AI Summary & Key Takeaways
AI InsightIt offers a practical alternative to traditional FEA for scenarios requiring fast and repeated evaluations across varying geometries.
📌 3 Key Takeaways
Abstract
In high-performance engineering systems such as turbomachinery and aerospace structures, accurately predicting structural dynamic properties like natural frequency is critical for avoiding resonance, ensuring fatigue life, and maintaining rotordynamic stability. While finite element analysis (FEA) offers high accuracy, it becomes computationally expensive when repeated across a broad design space defined by geometric variations. To address this, a deep learning-based surrogate model was developed to predict natural frequencies from basic geometric parameters such as length, thickness, and angle of attack. A dataset of 21 geometrically distinct configurations was created using coarse-mesh FEA simulations, each taking approximately 30 seconds. Twenty samples were used for training a deep neural network (DNN), with one sample reserved for validation. The trained model achieved prediction times around 1 millisecond and relative errors of 1.1–3.3% within the training range and 0.1–16.5% outside it. This represents a drastic reduction in computational cost while maintaining acceptable accuracy. The proposed method enables rapid design iterations, sensitivity analysis, and early-stage decision-making in structural dynamics. It offers a practical alternative to traditional FEA for scenarios requiring fast and repeated evaluations across varying geometries.