Smart vibration sensing and predictive analytics for intelligent textile manufacturing: An IoT-edge and machine learning method

Abstrak

This study aimed to propose an IoT-Edge method for detecting vibration abnormalities in Textile Manufacturing, specifically on Draw Texturing Yarn (DTY) machines using an ADXL345 sensor and a Machine Learning Algorithm. The proposed system incorporated wireless sensor nodes, the MQTT protocol, Fast Fourier Transform (FFT) analysis, and a tuned Random Forest (RF) classifier to enable real-time monitoring as well as predictive maintenance. During the analysis, vibration data were collected from 13 spindles, with features extracted in both time as well as frequency domains to distinguish between normal and abnormal machine conditions. The RF model, optimized through hyperparameter tuning, achieved an accuracy of 97%, significantly outperforming the Support Vector Machine (SVM) baseline, which reached 71%. Major results showed the effectiveness of energy and centroid features in fault detection, with the Z-axis vibration proving to be a good indicator of yarn defects. The system presented low latency (average 20.37 ms) in data transmission using the MQTT protocol, ensuring practical deployability. This study offered a scalable and cost-effective solution for industrial vibration monitoring, bridging gaps in real-time processing and seamless IoT incorporation to support predictive maintenance in textile manufacturing.