Structural Analysis Based On Testing Data Mining And Cellular Automata

The traditional structural analysis techniques, such as the finite element method, study structural performances all based on the constitutive condition and mechanism, which are applied to all areas of engineering. However, an issue has existed for a long time, that is, the difference between the structural response caused by the variability of the structure itself and the corresponding analytical result has not been studied fully in theory. At present, the variability of a structure is generally treated as a random factor and then connected with some physical parameters of the structure. This method on dealing with variability can only reflect the variability of structure limitedly and one-sidedly. In other words, this treatment does not present the structural variability in basis of theory and mechanism and is only based on the experience with statistical parametric analysis. Hence, it is difficult for the FEA to predict the structural response of the masonry structure close to experimental ones.During experimental and theoretical analysis of masonry structure, it was found that structural variability is not completely caused by random factors, but it exists in certainty, which is affected by the structural geometrical property and boundary constraints or others. Therefore, an idea was inspired in this study: If it is possible that the latest intelligent techniques, such as data mining and cellular automata, are applied to model this certainty existed in variability, and then the model is back fed to the existing analytical methods, it can improve the analytical methods so that the predicted structural response is more accurate.First, a comparative analysis of their theoretical and experimental results is done to study the tendency and distribution of the comparative values for a large number of masonry wall panels inclusive of different configurations. From the study, a regularity emerges: for the various zones within the same wall panel or different wall panels with different configurations, if the relative positions of two local zones are similar, and the constraints governing the two zones are also similar, the ratio between theoretical and experimental values are basically the same; besides, the distribution patterns of the ratio values at local regions are also similar. Accordingly, this paper presents the concept of the structural local property, which describes the similarity of the relative positions of the zones and the structural constraints. In this way, a prototype of the structural local property is formed corresponding to data mining. Furthermore, the method is proposed to obtain the local property coefficient of other structural systems based on the data mining technology.Second, this paper defines the local property coefficient as the ratio of the displacement values of the finite element analysis to the corresponding experimental displacement values at measured points. The significance of coefficient is in the similar property relating to the similar local zones within structures, that is, two similar zones in different structures have the same local property coefficient. Therefore, the finite element model of masonry wall panel can be modified by the local property coefficients of all local zones. The way for the modification is: the global elastic modulus is multiplied by the local property coefficients of all zones, so that the elastic modulus of the structure in different zones is different; then, the modified elastic modulus is used in the FEA of the structure. By three sets of test wallets and wall panels, the vertically loaded wall panel, the laterally loaded wallette and the test of laterally loaded wall panel, the application of the local property coefficients in their FEA can make the analytical results closer to the experimental results, especially in the center zones of the wall panel。In this way, it obtains the various local property coefficients about test structures, through a simple data mining. This method can extract a part of"variance"of the structure which is considered random in the past analytical tasks; and this part of certainty of the structural"variance"property can reflect in the quantitative analysis of the structure and makes the analysis results more accurately embody the real situation.Third, the paper studies the property of the similar zone and the corresponding rule for matching zone similarity. As it is difficult to describe the regional states and to automatically match similar zones, this paper extends Zhou's research results, the cellular automata model of masonry wall panel, to other structural systems and gives out the new algorithm of the zone state value. In other words, this paper presents a general formula for calculating the zone state value of various structures, and establishes the rule for matching similar zone by clustering analysis. Meanwhile, in order to discriminate the scope of rule for matching zone similarity, the concept of structural similarity level is proposed and the corresponding formula is given to compute the similarity level. Structural similarity level quantifies the accuracy of the matching results.Finally, it verifies the above new concept and new method by case study. Three sets of local property coefficients are applied to modify the corresponding finite element models. Besides, taking the fourth set of the test panel named Panel SB06 as the base panel, the FEA of the wall panel is done to predict the different deformation modes of other unknown panels. As the geometrical size of predicted panel different from that of the base panel, it is necessary to amend the initial values of the transfer function and transfer coefficients, in application of cellular automata model. This paper presents two formulas for modifying initial value in the directions of length and height. In the case of the constant boundary initial value and the transfer coefficient of the base panel, both values of the predicted panel are amended using the two formulas and then the similar zones between the two panels are matched using the proposed rule; thus, the deformation of unknown panel can be predicted accurately.The method that intelligent techniques are used to conduct data mining of local property of structures, to match zone similarity and improve the FEA model, can predict not only a deformation mode, but also a set of deformation modes by adjusting the values of the transition efficient.