Undetected partial discharges (PDs) are a safety critical issue in high voltage (HV) gas-insulated systems (GIS). While the diagnosis of PDs under AC voltage is well-established, the analysis of PDs under DC voltage remains an active research field. A key focus of these investigations is the classification of different PD sources to enable subsequent sophisticated analysis. In this paper, we present an analysis of a 1D-CNN-based approach for classifying laboratory PD signals caused by metallic protrusions and conductive particles on the insulator of HVDC GIS, under both negative and positive potentials. Most notably, our study demonstrates that this type of neural network, regardless of the training order, can generalize learnings to operating voltage multiples that it has not previously encountered. We evaluate this generalization performance under the presence of additional white Gaussian noise and investigate the influence of excluding the amplitude-related information in the signal. Further, we compare the network's performance when using input signals in both the time and frequency domain.