High Pressure Air Compressor Condition Monitoring Using Unsupervised Pattern Classification Based on Projection Networks
High pressure air compressors (HPAC) are a high maintenance machine for they break down more often than expected and they serve critical roles. This study established the utility of an unsupervised pattern classifier system integrating a clustering algorithm based on DBSCAN and a dynamic classifier based on projection network to classify the condition of a 4-stage high pressure air compressor. The clustering algorithm is used to form clusters from un-labeled data and eliminate outliers. Subsequently, a system of projection networks is established to recognize all the significant clusters. The compressor data is consisted of pressures and temperatures at all four stages taken under various conditions including different baseline conditions, 3rd stage suction valve fault, 3rd stage discharge valve fault, and cylinder pitting and corrosion. The clustering algorithm was able to form clusters that each individually contains data mostly from a single class, and the projection network was able to differentiate these clusters and therefore classify the condition of the compressor correctly about 94% of the time. The ability of unsupervised classification does come with a price of lower classification accuracy. It was about 5% lower than what was accomplished by supervised classification.