Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore

Text Mining Framework for Predictive Maintenance in Manufacturing

Duc-Minh Pham1,a and Su Myat Phyoe1

1Manufacturing Execution and Control, Singapore Institute of Manufacturing Technology (SIMTech), Singapore


In this paper, we propose a framework using integration of text mining algorithms for predictive maintenance in the cyber-physical system of manufacturing. By mining the text of log messages in the monitoring window, information about the coming failure can be predicted. The unstructured log messages are pre-processed and clustered into structured data that makes features can be extracted more efficiently. At the pre-processing stage, data is cleaned up by excluding unnecessary characters and words. A stop words list of system log files is built to remove unnecessary words in log files. The text mining can structure data from different systems version with different log files configuration. Beside features of text content in log messages, other statistical features of log messages are also extracted. The monitoring window is 24-hours, and prediction window time (warning time) is one hour. Maintenance records data is used to extract the failure data to label training data. Timestamp in message log files and maintenance records are used to match these two sets of data. Therefore, with labeling data supervised classification algorithms can be used to predict the failure output. The proposed prediction algorithms not only prediction failure in warning time but also give recommendation for proactive action to prevent failure or preparation before failure happening. The recommendation from data insight can be extracted from text mining in monitoring time and matching with text mining of maintenance records.

Keywords: Text Mining, Predictive Maintenance, Log files, Maintenance records

PDF Download