A convolution-based method to correct oscillations in dynamic stress–strain curves from split Hopkinson pressure bar tests

Oscillations in stress–strain responses obtained from Split Hopkinson Pressure Bar (SHPB) experiments can obscure the intrinsic material behavior, leading to misidentification of dynamic yield strength, distortion of strain-hardening characteristics, and inaccurate calibration of constitutive models. To mitigate oscillations in dynamic stress–strain curves obtained from split Hopkinson pressure bar (SHPB) tests, a convolution-based correction method is proposed. The formulation is derived from the dispersion of elastic waves propagating in slender bars, which arises from frequency-dependent wave velocities and is governed by material and geometric properties. As this behavior is time-invariant, the wave propagation process can be modeled as a linear time-invariant (LTI) system, allowing the one-dimensional stress wave evolution in SHPB bars to be uniquely characterized by an impulse response function. Subsequently, the waveform obtained using the conventional dispersion correction method is subtracted from the experimental signal to extract a time-domain residual term, upon which the convolution-based correction framework is established and implemented for oscillation suppression. To evaluate the applicability of the proposed approach, experimental results for DH36 steel are compared with predictions from the Johnson-Cook constitutive model, the traditional dispersion correction method, and the present method. The results demonstrate that the proposed method provides improved agreement in terms of oscillation amplitude reduction, peak-valley correspondence, and consistency with both experimental data and the Johnson-Cook constitutive response. Comparative analyses demonstrate that the proposed method effectively reduces oscillations in dynamic stress–strain curves of different materials obtained from SHPB tests, thereby enhancing the reliability of dynamic yield identification and constitutive modeling.