**4. Conclusions**

The composite filled with carbon nanotubes after the mechanical deformation was simulated. The nanotubes were modelled as randomly distributed non-overlapping ellipsoids. The mechanical stretching of the composite was introduced as the non uniformity of the angular distribution of the ellipsoids. It was shown that the uniform composite provides the higher conductivity and its percolation concentration is the lower, than that of the aligned composite. The anisotropy of the macroscopic properties was investigated. The percolation concentration and conductivity are lower along the direction of the partial orientation of the nanotubes in comparison with the perpendicular one. That can be understood in terms of the conductive paths tunnelling distance and the total number of the conductive paths formed in different directions.

The impact of the periodic boundary conditions was cleared. It was demonstrated, that very different behaviour of the conductivity upon the CNT alignment presented in literature may be explained by the boundary conditions. It was proved, that the model with the periodic boundary conditions provide more relevant conductivity results since the conductivity becomes unit cell size-independent.

The presented model may be used for strain sensor development. The utilisation of the model as pre-experimental step allows to find out the optimal conditions for the composite synthesis, taking into account the nanotube aspect ratio, concentration, the matrix-related properties (in terms of the tunnelling barrier value). At the same time, it allows to predict the behaviour of the main macroscopic parameters on the deformation.

**Author Contributions:** A.P. code developing, writing the draft; P.K. J.M. and M.Š. writing the final paper; D.L. and D.M. GPU computations; P.K., D.B., P.L. and J.B. conceptualization. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was partially funded by H2020 RISE 734164 Graphene 3D, H2020 RISE Project 823728 DiSeTCom. PK is supported by Horizon 2020 IF TURANDOT project 836816 and the Academy of Finland Flagship Programme, Photonics Research and Innovation (PREIN), decision 320166.

**Acknowledgments:** Massive GPU computations were performed on KAUST's Ibex HPC. We thank KAUST Supercomputing Core Lab team and especially to Mohsin Shaikh and Passant Hafez for assistance with execution tasks on Volta and Pascal GPU nodes.

**Conflicts of Interest:** The authors declare no conflict of interest.
