AIAA RP Page 5 In an optimized balance design, the strain-gages are located such that the interaction effects on the first bridge are opposite in sign to the interaction effects on the second which are wired in parallel to minimize interactions. Would prefer In an optimized balance design, the strain-gages within a bridge are located and wired such that the strain from all loads except the component to be measured will cancel. Or In a TASK balance design, the strain-gages are located such that the interaction effects on the first axial bridge are opposite in sign to the interaction effects on the second which are wired in parallel to minimize interactions in the output. 3.1.2.1 Does anyone still do this? 3.2 pg 17 Zero Load Outputs Another initial step in the process outlined in Figure 2 is to determine the zero load outputs for the balance. Put another way, this would correspond to the output of the balance bridges at the rated excitation voltage under a weightless condition. Since a weightless condition cannot realistically be achieved, these values are usually determined by averaging bridge outputs with the balance level in pitch, at four roll positions indexed by 90o, (e.g. 0o, 90o, 180o, and 270o of roll). It is also good practice to make one or more repeats of this group of four measurements to minimize errors in the average. Do not see taking repeats as a good use of time. Pg 18 If the two load measurements can be made reasonably close to one another in time, then it can be assumed that many factors (e.g. temperature, amplifier gain) that can cause the balance outputs to drift, will tend to cancel out. Should be amplifier offset since amplifier gain changes will not cancel. PG 30 Step 3 Tare Iteration Process In order to calculate the tare loads, the bridge outputs minus the zero load outputs arfe needed for the initial point in each load series. These data are shown in Table 7. 4.3 Pg 49 bottom In this case, only one of the two cross-component terms, NF1*RM or NF2*RM, can be computed. However, if the combinations loading of NF1 with RM and NF2 with RM are obtained independently by a different loading scheme, then both cross-component terms could be computed. In this case, only one of the two cross-component terms, NF1*RM or NF2*RM, can be computed. Without further information neither term should be computed. . However, if the combinations loading of NF1 with RM and NF2 with RM are obtained independently by a different loading scheme, then both crosscomponent terms could be computed. From AIAA-R-091-2003-5yr review Comments.pdf in your IBTWG folder Known issues with the example (wrong raw data points and wrong tare on series 6 table 18 ). Norbert’s comments on the summation index for most equations). I do not want to include the alternate iteration method Norbert commented on. Just adds confusion. 3.1.3 I think Norbert’s comments on the classical form are correct but I’m not sure anyone cares about the classical form versus common practice in the balance community. I’ve yet to hear of a 96x6 matrix being used. The matrix is usually presented in the transpose form for ease of viewing but I think the equations are correct. I have not seen the referenced document. Norbert’s new chapter Math Model Evaluation Do not understand the use of underlining in this document I cannot recommend the statistical methods for selecting math models at this time. SVD is fine for eliminating terms which are not supported. Using VIF to select terms to include seems to leave out significant terms and few in the group probably understand the process. Selection of signed terms should be done with great caution as they fit errors in applied loads very well. For example any difference in the X or Y location of a load point on the positive or negative side of a calibration body will appear to be a sign dependant interaction. I do not see the percent interaction matrix to be a pass fail test for selection of terms. The Hierarchy rule seems arbitrary at best. I do not understand the statement about a constant shift in the independent variables. No recommendation is made in this document and I see no reason to bring this into the document. I have never heard of the PRESS residual. Norbert’s new section OCT 28 2011 I do not break things into term groups or options. I really don’t like the odd and even function superposition section. No insight or value added. Norbert’s new chapter after 3.1.2.2 OCT 28 2011 The equations Norbert gives are overly detailed and can be simplified From equation 3.1.6 the percent interaction matrix can be computed using the vector {G} formed from the maximum component loads. The percent contribution of each term is then C i,j * Gj / Ci,iGi *100. For refence equation 3.1.6 is {R}n = {a}n + [C]n,m {G}m I do not know the origin of NASA’s .05% rule or if it is still used. Percent interaction matrix is useful to see if the balance behaves as expected. Pairs of large cross product terms which tend to cancel may indicate highly dependent terms. If elimination of one of the terms greatly changes the other term, then probably neither term is independently defined well enough to generate. With experience with a particular design the terms which should be relatively large can be known. For example most moment balances the interaction of PM2*RM onto YM2 is significant due to deflections. Percent interactions may also point to behavior not anticipated by the designer. This may be useful in improving the design.