Fault Diagnosis of Reciprocating Pumps and Compressors Using Condition Monitoring Signal Analysis Employing Machine Learning Techniques

This project addresses the critical need for predictive maintenance in industrial rotating machinery, such as pumps and compressors, which operate under severe conditions and whose failure can lead to significant economic losses. The focus is on condition-based maintenance, utilizing the analysis of physical signals, predominantly vibrations, to identify incipient failures. Traditionally, this diagnosis has relied on signal processing techniques and classification algorithms like Artificial Neural Networks (ANN) and Random Forest (RF). However, this work aims to overcome the limitations of conventional methods, such as manual feature extraction, by applying and comparatively evaluating Deep Learning (DL) architectures. The methodology involves gathering information on DL algorithms, defining a training protocol with signal database generation, and statistically evaluating three selected DL models to determine the most accurate approach for fault classification. The final objective is to document the findings and synthesize them into a scientific article for indexed journal submission.<br/><br/><b>Goal</b>: <br/>To diagnose failures in reciprocating pumps and compressors by analyzing condition monitoring signals, utilizing advanced machine learning and deep learning techniques.<br/><br/><b>Research lines</b>: <br/>Control engineering and automation technologies