Quantitative Characterization of Compressed Air Systems

Abstract

In industrial production plants, leaks in the compressed air system (CAS) lead to significant energy losses. In view of rising global energy costs, increasing the efficiency of such systems represents considerable costcutting potential. When analyzing CASs, it is often not possible to quantify the basic consumption and leakages from the available data. For companies seeking to evaluate potential savings and enhance energy efficiency, it is crucial to quantitatively assess the efficiency of CASs in terms of leakage rate. Determining the quality is typically done at a pressure of 1 bar, which is insufficient due to the volume influence of different compressor-units used in production plants. This work focuses on the algorithmic identification of pressure drop intervals as well as the determination and graphical representation of a quantitative trend visualization for industrial pressure drop intervals over a broad pressure range. For the study, the pressures time series of four production sites were recorded within one year. Shutdown periods were identified from the raw data using four machine-learning (ML) clustering and classification methods. By segmenting the pressure drop intervals, the shutdown periods were parameterized, from which quantitative trend visualization for the pressure drop periods due to compressor shutdown could be calculated and interpreted in the context of a graphical representation. The classification methods were compared as part of this work. With an F1-score of up to 0.99 being reached, only the variance differed among the algorithms. From the temporal observation of the calculated trend visualizations, rapidly falling curves suggest an increasing system leakage.