In the last few years, with the development of artificial intelligence and deep learning, there is an increasing number of different methods that can be used to solve previously difficult problems. One such method is the combination of meta-learning and transformer model, which we used in our final thesis. The latter enables fast and effective learning of models using a small number of input data. In our case, we wanted to predict several different types and severity of bearing defects based on a small number of previously performed measurements. By using the Python library PyTorch and taking into account the appropriate structure of the transformer model and the meta-learning process, we managed to achieve a satisfactory final accuracy. Through weighted voting, the learned models were able to predict with good accuracy which type of bearing defect the input measurement represents and also its severity. The performance of the predictions was also tested on measurements prepared exclusively for model testing and visually displayed in the form of a confusion matrix.
|
