Fault detection in induction motors is a useful practice when some critical processes depend on good machines performance. This work proposes a new computational model for detecting oil whirl faults in induction motors using orbital patterns. Signal vibrations are measured and pre-processed in order to obtain a characteristic orbit that represents the motor condition where an oil whirl fault is present. Through an artificial neural network, the orbital patterns are classified according to the motor condition: good or faulty. Experimental results show a good performance for the proposed model, providing a new tool for recognizing problems in induction motors.