In order to replenish the oil in the oil tank in time and reduce the loss caused by insufficient oil volume, we use wind turbines as an example to build a 1,300 oil tank picture dataset containing different oil levels. And the YOLOv5 network model based on the PyTorch framework trains the relevant dataset, and the oil tank and oil are detected through the training model to identify the oil volume of the oil tank, the model effect can meet industrial applications. The experimental results show that the YOLOv5 model established in this paper is 96% of the average accuracy of oil level recognition, which effectively solves the problem that the oil deficiency of oil tanks cannot be found in time.
The cost of operation and maintenance(O&M) has been one of the central budgets in the wind turbines' life cycle. In the final stage of the O&M work, the administrators must manually review the on-site photos to ensure the O&M is qualified, which is time-consuming and ineffective. To improve the efficiency and quality of O&M reviewing work while reducing its costs, we propose an auxiliary reviewing system to optimize the reviewing process. Our system architecture consists of data collection, analysis, and presentation modules. During day times, the data collection module will handle storing and organizing the photos of O&M. The data analysis module will perform duplicate image detection using the MessageDigest Algorithm 5(MD5) and missed image detections through the pre-trained ResNet50 deep learning model. The review results from the analysis module will be fully updated to the database after the analyzing process and rendered into graphs or tables on the webpage when required by the reviewers. Analyses are done in this paper to visualize the functions of each module of the proposed system. The experiment evaluates the performance of three deep learning models, including AlexNet, VGG16, and ResNet50, based on authentic on-site inspection images data from our "first annual inspection" dataset to determine which model yields the best image classification performance. The experimental result reveals that ResNet50 can reach the highest accuracy at 96.2% on the test dataset. Thus, we choose ResNet50 to train the image classifier of our data analysis module.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.