โครงการวิจัย

Detecting motor anomalies in heating, ventilation, and air conditioning (HVAC) systems remains challenging due to the high dimensionality and dynamic nature of real-world data. Developing a model that can predict anomalies without relying on manual labeling is crucial for enhancing predictive maintenance (PdM) and supporting smart manufacturing initiatives.

This research presents a method for developing a model to predict motor anomalies in a HVAC system that uses real vibration data collected from industrial sensors connected via the Internet of Things (IoT). The collected data is preprocessed using Principal Component Analysis (PCA) for dimensionality reduction and labeled through K-Means clustering to classify data as normal or abnormal. Then, three sequence-based deep learning (DL) models, consisting of Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM), are  trained with hyperparameter optimization using the Tree-structured Parzen Estimator (TPE) to forecast abnormalities up to 24 hours in advance. The results indicate that the GRU model performs best on the X-axis, achieving the highest accuracy, precision, recall, and F1 score, along with the narrowest uncertainty zone. On the Z axis, LSTM yields the most accurate predictions, with the fewest incorrect predictions in the uncertainty zone. These findings demonstrate that the proposed approach effectively supports PdM applications and aligns with  smart manufacturing concepts in Industry 4.0.

1. สามารถตรวจจับและทำนายความผิดปกติของมอเตอร์ได้ล่วงหน้าอย่างแม่นยำ

2. เพิ่มพูนองค์ความรู้และความเข้าใจในการประยุกต์ใช้เทคโนโลยี Deep Learning

3. ลดความเสี่ยงในการหยุดชะงักของกระบวนการผลิต และยืดอายุการใช้งานของอุปกรณ์

• -