Power capacitors are essential reactive power compensation devices in distribution systems, improving power factor, enhancing voltage quality, and reducing feeder losses. However, factors such as internal defects, poor connections, and overload operation can lead to insulation degradation and capacitance deterioration. In this study, we propose a fault detection method based on a 1D convolutional neural network that directly classifies capacitor conditions using time-domain charging harmonic current signals without the need for additional feature extraction. Low-voltage charging harmonic currents were generated using a power testing system, and high-frequency current sensors together with an oscilloscope were employed to acquire the waveform data. The developed model effectively learns the intrinsic characteristics of the current signals and accurately identifies capacitor operating states. Experimental results showed that the proposed method achieves a degradation-fault detection accuracy of 97.78%, demonstrating its effectiveness and practical value for the condition monitoring and preventive maintenance of power capacitors.