A Deep Learning Approach for Fusing Sensor Data from Screw Compressors.

Area of Science:

• Building energy systems
• Artificial intelligence in engineering
• Thermal management technologies

Background:

• Chillers are crucial for thermal regulation in large buildings but suffer from inefficiencies and limited optimization due to system accessibility.
• Data analysis and deep learning offer potential for transforming sensor data into actionable insights for real-time monitoring and anomaly detection.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting the control stages of a chiller's screw compressor.
• To enable real-time monitoring and early anomaly detection in chiller operations.

Main Methods:

• A 1D convolutional neural network (CNN) was employed for data fusion and prediction of four distinct control stages.
• The model was trained and evaluated using real-world data from a hospital building's chiller system.

Main Results:

• The proposed 1D CNN model demonstrated satisfactory performance and acceptable training times compared to existing methods.
• The model successfully predicted control states for screw compressors not included in the training set.
• Simulated failure cases showed the model's capability for early alarm detection through continuous misclassification.

Conclusions:

• Deep learning, specifically 1D CNNs, provides an effective approach for optimizing chiller performance and detecting anomalies.
• The developed model offers a robust solution for real-time monitoring and predictive maintenance in building thermal regulation systems.
• The model's generalizability across different screw compressors enhances its practical applicability in diverse building environments.