Automated constructionmanagement platform with image analysis using deep learning neural networks

As expansion of the power system is required to match the increase in demand, it becomes necessary to build power substations and, consequently, electrical power feeders. During construction, it is essential that qualified inspectors carry out monitoring. However, the professionals who manually perform inspection might make errors in assessment, therefore an automated solution could help them in performing the task more accurately. The objective of this work is to propose an automated solution for monitoring the construction of feeders in electric power substations, based on deep learning techniques. This proposal aims to meet the growing demand of the energy industry, improving efficiency and reducing dependence on human inspectors. To achieve the proposed objective, cameras were installed in different electrical power substations to collect images from a real environment. Then, three object detection methods (Faster R-CNN, SSD, and YOLO) were evaluated with different convolutional neural network architectures. In the results, considering the mAP (mean Average Precision) evaluation metric for object detection, we could achieve a value of 0.920 for an @[IoU = 0.50] using the Faster R-CNN method with a Resnet-50, which was the best result of all the compared methods. During the evaluation of the proposed solution, we noticed the contribution of the system to the monitoring of feeder constructions in substations. The tool was able to automate the monitoring process, directly helping the inspectors and the company's managers.