Materials like thermoplastic composites exhibit anisotropic behavior, and their characteristics may vary between the manufacturing samples due to inability to precisely control the manufacturing process . Due to lack of knowledge on this variation, one may incorporate the epistemic knowledge in the description of material properties by use of the probability theory. For this purpose one has to develop the stochastic model that may represent the material symmetry or its corresponding characteristics (e.g. constitutive parameters) as uncertain. To achieve this, one may model the material properties as symmetric and positive definite stochastic (SPD) tensors, the prior knowledge of which is described with the help of the maximum entropy principle. This talk focuses on the modeling of stochastic tensors of material properties such as thermal conductivity by use of the parametric approach [1]. The idea is to generate an ensemble of SPD tensors for which certain classes of spatial symmetries and invariances are prescribed together with the second order statistics of a possible higher spatial invariance class. We show that the parametrization of the model reduces to the so-called scaling and rotation related parameters that describe the uncertainties representing the corresponding symmetry class and orientation, respectively. The previously mentioned parameters are then modelled with the help of independent random variables on a manifold structure with the assigned probability distribution coming from the maximum entropy principle. To showcase the model, we apply the proposed approach on 2D and 3D numerical examples involving multiphysics problem such as induction welding of the thermoplastic composite.