Challenge
In automotive paint shops, ventilation is crucial for removing harmful chemicals and vapors, ensuring a safe work environment, and preventing defects on painted surfaces. Motors in ventilation units are prone to failure due to high heat, contaminants, and vibration during painting operations. Developing an AI model to predict anomalies in ventilation motors before they occur is essential for proactive maintenance and implementing this approach across multiple facilities.
Approach
Sensor data, including vibration and temperature, was loaded for the minimum duration required for successful exploratory data analysis (EDA). Domain-specific feature engineering was performed to develop an autoencoder-based deep learning model. This model was trained on a large amount of normal operational data using a semi-supervised learning approach. The trained model was deployed as a streaming service using an AI platform (Runway) to provide real-time anomaly detection, enabling factory managers to respond quickly to any anomalies.
Value Delivered
The AI-driven model improved the accuracy of motor anomaly detection in ventilation systems by 20% compared to traditional rule-based alarm systems, minimizing false positives. Accurate anomaly detection reduced facility maintenance time by over 30%, enhancing operational efficiency and reliability.
