Internal Leakage Diagnosis of a Hydraulic Cylinder Using C-LSTM Neural Network

Abstract

The leakage in hydraulic cylinders is of utmost importance due to its critical nature and potential consequences like reduction in system efficiency and performance. It results in the loss of hydraulic pressure which leads to diminished power output. The present manuscript discusses the leakage detection of a hydraulic cylinder utilising pressure signals. For this, pressure signals of a 2-stage hydraulic cylinder for no leakage and leakage condition (5% wear in the nominal diameter of the seal) are acquired. A total of 100 samples (50 for each condition) are divided into 80:20 for training and testing. The training data is used to train Convolution Long Short Term Memory (C-LSTM) neural network which captures spatial along with temporal features. The layer of 1-D convolution extracts the features from the signals and is then fed to the LSTM layer. The model is trained for 30 epochs and achieves the validation loss <0.2. The training accuracy of the proposed C-LSTM model is found to be 97.47% whereas test accuracy is ~95%. Thus, C-LSTM has proved to be a robust and reliable methodology for diagnosing internal leakage in hydraulic cylinders. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.