Fatigue damage prognosis of orthotropic steel deck based on data-driven LSTM

Deng, Peng-hao; Cui, Chuang; Cheng, Zhenyu; Zhang, Qinghua; Bu, Yizhi

doi:10.1016/j.jcsr.2023.107777

Cited by 7 publications

(2 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In a case study of the application of the data-based model, it was demonstrated that increased temperature and high traffic levels together had a significant negative impact on fatigue dependability, with a commonly used limit being achieved up to 40 years sooner than under the 'no change' baseline scheme. Nevertheless, the LSTM method performs well, as evidenced by validation results from the fatigue damage prognosis of an OSD, where the mean percentage absolute errors of the two types of fatigue susceptible features were less than 10% [86].…”

Section: Resultsmentioning

confidence: 97%

“…This ANN was then verified using a test set before being used to forecast the structural damage in an offshore jacket structure. Deng et al [86] performed a fatigue damage projection in OSDs with Long Short-Term Memory (LSTM). It was discovered that data-driven LSTM may greatly maintain the precision of fatigue damage predictions over time.…”

Section: Neural Network Based Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Advances in Machine Learning Techniques Used in Fatigue Life Prediction of Welded Structures

Gbagba,

Maccioni,

Concli

2023

Applied Sciences

View full text Add to dashboard Cite

In the shipbuilding, construction, automotive, and aerospace industries, welding is still a crucial manufacturing process because it can be utilized to create massive, intricate structures with exact dimensional specifications. These kinds of structures are essential for urbanization considering they are used in applications such as tanks, ships, and bridges. However, one of the most important types of structural damage in welding continues to be fatigue. Therefore, it is necessary to take this phenomenon into account when designing and to assess it while a structure is in use. Although traditional methodologies including strain life, linear elastic fracture mechanics, and stress-based procedures are useful for diagnosing fatigue failures, these techniques are typically geometry restricted, require a lot of computing time, are not self-improving, and have limited automation capabilities. Meanwhile, following the conception of machine learning, which can swiftly discover failure trends, cut costs, and time while also paving the way for automation, many damage problems have shown promise in receiving exceptional solutions. This study seeks to provide a thorough overview of how algorithms of machine learning are utilized to forecast the life span of structures joined with welding. It will also go through their drawbacks and advantages. Specifically, the perspectives examined are from the views of the material type, application, welding method, input parameters, and output parameters. It is seen that input parameters such as arc voltage, welding speed, stress intensity factor range, crack growth parameters, stress histories, thickness, and nugget size influence output parameters in the manner of residual stress, number of cycles to failure, impact strength, and stress concentration factors, amongst others. Steel (including high strength steel and stainless steel) accounted for the highest frequency of material usage, while bridges were the most desired area of application. Meanwhile, the predominant taxonomy of machine learning was the random/hybrid-based type. Thus, the selection of the most appropriate and reliable algorithm for any requisite matter in this area could ultimately be determined, opening new research and development opportunities for automation, testing, structural integrity, structural health monitoring, and damage-tolerant design of welded structures.

show abstract

Section: Resultsmentioning

confidence: 97%

Section: Neural Network Based Methodsmentioning

confidence: 99%