Artificial Intelligence (AI) put an increasing amount of strain on our total energy consumption and CO2 production. Not only is AI becoming increasingly more popular, but also AI models keep growing and thus need an increasing amount of computational resources. Recent research tries to mitigate this effect by creating more efficient hardware and by using Green AI, which is research in AI with additional focus on the computational resources a model requires. During this research one such Green AI method will be studied. This research focusses on the effects of memory usage on Canoncial Polyadic Decomposition (CPD) and Tensor Train (TT) decomposed Convolutional Neural Networks (CNN)s. These decomposed kinds of CNNs reduce the amount of parameters in the model, but may increase the amount of memory that is required to run the model. Therefore, first a theoretical analysis will be done on the memory used by these models. This analysis will then be validated by doing a real life scenario test and the effects of memory usage on inference time will be explored. Finally, regressions models will be made to see whether it is possible to predict the inference time. The results of these tests show that decomposed CNNs require more memory and the memory required in the real-life scenarios is higher than that was expected in the theoretical analysis. For the tested systems, it is shown that memory does influence the inference time negatively. Additionally, it was found that for very small kernels some initialization bias seems to be present, which makes the inference time larger,despite the CNN having less parameters and requiring less memory. Finally, it is shown that despite this inference bias, it is possible to predict whether the use of decomposed CNNs is beneficial to use compared to a regular CNN in terms of inference time.

In this thesis, a closer look will be taken into the design of a propulsion system of an eVTOL (electric Vertical Take-off and Landing aircraft). The primary goal of the eVTOL is to maximize the payload for a given maximum take-off weight and the eVTOL has to go into service in 2030. To make an as light weight as possible propulsion system, an investigation will be done for the propeller length and the number of rotors, which is present on a given eVTOL design. In addition to this, different drive choices will be discussed and an optimal drive choice will be made. For this drive, a closer look will be taken in efficiency optimisation. From these parameters a design example will be given using the current state-of-the-art drive technology. In addition to this possible future drive, developments will be taken into account.