AI-DRIVEN DIAGNOSTICS AND DIGITAL TWIN FORECASTING FOR 1-WIRE DS18B20 SENSOR BUSES USING CRC-AWARE DATA QUALITY METRICS

This study introduces an AI-driven research platform designed for integrity-aware analytics of 1-Wire DS18B20 temperature sensor buses. The platform integrates ROM-level device identification, real-time acquisition through the Linux w1slave interface, and CRC-aware data quality metrics to explicitly capture measurement reliability. A dedicated quality layer quantifies bus integrity using CRC pass/fail rates, missing-value statistics, and noise-based sensor health scoring, enabling early detection of unstable devices and communication issues. Building on this foundation, a cognitive diagnostics module performs robust anomaly detection on prediction residuals using MAD-based z-scores and evaluates temporal degradation via drift assessment (Page–Hinkley), supplemented by short-window trend drift detection when limited data are available. For predictive modeling, the platform formulates a next-step forecasting problem that estimates the target sensor temperature from its previous state and synchronous peer-sensor readings, supporting both one-step prediction and recursive digital twin simulation. An ensemble of ridge regression and extreme learning machine (ELM) is employed to combine linear robustness with nonlinear representational power, while permutation-based feature importance provides interpretable sensor-level contributions to forecasts. The platform automatically generates structured logs, visual diagnostics, and reproducible reports, offering a practical and research-grade framework for dependable monitoring, forecasting, and fault analytics in low-cost 1-Wire sensor networks.

Перевод пока недоступен