Industrial manufacturing relies on fully functional production machines that run under optimal conditions. Intelligent planning of the maintenance work on these machines is important in order to reduce the number and duration of production downtimes. Machine learning of models that predict maintenance is essential to reduce the risk of unwanted and unplanned production downtime.
Developing a high-precision machine learning model for predictive maintenance depends on the data available. The number of machine failures is low and the occurrence of important events in the training data is rare. Therefore, it is difficult to build a model with high precision and possible future machine repair needs go undetected until a machine outage occurs.
Federated learning can be used to train on a fleet of industrial machines in various production companies in order to expand knowledge about machine failures, repairs, or maintenance. This technology enables full data protection and individual machine information is treated confidentially in order to avoid conclusions about the production process. The production process is improved by predicting the necessary maintenance work, which allows the machines to operate under optimal working conditions. In addition, new production machines benefit from the existing knowledge of other companies and machines.