Vibration Based Fault Monitoring of a Compressor using Tree-based Algorithms

Abstract

Reciprocating air compressors are used as a part of manufacturing and engineering industries to offer pressurized air, which is utilized for different productive purposes. Compressors are trusted upon to be prepared and readily available as and when required and any interim stoppage or interruption will affect the manufacturing processes that are dependent on compressed air. From the reports of any maintenance engineer, one can find that in a reciprocating air compressor, components like bearings, valve blade, V-belt and piston rings add to a more noteworthy level of failure. Researchers do make attempts to find a suitable device that is profoundly welcome by the industry, for diagnosis of the fault that recommends a remedial action. Towards this direction, a study was attempted and vibration signals were collected from an experimental setup under supervised learning technique. Statistical features of the same were extracted for various combinations of fault conditions and analyzed using different tree-based algorithms with an intention to find the best one that will classify the fault with more accuracy and with the least computational time.