27.12.2012 Views

ARUP; ISBN: 978-0-9562121-5-3 - CMBBE 2012 - Cardiff University

ARUP; ISBN: 978-0-9562121-5-3 - CMBBE 2012 - Cardiff University

ARUP; ISBN: 978-0-9562121-5-3 - CMBBE 2012 - Cardiff University

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Figure 3 exemplifies for intradiscal pressure and the contact force in the left facet joint,<br />

the response can indeed vary widely, depending on the chosen model output and the<br />

load case. Common point estimates like sample means and sample standard deviations<br />

(given normally distributed results) describe the scatter in the model response. Yet, the<br />

computed point estimates are random variables themselves, whose reliability depends<br />

on the chosen sample size, 500 in our case.<br />

Figure 3: Model responses for all 500 variations for all load cases (Cmp: compression, Ex: Extension,<br />

Flex: flexion, LatL/LatR: lateral bending left/right, RotL/RotR: rotation left/right)<br />

One way of estimating the distribution (and therefore the reliability) of the means would<br />

be to repeat the whole aforementioned procedure times, i.e. for 100 this would<br />

require solving ⋅ 50,000 FE models. Fortunately bootstrapping the computed<br />

500 solutions provides reasonable precise distribution estimates at a fraction of the cost<br />

(with 1000 bootstrap runs). 11<br />

We can use this approach to estimate the reliability of the computed mean model responses<br />

depending on the sample size by sub-sampling before drawing each bootstrap<br />

sample. We found that a sample size of 100 already provides a reasonable stable<br />

approximation of the population mean (Fig. 4).<br />

7. SENSITIVITY ANALYSIS<br />

As a byproduct of the uncertainty analysis, we can use the results to also investigate the<br />

relative importance of each individual parameter for the model response. Scatter plots of<br />

input vs. output variables give first hints of possible relationships (Fig. 5); the correlation<br />

matrix (PPMCC) quantifies the degree of linear relationship between all parameters.<br />

By decomposing the total variance of the model response, we can compute the relative<br />

contribution of each parameter to the model response and hence its relative importance.<br />

12–14 We found that disc height and the position of the facet joints as well as the<br />

distance between articulating facet joint surfaces are especially important for determining<br />

the mechanics of the spine segment.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!