Insulated Piercing Connector (IPC) Torque Prediction Using Random Forest (RF) Model and Anomaly Detection Using Isolation Forest for Low Voltage Overhead Distribution Networks

Efficient electrical connections are critical for maintaining the reliability of aerial bundled cable (ABC) systems, where Insulation Piercing Connectors (IPC) play a vital role. This study focuses on
predicting torque values in IPC installations using machine learning techniques to enhance mechanical performance and reduce failure risks. Three mechanical tests were conducted, including
Shearhead, Continuity, and Body tests, using IPC samples provided by TeknoBumi. The data were analyzed using Random Forest, Decision Tree, and Artificial Neural Network models.Data
preprocessing included normalization, and model performance was evaluated using R-squared, Mean Squared Error, Root Mean Squared Error, and Mean Percentage Error. Among the models,
Random Forest showed the highest accuracy across all tests. In addition, anomaly detection methods including Isolation Forest, One-Class Support Vector Machine, and Local Outlier Factor were
applied to identify abnormal torque values that could indicate potential installation issues or defects. Isolation Forest proved to be the most consistent and reliable method for detecting outliers. The findings of this study demonstrate that integrating AI-based predictive modeling and anomaly detection can support early identification of mechanical inconsistencies, improve IPC installation
quality, and enhance the overall reliability of low-voltage overhead distribution systems.