What is it about?
Data analytics today mainly focuses on finding patterns in large data base and using this to predict future behavior of a system. This is not possible for new corrosion phenomena or for rare failure modes, where there is not enough prior data. This paper describes a framework to combine fundamental knowledge of corrosion with data to predict new corrosion phenomena.
Featured Image
Why is it important?
Many major industrial corrosion failures occurred because we did not anticipate new failure processes. This has been true of pipelines, nuclear power plants, chemical industries, and others. Once the failures occurred, we conducted a lot of research to understand them. Today, the success of software companies like Facebook and Google has encouraged a data analysis approach, but this is not sufficient to predict new phenomena that have not yet occurred, for which there is not a lot of data or the data needs to be linked and transformed by knowledge. This paper proposes a framework to go from a reactive mode to a proactive mode of assessment.
Perspectives
As systems age, new failure modes will crop up. In this regard, physical systems are just like living systems where aging occurs at the molecular level leading to build up of defects that then influence uncontrolled cell growth or cell repair leading to various new health problems. Just like living things, physical systems age because of various changes in the material and the interface between materials and the environment. Monitoring external data like corrosion rate is like monitoring the temperature of a patient - hardly enough to say something about the fate of that patient. We need a framework to know the totality of the history of a system and understand how all these forces can combine to produce a particular failure mode.
narasi sridhar
DNV GL AS
Read the Original
This page is a summary of: 2017 Frank Newman Speller Award: Knowledge-Based Predictive Analytics in Corrosion, CORROSION, February 2018, NACE International,
DOI: 10.5006/2635.
You can read the full text:
Contributors
The following have contributed to this page
