Performance Data Analytics for Contact Cooled Rotary Screw Air Compressors
Xavier Chelladurai1, Suraj S. Jain2, Aryan Soni Burman3, Sonal Kumar4, Sindhu Srinivas5

1Xavier Chelladurai*, Professor, Department of Computer Science, School of Engineering andTechnology, Christ (Deemed to be University), Bangalore, India.
2Suraj S Jain, Department of Computer Science, School of Engineering andTechnology, Christ (Deemed to be University), Bangalore, India.
3Aryna Soni Burman, Department of Computer Science, School of Engineering andTechnology, Christ (Deemed to be University), Bangalore, India.
4Sonal Kumar, Department of Computer Science, School of Engineering andTechnology, Christ (Deemed to be University), Bangalore, India.
5Sindhu Srinivas, Department of Computer Science, School of Engineering and Technology, Christ (Deemed to be University), Bangalore, India.
Manuscript received on March 12, 2020. | Revised Manuscript received on March 25, 2020. | Manuscript published on March 30, 2020. | PP: 3597-3602 | Volume-8 Issue-6, March 2020. | Retrieval Number: F8893038620/2020©BEIESP | DOI: 10.35940/ijrte.F8893.038620

Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: The operations management of air compressors gets transformed from the traditional reactive approach to the proactive data analytics based one. While large volume of data is analyzed, it is important to identify the parameters that cause performance deterioration. This helps in the preventive maintenance and increases the system availability to a great extent. In this research, we propose an approach to identify the causal parameters and test them in a practical customer environment. We have analyzed the data pertaining to Contact Cooled Rotary Screw Air Compressors from a customer’s site. The system is controlled by XE-90M/145M controller. The controller continuously monitors 85 parameters and values collected every minute and uploaded to the cloud every 15 minutes. We analyzed the data for all the parameters, identified the upper control limit (UCL), lower control limit (LCL) and mean values. Using the historical data of drip of compressors, we have analyzed the data and classified the parameters into two categories, parameters which causes the performance deterioration and the ones that goes abnormal as a consequence of performance deterioration. Also, the study has discovered that the causal parameters show performance deterioration symptoms during a window of one to three hours before the actual drip happens. This has helped the customer to proactively do the corrective actions in the window and avoid drip of the compressor.
Keywords: Predictive Maintenance, Data Science, Deep Learning, Compressors, Machine Learning, Data Analysis.
Scope of the Article: Machine Learning.