We invite interested scholars to work with us on developing new systems that apply machine learning to achieve accurate results in dynamic mechanical measurements. Machine learning provides a powerful means for performing deconvolution on sensor outputs to provide accurate measurements of dynamic inputs. Trained neural networks can provide an accurate system response in the absence of complete knowledge of the measurement system and accomodate nonlinear behavior often important in real physical systems. This work is centered on building physics-constrained neural networks representations of dynamic mechanical measurement systems, training the constrained networks using known calilbration inputs, and applying the trained networks to achieve accurate measurements of arbitrary-waveform unknown inputs. Transparency in the developed neural network models allow for physics discovery from the trained networks.