The condition monitoring and early fault diagnosis of motor drive systems play a critical role in ensuring the reliability and availability of industrial robots operating in smart manufacturing environments. Motor drives in robotic applications are exposed to nonlinear dynamics, variable mechanical loads, and progressive degradation, which significantly limit the effectiveness of conventional model-based fault detection and diagnosis (FDD) methods due to parameter uncertainty and unmodeled dynamics. This paper proposes a data-driven fault detection framework based on a hybrid Convolutional Neural Network–Recurrent Neural Network (CNN–RNN) observer for the continuous condition monitoring of motor drive control systems in industrial robots. The CNN component enables the automatic extraction of fault-sensitive features from multichannel sensor signals, while the RNN component captures temporal dependencies associated with fault evolution and degradation processes. The observer-based structure allows residual-like information to be implicitly learned from operational data without requiring an explicit analytical model of the system. The esults demonstrate that the proposed CNN–RNN observer achieves a fault detection accuracy of 98.4%, outperforming conventional observer-based diagnostic approaches by 12.5%. Moreover, incipient faults are reliably detected up to 350 operating hours prior to critical failure while maintaining a false-positive rate below 1.2%. These results confirm the effectiveness of deep learning-based observers as a practical solution for the condition monitoring and predictive maintenance of motor drive systems in industrial robotic applications.