Traditional fault diagnosis methods struggle to effectively process the nonlinear and non-stationary vibration signals of motor bearings under complex operating conditions. These limitations often result in reduced diagnostic accuracy and weak generalization performance. To address this issue, this paper proposes a cascaded fault diagnosis model that integrates Empirical Mode Decomposition (EMD), a one-dimensional Convolutional Neural Network (1D-CNN), and a Transformer Encoder. First, EMD is applied to decompose the raw vibration signal into multiple Intrinsic Mode Functions (IMFs), which are then reconstructed using residual techniques to form multi-channel time-frequency representations. These representations are fed into a two-layer 1D-CNN to extract local multi-scale features, followed by a two-layer Transformer Encoder to capture long-range temporal dependencies, thereby enabling efficient fault classification. Experimental results on the CWRU bearing dataset demonstrate that the proposed model achieves an average classification accuracy of 99.66%, outperforming conventional deep learning methods. This confirms the effectiveness of the integrated EMD, 1D-CNN, and Transformer architecture in enhancing feature extraction and classification of complex signals. Furthermore, grounded in TRIZ theory, the paper analyzes future trends in deep learning model development, particularly in terms of structural integration, intelligent evolution, and improved generalization capabilities.