YEAR 2024 / VOL. 1 / NO. 2
Internet of Things
Published on December 01, 2024
5 articles
Anomaly Detection in Smart Building Sensor Data Using Autoencoders
By Jan Novák
Abstract
A convolutional autoencoder is trained on 18 months of HVAC and occupancy sensor data from a university building to detect equipment faults and occupancy anomalies. The model achieves a precision of 89% with a false-positive rate under 3%.Digital Twin Architecture for Predictive Maintenance in Manufacturing
By Jan Novák
Abstract
We design a digital twin framework that ingests real-time sensor streams from a production line, models machine degradation using a physics-informed neural network, and triggers maintenance alerts 72 hours before predicted failure. A 14-month pilot at an automotive supplier shows a 40% reduction in unplanned downtime.Energy-Efficient Routing Protocols for Industrial IoT Networks
By Jan Novák
Abstract
We propose LEACH-E, an extension of the classic LEACH clustering protocol that incorporates residual energy and link quality metrics for cluster-head selection. Simulations on MATLAB show a 28% increase in network lifetime compared to LEACH on a 100-node topology.Federated Intrusion Detection for Edge-Connected Smart Grids
By Jan Novák
Abstract
Smart grid edge nodes are vulnerable to coordinated attacks that evade centralised IDS by staying below individual-node thresholds. We implement a federated IDS where each node trains a local LSTM on its traffic patterns and contributes model updates to a global aggregator, achieving 96.4% detection rate with no central data collection.Lightweight TLS Implementation for Constrained IoT Devices
By Jan Novák