Specific degradation modes like corrosion shorten the service life of equipment. However, using existing corrosion modeling approaches is difficult because the necessary variables are rarely collected, and traditional models do not readily elucidate service life impacts. There is a need for using more widely available data to measure the effects of such degradation mechanisms on the service life of facility components. This insight can improve the ability to plan repairs or develop effective mitigation programs.
Traditional corrosion models require features like time-of-wetness or ion concentration that are difficult to collect and associate with in situ equipment. This study instead utilized more widely available maintenance and inspection text data. Natural language processing techniques group the components into subpopulations of similar degradation modes. Reliability models for the subpopulations infer the impact of distress mechanisms on the service life of components.
Corrosion effects of building enclosures reduced the predictive mean time to failure by about 12 years. This result was validated with a statistically significant t-test and Bayesian analysis. Similar results were demonstrated on other component types in electrical, mechanical and structural systems. These findings help identify specific risks that may be targeted for mitigation policies without collecting degradation-specific data features.
The traditional approach requires collection of physical reliability modeling features like ion concentration or time of wetness, which are rarely collected for facility components. The developed model instead relies on more widely available maintenance text data to measure the impact of corrosion on facility equipment service life.
