Monitoring of compressor operations - A machine learning approach

Compressors are important components in many industries. Proper monitoring technologies are very important and demanding. Failure of compressors could be very costly. The quality of compressor performance maps is very important for the operational availability of compressed air because there is a huge use of compressed air in almost every branch of industry. A remarkable percentage of the maintenance cost could be saved with a proper monitoring technology and maintenance program. Caused by the variety of compressor types and operational conditions complex superpositions of vibrational states occur which makes automated evaluation of faults demanding. In addition to vibration data the study has been extended to ultrasound frequencies using a new sensor technology. The broadband-ultrasound sensors and the diagnostic system cover a frequency range up to about 100 kHz enabling the simultaneous acquisition of vibration and ultrasound. The higher frequency range enables often an approach to early indications of faults. The extension of the sensor and signal processing methodology towards higher frequencies provides some advantages for the earlier prediction of operational states and lifetime of compressor components due to its sensitivity to small scale vibrations and turbulences caused by vibration effects. An increase of friction and micro-shocks, often an indicator for inappropriate operation, does not provide intense vibration. For this study, a screw type compressor has been equipped with a set of new broadband ultrasound sensors. For comparison and complementary purposes, vibration sensors have been placed. The use of ultrasound and advanced data technology has been demonstrated for different operational states over a longer period. It has been shown that ultrasound can be a promising tool for condition monitoring and fault diagnosis of screw compressors. Amongst the new sensor technology, advanced data processing on the basis of machine learning techniques such as deep neural networks provides an advanced diagnostic network.