Green Edge AI: Constant-Time Memory Stabilization and Anomaly Preservation for Industrial Cyber-Physical Systems

Abstract

Continuous ingestion of high-dimensional telemetry forces predictive networks into a severe space-time trade-off, where conventional outlier rejection algorithms dilute mathematical gradients with benign idle data and induce thermal throttling on local hardware. This study introduces an $\mathcal{O}(d)$ ternary quantized framework that autonomously segregates sequential telemetry into redundant cache, degradation envelopes, and adversarial noise using parameter-free interquartile scaling and peak density thresholds. Empirical evaluations utilizing empirical machining datasets verify that topological isolation definitively eliminates gradient explosions inherent to raw data ingestion. The architecture achieves empirical risk minimization four times faster than robust Z-score baselines by discarding uninformative payload mass, while entropy evaluations prove anomalous amputations induce negligible structural information loss. Hardware simulations validate an absolute thermodynamic equilibrium, entirely preventing processor thermal throttling. Isolating pure physical degradation signatures from bulk transmission fundamentally dismantles the dependency on logarithmic cloud-indexing for automated diagnostics. Resolving this computational bottleneck establishes a thermodynamically sustainable foundation for decentralized operations, strictly fulfilling global mandates for energy-efficient computing infrastructure without compromising predictive integrity.