Modelling of Average Weight Loss in Welding Defects using Response Surface Methodology and Artificial Neural Network

Abstract

The study aims to bridge this gap by scrutinizing the impact of a specific nonelastic
factor, namely the average weight loss, on pipeline weldments and its
interaction with elastic properties. To fulfil this objective, a comprehensive
experimental inquiry is conducted, encompassing diverse welding methods,
materials, and environmental conditions to authentically replicate real-world
situations. This investigation unveils the intricate interrelation between elastic and
non-elastic facets, underscoring the necessity of encompassing the latter to ensure
the dependability of pipeline weldments across various operational contexts.
Cutting-edge techniques, such as machine learning algorithms and finite element
simulations, are harnessed to accurately predict and optimize these non-elastic
factors, thereby enhancing the overall strength and structural integrity of pipeline
weldments. The experimental setup adheres to the central composite design,
meticulously constructed utilizing design expert software (version 13.0). The
response surface methodology analysis yields optimal outcomes, suggesting a gas
flow rate of 14.667 liters per minute, a voltage of 21.280 volts, and a current of
160.000 amps. These parameters collectively yield a welded joint with an average
weight loss value of 0.236, achieving a desirability value of 0.918. Additionally, the
artificial neural network model is employed to predict output parameters and
compared against the RSM methodology. The findings underscore the pivotal role
of optimizing non-elastic performance factors in pipeline weldments. By
accurately anticipating and controlling the period of immersion, engineers and
professionals within the pipeline sector can design weldments capable of enduring
harsh conditions, thus, prolonging pipeline operational lifespans.