Sei sulla pagina 1di 4

PARAMETER IDENTIFICATION OF PLATE STRUCTURES USING SENSITIVITY ANALYSIS AND INVERSE STATIC APPROACH INCORPORATING DISPLACEMENT MATRIX ERROR

FUNCTION
D. Bandyopadhyay1, S. Ghosh2
1,2

Department of Construction Engineering, Jadavpur University, Kolkata 700098, India

Damage or stiffness changes in structures at local scale are very common, requiring accurate identification and estimation for ensuring safety, usefulness and restoration. Damage Identification is commonly undertaken using inverse dynamic methods. However, most of these dynamic identification techniques require sophisticated instrumentation and intricate mathematical treatment. Static method seems to be an alternative and attractive proposition being simple and feasible for practical implementation. A two stage analysis method adopting sensitive MDOF selection and damage identification based on inverse analysis using a displacement based error estimating function is proposed. The method is demonstrated to identify the structural parameters in plate structures. Numerical results efficiently predicted the simulated damage by the proposed method. These results have been compared with previous work and found greater accuracy. Keywords: plate structures, sensitivity analysis, inverse statics, displacement matrix error function 1 Introduction Structural damage identification by NDT is a significant area of research in structural engineering. The reduction of structural stiffness resulted from changes in material and geometric properties are considered as damage, which is very common during the design life of a structure. Accurate identification and estimation of damage is essential for ensuring structural safety, future usefulness, economic and effective restoration. [1] Both static and dynamic methods have been used for damage detection. Most of the research approaches is pursued with dynamic measurements due to some inherent advantages. However, techniques based on static measurements maybe an alternative proposition being simple and economical. For a large, complex structure, static identification technique is constrained with the requirement of large amount of measurements. Also the effect of damage at a particular location may be concealed in the responses at another location due to the limitation of load paths. Nonetheless static measurements can be used to evaluate any changes in stiffness parameters [2, 3, 4, and 5]. There is a limited availability of literature on parameter identification by static responses. Earlier work [6] minimized the difference between the actual and analytical stiffness matrices by adjusting the stiffness elements and displacement measurements were
1 2

taken at certain locations and spline functions were used for evaluating displacements where measurements were not taken. In [7 and 8] stiffness was estimated using least square minimization of the difference between the analytical and measured stiffness matrices. The displacement degrees of freedom (DDOF) must coincide with the force degrees of freedom (FDOF), which does not seem to be feasible in practice. In [3], a combination of static and modal responses to evaluate stiffness change on an element to element basis was performed. A method for parameter identification using static condensation technique with incomplete measurements and force matrix based error estimator was successfully developed and applied to truss and frame structures [4]. However, measurements taken from arbitrary MDOFs may not lead to accurate estimation of parameters and hence proper selection of DOFs is essentially significant [1, 10]. A two stage analysis procedure incorporating sensitivity analysis for proper selection of MDOF and damage identification using static test data from the selected MDOF in frame structures was developed using force matrix based error estimator [10]. Further enhancement in formulation was carried out using a displacement matrix error estimator [1]. In this proposed paper the approach is extended to system identification in plate type structures.

Associate Professor, dban65@yahoo.com Student, swarnava.jukol@gmail.com

2 Theoretical Formulation 2.1 Sensitivity Analysis With applied set of force at certain DOFs, change in particular stiffness parameter Pi produce measurable changes in displacements only at some DOFs. Hence measurements from insensitive DOFs will not give accurate estimation of that Pi. Sensitivity analysis is a study on how the variation in output (displacements) of the model can be attributed to the variations in the inputs (forces and stiffness); determining the sensitive locations to be taken for measurement. Sensitivity analysis can be carried out on the model using forward problem analysis from the equation {f}=[K]{u} (1 )

can be partitioned into sets of sub matrices corresponding to measured and unmeasured DDOFs as:

(5)
Where [Ua] is the measured displacement matrix of size Number of measured displacement (NMD) X Number of Sets of Applied Forces (NSF), [Ub] is the measured displacement matrix of size Number of Unmeasured Displacement (NUD) X Number of Sets of Applied Forces (NSF) and [Fa] and [Fb] are the force matrices corresponding to [Ua] and [Ub] of sizes NMD X NSF and NUD X NSF respectively. From equation (5), eliminating [Ub] ; [Ua] is represented as: [Ua ]= [[Kaa ]-[Kab ] [Kbb ] -1[Kba ]] -1 [[Fa ]-[Kab ] [Kbb ] -1[Fb ]] (6) Sanayei & Onepede[4] used an error function based on force matrix. Here Displacement Matrix Error Function (DMEF) is used and which is the difference between analytical displacement matrix calculated from equation (6) and the measured output displacement matrix corresponding to measured DDOFs is given by: [E(p)]= [[Kaa]-[Kab][Kbb] -1 [Kba]] -1[[Fa]-[Kab] [Kbb] -1[Fb]]-[Ua] (7) If all the parameters are undamaged, [E(p)] is zero. [E(p)] is of size NMD X NSF and is a nonlinear function of the parameters p due to the inversion of the stiffness matrices. In order to linearize [E(p)] , it is vectorized by concatenating all its columns vertically producing an error vector {e(p)} of size NM x 1 where NM is the number of measurements and is numerically equal to NMD by NSF. The parameters can be adjusted by linearization using a first order Taylor series expansion of {e(p)}: {e(p+p)}={e(p)}+[S(p)]{p} Where [S(p)] is the sensitivity matrix given by (9) Evaluating [S(p)] requires the differentiation of the error matrix with respect to each parameter ( pj) to form [S*(pj) ] of size NMD x NSF (8)

Pre-multiplication by [K] -1 on both sides and expressing [K] as a function of stiffness parameters : [K(P1,P2,P3,,Pk,Pn)] -1{f} = {u} (2)

Sensitivity analysis for stiffness parameter Pi with application of jth set of force {fj } at respective FDOFs is performed by perturbing Pi by Pi and calculating the changed measurable displacement vector {u} from the static equilibrium equation (2) as: [K(P1,P2,P3,,Pi,+PiPn)] -1{fj}={u}+{u} (3)

The DOFs corresponding to which there are no measurable changes in displacement (u=0) are the insensitive degrees of freedom for stiffness parameter Pi and a particular set of force {fi}. Sensitivity analysis for Pi and {fi} should be carried out with various increments or decrements of Pi. 2.2 Parameter Identification Inverse analysis technique can be used to calculate any changes in the material properties (Parameters) of the structure from known sets of forces and measured sets of displacements. However it is impractical to measure displacements at all degrees of freedom in the entire structure for all sets of forces. Sanayei & Onepede [4] proposed a method for parameter identification using static condensation technique using sets of static forces at one subset of degrees of freedom (FDOF) and measuring displacements at another subset of degrees of freedom (DDOF). FDOFs and DDOFs may completely overlap, partially overlap or not overlap at all. The main advantage of this method is that displacement measurements need not be taken at all DOFs. For more than one set of forces, the force and displacement vectors can be concatenated horizontally in (1) and the resulting equation is: [F]=[K][U] (4)

Where [F] is the force matrix and [U] is the displacement matrix.[K] ,[U] and [F] in equation (4)

(10)
[S*(pj)] is vectorized similarly as [E(p)] into {S*(pj)} and is then evaluated for j=1 to NSF to form the full sensitivity matrix [S(p)] of size NM x NUP. The error function {e(p)} is reduced to a scalar performance error function J(p+p) as: J(p+p)={e(p+p)}T {e(p+p)} (11)

Change in stiffness can be identified by minimizing J(p+p) with respect to stiffness parameter of each element yields: [S(p)] T [{E(p)}+[S(p)]{p}]={0} (12)

Three cases arise out of equation (13). In first case, the number of independent measurements (NIM) is same as the number of unknown parameters (NUP). In such case direct inversion technique can be used to solve equation (13): {p}=-[S(p)] -1{E(p)}r (13a)

In second case, NUP is greater than NIM, the system is unsolvable. In the third, NUP is less than NIM, then [S(p)] is not a square matrix and least square method is used to solve equation (13): {p}=-[[S(p)] T [S(p)]] -1 [S(p)]{E(p)}r (13b)

and elastic modulus of 2000 unit force/unit length2. The plate is modelled considering fixed support on all sides. Finite element model is prepared by discretizing the problem geometry with four noded isoparametric type plate elements of size 2.5 by 2.5 unit length2 and are numbered from 1 to 64 (shown in Figure 1). Element number 21 (hatched in Figure 1) is damaged with 10% reduction of elastic modulus. Due to symmetric nature of the problem geometry, sensitivity analysis is performed only in the lower left quadrant (16 elements) in the present example. Figure 2 shows a particular insensitive case which will not be able to detect change in elastic modulus of element number 21. Figure 3 shows the sensitive case which will be able to detect change in elastic modulus of element number 21. Similar cases must be considered for all other cases of FDOFs, DDOFs and elements. The DDOFs which are capable of detecting changes in stiffness parameter of all elements and are also common to the maximum number of sets of forces are taken for the first set of measurements (marked as 1 in circle in Figure 1). Number of measured displacements (NMD) in such case is 4. Force of 1 unit is applied at each FDOF (marked as crosses in Figure 1) and the number of sets of forces (NSF) in this case was 40. For second case, NMD is increased to 8 and for third case NMD is further increased to 12 by taking additional set of displacement measurements from sensitive degrees of freedom in second case (marked as 2 in circle in Figure 1) and third case (marked as 3 in circle in Figure 1). These are used as inputs in the parameter identification algorithm and is evaluated for the three cases (NMD = 4, NMD = 8 and NMD = 12).

After evaluating the vector {p} from equation (13a) or (13b), an iterative process is used for parameter identification as: {p}r+1={p}r+{p} (14)

With the new parameters from equation (14), the parameter identification algorithm is repeated unless it converges. 3 Results and Discussion

Figure 2: Insensitive MDOF at 38 for FDOF at 11 (Damaged Element 21)

Figure 3: Sensitive MDOF at 31 for FDOF at 11 (Damaged Element 21) Figure 1: Geometry and FE model of Plate To illustrate the proposed method a numerical experiment is conducted with an isotropic square plate of side length 20 units, thickness of 0.25 unit lengths Parameter Identification algorithm using inverse analysis technique incorporating displacement based error estimator as discussed in the earlier section is used for analysis. Figure 4 and Figure 5 shows the variation of average percent error in parameter estimation and condition number of sensitivity matrix with increase in

number of measured displacements respectively. The present work is also compared with analysis performed using a force matrix based error estimator function published in [4]. Figure 4 shows that % error in estimation by inverse analysis with DMEF in the present work is always less than those estimated using force based error function in [4]. Also % error is observed to decrease from 0.2% at NMD= 4 to less than 0.01% at NMD=8 and NMD=12. From Figure 5, Condition Number of sensitivity matrix is also observed to decrease with increase in NMD. Number of digits of CN can be directly correlated with loss in accuracy in matrix inversion. Analysis with NMD=4, CN had 3 digits before decimal suggesting reduction in accuracy loss due to matrix inversion when NMD is increased from 4 to 8. With NMD=12, CN in present work was 3 digits but 2 digits in analysis with force based error estimator. Therefore when NMD is increased from 8 to 12, almost no change in accuracy loss due to inversion is observed. Table 1 shows the values of average numerical undamaged parameter (PN) and average numerical damaged parameter (PD) for the three cases of NMD. PN in all the three cases were equal to 2000. PD is also well estimated having very negligible error using the present analysis method. Comparison with analysis by algorithm in [4] shows that displacement matrix based error estimator function is much more accurate than force matrix based error function.

4 Conclusion Parameter identification using incomplete static measurement data from sensitive locations can be suitably applied for detecting structural damage (reduction in stiffness parameter) in plate type structures. Numerical experiments demonstrate that inverse analysis performed with input measurements taken from sensitive degrees of freedom gives accurate prediction of parameters. The proposed displacement based error estimating function lead to a much better estimation with very small percentage of inaccuracy when compared to analysis with force based estimator from earlier research. Additional measurements taken from other sensitive locations further reduce errors in identified parameters. Accelerated convergence of inverse algorithm is also achieved, facilitating reduction in computational cost with remarkable gain in accuracy. References [1] D. Bandyopadhyay and S. Roy, System Identification of Structures from Limited Measured Data Based on Inverse Static Approach, International Symposium on Engineering under Uncertainty: Safety Assessment and Management, January 4-6, 2012. [2] S.J. Clark, Member Stiffness Estimation in Linear Elastic Structures from Static and Vibrational Response, M.S. Thesis, UIUC, 1989. [3] P. Hajela and F.J. Soeiro, Structural Damage Detection Based on Static and Modal Analysis, AIAA Journal vol. 28,1990,pp. 1110-16 [4] M. Sanayei and O. Onipede, Damage Assessment of Structures Using Static Test Data AIAA Journal vol. 29, 1991, pp. 1174-79 [5] M. Sanayei and M.J. Saletnik,Parameter Estimation of Structures from Static Strain Measurements I: Formulation J.Struct. Engg, vol. 122, 1996, pp. 555-61 [6] Z. Sheena,A. Unger and A. Zalmanovich, Theoretical stiffness matrix correction by using static test resultsIsrael Journal of Technology, vol. 20, 1982,pp. 245-53 [7] M. Sanayei and S. Scamolpi, Structural element stiffness identification from static test data, Journal of Engineering Mechanics Division, ASCE, vol 117,1991(EM5) [8] M. Sanayei and R.B. Nelson,Identification of structural element stiffness from incomplete static test data.Soc. of Automotive Engg. Technical Paper,SAE-861783 [9] K.D. Hjelmstad,S.L. Wood and S.J. Clark, Mutual residual energy method for parameter estimation in structuresStructural Engineering, ASCE, vol 118(1), 1992, pp. 223-42 [10] D. Bandyopadhyay and S. Roy, Structural Damage Identification of Truss from Simple Static Test Data by Sensitivity Analysis, ICTACEM, 2010

Figure 4: Average percent error in estimation

Figure 5: Condition Number of Sensitivity Matrix Table 1: PD and PN for three cases Present Work Sanayei et.al. [4] PD PN PD PN 1791.6 2000.1 1771.8 1995.5 1791.6 2000.0 1775.7 1994.5 1791.5 2000.0 1775.5 1996.6

NMD 4 8 12

Potrebbero piacerti anche