Predictive Maintenance in Twin Screw Air Compressors: a Case Study
Silvia Maria Zanoli, Crescenzo Pepe, Mahmoud Sfar Hancha
Abstract
The present paper proposes a predictive maintenance application to twin screw air compressors. An experimental setup was designed to acquire compressor operation data under different operating conditions. To detect the operating parameters of the compressor, the data acquisition system was realized exploiting Industry 4.0 concepts. An in-depth data analysis phase represented the initial point for the application and comparison of supervised machine learning techniques. The designed tool allows four operating conditions to be classified concerning the state of degradation of the oil used by the compressor, of the filters, of the separator, and of the power circuit. The results obtained from the experimental tests allowed to conclude that innovative techniques based on supervised machine learning can be applied to twin screw compressors, which are widely used in the industry, being able to perform predictive maintenance policies replacing the currently adopted periodic and corrective maintenance policies. The results obtained showed good values in terms of accuracy of the adopted models.