Zhou, S; Coleman, D; Turnbull, A (2001) Application of neural networks to predict sulphide stress corrosion cracking of duplex stainless steels. NPL Report. MATC(A)21

Text MATC21.pdf Download (291kB)

Abstract

The development of the “Corrosion Database of Duplex Stainless Steels” was a significant advance in encouraging the safe use of these materials. However, as corrosion is affected by many variables such as temperature, pressure of H2S, yield strength of material, there are many conditions for which specific data are not available and prediction of susceptibility using traditional statistical methods is not possible. Neural networks have been used to achieve such predictions.
Algorithms have been developed using multi-layer feed-forward neural networks to predict the susceptibility to sulphide stress corrosion cracking of both standard 22 Cr and high alloy duplex stainless steels. Kohonen neural networks have also been created to aid the development of these algorithms. The latter also serve as a search engine enabling the nearest neighbours that exist in the “Corrosion Database of Duplex Stainless Steels” to be accessed. When used together, these techniques have the potential to provide a fully functional software package capable of accessing existing information and prediction based on the database data.
Finally, the networks have been analysed using network weights. These indicate the relative importance of parameters such as partial pressure of H2S, NaCl content and temperature. Neural-fuzzy algorithms have been developed that “mimic” the multi-layer feed-forward networks and allow some “rules” to be constructed.

Item Type:	Report/Guide (NPL Report)
NPL Report No.:	MATC(A)21
Subjects:	Advanced Materials Advanced Materials > Corrosion
Last Modified:	02 Feb 2018 13:17
URI:	http://eprintspublications.npl.co.uk/id/eprint/1953

Actions (login required)

View Item