Diseño óptimo de redes de sensores inteligentes en procesos industriales: una estrategia híbrida de aprendizaje probabilístico y búsqueda local orientada a confiabilidad y valor de información

Abstract

Sensor network design is a strategic decision for digital industrial process operation because it determines which information will be available for monitoring, data reconciliation, advanced control, fault diagnosis, predictive maintenance, and digital twins. In current industrial scenarios, sensor selection cannot be reduced to an acquisition-cost minimization problem: it must account for reliability, degradation, maintenance, installation constraints, model uncertainty, life-cycle cost, and the economic value of information. This paper presents an optimal design methodology based on a hybrid PBIL-SOTS strategy. The proposal combines population-based probabilistic learning, which is useful for exploring diverse configurations, with Tabu Search and strategic oscillation, aimed at improving solutions close to the feasibility boundary. The evaluation function is reformulated to integrate precision, estimability, system-wide reliability, information contribution, and operational value of measurements. A reproducible validation protocol is also proposed, including statistical analysis, ablation study, parametric sensitivity, and parallel execution. The resulting approach understands the sensor network as an information infrastructure rather than only as instrumentation equipment.