Scribd
Upload
13 views1 page

Optimizing Differential Evolution For Fitting Constitutive Models

This study compares Differential Evolution (DE) with traditional LM-TRR algorithms for fitting constitutive models to mechanical stress-stretch data in soft tissues. Results indicate that DE, particularly with extended objective functions, significantly reduces error and improves model fitting compared to LM-TRR, achieving better RMSE and R2 values. The findings suggest that DE with hyperparameter tuning is a more effective method for optimizing models in tissue biomechanics.

Uploaded by

zachtoom
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
13 views1 page

Optimizing Differential Evolution For Fitting Constitutive Models

This study compares Differential Evolution (DE) with traditional LM-TRR algorithms for fitting constitutive models to mechanical stress-stretch data in soft tissues. Results indicate that DE, particularly with extended objective functions, significantly reduces error and improves model fitting compared to LM-TRR, achieving better RMSE and R2 values. The findings suggest that DE with hyperparameter tuning is a more effective method for optimizing models in tissue biomechanics.

Uploaded by

zachtoom
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Sayed Ahmadreza Razian, Alxexy Kamenskiy, and Majid Jadidi - Optimizing Differential Evolution for

Fitting Constitutive Models of Bi-Directional Mechanical Stress-Stretch Data in Soft Tissues

Abstract

Background: In tissue biomechanics, the Levenberg-Marquardt and Trust-Region-Reflective (LM-TRR)


algorithms are commonly used for nonlinear least square optimization. These methods aim to find the
parameters of constitutive models that provide the best fit to experimental data and minimizing errors.
However, they may not always achieve the optimal fit.

Differential Evolution (DE) is an evolutionary method used to iteratively improve the fit of models by
minimizing an objective loss function and finding the global minimum. This study utilizes an extended
version of objective loss function for DE and compares it with the MATLAB lsqnonlin function, which is
based on LM-TRR algorithms, in analyzing biaxial tests of femoral arteries.

Methods: Extended versions of Minkowski Distance and Chebyshev Distance with various power values
are employed as objective functions for DE. The dataset was comprised of planar biaxial tests conducted
on human femoral arterial specimens sourced from 50 donors, aged between 14 and 85 years (43 ± 18
years old, 23 males). The Four-Fiber Family constitutive model was used for optimization. Root Mean
Square Error (RMSE) and R2 are chosen to evaluate the error between experimental and fitted data for
both algorithms. The LM-TRR algorithms, implemented through the MATLAB lsqnonlin function, fit the
Four-Fiber Family model to the datasets based on the RMSE objective relationship. Subsequently, DE
was utilized for fitting using hyperparameter tuning for both Minkowski and Chebyshev distance
functions. The best results for both algorithms are determined by selecting a group of parameters with
the lowest RMSE.

Results: The results showed that the extended version of DE improved RMSE from 0.006 to 7.6 kPa for
all specimens in comparison to lsqnonlin, with an average error reduction of 1.92 kPa. Additionally, R2
values increase for all specimens. Also, in 60% of the specimens the Chebyshev Distance objective
function produced lower error than the Minkowski Distance. The hyperparameter tuning showed that
the power of 1.0, 1.5, 2.0 and 4.0 in the objective function improved the fit in 22%, 34%, 38% and 6% of
the specimens, respectively. Also, 22% and 18% of the best results were from using power of 2.0 and 1.5
by Chebyshev Distance, respectively.

Conclusions: This study demonstrates that Differential Evolution by using the extended objective
functions, outperforms Non-linear Least Squares in fitting the experimental human artery data.
Moreover, by employing DE with hyperparameter tuning, a significant decrease in error between
experimental and fitted data is achieved.

You might also like