Providing more reliable warnings based on monitored Partial discharges (PDs) is one of the main issues in PD diagnostics at this time. Due to noise, sometimes it is difficult to recognize PD activities from a circuit with huge noise or noise may be interpreted as PDs by current operational knowledge rules. Service quality is not optimal due to those reasons. To solve these issues and give a better translation of the PD activities measured by SCG, this paper focus on creating a better approach of detecting and analysis PDs by improving current operational knowledge rules. The first part of this paper explores a new model for unit blocks in PD raw data preprocessing stage, which is able to enhance the difference of the output of a statistical analysis of a PD chart to distinguish between areas with events from real PDs and areas strongly impeded by noise. The second part of this paper focusses on implementing artificial neural network (ANN) to detect real PDs. The performance analysis in the end of this part indicates a good reliability of this method. Subsequently the output of ANNs is combined with Weibull fit curve fitting to analyze the risk of detected PDs. The data were obtained from historic data of failures/defects, measured with the Smart Cable Guard, a PD detection and location system available for medium-voltage cables.