Asistente virtual basado en Inteligencia Artificial para la obtención de indicadores de desempeño

Abstract

CONDITION FOR PUBLICATION OF PROJECT. In the Ecuadorian industry, access to key performance indicators (KPIs) is often limited due to the use of manual records and isolated platforms, which cause delays in decision-making and increase operating costs. This project proposes the design of a web-based platform with a virtual assistant powered by artificial intelligence, capable of responding to queries about reliability and availability metrics while also predicting failures in critical equipment. The objective is to reduce consultation times and provide reliable real-time information to strengthen the plant’s strategic management. The system was developed by combining a centralized database, an intuitive web environment, and a natural language processing engine that interprets user questions. In addition, a Random Forest machine learning model was trained with historical operating data and integrated into the assistant to generate timely predictions. The results showed a reduction in the average consultation time from nearly one hour to eleven minutes, together with a reliable predictive performance, achieving F1=0.9262 and ROC AUC=0.9567. It is concluded that the solution is a viable alternative for the digitalization of industrial processes, enhancing efficiency, traceability, and productivity in manufacturing environments. Keywords: Industry 4.0, digitalization, machine learning, productivity