Health classification of pumps using transformer-based deep learning

Abstract

This paper develops a health classification system for pumps to enhance operational efficiency and reduce unplanned downtime, crucial for manufacturing and water treatment industries. Leveraging real-time data from temperature sensors and industrial accelerometer, the system captures vital pump health indicators. Data is collected via Data Acquisition (DAQ) modules and by using Deep Learning (DL) techniques such as Long Short-Term Memory (LSTM) networks and Transformers; the pump health classification is achieved. These DL models excel at understanding complex temporal and spatial patterns in sensor data, essential for accurate fault detection. Through a comparative analysis of LSTM and Transformer models, their efficacy in pump health classification is assessed. This approach emphasizes the importance of sophisticated data analysis and deep learning in industrial maintenance practices. By providing fault detection, the system aims to significantly reduce maintenance costs, optimize resource usage, and enhance the safety and reliability of industrial operations.