- author = "Yadukrishnan Satheesan and Til Baalisampang and Hamid Ahmadi and Ehsan Arzaghi and Vikram Garaniya and Rouzbeh Abbassi",
- title = "A Clustering-Informed {AutoML} Framework for Pitting Corrosion Damage Classification for Offshore Structures",
- booktitle = "World Renewable Energy Congress",
- year = "2026",
- editor = "Tania Urmee",
- pages = "Paper No. 83",
- address = "Perth, Western Australia",
- month = "8--11 " # feb,
- publisher = "SSRN",
- keywords = "genetic algorithms, genetic programming, TPOT, stainless steel, AISI 316L, Pitting Corrosion, Machine Learning, Offshore Renewable Energy Assets",
-
URL = "
https://hdl.handle.net/102.100.100/734007",
-
URL = "https://figshare.utas.edu.au/articles/conference_contribution/A_Clustering-Informed_AutoML_Framework_for_Pitting_Corrosion_Damage_Classification_for_Offshore_Structures/31840084?file=63035872",
-
URL = "https://figshare.utas.edu.au/ndownloader/files/63035872/1",
-
URL = "
https://www.wrec2026.com/Abstract-Book-as-at-070226.pdf",
-
URL = "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6871218",
-
DOI = "
10.2139/ssrn.6871218",
- size = "10 pages",
- abstract = "Austenitic stainless steels such as AISI 316L are widely used in offshore renewable energy structures due to their favourable strength-to-weight ratio and corrosion resistance. However, prolonged exposure to chloride-rich marine environments promotes localised pitting corrosion, which significantly influences structural integrity and fatigue life. The stochastic nature of pit morphology limits the reliability of conventional corrosion severity classification methods based on deterministic thresholds or expert judgment. This study proposes an automated machine learning (AutoML) framework for objective corrosion severity classification of AISI 316L based on pit morphology. Key morphological features, including pit depth, metal loss, and pit diameter obtained from optical profilometry, were used as model inputs. Unsupervised 𝑘-means clustering was first applied to identify intrinsic groupings in the morphology space and to define three physically interpretable severity classes: mild, moderate, and severe. These cluster-derived labels were then used to train a supervised multiclass classifier using the Tree-based Pipeline Optimisation Tool (TPOT) AutoML framework. The optimal AutoML pipeline achieved a classification accuracy of 87.5 percent, with balanced precision and recall across all severity classes. The results demonstrate that the proposed data-driven methodology reduces subjectivity in corrosion severity assessment and provides a robust and reproducible framework for corrosion evaluation in offshore renewable energy infrastructure",
-
notes = "WREC 2026 https://www.wrec2026.com/abstract.php
To be published after the conference
Australian Maritime College, College of Sciences and Engineering, University of Tasmania, Launceston, Australia",
