Big Data Study on Two Industrial Machines (ZF) (07/2019-12/2020)

Description: The aim of this project is to improve the technical availability of industrial plants by means of prediction algorithms and big data technologies. In this context, component failures are to be detected and predicted with different time horizons so that the overall condition of the plants can be evaluated and, for example, proactive maintenance work can be initiated.

ZF Friedrichshafen AG has captured sensor data since December 2015. It regularly stores the machine's executions with a resolution of one millisecond. In addition to the sensor data, maintenance reports and production data are available to match the sensor data with produced parts and downtimes.

To this end, two algorithms are being developed. The first one considers only statistical properties of the measured data and provides an anomaly score to evaluate the current state of the machine. The second algorithm integrates a neural network to learn the relationship between the features derived from the sensor data and the time to the next failure. Thus, the second algorithm predicts the time horizon for the next machine downtime.

People involved (from Descartes Research Group): Samuel Kounev, Marwin Züfle

Project start: 01.07.2019

Project end: 31.12.2020