Performance Comparison Between a Statistical Model, a Deterministic Model, and an Artificial Neural Network Model for...

Area of Science:

• Materials Science
• Corrosion Engineering
• Computational Modeling

Background:

• Pitting corrosion poses a significant threat to the integrity of metals and alloys.
• Accurate prediction of pitting damage is crucial for the safety and longevity of engineering structures.
• Existing models for predicting pitting corrosion fall into statistical and deterministic categories.

Purpose of the Study:

• To compare the effectiveness of statistical, deterministic, and artificial neural network (ANN) models in predicting pitting corrosion.
• To identify the advantages and disadvantages of each modeling approach.
• To guide the selection of reliable methods for future algorithms predicting pitting damage.

Main Methods:

• Comparison of three distinct modeling approaches: extreme value statistics (empirical), deterministic (mechanism-based), and Artificial Neural Networks (ANNs).
• Utilized a laboratory-collected dataset of pitting corrosion to evaluate and contrast the models.
• Analysis focused on the predictive capabilities for cumulative pitting damage progression.

Main Results:

• The study illustrates the challenges in accurately predicting cumulative pitting damage.
• Each approach (statistical, deterministic, ANNs) demonstrates unique strengths and limitations in modeling pitting corrosion.
• The comparison provides insights into the reliability of different methods for predicting pitting damage functions.

Conclusions:

• No single model universally outperforms others; the choice depends on specific application requirements.
• Understanding the trade-offs between empirical, mechanistic, and data-driven approaches is key for developing robust predictive algorithms.
• Further research should focus on integrating the best aspects of these models for enhanced pitting corrosion prediction in engineering structures.