As a critical component of rotating machinery, the operating status of rolling bearings is not only related to significant economic interests but also has a far-reaching impact on social security. Hence, ensuring an effective diagnosis of faults in rolling bearings is paramount in maintaining operational integrity. This paper proposes an intelligent bearing fault diagnosis method that improves classification accuracy using a stacked denoising autoencoder (SDAE) and adaptive hierarchical hybrid kernel extreme learning machine (AHHKELM). First, a hybrid kernel extreme learning machine (HKELM) is initially constructed, leveraging SDAE's deep network architecture for automatic feature extraction. The hybrid kernel functions address the limitations of single kernel functions by effectively capturing both linear and nonlinear patterns in the data. Subsequently, the hierarchical hybrid kernel extreme learning machine (HHKELM) is refined through an enhanced Aquila Optimizer (AO) algorithm, which iteratively optimizes the kernel hyperparameter combination. The AO algorithm is further enhanced by incorporating chaos mapping, implementing a refined balanced search strategy, and fine-tuning parameter [Formula: see text], which collectively improve its ability to escape local optima and conduct global searches, thus strengthening the robustness of the model during parameter optimization. Experimental results on the CWRU , MFPT and JNU datasets demonstrate that stacked denoising autoencoder-adaptive hierarchical hybrid kernel extreme learning machine (SDAE-AHHKELM) has better fault classification accuracy, robustness, and generalization than KELM and other methods.