Duc Chinh Hoang

Prolonging the lifetime of wireless sensor networks are commonly achieved either through clustering paradigms or topology control. Under extreme deployment conditions, in which low node density deployment and sensor nodes have the finite communication range, these strategies exposed their weakness. Pure topology-control games often drain the energy quickly from the network connectivity, whereas clustering protocols extend lifetime at the cost of sink connectivity. This study proposes GT2, a two-stage game-theoretic framework that couples a mixed-strategy clustering game with a pure-strategy intra-cluster power-control game. Ten clustering protocols and topology control algorithms are tested across six deployment scenarios spanning free-space, target-region, and obstacle fields. Across the six scenarios, GT2 sustains the network lifetime above ten thousand rounds while more than doubling the packet-delivery ratio of the best competing clustering protocol in most cases. The simulation experiment shows that GT2 works to balance the two competing objectives instead of favoring just one, thereby prolonging network lifetime while delivering the highest reliability among clustering algorithms.

Multi-sensor vibration monitoring is important for industrial electric motors operating within coupleddrivetrains (e.g., motor–pump assemblies) because faults affect not only local vibration energy but alsotransmission relationships between measurement locations. However, many existing approaches eitherrely on isolated handcrafted features or use deep models that treat sensor channels as independentstacked inputs, without explicitly modeling inter-sensor coupling. To address this limitation, weintroduce a relational spectral descriptor that represents each vibration window using compact node-level statistics together with pairwise descriptors derived from multi-band magnitude-squared coherenceand cross-spectral phase. Speed-aware frequency bands and coherence-based gating are incorporatedto preserve diagnostically meaningful coupling while suppressing weak interactions. The resultingrepresentation can be used by both vectorized and structured downstream learners, allowing thecontribution of the descriptor itself to be separated from that of the downstream learning architecture.We evaluate the proposed descriptor under a strict record-level anti-leak protocol on the public 4TUmotor–pump dataset and the Machinery Fault Database (MAFAULDA), a public multi-sensor vibrationdataset. Under in-distribution evaluation, the relational descriptor reaches 99.37% test macro-F1 in theauxiliary 4TU evaluation and 99.19% on MAFAULDA. Under a harder 4TU cross-folder protocol, theflat relational descriptor achieves 72.35% macro-F1, outperforming a raw-signal 1D-CNN baselineevaluated under the same protocol. These results indicate that explicit spectral coupling provides acompact, interpretable, and practically useful representation for multi-sensor industrial fault diagnosis.

To ensure safe and comprehensive power monitoring within equipment limits, Non-Intrusive Load Monitoring (NILM) is a leading solution due to its convenience and cost-effectiveness. Various NILM approaches exist, including statistical methods, non-electrical variable integration, and machine learning models. However, these methods often fail to fully capture load signal characteristics, particularly currentvoltage relationships. The Convolutional Neural Network (CNN) model is the most effective for monitoring and predicting load combinations using V-A (voltage-ampere) trajectory graphs. However, transient events can distort these trajectories, leading to incorrect predictions. This study enhances CNN performance by integrating time-series features to improve accuracy and handle switching events. Input data is structured as sequential images representing circuit states over time. A baseline model is designed, with multiple variations tested to analyze the correlation between parameters and accuracy. The study successfully applies a CNN model with time-series characteristics for load identification, even during switching events. Experimental results determine the most optimal model for practical applications.