Predictive maintenance using Advanced Cluster Analysis (PACA)
Reference number | |
Coordinator | Chalmers Tekniska Högskola AB - Industrial and Materials Science |
Funding from Vinnova | SEK 4 994 976 |
Project duration | March 2019 - August 2022 |
Status | Completed |
Venture | The strategic innovation programme for Production2030 |
Call | Strategic Innovation Programme Produktion2030, call 11 for proposals within research and innovation. |
Important results from the project
The main objective of the PACA project was to develop new PdM algorithms to predict future maintenance needs using ML techniques and advanced cluster analysis. The project objectives were achieved by developing PdM algorithms to demonstrate their effects in industrial applications. Furthermore, the dissemination of the knowledge acquired and the research conducted was achieved by publishing scientific material together with the design of educational materials and theses/coursework that the project partners can use for educational purposes.
Expected long term effects
The results of the PACA project are identified patterns/models indicating machine health and proof of concept predictive maintenance (PdM) demonstrations via the designed software frameworks based on advanced ML/cluster analysis. The purpose of these frameworks was to show how they can be used as a feasibility study for decision support for PdM in the industrial parties´ real production environments. The expected effects are increased OEE, resource efficiency, competence in smart maintenance, and advanced data analysis for the competitiveness of the Swedish industry.
Approach and implementation
The PACA project was implemented in five main work packages: data collection and preprocessing, pattern identification, algorithm design and evaluation, and decision support systems for maintenance and dissemination. The main research method was designed on the basis of exploratory data analysis. Data provided by the industry partners was analyzed to identify interesting patterns and to create an understanding of how different patterns correlate with machine breakdown and can be used to predict future machine failures or anomalies.