Image based surface damage detection of renewable energy installations using a unified deep learning approach. (November 2021)

• Record Type: Journal Article
• Authors: Shihavuddin, ASM; Rashid, Mohammad Rifat Ahmmad; Maruf, Md Hasan; Hasan, Muhammad Abul; Haq, Mohammad Asif ul; Ashique, Ratil H.; Mansur, Ahmed Al
• Is Part Of: Energy reports. Volume 7(2021)
• Page Start: 4566
• Page End: 4576
• Publication Date: 2021-11
• Subjects: PV panel -- Wind turbine -- Structural health monitoring -- Deep learning -- Damage detection -- Drone inspection; Power resources -- Periodicals; Energy industries -- Periodicals; Power resources; Periodicals; Electronic journals; 621.04205
• DOI: 10.1016/j.egyr.2021.07.045
• Languages: English
• ISSNs: 2352-4847

Abstract

Abstract: Smart Grid technology as a platform can encompass several advanced technological features across the spectrum of the power system. Smart Grid may introduce an early warning system for structural damage on the surface of large-scale power utility facilities providing crucial information to maintain the safety of the plant. To achieve such a cost-effective structural health monitoring system, a holistic smart inspection implementation framework is required. Using recently available sophisticated inspection technologies, high-resolution images covering the structural condition of large infrastructure can be routinely acquired as a remote monitoring process. Automated analysis of these inspection images can significantly reduce the inspection cost, provide an effective detection mechanism, and shorten reporting time, as a result, reducing overall maintenance costs, and improving safety measures. In this work, we have applied state of art deep learning based inspection image analysis methods for surface damage detection of various renewable energy power plants with a single unified model. We have achieved the state-of-the-art accuracy of 0.79 mean average precision on average even where input images are of varied modalities: from thermal images to visual images, from high- to low-resolution images, and from PV panels to wind turbines. All variations have been tackled with one single deep learning model to detect surface damages. Our results demonstrate the promise of effectively deploying a single trained model to inspect a wide range of energy installations while reducing the monitoring cost significantly. In addition, this work also published the reported dataset comprising four specific image sets for the research community. Highlights: Utilization of drone images of PV and Wind installations for health inspection. Novel approach in PV and Wind turbine damage detection by object detection methods. Performance comparison of recent object detection techniques for damage detection. A trained model combining Infrared and drone imagery providing accurate results. A dataset of annotated damages in Solar and Wind power installations.