Sources of Error and Modeling Inaccuracies

As previously discussed, the modeling of the full-scale test unit required consid­eration of the many different variables and how they interacted with one another. Given these possible variables and interactions, sources of error and modeling inaccuracies are inevitable. The research team sought to mitigate as many of these factors as possible through the testing regime and through

Copyright by ASTM Int’l (all rights reserved); Tue May 6 12:07:08 EDT 2014 Downloaded/printed by

Rochester Institute Of Technology pursuant to License Agreement. No further reproductions authorized.

identifying potential areas of inaccuracy. The warped shape of the glass and modeling of the framing connections are two such areas of possible error.

The dual deformation modes of the glass proved to be one main area of uncertainty in the results. Because of the complex deformation of the glass, the readings of the rosette strain gauges did not show a close correlation to the strains in the numerical model. Buckling of the glass can cause a dramatic change in stress values which could not be obtained in the FE model. However, the linear strain gauges showed a much closer correlation to the tested unit (see Figs. 16 and 17), so it was reasonable to consider the model a good representa­tion of the behavior of the full-scale test unit.

As previously stated, the connections between the framing members in the full-scale testing consisted of several machine screws fastening the horizontal members to the vertical members. In the FE model, these connections are mod­eled as pins (not restraining any moment) or fixed (restraining relative rota­tion). However, the actual connections are able to transfer some amount of moment before there is enough rotation to consider the connection pinned. This difference between the actual connection and the modeled connection can impact the correlation between recorded and modeled strains. The stiffness of the connections is unknown and it is difficult to predict without additional tests. It has been decided that the assumption of a pinned connection is the closest prediction of a real behavior because it ultimately led to a closer correlation in data.

To illustrate the effect of the stiffness of the connections on the behavior of the model, two graphs are presented (Figs. 18 and 19). These graphs show uni­directional strain gauges #10 and #11 (see Figs. 4 and 8) readings for pinned and fixed conditions of the connections respectively. Graphs noted as FE are

FIG. 18—Comparison of strains at unidirectional strain gauges locations for pinned frame connections. TOP XX are strains in the location and direction of strain gauge #11 and SIDE YY of strain gauge #10.

Copyright by ASTM Int’l (all rights reserved); Tue May 6 12:07:08 EDT 2014 Downloaded/printed by

Rochester Institute Of Technology pursuant to License Agreement. No further reproductions authorized.

FIG. 19—Comparison of strains at unidirectional strain gauges locations for fixed frame connections. TOP XX are strains in the location and direction of strain gauge #11 and SIDE YY of strain gauge #10.

numerical results, and tests 1 and 3 are physical results of the two tests. Note that the fixed model captures an event at 6 in. (150 mm) of applied bending. This is possibly a location where the system deforms in one of two possible states. Also, the outcome of two very similar tests being different past this point reveals a potential instability of the system. Note, that non-zero initial strains in test 1 are residual strains after the specimen had been loaded to 4-in. (100-mm) displacement for the first time. A reset in the test procedure was required to cor­rect the loading mechanism. The graph shows strains after this process.