Damage Identification For Weld In Members Of Steel Structure Based On Neural Networks Method

Structures are always subjected to various adverse factors during its long-time service, such as fatigue, environmental corrosion, material aging, etc. These factors will bring to structural damage accumulation. Because the accumulate damage cannot detected conveniently, structural safety will be threatened severely. Therefore, it is very important for the research on structural damage automatic identification.The issue on identifying structural weld damage in the members of steel structures automatically is investigated intensively in present paper, by using neural networks method and referencing the related theory on structural strain mode analysis. In view of there is notable difference between weld line and main structural material on microstructure, at the same time, its mechanical capability is affected by manufactured techniques greatly and its mechanical status is more complex. Accordingly, it is more possible for weld line to be damaged. However, the issue on identifying weld damage in steel structures automatically is seldom discussed during the existing research by far.The main content of the paper is as follows:(1) The primary theory and characteristic on various structural damage identification methods based on vibrational parameters is reviewed respectively. It is pointed that the parameters based strain is more sensitive to structural local minute damage, so it is very ideal signature for structural weld damage identification.(2) Some related theory on neural networks technique is introduced briefly, then the principle and realized process on structural damage identification based on neural networks method is expounded.(3) The strain mode based approach to identify weld damage in members of steel structure automatically is demonstrated in present paper. By using the method based on structural vibration characteristics and neural network methods, the approach to locate single and double weld damage in a simply support steel beam with open I-formed is investigated respectively, and the effective measure to resisting noise distribution for networks is presented. It is shown that the method combination neural networks and strain mode difference results in a satisfactoryidentification consequence to weld damage, and trained-over networks has a strong capability to resist distribution to the test samples contaminated by noise.(4) The intelligent inspect system for roof latticed truss structure in Shenzhen citizen center is selected as a practice background, and the intelligent inspect method to the steel bracket of roof truss is researched in present paper, in which includes several tasks as follows: automatic load identification for bracket based on neural network method, prompt calculation for the stress of key points on bracket, and its service status evaluation and stress alarm. The simulated result shows that the proposed method can be applied to actual engineering availably.