Abstract
Compensating thermal errors using predictions from temperature-based empirical thermal error models is a widely used and convenient approach for mitigating the effects of the errors. However, thermal error models are often trained on small datasets that are not representative of all situations that the machine, and so model, may encounter during online use. Various adaptive strategies have been proposed to help models overcome this problem, such as intermittent model update strategies. One challenge that remains open is how such strategies can be automated.
This work proposes the use of modal analysis through Proper Orthogonal Decomposition (POD) and Hidden Markov Models with Gaussian Mixture Model emissions (GMM-HMM) in characterizing the thermal state of a machine tool. Performing a windowed POD analysis on the temperature data results in POD modes that define the heat cycles experienced during the windowed period. GMM-HMM are then used to cluster the POD modes. The approach correctly classifies various temperature data from all but one of the test datasets with True Positive Rate (TPR) value of over 61%. The presented approach can achieve higher accuracies through implementing the discussed improvements and be incorporated into the thermal error modelling strategy to inform adaptive modelling strategies
This work proposes the use of modal analysis through Proper Orthogonal Decomposition (POD) and Hidden Markov Models with Gaussian Mixture Model emissions (GMM-HMM) in characterizing the thermal state of a machine tool. Performing a windowed POD analysis on the temperature data results in POD modes that define the heat cycles experienced during the windowed period. GMM-HMM are then used to cluster the POD modes. The approach correctly classifies various temperature data from all but one of the test datasets with True Positive Rate (TPR) value of over 61%. The presented approach can achieve higher accuracies through implementing the discussed improvements and be incorporated into the thermal error modelling strategy to inform adaptive modelling strategies
Original language | English |
---|---|
Title of host publication | Laser Metrology and Machine Performance XV |
Subtitle of host publication | 15th International Conference and Exhibition on Laser Metrology, Machine Tool, CMM & Robotic Performance: LAMDAMAP 2023 |
Publisher | euspen |
Number of pages | 10 |
ISBN (Electronic) | 9781998999125 |
Publication status | Published - 15 Mar 2023 |
Event | 15th International Conference and Exhibition on Laser Metrology, Coordinate Measuring Machine and Machine Tool Performance - Edinburgh, United Kingdom Duration: 14 Mar 2023 → 15 Mar 2023 Conference number: 15 |
Conference
Conference | 15th International Conference and Exhibition on Laser Metrology, Coordinate Measuring Machine and Machine Tool Performance |
---|---|
Abbreviated title | Lamdamap 2023 |
Country/Territory | United Kingdom |
City | Edinburgh |
Period | 14/03/23 → 15/03/23 |