ABSTRACT Current research on impact equivalence prediction for packaged goods primarily focuses on drop impacts. However, in express logistics, horizontal impacts on package faces or edges—caused by vehicle acceleration, cornering and braking—are more prevalent. Through horizontal impact tests, this study examines the effects of impact orientation, number of impacts and initial impact velocity on velocity change ( ΔV ) and impact angle inference. Results indicate that ΔV exhibits nonlinear growth with initial velocity and number of impacts across different orientations, stabilizing after approximately four impacts with accelerated stabilization at higher velocities. Impact angle inference is predominantly influenced by orientation. Accordingly, a γ ‐based impact angle inference method (utilizing normalized peak acceleration ratio) is proposed, achieving < 2° mean absolute error (MAE) in additional impact tests. Furthermore, five artificial neural network (ANN) prediction models were developed, utilizing impact type, impact angle, number of impacts and ΔV as input features, with initial impact velocity as the output target. Among these, the Particle Swarm Optimization‐Backpropagation Neural Network (PSO‐BPNN) demonstrated optimal performance: internal validation (696 datasets) yielded R 2 = 0.9280 and RMSE = 0.0547; external validation (24 independent datasets) showed maximum relative error < 3% with mean relative error of 0.83%. In six characteristic impact tests simulating postroad‐transport conditions, the PSO‐BPNN model predicted initial impact velocity with a mean error of 4.99%, significantly outperforming the effective drop height (EDH) method (15.37%). Furthermore, the model identified impact angles within < 3° mean error—a capability unattainable with EDH—establishing a novel methodology for characterizing transportation shock responses through standardized horizontal impact testing.
Chen et al. (Fri,) studied this question.