This paper presents an innovative approach to creating an integrated neural network system that combines the functions of adaptive control and fault diagnosis of permanent magnet synchronous motors (PMSM). Unlike traditional methods that consider control and diagnostic tasks in isolation, the proposed architecture is based on a common latent representation of the motor state, which provides synergy between subsystems and optimization of computational resources. The developed system includes a multi-level signal analysis mechanism that combines the advantages of spectral and temporal analysis using convolutional and recurrent neural networks. A key feature of the proposed approach is the adaptation of the control strategy depending on the detected anomalies, which minimizes the negative impact of incipient faults on drive operation. Experimental studies demonstrated a significant improvement in diagnostic accuracy compared to traditional methods. The proposed approach is particularly valuable for robotic systems and electric transport, where the reliability and fault tolerance of electric drives are critical for safe and efficient operation.