Leakage fault detection in Electro-Hydraulic Servo Systems using a nonlinear representation learning approach

Siavash Sharifi; Ali Tivay; S Mehdi Rezaei; Mohammad Zareinejad; Bijan Mollaei-Dariani

doi:10.1016/j.isatra.2018.01.015

Leakage fault detection in Electro-Hydraulic Servo Systems using a nonlinear representation learning approach

ISA Trans. 2018 Feb:73:154-164. doi: 10.1016/j.isatra.2018.01.015. Epub 2018 Jan 3.

Authors

Siavash Sharifi¹, Ali Tivay², S Mehdi Rezaei³, Mohammad Zareinejad⁴, Bijan Mollaei-Dariani⁵

Affiliations

¹ Department of Mechanical Engineering, Amirkabir University of Technology, Tehran, Iran. Electronic address: siavash.sharifi@aut.ac.ir.
² Department of Mechanical Engineering, Amirkabir University of Technology, Tehran, Iran. Electronic address: alitivay@aut.ac.ir.
³ Department of Mechanical Engineering, Amirkabir University of Technology, Tehran, Iran. Electronic address: smrezaei@aut.ac.ir.
⁴ New Technologies Research Center, Amirkabir University of Technology, Tehran, Iran. Electronic address: mzare@aut.ac.ir.
⁵ Department of Mechanical Engineering, Amirkabir University of Technology, Tehran, Iran. Electronic address: dariani@aut.ac.ir.

PMID: 30686294
DOI: 10.1016/j.isatra.2018.01.015

Abstract

Electro-Hydraulic Servo Systems (EHSS) are employed as actuators to track the desired trajectory and exert force in heavy-duty industrial applications. The EHSS is often prone to problems such as leakage and actuator seal damage during the course of its utilization. These faults which cannot be directly detected from current sensor values, can eventually result in complications and degrade control performance. The goal of this research is to use representation learning concepts to detect these faults with decreased complexity. The objective is to find a nonlinear mapping to transform raw data into another space in which classification becomes easier. The data are driven from the hydraulic supply pressure signal. To find the mapping, a custom-built optimization algorithm is proposed along with a suitable cost function to carry out the search for the new representation. The performance of the resulting transformation is tested in an experimental setting to show the merits of the proposed method.

Keywords: Classification; Fault detection; Hydraulic; Nonlinear mapping; Representation learning.