Desarrollo de algoritmo de detección de nivel de saturación para la determinación del estado de filtros en mascarillas industriales.
Loading...
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
La determinación del estado de saturación de filtros en mascarillas industriales se realiza mediante criterios subjetivos los cuales no reflejan el desgaste real del medio filtrante. Esta tesis propone el desarrollo y validación de un algoritmo capaz de estimar en tiempo real el nivel de saturación de filtros respiratorios a partir del análisis de señales respiratorias medidas dentro del respirador y se enmarca en el proyecto FONDEF código ID24i10487, titulado “Sistema de monitoreo de condiciones de respiradores industriales”.
Se realizó un análisis exploratorio y estadístico de las variables respiratorias, evidenciando que el proceso de saturación es progresivo, no lineal e irreversible. Los resultados mostraron que la presión respiratoria es el indicador más robusto y sensible frente a la degradación del filtro, mientras que la humedad cumple un rol complementario y la temperatura presenta una baja capacidad discriminante.
El problema fue formulado como una tarea de regresión supervisada para la estimación continua del nivel de saturación. Se evaluaron distintos modelos de regresión, seleccionándose un modelo basado en Extreme Gradient Boosting (XGBoost) por su mejor desempeño global y en la zona crítica de operación. El algoritmo incorpora ponderación de muestras, suavizado temporal y restricciones de monotonicidad, garantizando coherencia física en las estimaciones.
La validación experimental incluyó pruebas en tiempo real bajo condiciones promedio, saturación acelerada y pruebas de larga duración. En escenarios representativos de uso industrial, el sistema mostró una estimación estable y consistente con el ground truth, mientras que en escenarios extremos presentó una respuesta conservadora adecuada desde el punto de vista de seguridad. La evaluación como clasificador evidenció una alta sensibilidad (0.997) y un AUC cercano a 0.96, demostrando su capacidad para detectar oportunamente condiciones críticas.
The determination of the saturation state of filters in industrial respirators is commonly based on subjective criteria or usage time, which do not accurately reflect the actual degradation of the filtering medium. This thesis proposes the development and validation of an algorithm capable of estimating, in real time, the saturation level of respirator filters through the analysis of respiratory signals measured inside the respirator, within the framework of the FONDEF project ID24i10487, “Monitoring System for Industrial Respirator Conditions”. An exploratory and statistical analysis of the respiratory variables was conducted, revealing that the filter saturation process is progressive, non-linear, and irreversible. The results showed that respiratory pressure is the most robust and sensitive indicator of filter degradation, while humidity plays a complementary role and temperature exhibits limited discriminative capability. The problem was formulated as a supervised regression task aimed at continuous saturation estimation. Several regression models were evaluated, with an Extreme Gradient Boosting–based model selected due to its superior overall performance and accuracy in the critical operating region. The algorithm incorporates sample weighting, temporal smoothing, and monotonicity constraints to ensure physically consistent saturation estimates. Experimental validation included real-time tests under average conditions, accelerated saturation, and long-duration scenarios. Under representative industrial conditions, the system produced stable estimates consistent with the ground truth, while under extreme scenarios it exhibited a conservative response suitable from a safety perspective. When evaluated as a classifier, the system achieved high sensitivity (0.997) and an AUC close to 0.96, demonstrating its ability to reliably detect critical saturation conditions.
The determination of the saturation state of filters in industrial respirators is commonly based on subjective criteria or usage time, which do not accurately reflect the actual degradation of the filtering medium. This thesis proposes the development and validation of an algorithm capable of estimating, in real time, the saturation level of respirator filters through the analysis of respiratory signals measured inside the respirator, within the framework of the FONDEF project ID24i10487, “Monitoring System for Industrial Respirator Conditions”. An exploratory and statistical analysis of the respiratory variables was conducted, revealing that the filter saturation process is progressive, non-linear, and irreversible. The results showed that respiratory pressure is the most robust and sensitive indicator of filter degradation, while humidity plays a complementary role and temperature exhibits limited discriminative capability. The problem was formulated as a supervised regression task aimed at continuous saturation estimation. Several regression models were evaluated, with an Extreme Gradient Boosting–based model selected due to its superior overall performance and accuracy in the critical operating region. The algorithm incorporates sample weighting, temporal smoothing, and monotonicity constraints to ensure physically consistent saturation estimates. Experimental validation included real-time tests under average conditions, accelerated saturation, and long-duration scenarios. Under representative industrial conditions, the system produced stable estimates consistent with the ground truth, while under extreme scenarios it exhibited a conservative response suitable from a safety perspective. When evaluated as a classifier, the system achieved high sensitivity (0.997) and an AUC close to 0.96, demonstrating its ability to reliably detect critical saturation conditions.
Description
Tesis presentada para optar al grado de Magíster en Ciencias de la Ingeniería con mención en Ingeniería Eléctrica.
Keywords
Algoritmos, Filtros, Dispositivos de protección respiratoria, Industrias