Sunteți pe pagina 1din 11

Reliability Modeling Using Finite Element Response Surface Techniques

Robert C. Patev1, Randy M. James2, and David M. Schaaf3


U.S. Army Corps of Engineers, New England District, 696 Virginia Road, Concord, MA 01742 2 ANATECH Corp., 5435 Oberlin Drive, San Diego, CA 92121 U.S. Army Corps of Engineers, Louisville District, 611 Martin Luther King Boulevard, Louisville, KY
1

Abstract: The paper will discuss the use of ABAQUS in the development of response surface models for use in time-dependent reliability modeling of navigation structures. The paper will focus on how to utilize finite elements to assist in developing or defining the response failure surface for structural components. The concepts and methods of applying statistics to build response surface models will be addressed. The bridging of ABAQUS finite element modeling into the development of Monte Carlo Simulation programs to perform the reliability calculation will be discussed. Case studies of the overall methodology and modeling sequence will be highlighted and the application to the rehabilitation of navigation structures will be presented. Conclusions of the methods presented and recommendations for application to other engineering disciplines will be addressed.

1. Introduction
This paper describes the use of ABAQUS finite element models in conjunction with response surface modeling techniques to assist in performing time-dependent reliability analysis for infrastructure components. Response surface models permit the development of reliability models where a closed form solution for the limit state surface does not exist. Using both global and local finite element modeling can greatly assist with defining specified points on the limit state surface. This surface may be a highly nonlinear surface as well as multidimensional in random variable space. By dimensionalizing the randomness of the variables through the finite element models, a discrete surface can be approximated at points that can be fit using either regression or interpolation methods. This paper will present some basics behind response surface techniques and how finite element modeling is used to develop time-dependent reliability models. These steps will be demonstrated through the use of finite element examples for infrastructure components that were analyzed using ABAQUS. These examples will show how response surfaces can be developed for the limit states and how this process can be applied to other types of engineering reliability models.

2002 ABAQUS Users Conference

2. Response Surface Techniques


2.1 Introduction

The concept of response surface techniques has further developed as the use of finite element models and programs in all forms of engineering has increased. This process uses a rather basic approach, based on finite element analysis, to solving the limit state or failure surface that is used in reliability analysis. Faverelli (1989) and Bucher and Bourgund (1990) were some of the first in the field of reliability to incorporate this technique into a formal reliability analysis. With the increase in computer capabilities and advances in finite element coding over the past century, response surface techniques permit a highly accurate and expedient way to analyze critical infrastructure components for reliability. 2.2 Reliability

An introduction of the concepts of reliability and time-dependent reliability will be needed to tie together the use of these concepts with response surface modeling. Reliability analysis tries to analyze if a component or system is safe (satisfactory) or unsatisfactory (i.e., fails). This relationship is traditionally shown as a capacity, C, and demand, D, relationship. If C > D, then the component or system is considered safe, however, if the D > C then the component or system is considered to be unsatisfactory performance or has failed. This limit state surface is the boundary been safety and failure and is shown in Figure 1 in the space for two random variables, X1 and X2. Reliability is generally expressed as a limit state function, Z, which is a function of random variables, Xi, where, Z = g (X1, X2,., XN) and the limit state is expressed as Z =D-C>0 Limit states may be defined as a complete or near complete failure of a system or component. These might be based on stress, strain, moment, or shear. Other limit states, which might be used for navigation structures, can be performance-based such as deflections, critical crack lengths, or cracking percent. Typically, the limit state surface is highly non-linear and contains multidimensional aspects that are based on the number of random variables used to define the problems. These aspects make it not very simple to quantify a closed form solution for the reliability problem. This is where response surface techniques can greatly assist in approximating the limit state surface for use in reliability analysis.

2002 ABAQUS Users Conference

X1 Safe C < D Limit State Surface

Fail D > C X2

Figure 1. Limit State Surface

2.3

Time-Dependent Reliability

The degradation of infrastructure components due to age, environment, and degradation lends itself to a time-dependent reliability problem. Time-dependent reliability concepts focus on the degradation of strength and represented in terms of its variability as fR(r) (or capacity) versus the load represented in terms of its variability, fS(s) (Patev 1998). At some point in time as the degradation of the strength occurs, the capacity of the structure will be less than the demand. This is the point of assumed unsatisfactory performance or failure. This relationship between strength and load over time can be demonstrated in Figure 2.

fR(r)

fS(r)

time, t
Figure 2. Time-Dependent Relationship of Strength and Load The limit state for a time-dependent structure or component can be shown as R(t) - S(t) < 0 and the resulting probability of failure at time t is 2002 ABAQUS Users Conference 3

Pf(t) = P[R(t) < S(t)] In addition, the hazard function is important for the use in the economics analysis for decision-making purposes, the hazard function or conditional failure rate can be expressed as h(t) = P[fail in (t,t+dt)| survived (0,t)] Also, the time-dependent reliability can be represented in mathematical form by the equation where,
t

1 L(t ) = exp[ t[1 ( ) FS ( g (t )r )dt ]] f R (r )dr t

FS = CDF of load g(t)r = time-dependent degradation fR(r)dr = pdf of initial strength = mean rate of occurrence of loading

Closed form solutions for this double integral are not available except in a few cases so we need to utilize Monte Carlo Simulations to examine the life cycle (Patev, 1998) for a component or the entire structure. 2.4 Response Surfaces

Finite element analysis (FEA) is currently used in practice every day for structural design and analysis of engineering structures. However, constitutive models that define most failure modes are not readily available for these types of reliability problems. So using tools such as FEA is very valuable in estimating defined points on a response surface for the limit state function of random variability. While they cannot be used to determine a closed form solution to a limit state function, they can be easily used to define points that can then be used as an approximation for the limit state surface. The concept of a response surface (points) for a non-linear limit state of two random variables, X1 and X2 is shown in Figure 3. X1 Safe C < D Limit State Surface

Response Points Fail D > C X2 Figure 3. Response Surface Points 4 2002 ABAQUS Users Conference

The response surfaces are generated by selecting important points in the domain of random variables. This domain may be a mean, minimum, or maximum of each random variable. In addition, these curves must also investigate and define any correlation between random variables on the limit state surface. Once a number of FEAs are performed, the response curves for the random variable are then fitted via regression or interpolation analysis for use in the reliability analysis. Also, the response surface technique will have some statistical error. Since certain points only define the limit state surface, the in-between points defined by a regression curve can only be considered approximate. A non-linear regression on the surface points should produce the best fit statistically but residual errors will exist and can be accounted for as model variability within the reliability analysis itself.

3. Case Examples
3.1 Alkali Aggregate Reaction of Mass Concrete Structures

Alkali Aggregate Reaction (AAR) problems found in mass concrete at navigation structures is an example of reliability problems where no stress-strain constitutive model is available in the literature. AAR growth is caused by chemical reaction in concrete between the aggregate (typically silica constituents) and the cements (alkalis). This is the reason why the term is called alkali-aggregate reaction. AAR creates an unlimited swelling and volumetric expansion of mass concrete, which causes substantial cracking within mass concrete structures. Special curves and modeling were developed using FEA and response surface techniques to assist with the reliability analysis of a chamber at a lock structure. The calibration of the stresses and strains in the concrete utilized three different FEMs. These include an overall global model and two local models of a navigation lock structure. The FEMs were generated and analyzed using ABAQUS 5.8 in combination with the concrete analysis program called ANACAP-U (James, 1997). The growth rate for the concrete was modeled as a swelling factor (reversed shrinkage) in the ANACAP-U software package. The first FEM is a global model that defined the entire river wall monolith (symmetric about the axis-of-the-dam). This was developed and calibrated to match the existing growth of the lock in three dimensions taken by both instrumentation and field measurements. The global model is shown in Figure 4.

2002 ABAQUS Users Conference

Figure 4 Global Finite Element Model The local FEMs for the miter gate monolith (Block 47) was developed to calibrate both the strains measured from instrumentation and the split tensile cracking patterns that exist in the monoliths. The local model is shown in Figure 5.

Figure 5 Local FEM

2002 ABAQUS Users Conference

These FEMs were used to develop two factors that were important to the reliability model. Since there were no closed form solutions of the limit state surface to predict the effects of the tensile cracking in the mass concrete structure, these needed to be developed using response surface modeling. The critical factors were the growth rate of the AAR concrete over time and compressive strength versus strain of the mass concrete over time. These curves were developed using a minimum of twenty-four FEAs for random variables to assist in defining the limit state surface. Representative curves for these random variables on the response surface are shown in Figure 6 (growth rate) and Figure 7 (compressive strength). From these response surface curves a Monte Carlo Simulation program was developed to assist with the reliability calculations. Additional details on the reliability analysis and program can be found in Patev 2000.

Figure 6 Response Curve for Growth Rate

Figure 7 Response Curve for Compressive Strength

2002 ABAQUS Users Conference

3.2

Fatigue and Corrosion of Miter Gates

Another example of using response surfaces in reliability analysis is for the hydraulic steel structures. Miter gates are a very important component of a navigation lock and dam. Miter gates are large structures (sometimes exceeding 80 feet in height) and are typically built of a series of welded steel components (plates, girders, intercostals, etc..). The gates assist with retaining the upper pool at a lock (i.e., lift or height of water being pooled) and permit the safe navigation of barge traffic through the lock. A typical set of miter gates is shown in Figure 8.

Figure 8 Miter Gates at a Lock and Dam

However, miter gates run through a number of cycles due to differential water pressure caused during the opening and closing of the gates for navigation traffic. This cycling creates fatigue problems at the welded connection in particular areas of the gate structure. In addition to fatigue, most of the time gates are submerged below the water line, which make them subjected due to corrosion as well. This degradation due to both fatigue due to cycling and section loss due to corrosion cause cracks in the highly stressed areas of miter gates. Typical cracking patterns at horizontally-framed miter gates are shown in Figure 9.

2002 ABAQUS Users Conference

Figure 9 Fatigue and Corrosion in Miter Gates (Arrows define the cracks locations, corrosion is visible)

The process used to develop the response surface is summarized here but can found in more detail in James (2001). The process used to develop the response surface involved many steps and FEM models to develop the final limit state. This included evaluating the load distributions including the effects of component wear, performing residual stress distribution which involved the welding process used on gates and the model restraints, developing model for crack initiation which is dependent on stress difference in load cycle, derive a model for crack growth rates which included a model for stress intensity versus crack length, and finally to define the limit state and evaluate the response surface for parameter variations. This process involved running numerous FEM runs to account for variability in the final responses for the limit state. To perform these steps a global FEM model was developed in ABAQUS for the gates to assist with determining the load distribution for normal load cycle and operation conditions and the exact stress distributions for a local model of the welded flange connections. The global model of the miter gate is shown in Figure 10.

2002 ABAQUS Users Conference

Strut Arm Anchor Arms

Pintle

Local Model of Flange Connection

Figure 10 Global Model of Miter Gate

A final local model FEM was then developed to define the limit state to determine the extent of cracking needed to compromise structural integrity. This FEM incorporated the lower three girders of the miter gate as shown in Figure 11. This local model used eigenvalue calculations to find the critical factor on the operating loads that would cause buckling. From increasing the levels of cracking, damage would occur to the lowest girder until the buckling factor was at unity. This was the limit state for the response surface. From this limit state analysis, random response surfaces due to fatigue and corrosion, welding stress, and buckling and material properties were developed from multiple FEAs . Non-linear regression curves were then fit to the data from the FEM model to predict buckling. A Monte Carlo Simulation program was then developed around these equations that incorporated the statistical sampling to develop the reliability for the miter gate.

Figure 8 Limit State FEM for Miter Gate (Note: buckling is in lower left hand corner)

10

2002 ABAQUS Users Conference

4. Conclusions and Recommendations


Response surface techniques are a valuable tool in assisting with reliability analysis for infrastructure components. These techniques permit realistic reliability modeling to be conducted for limit states that are highly non-linear or have multidimensional random variable aspects. While these techniques are powerful, they still need a watchful eye during response surface development to insure that the statistical results from the final reliability model do not give misleading information. This is why extensive and often painstaking calibration is needed with both the response surface and the reliability models.

5.

Acknowledgements

The authors greatly acknowledge the support and funding from the Ohio River Main Stem System Study and the Chickamauga Lock Rehabilitation Study. The view of the authors does not purport to reflect the position of the Department of the Army or Department of Defense.

6. References
1. Bucher, C. and Bourgund, U. A Fast and Efficient Response Surface Approach for Structural Reliability Problems, Structural Safety, Vol. 7(1), 1990. 2. Faravelli, L. Response Surface Approach for Reliability Analysis, American Society of Civil Engineers, Journal of Engineering Mechanics, Vol. 12(4), 1989. 3. James, Randy M. ANACAP-U, ANATECH Concrete Analysis Program Users Guide, Version 2.5, San Diego, CA, 1997. 4. James, Randy M., Schaaf, D.M., Werncke, G.M., and Patev, R.C. Finite Element Modeling for Reliability Evaluation of Fatigue Cracking at Welded Connections, Eighth International Conference on Structural Safety and Reliability, ICOSSAR 01, Newport Beach, CA, 17-22 June 2001. 5. James, Randy M., Schaaf, D.M., and Patev, R.C. Finite Element based Reliability Models for Concrete Deteriorization due to Alkali Aggregate Reaction, Eighth International Conference on Structural Safety and Reliability, ICOSSAR 01, Newport Beach, CA, 17-22 June 2001. 6. Patev, R.C., Schaaf, D.M., Riley, B.C., and Kathir, N. Time-Dependent Reliability Analysis for Navigation Structures, ASCE Proceeding of the Engineering Foundation Conference on Risk-Based Decision-Making in Water Resources VIII, Santa Barbara, CA, Oct 1998 7. Patev, R.C., Schaaf, D.M., Riley, B.C., and Kathir, N. Time-Dependent Reliability for Structures Subjected to Alkali-Aggregate Reaction, ASCE Proceeding of the Engineering Foundation Conference on Risk-Based Decision-Making in Water Resources IX, Santa Barbara, CA, Oct 2000.

2002 ABAQUS Users Conference

11

S-ar putea să vă placă și