Application of Machine Learning Algorithm for Fault Detection in Pump

Anomaly detection in pumps is critical, especially in industrial applications. Industrial-scale pumps need to be replaced and maintained regularly. Anomaly detection can be used to forecast pump failure before it happens. Vibrational analysis has been utilized to non-intrusively ensure the effectiveness and safety of the machine’s functioning. When certain machine defects occur, abnormal vibrating patterns can be observed. Unusual vibration is also caused by flaws in hydro-mechanical parts, such as shaft misalignment, bearing instability, fractured blades, cavitation, and more. In this paper, a machine learning approach for failure detection is used to analyze the vibrational signal of a normal and faulty pump. Convolutional Neural Network (CNN) and Artificial Neural Network (ANN) algorithms are used to classify vibration signals. The dataset is extracted from the ADXL 335 accelerometer installed in the pump shaft. The pump’s vibration signal in the X and Y axes was recorded. These datasets are treated as a time sequence problem so, the sliding window technique is used. Two machine learning models: ANN and CNN are created to accurately categorize the impeller’s state. The study suggested CNN be the best model for the prediction of the broken blade of the impeller with 99.89% accuracy in validation. The generated CNN model can be utilized in the industry for real-time fault detection of pumps.