Semi-Supervised Approach Using Transductive Support Vector Machine for Internal Leakage Detection in Two-Stage Hydraulic Cylinder

Abstract

Hydraulic cylinders with higher stages of extraction are extensively used in earthmoving and heavy machines due to their longer stroke, shorter retracted length, and high-end performance. The rigorous and long hours of operations make cylinders prone to internal leakage, which visually remains unnoticeable. This paper presents the conceptualization and realization of a newly developed 210 bar high-pressure hydraulic test rig actuated by a two-stage hydraulic cylinder. Experiments have been carried out to acquire pressure signals for two different leakage conditions (3% and 5% for moderate and severe leakages respectively) in the ramp wave motion of the cylinder. A decline in the working pressure and the piston velocity by approximately 10% and 45% for these leakage conditions respectively is noted. The time–frequency analysis infers these signals contain low-frequency components. For the automated leakage detection, a new iterative probability-based, transductive semi-supervised support vector machine (TS-SVM) is proposed capable of learning with limited datasets in several iterations. TS-SVM classifies the internal leakage with 100% accuracy in four iterations and utilizes only 64% of the total training data.