Computer Vision News - September 2022

53 Nadine Rücker hardware is key and log files hold non- patient adjustment measurement Key Performance Indicators (KPIs). Those were usedtotrainamodel calledLSTMFCN(LSTM concatenated with a Fully Convolutional neural Network) in order to learn from those KPIs about hardware conditions. Our model could predict hardware failures with an AUROC (area under the receiver operating characteristic curve) of 99%. resulting in F1-Scores of 92% or higher in all studied data sets and scenarios and ensures high data quality for subsequent analyses. Hardware Failure Prediction As soonasparsingchallenges areovercome, manifold opportunities for knowledge gain arise. In our studies, we employed log files for automated hardware failure detection. In the context of medical imaging, reliable Log files are produced by almost any modern system and contain vast amounts of information. With updates, also log file structures can change. In order to parse log files successfully, we propose “FlexParser” which extracts desired events reliably despite structural changes. This enables manifold use cases, e.g. hardware failure prediction.