Condition monitoring of centrifugal pumps is vital due to their crucial role in industries. One of the most prevalent faults in pumps is cavitation, which can cause mechanical faults or even failure in the pump. In this paper, an approach is suggested to detect cavitation in a centrifugal pump using time-domain analysis of the pressure signal residual. First, pressure and torque signals are obtained using a model of the electro-pump, and then pressure deviation from the pump performance curve is defined as a residual. The residual time-domain features are extracted and applied as inputs to a self-organizing map (SOM) neural network to classify the system modes. The results indicate that the suggested method is capable of detecting incipient cavitation. Furthermore, the suggested method demonstrates robust performance against disturbance.