Data Mining for Intelligent Green Monitoring of Machine Centers - Backlash Error Analysis for Condition Monitoring
MetadataShow full item record
The European Union has pledged to limit climate change by reducing energy consumption and the manufacturing industry is among the biggest energy consumer in Europe. A project named Intelligent Green Monitoring (IGM) is developing an intelligent predictive maintenance system to reduce energy consumption by improving maintenance management on machine centers. This report describes this system and its theoretical background which uses condition monitoring, key performance indicators and knowledge discovery technology in the maintenance decision making for prognostics and diagnostics. A task in the IGM project is to improve monitoring of the degradation of the axis linear position accuracy in the feed drive system of the machine center. Monitoring the axis linear position accuracy today is performed periodically every 3-6 months and a lot of resources are used on maintenance to maintain the accuracy. To improve maintenance management of the axis linear position accuracy a data acquisition system is proposed to collect different key performance indicators of the feed drive system in real-time. The historical data is used to model, predict and estimate the degradation of the system and maintenance is performed accordingly. This thesis address the task above and investigates a method of collecting the backlash error which is a key performance indicator of the axis linear position accuracy and feed drive system degradation. A method of extracting the backlash error from displacement data has been developed and applied to specially designed test cycles which in theory should contain the error. The analysis of the data shows the error but the results from the measurement method is varying. The proposed method of extracting the backlash error from the reference machine center seems to be achievable, but has to be further developed and tested. The second part of the IGM project task is to model the extracted backlash error in an Artificial Neural Network (ANN). This report describes the theoretical background of ANN and how to develop an ANN to model a backlash error. From the development of the ANN, an automated regularization training function seems adequate to model the backlash error and should be considered when developing future backlash error models.