Thesis Integración de machine learning para la interpretación y explicación de los sistemas de control en DCS
Loading...
Date
2026-03
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Electrónica
Departament
Campus
Campus Casa Central Valparaíso
Abstract
La creciente complejidad de los sistemas de control distribuido (DCS) en la industria minera moderna ha generado una brecha significativa entre la lógica de control realmente implementada en los controladores y la documentación técnica disponible para su comprensión y mantenimiento. Esta problemática impacta directamente en los tiempos de diagnóstico, la seguridad operacional y los costos asociados a detenciones no programadas, especialmente en operaciones de gran escala como la Compañía Minera Doña Inés de Collahuasi (CMDIC). El presente trabajo de titulación propone el diseño e implementación de un asistente virtual inteligente, denominado Agente DCS, orientado a la interpretación y explicación de la lógica de control en sistemas DCS, utilizando modelos de lenguaje grandes (LLMs) ejecutados de forma local y técnicas avanzadas de extracción de información estructurada. La solución se enfoca específicamente en archivos de exportación FHX de la plataforma Emerson DeltaV, los cuales son transformados desde su representación textual jerárquica hacia un modelo basado en grafos que permite capturar explícitamente las relaciones lógicas entre bloques funcionales, señales y enclavamientos. Sobre esta base estructurada, se implementa una arquitectura de recuperación de información y generación aumentada (RAG), que permite realizar consultas en lenguaje natural y generar explicaciones técnicas coherentes, actualizadas y contextualizadas, sin requerir acceso directo a las estaciones de ingeniería ni a infraestructuras externas de cómputo. El sistema ha sido diseñado para(...).
The increasing complexity of Distributed Control Systems (DCS) in the modern mining industry has generated a significant gap between the control logic implemented in controllers and the technical documentation available for its understanding and maintenance. This gap directly affects fault diagnosis times, operational safety, and costs associated with unplanned downtime, particularly in large-scale operations such as Compañía Minera Doña Inés de Collahuasi (CMDIC). This thesis presents the design and implementation of an intelligent virtual assistant, named DCS Agent, aimed at interpreting and explaining control logic in DCS environments using locally deployed Large Language Models (LLMs) and advanced structured information extraction techniques. The proposed solution focuses on FHX export files from the Emerson DeltaV platform, which are transformed from their hierarchical textual representation into a graph-based model that explicitly captures logical relationships among functional blocks, signals, and interlocks. Based on this structured representation, a Retrieval-Augmented Generation (RAG) architecture is implemented, enabling natural language queries and the generation of coherent, up-to-date, and context-aware technical explanations without requiring direct access to engineering workstations or external computing infrastructures. The system operate(...).
The increasing complexity of Distributed Control Systems (DCS) in the modern mining industry has generated a significant gap between the control logic implemented in controllers and the technical documentation available for its understanding and maintenance. This gap directly affects fault diagnosis times, operational safety, and costs associated with unplanned downtime, particularly in large-scale operations such as Compañía Minera Doña Inés de Collahuasi (CMDIC). This thesis presents the design and implementation of an intelligent virtual assistant, named DCS Agent, aimed at interpreting and explaining control logic in DCS environments using locally deployed Large Language Models (LLMs) and advanced structured information extraction techniques. The proposed solution focuses on FHX export files from the Emerson DeltaV platform, which are transformed from their hierarchical textual representation into a graph-based model that explicitly captures logical relationships among functional blocks, signals, and interlocks. Based on this structured representation, a Retrieval-Augmented Generation (RAG) architecture is implemented, enabling natural language queries and the generation of coherent, up-to-date, and context-aware technical explanations without requiring direct access to engineering workstations or external computing infrastructures. The system operate(...).
Description
Keywords
Inteligencia artificial, Sistemas de control distribuido, Modelos de lenguaje grandes, Recuperación aumentada por generación, Automatización industrial
