Session: HT-01-01 Design, Analysis and Life Prediction of High-Pressure Vessels and Equipment

Paper Number: 154566

154566 - Deep Learning Applied to Fracture Mechanics in Boundary Pressure Vessel and Support 

Abstract: 

This comprehensive study introduces a groundbreaking approach to applying deep learning algorithms in industrial applications, specifically within the realm of fracture mechanics. The primary aim is to significantly enhance the efficiency of sensitivity analysis for structural systems, prioritizing accuracy and secure decision-making processes. The research's core is centered on training sophisticated neural networks to accurately predict the evolution of Stress Intensity Factors (SIF) in pressure vessels under the influence of time-dependent thermal and mechanical loads.

 

To assemble the necessary datasets for this endeavor, traditional fracture mechanics calculations are conducted in line with the RCC-M code, as referenced in [5]. One of the key challenges addressed in this study is the irregularity of time steps and the varying durations of loads applied to the system. Addressing this complexity requires extensive data preprocessing and the development of a custom network architecture, tailored to handle these irregularities efficiently.

 

In the quest to achieve optimal performance, the study introduces a novel model that integrates convolutional and recurrent neural network layers. This hybrid architecture demonstrates superior predictive capabilities, marking a significant advancement in the field. The decision to utilize neural networks is further justified by the development of a model that accounts for geometric variabilities. This enhancement significantly improves the network's adaptability and versatility, making it more suited for real-world industrial applications.

 

Further adding to the model's robustness, the study incorporates insights from intricate 3D finite element simulations into the training process of the final neural network. This integration enriches the network's ability to make nuanced predictions, capturing the complexities of fracture mechanics in a more comprehensive manner.

 

The study also delves into the implications of these findings for the industry. The integration of deep learning into fracture mechanics can revolutionize the way structural assessments are conducted, offering a more rapid yet equally reliable alternative to conventional methods. This advancement not only promises to improve the safety and durability of industrial structures but also has the potential to streamline the design and maintenance processes, leading to cost-effective and time-efficient practices.

 

In conclusion, this research paves the way for a new era in industrial applications of fracture mechanics, leveraging the power of deep learning to offer transformative solutions. The integration of sophisticated neural networks, capable of handling complex data sets and simulating real-world scenarios, marks a significant milestone in the field. The findings of this study have far-reaching implications, promising to enhance the safety, efficiency, and reliability of structural systems across various industries.

Presenting Author: Abdelhak Benrabia Framatome

Presenting Author Biography: Professional Biography
I am an experienced engineer with a diverse background in mechanical engineering, data science and currently focused on leading innovative machine learning projects in the nuclear industry. With over seven years of experience at Framatome, a renowned company in the nuclear sector, I have honed my skills in both the technical aspects of engineering.
Mechanical Engineer (October 2017 - Present)
In my role as a Mechanical Engineer in the steam generator department, my focus has been on the mechanical design and justification of primary/secondary components of pressurized water reactors in the French nuclear fleet and international EPR (Evolutionary Power Reactor) projects. My responsibilities include equipment sizing through finite element and analytical calculations, conducting mechanical justification studies (elastic, non-linear, thermo-mechanical), and engaging in discussions with customers, safety authorities, and other stakeholders. I also handle fabrication anomalies and provide technical consultation, drawing upon my expertise in data science and programming in Python.

Authors:

Abdelhak Benrabia Framatome
Pascal Duranton Framatome
Olivier Vernhet Framatome
Amro El Betepasawy Framatome
Remi Bessonies Framatome