Building Intelligent Digital Twin Models Using Neural Networks

This whitepaper presents a framework for building intelligent digital twin models using neural networks within Simio software. It outlines the capabilities of Simio as a discrete-event-based digital twin simulation tool that integrates AI features, enabling users to define and train regression neural networks without requiring external programming. The document details the process of generating synthetic training data, which addresses the challenge of obtaining labeled data for AI applications. It explains how neural networks can simplify complex decision-making in simulation models by predicting key performance indicators (KPIs) such as makespan. The paper also describes the relationship between neural network models and elements in Simio, emphasizing the automated recording of training data and the training process using TensorFlow. Additionally, it discusses the importance of separating training records into distinct datasets for effective model training and validation, along with strategies to prevent overfitting during the training phase.