EARLY DETECTION OF LEAKAGES IN THE EXHAUST AND DISCHARGE SYSTEMS OF RECIPROCATING MACHINES BY VIBRATION ANALYSIS

This paper discusses the early detection of leakages and more generall y the problem of predictive diagnosis in the exhaust and discharge sys tems of reciprocating machines. Vibration transducers, which are easy to install on the machine, are the basic tool needed for devising a me thod of detection. The vibrational signals picked up on these machines are highly non-stationary. However it is possible to characterise any defect by using signal processing involving techniques that can accom modate this aspect [time-frequency analysis, non-stationary auto-regre ssive modeling]. Pattern recognition methods and more particularly, me thods of classification, are then used, notably involving neural netwo rks which make it possible to 'automate' the diagnosis.