R. Kosturkov. Model-Based Diagnosis of the Pneumatic Systems Condition

Key Words: Model-based diagnostics; pneumatic systems; pressure drops; time series; coefficient of determination; Pearson correlation coefficient.

Abstract. Faults are adverse events in any industrial production system. Their occurrence affects the efficiency of the system and reduces the competitiveness of production. Early detection and diagnosis of faults in automated systems is important to prevent equipment damage and loss of performance. For this purpose, more and more sophisticated systems for observation and monitoring of basic characteristics in automated processes are being built. A prerequisite for increasing their efficiency is the use of additional sensory information, modeling and intelligent information analysis to detect faults. The paper explores the possibility of diagnosis of unwanted pressure drops in pneumatic systems. These model-based diagnostic methods aim to distinguish the causes of their occurrence or location. The objects of diagnosis are pressure drops in the supply line or those in the branch, main lines. The presented formulation of the problem and task are dictated as a result of inspection and analysis of operating pneumatic systems of industrial enterprises in the country. It is the pressure drops that are defined as the main and most frequently occurring problem in the operation of the system, and for the resource optimization of the distribution network their localization is of special importance. The paper proposes an approach for the use of load diagrams (time series) with two measurable variables – instantaneous flow and pressure. Based on continuous monitoring and a known model relationship between the two quantities, indicators for detection and localization of pressure drops are determined, reducing the efficiency in the components of the pneumatic system – the main line, local stations or the compressor installation. For the purposes of verification of the proposed approach and the performed analysis – in general, real system data from 13 specific production machines were used.