Fault Detection in The Rotatory Machine Using Machine Learning

The efficient operation of rotary machines is essential for ensuring optimal performance in various industrial applications. The early detection of faults in these machines is of paramount importance to prevent significant damages and reduce downtime. This paper presents a novel approach for fault detection in rotary machines utilizing machine learning (ML) techniques. By employing vibration signal data, the ML models were trained to recognize patterns indicative of common faults such as misalignment, imbalance, and bearing defects. Various features were extracted from the raw signal, including time-domain statistics, frequency-domain components, and waveform characteristics. These features were fed into different machine learning algorithms, including decision trees, support vector machines, and deep neural networks. The models' performance was evaluated using accuracy, precision, recall, and F1-score metrics. Among the models tested, deep neural networks exhibited superior performance with an accuracy exceeding 95%. The results demonstrate that machine learning, especially deep learning techniques, can provide a robust tool for automated, real-time fault detection in rotary machines. Furthermore, this approach can be generalized to different types of machinery and operational conditions, paving the way for the broader adoption of ML techniques in the domain of machinery maintenance and monitoring.

Keywords: rotary machines, machine learning, fault detection, vibration signals, deep neural networks, time-domain statistics, frequency-domain, machinery maintenance.