The alloy Ti-43.5Al-4Nb-1Mo-0.1B, known by the brand name TNM, is a material for light weight constructions. It is designed for heavy-duty components, that are used at elevated operation temperatures. With a density of 3.9g/cm³ and high strength even at temperatures over 700°C this material is suitable for applications in aviation industry. TNM is used as a material for turbine blades of the low pressure stage of the geared turbo fan (GTF) turbine in aircraft. These components are produced by powder metallurgy processes. The powder is manufactured in a process called EIGA (Electrode Induction melting Gas Atomization), where a TNM electrode is melted continuously by means of inductively generated heat. Subsequently the liquefied material is atomised through a nozzle. The optimisation of this process requires the knowledge of the time varying temperature and stress field of the electrode. An experimental characterisation is too complex, therefore a numerical model to represent this process is presented. For this purpose three manufacturing phases have to be modelled: (i) initially a centrifugal cast electrode cools down and residual stresses appear due to temperature gradients. (ii) Subsequently the electrode is heated up until melting, where the heat source distribution is determined by the arrangement of the induction coils. These heat sources move controlled by time with the velocity of the melting process along the electrode. (iii) Thermo-mechanical stresses occur due to the resulting temperature field and are superimposed with the stresses calculated in phase (i). The complex temperature fields in phase (ii) cannot be modelled with the standard functionality of commercially available finite element software. To accomplish this additional user-subroutines have to be coded. This model allows to determine the stress field for variable geometries and arrangements of the induction coils. Given the objective to minimise the maximal occurring stresses, guidelines for the optimal choice of geometry and process parameters can be deduced.