Hybrid feedforward neural network for pressure vessel internal corrosion prediction: integrating chemical models with inspection data for structural integrity assessment

This study presents a hybrid framework integrating a physics-based corrosion model with a feedforward neural network (FNN) to predict corrosion rates and estimate the remaining useful life (RUL) of industrial pressure vessels for condition-based maintenance. Using non-destructive evaluation (NDE) wall thickness measurements from 24 inspection points over multiple years (2002–2008) and physics-based training data, a three-layer FNN with Monte Carlo dropout predicts localized corrosion rates, while exponential and linear degradation models project future wall thickness. The FNN achieved a coefficient of determination (R²) of 0.975 for corrosion rate prediction and a mean absolute error (MAE) of 0.1204 mm/year. For thickness prediction, the exponential model achieved R² = 0.99 with MAE = 0.0389 mm, outperforming the linear model (MAE) = 0.1350 mm. The framework was integrated with Fitness-for-Service (FFS) assessment based on API 579-1/ASME FFS-1 standards, enabling classification of vessel components and identification of sections requiring maintenance. This hybrid approach translates predictive analytics into standards-compliant engineering decisions for structural integrity management.