Image-Based Crack Detection Method for FPSO Module Support

• Published in: Buildings (Basel) Vol. 12; no. 8; p. 1147
• Main Authors: Su, Xin, Jia, Ziguang, Ma, Guangda, Qu, Chunxu, Dai, Tongtong, Ren, Liang
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01-08-2022
• Subjects: Accuracy; Algorithms; Artificial intelligence; Cracks; Deep learning; Feature maps; FPSO; MobileNetv2; Modules; Neural networks; Ocean engineering; Offshore engineering; Offshore equipment; PSPNet; semantic segmentation; Semantics; Stress concentration; Support vector machines
• ISSN: 2075-5309; 2075-5309
• DOI: 10.3390/buildings12081147

Floating Production Storage and Offloading (FPSO) is essential offshore equipment for developing offshore oil and gas. Due to the complex sea conditions, FPSOs will be subjected to long-term alternate loads under some circumstances. Thus, it is inevitable that small cracks occur in the upper part of the module pier. Those cracks may influence the structure’s safety evaluation. Therefore, this paper proposes a method for the FPSO module to support crack identification based on the PSPNet model. The main idea is to introduce an attention mechanism into the model with Mobilenetv2 as the backbone of the PSPNet, which can fuse multiple feature maps and increase context information. The detail feature loss caused by multiple convolutions and compressions in the original model was solved by applying the proposed method. Moreover, the attention mechanism is introduced to enhance the extraction of adequate information and suppress invalid information. The mPA value and MIoU value of the improved model increased by 2.4% and 1.8%, respectively, through verification on FPSO datasets.