From Vibrations to Computation: Numerical and Experimental Investigation of a Mechanical Duffing Oscillator for Physical Reservoir Computing

Physical reservoir computing (PRC) is a neuromorphic computing paradigm that leverages the intrinsic dynamics of a nonlinear physical system (the reservoir) as computational resource. We propose a PRC implementation based on a doubly-clamped aluminium beam exhibiting geometric nonlinearity and behaving as a Duffing oscillator. We address the question: when is nonlinear dynamics an advantageous resource for computing? To investigate this question, we experimentally evaluate the computational capabilities of the proposed platform using two established benchmark tasks: parity computation and MNIST handwritten digit recognition. Under appropriately tuned driving conditions, the intrinsic nonlinearity of the system is sufficient to retain temporal correlations between present and past inputs, even without external feedback. This establishes a physically interpretable mechanism for memory, directly linked to the system’s operating regime. The results demonstrate that effective physical reservoir computing can be achieved with minimal mechanical complexity, highlighting the potential of nonlinear mechanical systems as platforms for mechano-intelligence.