Accurate fault diagnostics in rotary machines is generally approached by developing a physical model of faults and understanding the relationship between faults and measurable signals captured by a variety of sensors. Classical fault identification and classification models employ analytical, signal processing and statistical-based features of the sensor signals combined with suitable classifiers. These feature-classifier combinations are engineered by incorporating expert-based knowledge about characteristic signatures related to faults. The engineered-features have shown success for fault diagnostic in mechanical systems that exhibit similar signal characteristic, however the fault-specific nature of these features limits their performance for general signal monitoring. The goal of this thesis is to investigate and compare the performance of automatic feature extraction through unsupervised learning instead of feature-engineering.
The Tightening Technique group is currently supporting the Data Driven Services in the project “Predictive maintenance”. The student will contribute to the data analysis conducted by the team. The students will also use a graphical machine learning software and investigate its potential for feature learning.
The performance of the proposed feature extraction model will be validated on sensor data collected from an experimental test-rig specifically designed to study characteristics of bearing and gearbox related faults. Feature learning on raw vibration signal possibly will extract vibration-features that can improve fault identification performance of subsequent classifier. Consequently, instead of feature-extraction, the feature-learning approach will be utilized to capture domain specific failure features. The feature-learning approach alleviates dependence on prior knowledge of the problem and proves beneficial in tasks where it is challenging to develop characteristic features”.
This topic requires programming experience in Python, R or Matlab, as well as knowledge of machine learning methods and a strong interest in applying these techniques on real-world data.
Our office is located in Sickla, Stockholm.
As this is a project position for studies and not an employment, it does not qualify for seeking a work permit. We can therefore only accept applications from students who are either attending Swedish universities (i.e. already have a student visa) or, if they are attending universities abroad, are EU-citizens.
Great ideas accelerate innovation. At Atlas Copco Industrial Technique we team up with our customers to turn industrial ideas into smart manufacturing assembly solutions and innovative industrial tools. Our passionate people, expertise and service bring sustainable value to industries everywhere. Atlas Copco is based in Stockholm, Sweden with customers in more than 180 countries and about 37 000 employees. Revenues of BSEK 95/ 9 BEUR in 2018. For more information: www.atlascopcogroup.com