Big Data Analysis Procedure Model for Manufacturing and Logistics: Strategies and Tools for the Practical Application

Big Data Analysis Procedure Model for Manufacturing and Logistics: Strategies and Tools for the Practical Application

Kategorien Beiträge in Büchern (reviewed)
Jahr 2018
Autoren Hübner, M.; Jahn, P.; Tewaag, G.
Veröffentlicht in In: Advances in Production Research Proceedings of the 8th Congress of the German Academic Association for Production Technology (WGP), Aachen, 2018, S. 52 - 62

In times of a gradual digitalisation of the production, more and more data is collected from diverse sources of a production process. The awareness for the potential insights generated from this data has increased massively in recent years. By now, even small-scale enterprises have the capacities to store the routinely incoming data. Crucial for the success of Big Data projects is the proficient reprocessing.

In this paper a model for the general approach to Big Data Analysis will be presented and the essential mathematical tools will be described in performance and suitability. For this, multiple data-mining models have been analysed and joined in a holistic approach. Additionally, different analysis strategies (e.g. correlation, regression, clustering and decision-trees) have been evaluated regarding their uses and limitations.

For verification, the derived model has been tested in collaboration with an industry partner on a multistage production process. Prediction models were developed and verified on a test group of data. For the preparation and analysis of the population, the data-mining workbench KNIME has been used.

It was possible to show, that multivariate linear correlations can be detected and examined using different analysis tools like matrices of the correlation coefficients, principal component analysis (PCA) or multidimensional scaling (MDS). Clusters and rule based decision tree models could be found as well. Based on the findings an optimisation of the assessed production process could be realised. Due to the derived structure and plan of procedure, the advantages of aforesaid models could be concentrated. A reduction of the processing time and an improved error prediction were made possible. Additionally a number of prior unknown factual contexts could be discovered between the collected parameters.

ISBN 978-3-030-03451-1
DOI 10.1007/978-3-030-03451-1_6