Data-Driven Analysis of Industrial Printers’ Error-Handling Processes
Student: Harry Herbert Beyel
Title: Data-Driven Analysis of Industrial Printers’ Error-Handling Processes
Supervisor: Dr. Sebastiaan J. van Zelst
1st Examiner: Prof. Wil van der Aalst
2nd Examiner: Prof. Thomas Rose
Summary
Modern industrial printers, like other machines, record an enormous amount of data. These data are based on sensors and embedded software. Each datum consists of a timestamp and an activity description. To analyze the processes at machines, recorded data have to be translated into event data. However, after translating the data, there is a granularity gap between the level of translated event data and the level at which conclusions about processes can be made. The limited view of data recording causes this gap. The prediction of component failures is valuable. Nevertheless, a limited amount of component failures might be known by companies. In this thesis, we propose a framework that overcomes the former introduced issues. We propose a data translation technique to translate printer data into low-level event data. To overcome the granularity gap, we propose a rule-based framework consisting of activity categorization to abstract low-level event data to high-level event data. By doing so, we convert domain knowledge into rules. We propose a technique to generate more data to overcome the limited available data for component failure prediction. We evaluate our framework on real-life data. By doing so, we demonstrate that our framework is usable in practice.
Note
This work was performed in collaboration with an industrial partner. Therefore, the final thesis is not shared publically. For more information, contact Dr. Sebastiaan J. van Zelst