4.1 Experiments and results with the tensegrity unit and GA
In the initial experiments step of modeling and simulating a tensegrity unit, various different tests took place until the intended level of accuracy and validity to results was reached. Most of the effort was about setting the whole simulation environment in such a manner that both tensegrity modeling would be as closer to real implementation as possible as well as the correct setup for Genetic Algorithms to operate was built.
4.1.1 Experiments with the number of the struts (porosity)
The first level of experimentation with the tensegrity unit was focused around the number of struts that should be used. After some initial experiments, which clearly proved that the number of struts is a parameter closely related to the tensegrity unit porosity percentage, it was decided to define a variable called ‘numberOfStruts’ that controlled the specific property. Three different experiments with different number of struts (5,7, 9) are tested and simulate different porosity percentages (30%, 50%, 70% respectively). The diagram below shows the different attempts and the correspondence configurations.
4.1.2 Experiments with the size of voxels
Concerning the GA in tensegrity unit scale two different experiments took place. At first, different size of voxels were tested as the diagram below represents. The tests concerned the time efficiency of the simulation in order to converge to the optimum result and ranged among 2 to 5 size value of voxel. The size that was used in the final experiment is the value of 3 considered as most appropriate for this solution because it combines time saving and relative accuracy comparing with the value 2, 4, and 5. The higher values exponentially increased simulation time in order to give a result giving in slightly more accurate results and hence they were rejected.
4.1.3 Experiments with fitness evaluation
The second experiment refers to the different measurement ways of tensegrity unit fitness. Having said that the fitness function is the calculation of the absolute velocity behind the tensegrity unit, two attempts were applied. The first used as fluid container same size boxes (the size it depends to the unit with the biggest volume) for all the units regardless their volume and their configuration having the same size calculation area. With this effort it was observed that it was difficult to calculate the absolute velocity and reach to ‘fair’ results because this method did not take into account any of geometrical properties of the units. Following this method, smaller tensegrity units had always bigger amount of wind velocity behind them regardless of their porosity. For this reason another method was implemented. The fluid container box was adjusted to the size and configuration of each unit and margins were placed in the right and left side of the unit as behind it. The new calculation of fitness referred to the calculation of the average velocity per unit volume, with other word per voxel. This way of measuring proved efficient and ‘fair’.
Finally, in this step of experiments and after the implementation of the GA, the three most fit units from the categories ‘number of struts 5,7,9’ are preserved and their data are kept in order to be used in the third step.
In the initial experiments step of modeling and simulating a tensegrity unit, various different tests took place until the intended level of accuracy and validity to results was reached. Most of the effort was about setting the whole simulation environment in such a manner that both tensegrity modeling would be as closer to real implementation as possible as well as the correct setup for Genetic Algorithms to operate was built.
4.1.1 Experiments with the number of the struts (porosity)
The first level of experimentation with the tensegrity unit was focused around the number of struts that should be used. After some initial experiments, which clearly proved that the number of struts is a parameter closely related to the tensegrity unit porosity percentage, it was decided to define a variable called ‘numberOfStruts’ that controlled the specific property. Three different experiments with different number of struts (5,7, 9) are tested and simulate different porosity percentages (30%, 50%, 70% respectively). The diagram below shows the different attempts and the correspondence configurations.
4.1.2 Experiments with the size of voxels
Concerning the GA in tensegrity unit scale two different experiments took place. At first, different size of voxels were tested as the diagram below represents. The tests concerned the time efficiency of the simulation in order to converge to the optimum result and ranged among 2 to 5 size value of voxel. The size that was used in the final experiment is the value of 3 considered as most appropriate for this solution because it combines time saving and relative accuracy comparing with the value 2, 4, and 5. The higher values exponentially increased simulation time in order to give a result giving in slightly more accurate results and hence they were rejected.
4.1.3 Experiments with fitness evaluation
The second experiment refers to the different measurement ways of tensegrity unit fitness. Having said that the fitness function is the calculation of the absolute velocity behind the tensegrity unit, two attempts were applied. The first used as fluid container same size boxes (the size it depends to the unit with the biggest volume) for all the units regardless their volume and their configuration having the same size calculation area. With this effort it was observed that it was difficult to calculate the absolute velocity and reach to ‘fair’ results because this method did not take into account any of geometrical properties of the units. Following this method, smaller tensegrity units had always bigger amount of wind velocity behind them regardless of their porosity. For this reason another method was implemented. The fluid container box was adjusted to the size and configuration of each unit and margins were placed in the right and left side of the unit as behind it. The new calculation of fitness referred to the calculation of the average velocity per unit volume, with other word per voxel. This way of measuring proved efficient and ‘fair’.
Finally, in this step of experiments and after the implementation of the GA, the three most fit units from the categories ‘number of struts 5,7,9’ are preserved and their data are kept in order to be used in the third step.