摘要
Heterogeneous rack systems combining servers, storage arrays, and switches are common in modern data centers, yet hard disk drives (HDDs) within such configurations face performance degradation from complex structural and airflow-induced vibrations. This work proposes a slot-based, machine learning framework to predict HDD degradation risk and guide layout optimization. Each rack slot is characterized by a feature vector comprising modal frequency, local stiffness, airflow excitation, and vibrational coupling, estimated via representative simulations. Decision tree and neural network models are trained to predict IOPS degradation based on these descriptors. A reference configuration-switches at the top, servers in the middle, storage at the bottom-is used to demonstrate rack-level risk mapping. The framework supports explainability through SHAP analysis and enables layout recommendations that reduce slot-level exposure to high-risk conditions. The proposed method is modular, generalizable across rack designs, and suitable for deployment-oriented optimization of HDD placement strategies.