Session: 14-06-01: Machine Learning and Deep Learning in Safety, Reliability, and Maintenance I

Paper Number: 143109

143109 - Enhancing Tubular Structure Integrity Assessment: A Machine Learning-Integrated Failure Assessment Diagram Approach 

Structural components, particularly tubular ones, are critical in several industries, including oil and gas. These components are often exposed to harsh conditions that can lead to crack-like flaws, threatening their integrity and potentially leading to safety-critical failures. Understanding and managing these risks is crucial for the reliability and safety of the systems that depend on them. Defects in these components typically result in two main failure modes: brittle fracture and plastic collapse. For brittle materials, linear elastic fracture mechanics is used, considering factors like applied stress, defect size, and material fracture toughness. On the other hand, plastic collapse occurs when the stress across the remaining uncracked section reaches the material's flow stress, which is usually estimated by averaging the yield and ultimate tensile strengths.

Elastic-plastic fracture mechanics becomes relevant for structures that undergo significant plastic deformation before failing. A key element of this approach is the failure assessment diagram, which helps analyze the likelihood of failure in flawed structural components. This methodology involves determining a failure boundary and plotting an assessment point that reflects the current state of the component in terms of stresses and material properties. The diagram includes a cutoff value to represent a criterion for plastic collapse and offers a visual representation of the failure modes and the concept.

The integrity of a structure is evaluated using two main parameters in the failure assessment diagram. The load ratio indicates how close the structure is to plastic collapse, while the stress intensity ratio gauges the likelihood of brittle fracture. If the assessment point falls outside the acceptable range, it suggests a higher risk of structural failure.

Previous studies have used various analytical and numerical techniques to establish the critical boundary of the failure assessment diagram. The initial boundary was shaped using a strip-yield model with adjustments for plastic behavior. This was enhanced by incorporating the J-integral in elastic-plastic calculations to improve accuracy. Recent advancements include the use of machine learning to enhance failure assessment techniques, with artificial neural networks being employed for fitness-for-service evaluations and predicting the lifespan of critical systems under various conditions.

This research aims to improve the methods used in structural integrity assessments within the failure assessment diagram framework, focusing on the effects of secondary stresses on tubular structures. It proposes a new approach that utilizes machine learning to refine the analysis of the failure assessment diagram, using extensive data from numerical analyses to determine the plastic collapse limit load, define the assessment curve, and evaluate the residual stress parameter.

 

Presenting Author: Mohamed Elkhodbia Khalifa University

Presenting Author Biography: Mohamed Elkhodbia, a PhD candidate at Khalifa University's Mechanical Engineering department, specializes in finite element analysis, solid mechanics, and machine learning. His B.Sc. and M.Sc. degrees were also earned at Khalifa university. For his M.Sc., he worked on coupled electro-mechanical finite element modeling in tactile sensors using Abaqus, Python, and machine learning. His current research, sponsored by ADNOC, explores new casing designs for sour service conditions.

Authors:

Mohamed Elkhodbia Khalifa University
Alok Negi Khalifa University of science and technology
Imad Barsoum Khalifa University of Science and Technology
Akram Alfantazi Khalifa University of Science and technology