Corrosion of bolted joints – analysis and smart monitoring

Abstract

This study examines corrosion of flange faces in bolted joints with and without gaskets used in industrial applications like pressure vessels and piping, wind turbines, and bridges. A test rig based on a bolted joint fixture were developed to simulate real-world conditions, allowing continuous monitoring of factors influencing corrosion, such as temperature, fluid flow, pressure, conductivity, pH, current, and potential. Electrochemical techniques, including potentiostatic and potentiodynamic polarization, electrochemical impedance spectroscopy, and electrochemical noise measurement, were used to characterize corrosion. The results show that graphite gaskets increase the susceptibility to localized corrosion on ASTM A182 F321 stainless steel flange surfaces under the same working conditions, compared to graphite gaskets with metal foil inserts and virgin polytetrafluoroethylene (PTFE) gaskets. The mechanism of flange face corrosion is that, for PTFE gaskets, corrosion propagation mainly occurs at the gasket inner diameter and propagates through the depth of the flange while, for graphite gaskets, corrosion occurs on the whole contact surface of the flange and the gasket. This study also examined the effect of gap dimensions between the gasket and flanges, which influence crevice thickness and depth, on corrosion behavior. It was found that increasing the gap thickness from 1.58 to 6.35 mm raises the general corrosion rate of the flange surface from 0.09 to 1.03 mm. y−1, and crevice corrosion initiation time increases from 0.23 to 3.12 h. Machine learning algorithms, including decision tree, support vector machine, random forest (RF), and bagging classifier, were compared for corrosion behavior prediction. Among these algorithms, bagging classifier achieved the highest accuracy of 94.4%. Additionally, recurrent neural networks, particularly long short-term memory (LSTM) models, were used to classify corrosion types using electrochemical noise data, achieving 93.62% accuracy, with a hybrid RF-LSTM approach reaching 97.85%. An unsupervised LSTM model with principal component analysis and k-means clustering offers potential for real-time corrosion monitoring. This work advances the field by demonstrating the effectiveness of machine learning and deep learning models in predicting and classifying corrosion behavior, offering a more efficient alternative to traditional corrosion testing. The insights gained on material selection, gasket design, and gap dimensions contribute to safer and more reliable flange designs in industrial applications, potentially reducing costs associated with maintenance and unexpected failures. These findings pave the way for real-time monitoring systems that can proactively manage corrosion risks, improving overall safety and operational continuity.

Date	14 Jan 2025
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Lucas Hof (Supervisor) & Hakim A. Bouzid (Co-supervisor)