Thermo-mechanical Analysis of the Joining Process in Friction Stir Welding of High-Density Polyethylene

Abstract

High-density polyethylene (HDPE) is a versatile thermoplastic with key applications across various industries, particularly in oil and gas, construction and water supply, where HDPE pipes are essential. Welding is the primary method employed to form strong, leak-proof joints in HDPE components. This research investigates the material behavior and defect formation in the Friction Stir Welding (FSW) of HDPE using numerical and machine learning approaches. Key to the building of an accurate numerical model of the welding process is the application of an accurate material model. Therefore, the research begins with the development of constitutive models for HDPE, considering three models, Johnson-Cook (JC), Three Network (TN) and Three Network Viscoplastic (TNV). These models were derived from tensile testing data at various strain rates and temperatures, and are validated through Finite Element Analysis (FEA) of drop weight impact tests. The numerical findings indicated that the TN constitutive model performed considerably better compared to other models between experimental impact tests and numerical predictions across all testing speeds. This research also addresses the crucial role of temperature in determining joint quality and strength by focusing on thermal modelling for thick HDPE plates using direct heat conduction analysis. It was shown that the utilization of a mixed heat generation model consisting of both solid state and viscous shear flow significantly improves the thermal model predictions. For the thermo-mechanical modelling of FSW, the Coupled Eulerian Lagrangian (CEL) technique was employed using the calibrated JC model. Four sets of process parameters were simulated and validated against experimental data, emphasizing material flow patterns, defect formation and associated temperature and axial force histories. Insufficient thermal gradients at lower rotational speeds led to inconsistant material flow and the formation of voids and wormholes. Furthermore, artificial neural networks (ANN) were trained to predict joint performance indicators of process temperatures and axial forces. The ANN models were further employed to forecast the same within the range of FSW parameters, enabling better control and identification of favorable weldability conditions. The numerical weld quality predictions from both CEL-based thermomechanical modelling and ANN approaches showed strong agreement with the experimental FSW results, validating their effectiveness and reliability.

Date of Award	5 May 2025
Original language	American English
Supervisor	Imad Barsoum (Supervisor)