Description
This research presents a method for vibration-based fault detection in hydraulic pumps using a convolutional neural network (CNN). The reliability and efficiency of the hydraulic pump are crucial for the effectiveness of the hydraulic system. Hence, in order to ensure optimal performance of the hydraulic pump, it is imperative to implement problem detection and monitoring. Initially, by employing the piezoelectric sensors positioned in proximity to the pump, vibration signals were acquired from both the operational and faulty conditions of the pump. The MATLAB software is utilized to preprocess the time domain vibration data before its input into the CNN. There are two types of CNN models commonly utilized in deep learning for fault detection: 1D CNN and 2D CNN. The comparison is conducted on both the frequency domain and time series vibration data, utilizing a 1D CNN architecture. Subsequently, the data graphs in the time domain and frequency domain are compressed to a reduced resolution size prior to being inputted into a 2D CNN. In order to make comparisons, the vibration signals are further analyzed using principal component analysis (PCA), which is a method for reducing the dimensionality of the data. The results indicate that utilizing the frequency spectrum as the input for the CNN leads to the highest accuracy of 100% for both 1D and 2D CNN models. Using the time series as the input for the CNN results in a decrease in defect identification accuracy to 96%. These results also show that the accuracy drops even more when PCA is used to compress time series or frequency spectrum data before it is fed into the CNN model.
