Search

In this contribution, a data-driven approach towards the prediction of maintenance for the critical component of an injection molding machine is presented. We present our path from exploring and cleaning the data towards the implementation of a prediction algorithm based on kernel density estimation. We give first analytical evidence of the algorithms potential. Moreover, we compare the approach described here with our previous work where we went a model-based approach and present advantages and disadvantages of the two approaches. We try to contribute to a non-comprehensive guide on the implementation of predictive maintenance systems for industrial mass production facilities.

In this contribution we present a model based approach towards predictive maintenance of the hydraulic pump of an injection-moulding machine for a system that remains in production. We present a methodology to iteratively gain a detailed understanding of a production facility. We describe our approach from problem evaluation, gathering of expert knowledge, data access, exploration of and knowledge mapping to the data, towards a first model. We discuss the advantages of the implementation of expert knowledge to define normal machine behaviour in terms of data-quality control and resource-friendly modelling. We present first results from a physical model of a subprocess and from a model based on vibration analysis and discuss their interplay and possible benefits from this different models.