Conclusion

For operators of industrial systems wishing to optimize the availability and dependability of their hardware investments, it is essential to monitor their behavior in order to detect degradation and predict its evolution before failure occurs. Data-driven diagnostic and prognostic methods offer possible alternatives for proposing solutions to this problem when data from sensors is only available. The diagnostic and prognostic procedure is made up of essential sequential phases: signal acquisition, information processing and validation, detection of threshold crossing followed by decision making, fault diagnosis, prognosis of degradation evolution, DEFAD estimation and strategic choice of maintenance policy. To ensure that each phase is carried out without risk of error, it is essential to implement the appropriate mathematical tools presented in this article. Data-driven diagnostic and...