Integration of Piezoceramic IoT Monitoring Systems with Amazon SageMaker for Predictive Analysis

Abstract

The article addresses the scientific and practical problem of transitioning from reactive monitoring to predictive analysis for IoT systems based on highly sensitive piezoceramic sensors. Unlike solutions focused solely on data collection and storage, this work proposes a comprehensive solution ensuring deep integration with the Amazon SageMaker machine learning platform. The proposed approach is based on three parallel flows: the Data Ingestion Flow for creating a «data lake» in Amazon S3, the Model Training Flow using AWS Glue for feature engineering, and the real-time Inference Flow based on SageMaker Endpoints. The methodology for constructing the machine learning components is described, including the critical stage of Feature Engineering, which transforms complex «raw» piezosensor signals (e.g., RMS, FFT) into analysis-ready data. The application of algorithms is analyzed, particularly the Random Cut Forest (RCF) for unsupervised anomaly detection and XGBoost for supervised fault classification. The architecture’s effectiveness was experimentally validated on a testbed within the AWS Academy Learning Lab environment, simulating both systemic (wear-and-tear) and sudden shock anomalies. The RCF model demonstrated high performance, achieving 91% precision and 88% recall in fault detection. The results demonstrate the suitability of the proposed solution for building scalable predictive diagnostics systems