Artificial Intelligence Experimental Environment For Structural Analysis

For a long time, all sorts of analytical theories and methodologies were constantly proposed and improved, aiming at analyzing the performance/behavior and response of structures precisely. These reseach achievements have resulted in increasingly powerful capacities of numerical simulation techniques such as finite element analysis technique. However, two issues have been challengingthe current analytical theories and methodologies. One is that both empirical formula and widely used FEA methods are all established on many a hypothesis, which causes natural errors between numerical simulation and actual behavior/response of structure. In many cases, these hypotheses even lead to the invalidation of numerical simulation when a structure is complex. The other is that a large quantity of experimental data accumulated over a long period of time is only applied in regression analysis and precision estimation of structural numerical simulation; thus, this results in enormous waste of abundant precious information about structural working performances/responses included in these data. Therefore,in order to avoid errors from hypothesis and enhance accuracy of structural analysis, it is needed to develop some new analytical methodologies that could predict structural behavior/response based on structural real behaviors/responses. For a full application of experimental data, more capable knowledge mining techniques should be developed .In order to addressthe issues elaborated above, this study proposes a new concept of Artificial Intelligent"Experimental Environment for Structural Analysis (AIEESA)"and esteblishes the corresponding system. The AIEESA consists of AITs, data mining techniques and experimental data, along with numerically modeling methodologies of structural configuration and a series of criteria for matching similarity of structural properties and mapping/predicting structural behavior /response. The experimental data is subjected to a process of data mining and then treated as the numerical modes suitable for AIT functions. When a new/unseen structural model is placed in this AIEESA, the behavior or response of the model, like its experimental expression, could be mapped/predicted based on the existing experimental data or the data from the site measurements. Each part of AIEESA is investigated after concept of AIEESA is illuminated.First of all, the data base in AIEESA is built as three parts: (1) Behavior of structure. Behavior of structure refers to failure patterns and failure loads of laterally loaded wall panels in this text; (2) Normalized behavior of structure. The normalization underlines themain features of failure patterns of laterally loadedwall panels with similar configurations and deletes some noise. (3) Structural response. Structural response in this text is the displacement values at the measuring points on experimental wall panels subjective toloading increment. Thus, original information resource of knowledge discovering is constituted in consideration of conveniently data mining.Secondly, two numerical modeling techniques in AIEESA are studied in this dissertation: one is the numerical modeling technique of failure pattern and the other is the numerical nodeling technique of structural configuration. A new concept of generalized wall panel is put forward to developthe numerical modeling technique of failure pattern. This concept enriches the special content of the similarity level and provides a quantitative method to compare the failure patterns of both base and new models. For the numerical modeling technique of structiral configuration, a CA modeling technique for masonry panels constrained at four edges and a dimensionless technique based on the conventional FEA are raised. The dimensional technique contributes more physical meaning of the numerical modeling of structural behavior in AIEESA.Thirdly, criterion forboth similar zone matching and behavior matching are investgated. Three weighted criteria for similar zone matching are proposed, and a comparison is made between applications effects of three criteria to find out the optimal criterion for panels with different configuration.Then, modeling methods of both global property variation and property variation with CA technique are studied. On one hand, global property variation could be described by transition coefficient in CA model. Meanwhile, a SVM model is built up to obtained optimal value of transition coefficient. Once the range of transition coefficient is determined, failure pattern of new model could be predictedd with AIEESA. On the other hand, this study also put forward that varying of initial value in CA model could reflects property variation properly.Moreover, three sorts of neutral networks: BP, RBF and RA are established to predict failure loads account of corresponding failure patterns, which enable function to predict failure loadsof AIEESA.Finally, a comparison is made between the a series of implementing results of AIEESA and corresponding experimental behavior/response to verify validation of AIEESA.In sum, the analytical system AIEESA built up in this study realizes predicting structural behavior/response directly from existing real structural behavior/response; and it also achieves mining out plenty knowledge contained in now existing experimental data. Therefore, the AIEESAnot only overcomes inherent defect of conventional analyzing technologies and methodologies, but alsoprovides a new way to mine out knowledge contained in existing experimental data and apply to analysis of new structures. Consequently, AIEESA provides researchers and engineers an effective structural analysis tool.