Name of the participant: Sandra Zilker
Description of the IT-research project: The goal of the selected research project is to develop a machine learning method that is able to make predictions about certain factual contexts. More precisely, it should be able to predict possible machine malfunctions and failures. The procedure is to be trained with existing (log) data. This approach can also be subsumed under the terminology Predictive Maintenance. This means that machines should not be repaired if they are already defective, but rather faults should be predicted on the basis of data analysis and appropriate maintenance steps taken before the machine fails. This proactive approach differs fundamentally from previous approaches (repair on failure). This is made possible on the basis of (sensor) data, which are continuously collected. After the data has been collected and stored, it is analyzed. The analysis approaches used so far in the area of predictive maintenance are mainly statistical in nature. In the present project a machine-learning approach is to be pursued. The core idea of machine learning is that procedures are trained on the basis of data in such a way that patterns can be recognized. In concrete terms, the solution is to be implemented with a deep learning process. Deep learning is based on neural networks. The difference to “conventional” neural networks is the higher number of hidden layers. Thus, the results achieved can be improved considerably (especially with regard to larger amounts of data). Since the data in the present case are signal courses, a sequence prediction is to be made. At the current state of the art, the deep-learning procedure is to be based on recurrent neural networks in order to map the “n to n” relationship between input (data) and output (prediction).
Software Campus partners: FAU, Trumpf
Implementation period: 01.01.2020 – 31.12.2021
