During explosion events, the flame encounters obstacles which result in the enhancement of turbulence. This phenomenon is highly unsteady and both temporally and spatially localized behind obstacles due to the propagation of the flame. The use of the Adaptive Mesh Refinement method is ideal for easily remeshing regions of interest without incurring significant CPU costs. Recently, a new vortex identification method was proposed, which defines a criterion Omega that identifies turbulence when the rotation rate overcomes the deformation rate. The method has been recently enhanced to better identify turbulence in confined reactive situations. This method relies on flame quantities and is therefore not applicable for general flows, reacting and non-reacting. Based on flow quantities of interest, our goal is to establish a robust AMR approach for unsteady cases. To achieve this, we investigate the Liu criteria for its standardised formulation. However, the formulation makes use of a case-specific parameter epsilon, which denotes a vortex intensity level and guides the selection of vortices. To improve the Omega criterion, we propose an epsilon formulation based on Taylor microscale. Using the suggested approach, we formulate a novel AMR criterion to target turbulence in LES. Additionally, this criterion defines the AMR level automatically using a Reynolds formulation. The strategy is first applied to a triangular bluff-body case, and then the approach is validated by analysing the vortical structures behind obstacles on the H2-air GraVent explosion channel. Using pressure probes at different locations and flame tip velocity measurements, we compare experimental data, a refined static mesh and our improved Liu model. Additionally, we examine the computational CPU improvement resulting from our AMR approach. Future work will focus on verifying the robustness of our methodology across various cases.