Detection of Steel Wire Rope Surface Defects On Elevators: A Transfer Learning Approach

Elevator safety critically depends on the condition of steel wire ropes, whose failure can lead to catastrophic incidents. Conventional inspection methods rely heavily on manual visual checks conducted infrequently, making early defect detection challenging. This study proposes an automated defect detection system utilizing deep learning and transfer learning to enhance the reliability and efficiency of steel rope inspections. A dataset comprising 2,000 images—categorized into four classes: normal, broken, worn/fatigued, and rusted— was collected from multiple elevator installations and annotated by industry experts. The Inception V3 Convolutional Neural Network (CNN) model, known for its high accuracy in image recognition, was employed using transfer learning. Image preprocessing included resizing, background standardization, and data augmentation to improve model generalization. Several configurations of hyperparameters were tested, including variations in learning rates, batch sizes, and training epochs. The optimal model was achieved using a learning rate of 0.00001, a batch size of 32, and 50 epochs, resulting in a classification accuracy of 90%. Evaluation metrics, including precision, recall, and F1 score, confirm the model’s robustness and generalization capability. The system effectively distinguishes between different types of surface defects, providing actionable insights for predictive maintenance. By reducing reliance on manual inspections and offering early detection, this approach significantly contributes to elevator safety. This research demonstrates the potential of integrating deep learning and transfer learning in industrial safety applications. Future improvements could include real-time implementation, integration with IoT-enabled monitoring systems, and expanded datasets to further enhance accuracy and reliability in diverse operational conditions.