Detection and Classification of Internal Leakage in Hydraulic Cylinders

TL;DR

This paper presents an LSTM-based algorithm for real-time detection and classification of internal leakage in hydraulic cylinders, achieving 96% accuracy and enabling proactive maintenance to improve system reliability.

Contribution

It introduces a novel LSTM neural network approach for internal leakage detection in hydraulic cylinders, enhancing detection speed and accuracy over traditional methods.

Findings

Achieved 96% accuracy in classifying leakage types.

Enabled real-time, online fault diagnosis for hydraulic systems.

Reduced maintenance costs and extended system lifespan.

Abstract

Hydraulic systems have been one of the most used technologies in many industries due to their reliance on incompressible fluids that facilitate energy and power transfer. Within such systems, hydraulic cylinders are prime devices that convert hydraulic energy into mechanical energy. Some of the genuine and very common problems related to hydraulic cylinders are leakages. Leakage in hydraulic systems can cause a drop in pressure, general inefficiency, and even complete failure of such systems. The various ways leakage can occur define the major categorization of leakage: internal and external leakage. External leakage is easily noticeable, while internal leakage, which involves fluid movement between pressure chambers, can be harder to detect and may gradually impact system performance without obvious signs. When leakage surpasses acceptable limits, it is classified as a fault or failure. In such cases, leakage is divided into three categories: no leakage, low leakage, and high leakage. It suggests a fault detection algorithm with the basic responsibility of detecting minimum leakage within the Hydraulic system, and minimizing detection time is the core idea of this paper. In order to fully develop this idea, experimental data collection of Hydraulic systems is required. The collected data uses pressure sensors and other signals that are single-related. Due to the utilization of Long Short-Term Memory (LSTM) recurrent neural networks, more complex data analysis was enabled, which the LSTM-based leakage detection algorithm successfully achieved, providing almost 96% accuracy in classifying leakage types. Results demonstrate that the proposed method can perform real-time and online fault diagnosis for each cycle, reducing maintenance costs and prolonging the hydraulic system's lifespan.