A Data-Driven Holistic Approach to Fault Prognostics in a Cyclic Manufacturing Process

Abstract

The complexity of manufacturing systems is increasing due to the increased requirements related to the variety and quality of the products, their complexity, and due to the general technological developments. In turn, the data related to the manufacturing processes is growing in size and in complexity. This presents new challenges for real-time monitoring, diagnostics, and prognostics of the processes. The challenges are addressed by new tools, methodologies, and concepts, collectively referred to as Big Data. The paper deals with the use of advanced methods for prognostics of infrequent faults on available but highly dimensional manufacturing process data. A holistic approach, which includes data generation, acquisition, storage, processing, and prognostics, is shown in a case of a plastic injection moulding process. Real industrial data acquired from five injection moulding machines and the Manufacturing Execution System within a period of six months is used. It is shown how the approach is able to tackle the high dimensionality and the large size of the data to create and evaluate prediction models for prognostics of the unplanned machine stops.