Accelerating Induction Motor Modeling with AI-Based Reduced Order Models

📌 Introduction

Accurately predicting the performance of induction machines across varying operational conditions is essential for robust electric drive design—especially in automotive and industrial applications. However, full Finite Element Analysis (FEA) simulations, while accurate, can be extremely time-consuming and computationally heavy.

This article introduces a hybrid solution developed by the Altair team that combines FEA with Artificial Intelligence (AI) to create a Reduced Order Model (ROM) of an induction machine. This approach achieves both speed and accuracy, enabling rapid design iteration and performance evaluation.

💡 Why Traditional Methods Fall Short

While T-equivalent circuit models are widely known and documented, they require labor-intensive testing or simulations for parameter identification, especially across diverse frequency and voltage settings. Classic reduced models based on FEA often fail to generalize outside the specific conditions they are trained on.

Moreover, traditional use of AI in this field has mostly been limited to fault diagnosis and control—not for creating predictive models of machine behavior over a wide range of operating points.

🚀 Our Approach: AI-Powered Reduced Order Modeling

Our method uses data from FEA simulations of an inner-rotor, 4-pole induction machine (225 kW) under multiple voltage, frequency, and slip configurations. The machine features:

• 60 stator slots and 74 rotor bars
• M235-35A steel for lamination
• Rated for 430 Nm torque at 15,000 rpm

✳️ Key Features:

• Only 360 FEA simulations were needed—strategically selected using slip sensitivity analysis
• Extracted outputs include torque, losses, current, power factor, and reactive power
• Data used to train an AI model using Altair® romAI™ in Twin Activate

📈 Results: Performance & Accuracy

Once trained, the ROM was tested on over 6,000 new operating conditions. Results were compared against reference FEA data and showed excellent predictive accuracy:

This level of accuracy, with orders-of-magnitude lower computation time, makes the ROM suitable for integration into system-level simulations, control algorithm validation, or efficiency map generation.

📂 Access the Full Paper

For more technical details, including model architecture, training methods, and simulation parameters:

🔗 Read the full IEEE publication here: AI Reduced Order Model of Induction Machine (IEEE Xplore)