2D-CNN Based Classification on Water Leakage Identification in Automatic Pump

This study proposes an approach for classification of unusual in automatic water pump systems, especially on a water leakage, one of the key issues that can significantly affect system efficiency and cause serious damage. The method utilizes motor current signals, which are transformed into 2D spectrograms using the Short-Time Fourier Transform (STFT). These spectrograms are then classified by a 2D Convolutional Neural Network (2D-CNN) Designed to determine between usual and unusual operating conditions with high accuracy. Experimental results indicate that the model can effectively detect unusual and leakage events when trained with appropriate parameters, such as a learning rate of 0.001 and 60 training epochs. This model serves as an efficient tool for preventing system failures and reducing maintenance costs in automatic water pump systems.