Predictive Maintenance in Twin Screw Air Compressors through Unsupervised Learning

A predictive maintenance module for oil-injected twin screw compressors is proposed in the present paper. Oil-injected twin screw compressors are widely used in the pneumatic drive industry and their maintenance represents an important issue. In fact, periodic maintenance policies are exploited in this field; this practice can conduct to a decrease of the efficiency of the plant units where these devices are installed. In order to solve this problem, an unsupervised learning approach is proposed in order to design a predictive maintenance module. Reliable and significant data are acquired through an ad hoc data acquisition and storage system, based on Industry 4.0 principles. Data analysis represented a first crucial step for the exploitation of unsupervised learning methods. Three different operating conditions were taken into account, associated to the compressor's oil degradation. Satisfactory results in terms of accuracy were obtained in the tests of the developed module on real data.