Classification of Pump Failure Using a Decision Tree Technique

Abstract

The oil and gas sectors employ pumps extensively. The dependability and good operation of the pumps contribute significantly to the manufacturing line’s efficiency. Nevertheless, breakdowns can occur for a variety of causes, and a single failure can result in severe production delays and economic losses. Therefore, it is of the highest importance to anticipate future errors and swiftly address their root causes. Traditional pump maintenance and fault detection methods are ineffective in detecting possible defects. However, with the rapid rise of industry 4.0, the growing use of sensors, and the use of artificial intelligence techniques, smart plants may automate their operations to greatly enhance their efficiency and quality of output. Given that, Prognostics and Health Management (PHM) is essential for optimal machine performance. Meanwhile, Predictive Maintenance (PdM) is an emerging topic within maintenance methodologies with the objective of predicting failure prior to its occurrence in order to schedule maintenance only when it is necessary. These solutions have benefited pump maintenance decision-making by addressing complexity issues, reducing downtime, enhancing overall reliability, and reducing pump operating costs. This study examines the use of the classification learner to predict pump failure. The vibration sensor data for the pump were employed for prediction in the MATLAB software using the classification learner machine learning method. According to the dataset, the variable of interest is the machine status, which is categorized as normal, broken, and recovering. On the basis of the confusion matrix and the evaluation of the true positive rate (TPR), the false negative rate (FNR), the positive predicted values (PPV), and the false discovery value, the actual result was anticipated (FDR). The experimental result reveals that the accuracy of the training model was 91.94%, whereas the accuracy of the testing model was 74.4%.