Proposed Outline for LEAP Verification and Validation Processes

The Liquefaction Experiments and Analysis Project (LEAP) is an international collaboration between universities in the US, Japan, UK, Taiwan and China to evaluate the capabilities of constitutive models for liquefaction problems. In developing this international collaboration, a key step is to develop and define the verification and validation procedure that will be used in this study. This paper will outline the current state of thinking of the validation and verification procedure to be used in LEAP and will include description of the procedure for the constitutive model and numerical model verification process as well as validation process using element and centrifuge test data. Specific details, requirements, and purposes of each of the exercises within the verification and validation procedure is described. of these models can accurately and practically assess their capabilities and limitations. This is important for decision makers, such as leaders of regulatory agencies that determine the safety of infrastructure and allocate resources accordingly. It is envisioned that the verification and validation process not only verifies and validates – it also must illustrate the quality of the verification and validation to users and to decision makers that are not necessarily numerical modeling experts. Thus, the suggested LEAP V&V process may be outlined as follows.  Constitutive model verification: description of physics intended to be modeled with graphical presentations to illustrate predicted constitutive behavior for specific standard test paths.  Numerical model verification: description of discretization and solution schemes with graphical comparisons to illustrate their stability and accuracy in selected numerical tests.  Element test validation: predictions of constitutive model are compared to data from laboratory test data (e.g., triaxial, simple shear, and onedimensional consolidation tests).  Class C validation by comparing predictions of numerical models to data from:  Benchmark centrifuge model tests (relatively simple geometry such as 1-D saturated sloping ground)  System physical model tests (e.g., more complex problems such as dams, cases involving soil-structure interaction, etc.)  Benchmark simulations of liquefaction in the field (e.g., validation against observed liquefaction in the Christchurch earthquakes of 2010-2011)  Class A validation:  New benchmark centrifuge model tests (relatively simple geometry such as 1-D saturated sloping ground)  New system centrifuge tests (e.g. dams, soilstructure interaction, etc.) Similar to Lambe’s (1973) explanation, we differentiate the classes of validation as follows. Class A is a true prediction of an event made prior to the event. A Class B prediction is after the event, but with results unknown to the predictor. A Class C validation is after the event, with results known to the predictor. For class C validation, the individual(s) conducting the modeling (herein called “predictor(s)”) may or may not iteratively adjust the model parameters to improve the quality of the agreement between calculations and observations. The verification steps (first two bullets in the above list) will identify which models have which capabilities, often with a yes/no or black/white binary classification. The verification steps do not require comparison to experimental data; but they do require demonstrating the model's ability to provide consistent results. However the validation against experimental data (the last three bullets in the above list) requires metrics for assessing the quality of validation. Because validation involves comparison to experimental data, there will be inevitable questions about data consistency and accuracy. The LEAP process will endeavor to establish relaiable calibration data and to develop reasonably robust validation metrics. At the time of this paper submission, the verification steps (first two bullets in the above outline) have been developed to a greater extent than the validation steps (last three bullets). Hence this paper focuses more attention to verification than to validation. 3 CONSTITUTIVE MODEL VERIFICATION Constitutive model verification is intended to develop a clear understanding of the range of behaviors that are intended to be simulated by the code and an illustration of the ability of the code to predict this behavior. Data will be collected in a standard format to allow systematic comparisons of the predictions from multiple constitutive models. Blank tables will be provided to encourage predictors to provide results in a format for easy crosscomparison of their constitutive model results with those of other researchers. Verification of the constitutive models will involve the developers answering several key questions about their constitutive models (3.1) and conducting a series of specific verification tests (3.2) using a generic set of soil data like that found in Table 1. Table 1. Default hypothetical soil descriptions for verification of constitutive model. Soil Type Dense (D) Medium Dense (M) Loose (L) Relative Density 80 50 20 Init. void ratio 0.55 0.7 0.85 Permeability 0.015 mm/s 0.03 mm/s 0.05 mm/s Method of Placement Dry pluviation Dry pluviation Dry pluviation