Design and Integration of a Scalable Anomaly Detection Microservice Using AWS SageMaker for IoT Sensor Data

Abstract

With rapid iteration of AI technology and the digitization of traditional industries nowadays, more and more advanced and convenient features for monitoring technologies are being developed. One important development is anomaly detection in condition monitoring. By applying this technique, condition monitoring system can analyze machinery vibration data in real time to detect potential malfunctions early. It then notifies the administrator, allowing for predictive maintenance, which reduced costs and prevents possible damage, even during the machine's runtime.

However, most existing research focuses only on algorithm performance under laboratory conditions. It often fails to address the complex challenges of integrating these methods into real, multi-tiered condition monitoring platforms. This creates a gap between theoretical models and practical industrial applications.

To address this gap, this work combines quantitative performance evaluation of random cut forest based anomaly detection with real world CM system to test reliability of proposed integration architecture.

By demonstrating seamless integration into an existing condition-monitoring pipeline and validating its performance under real-world load, this thesis will bring a reusable pattern and evidence for scalable, unsupervised anomaly detection solution in complex systems, transforming from theoretical models to real world application.