Automated machine learning and predictive maintenance have both become prominent terms in recent years. Combining these two fields of research by conducting log analysis using automated machine learning techniques to fuel predictive maintenance algorithms holds multiple advantages, especially when applied in a production line setting. This approach can be used for multiple applications in the industry, e.g., in semiconductor, automotive, metal, and many other industrial applications to improve the maintenance and production costs and quality. In this paper, we investigate the possibility to create a predictive maintenance framework using only easily available log data based on a neural network framework for predictive maintenance tasks. We outline the advantages of the ALFA (AutoML for Log File Analysis) approach, which are high efficiency in combination with a low entry border for novices, among others. In a production line setting, one would also be able to cope with concept drift and even with data of a new quality in a gradual manner. In the presented production line context, we also show the superior performance of multiple neural networks over a comprehensive neural network in practice. The proposed software architecture allows not only for the automated adaption to concept drift and even data of new quality but also gives access to the current performance of the used neural networks.
Log-File Analysis to Identify Internet-addiction in Children
