This thesis investigates the feasibility of the Finite-Element software DEFORM® in multi-layer arc welding by applying Design of Experiments to statistically analyse welding parameter influences. For this purpose, a manufacturing experiment was conducted, in which 32 layers of Ti-6Al-4V were built onto a substrate plate fitted with thermocouples to log the temperature history. The obtained real process data was then used to extract the layer timing and detailed movement data of the welding torch. A heat source model was then programmed in MATLAB® based on the normalised layer height and volumetric heating power, controlling thermal power depending on real deposition settings. Dummy heat sources were introduced to correctly activate each layer due to restrictions inside the DEFORM® arc welding environment. All accomplished results were combined in one time-synchronised heat source model to ensure convenient timetable import in simulations. Furthermore, the standardised Ti-6Al-4V material model was improved and extended to ensure accurate and stable calculations.
A parameter analysis study was conducted by applying statistical Design of Experiments, in which the absolute deviation between the temperature response curves and their thermocouple counterparts served as a comparative basis. A total of 8 parameters were identified as interesting and grouped in three experimental runs. Each run focused on different emphases to reduce the individual parameter count and thus the number of experiments needed for meaningful results. Individual timestamps were investigated through full quadratic response surface models, finding theoretical optimal process parameters for each subsequent data point. For this purpose, a custom MATLAB® code was developed for enhanced analysis, chaining the individual results to form a time series based analysis. Thermal conductivity k and heat source power efficiency η were found to be most influential by far, and the front and rear ellipsoid apex showed to be most influential among geometrical heat source parameters. Due to their small deviation from values suggested in literature, normalised heat source dimensions were used to focus on energetic parameters, enabling the proposal of time-dependent functions for heat source power and conductive boundary conditions.
An improved Finite-Element model was built using all gathered information, which follows boundary conditions and timings as realistically as possible. A very accurate approximation of real temperature measurements was possible in thermal analysis, however, the certainty of a realistic development remains to be confirmed by subsequent experiments.