In view of the development of electric drives, they manage the operation of motors, in addition to ensuring the properties of the drive, and perform the function of monitoring the machine's technical condition. In the case of popular industrial applications induction motors, electrical circuit damages are more than half of all appearing faults. In connection with the above, the task of early detection of defects becomes a priority in drive systems. Increasingly, the diagnosis of electric motors uses artificial intelligence techniques, in particular, neural networks in the form of classic or deep structures. However, providing useful functions requires the development of many diagnostic patterns that carry information about the technical condition of the machine. Therefore, expanding the scope of the system to include new types of defects requires a reimplementation of the neural structure. The solution to the problem of the universality of features is the use of transfer learning. To demonstrate the advantages of transfer learning, a fault diagnostic system for stator windings and rotor cage bars of an induction motor was developed. The developed system was based on direct analysis of the axial flux signal, bypassing the classical methods of symptom extraction. Particularly noteworthy is the fact that the system can detect two types of defect based on the symptoms acquired for one type of defect. Verification was carried out in the steady and transient states for the full range of load torque. Analysis of the detection of rotor cage bar defects in the absence of load is of particular importance due to the absence of the motor slip parameter, which limits the use of classical diagnostic methods. In addition, thanks to the use of direct signal processing by a convolutional neural network, it was possible to repeatedly reduce the response time to an emerging defect.