Shock modelling through FEA has been successfully accomplished and validated with benchmarks for several SLB applications [1]. However, the 3D shock modelling computational cost is high (hours to days for each run) limit the use of FEA for sensitivity study and field operational applications. The objective of this work is to develop computationally efficient ROM which can predict the shock & vibration behavior of electronics/chassis (at specific points of interest) in response to external dynamic excitation. We are presenting 2 complementary approaches and a workflow to combine those into a general multi-fidelity ROM: • Linear State Space (LSS) approach for low fidelity (LF) collar/housing/chassis vibration transfer function • Proper Orthogonal Decomposition (POD) approach and Machine Learning (ML) time stepping for high fidelity (HF) shock response. Demonstrate the capacity of POD projection and time stepping ML approach to reproduce shock response on simple downhole tool. This innovative dynamic ROM framework could deliver customized-fidelity downhole shock and vibration response at a fraction of the computational cost.