Cooking-related fires and combustible-gas leaks remain recurring domestic hazards, while lights and ventilation fans are often left running in empty rooms. This paper presents the design and experimental validation of a low-cost retrofit IoT node that integrates occupancy-aware actuation with early smoke and gas monitoring under a safety-first policy. An Arduino UNO executes time-critical sensing and relay control, and an ESP8266 provides Wi-Fi connectivity and a lightweight smartphone interface. Occupancy is inferred using a passive infrared (PIR) sensor to gate a lamp and fan, while an MQ-2 module monitors smoke and combustible gases. The control logic is implemented as an event-driven state machine that prioritises safety events, enforces minimum on and off timing to suppress relay chattering, and stabilises the gas channel using clean-air baseline normalisation (R/R0) with hysteresis. Bench verification confirmed I/O mapping and electrical isolation via an opto-isolated relay stage, and repeated switching did not reveal relay instability under the prototype loads. Scenario trials in a two-zone mock-up demonstrated reliable manual overrides, motion-triggered actuation without oscillation, and consistent alert generation during staged smoke exposures. The results support feasibility for incremental residential retrofits and identify deployment priorities, including sensor drift management, power integrity, and installation practice.
Logic-based machine learning models such as the Tsetlin Machine (TM) have recently gained attention for their energy efficiency and inherent interpretability. However, existing TM-based architectures remain limited in their ability to perform hierarchical feature learning, adapt dynamically to task complexity, and process temporal data efficiently. This paper proposes the Adaptive Logic Learning Architecture (ALLA), a novel hierarchical and energy-aware logic learning framework that addresses these limitations through adaptive clause networks (ACNs), multi-layer logical composition, and TLUs. ALLA enables dynamic clause growth and pruning, supports hierarchical abstraction, and integrates temporal reasoning within a unified propositional logic framework. Experimental results across image classification and sequential recognition tasks show that ALLA improves accuracy over conventional TM models while maintaining substantially lower energy consumption than deep neural network baselines. Hardware synthesis results further confirm the suitability of ALLA for low-power and edge-intelligent systems.
Open Access
News 
Journal Metrics
Journal Overview
Associated Journals