to traditional reliability analyses. Many degradation processes occur in two distinct stages or phases. This is not unusual or unexpected because (a) often the design includes protective coverings or coatings that beneficially delay the onset of a degradation process, (b) some physical or chemical processes naturally do not occur until there has been a time-delay of some type, or (c) after progressing gradually for some time, a degradation process may become more aggressive and fundamentally change.
There have already been several useful two-stage degradation models in the literature including the gamma-gamma and Weiner-Weiner models and others. In this talk, we introduce a new two-stage degradation model called the Weibull-gamma model. The first stage is modeled by a time-to-event distribution, instead of a stochastic process, to more closely resemble the physical behavior of some delayed failure mechanisms, while the second stage is a stochastic process similar to other models. In the new model, both stages are modeled as a function of stress covariates, which can then be used to conduct accelerated testing. The model is demonstrated using an example of a steel rebar used in reinforced concrete to build more reliable bridges. Testing data was collected for three years under accelerated conditions to evaluate the feasibility and reliability of new steel materials exposed to more stressful conditions. While the new model to useful to predict reliability, it is not particularly useful to forecast future degradation. In the final part of the talk, various machine learning forecasting methods are discussed and demonstrated.
Short bio
David Coit is a Professor in the Department of Industrial & Systems Engineering at Rutgers University, Piscataway, NJ, USA. He has also had visiting professor positions at CentraleSupelec Université Paris-Saclay, Paris, France, Tsinghua University, Beijing, China, and others. His current teaching and research involves system reliability modeling and optimization, and energy systems optimization. He has over 140 published journal papers and over 100 peer-reviewed conference papers (h-index 65). He is currently an Associate Editor for IEEE Transactions on Reliability and Journal of Risk and Reliability and for 15 years was a Department Editor for IISE Transactions. He has been a special issue editor for IISE Transactions, Reliability Engineering & System Safety (RESS) and Computers & Industrial Engineering. His research has been funded by USA National Science Foundation (NSF), U.S. Army, U.S. Navy, industry, and power utilities. His NSF grants included a CAREER grant to develop new reliability optimization algorithms considering uncertainty. He was also the recipient of the P. K. McElroy award, Alain O. Plait award and William A. J. Golomski award for best papers and tutorials at the Reliability and Maintainability Symposium (RAMS). Prof. Coit received a BS degree in mechanical engineering from Cornell University, an MBA from Rensselaer Polytechnic Institute (RPI), and MS and PhD in industrial engineering from the University of Pittsburgh. He is a fellow of Institute of Industrial & Systems Engineers (IISE).
Sandwiches are offered at the end of the seminar.
Practical information
- Informed public
- Free
Organizer
- Prof. Olga Fink (IMOS), Prof. Alexandre Alahi (VITA), Prof. Dusan Licina (HOBEL), Prof. Alain Nussbaumer (RESSLab)
Contact
- Olga Fink