Discovery and monitoring of the activities, based on historical data, followed by optimization of the activities, based on future predictions.
Context and challenges
Recent advances in the interconnectedness and digitization of industrial machines, known as Industry 4.0, pave the way for new analytic techniques. Indeed, the availability and the richness of production-related data at every step of the production is increasing.
As such, an emerging research area, known as process mining, enables the analysis of process-related data stemming from information systems supporting business processes. With the use of derived ad-hoc data mining methods, activities such as bottleneck analysis, resource networks, remaining flow time prediction, or deviance detection become possible.
Two main parts in this project have been identified:
- the discovery and monitoring of the activities, based on historical data;
- the optimization of the activities, based on future predictions.
On one hand, a series of dashboards has been developed for the extraction of information from the traces. These dashboards offer custom detailed views on Heraeus’ activities and provide the monitoring of key performance indicators. The dataset represents about 5 years of activities with around 300k events. For instance, we may cite a steering production view, a view on the daily workloads, or the control-flow analysis of the whole process filtered for a given product.
On the other hand, predictive models have been devised for providing insights related to the sales and the production. These models are trained with the available traces and can predict the future orders, the delay required to produce a specific article at any given time, the detection of some abnormal cases, or workload prediction. Next-step recommendations and real-time bottleneck detection will also be considered.
Involved methods include the usage of recurrent neural networks for the prediction of time series, regression trees for the prediction of the remaining time, and local outlier factor for the detection of abnormal cases.
Partners and funding
Funded by Innosuisse – Swiss Innovation Agency
A series of dashboard predicting the future workload and supporting tactic decisions has been implemented. A complete pipeline for supporting decision on workload planning is in progress.
F. Albertetti, H. Ghorbel (2018). Analytics in Industry 4.0: Improving Business Processes with Process Mining. FTAL 2018 – Industrial Applied Data Science