|
Understanding the crushing behavior of thin-walled energy absorbers and accurately predicting the maximum crushing load (MaCL) are critical to crashworthy design. This study developed an interpretable Symbolic Neural Network (SNN) for predicting the MaCL of hole-perforated tubes (HTs). Unlike black-box machine-learning models, the SNN yielded an explicit closed-form equation that enables rapid estimation and transparent interpretation of the input–output relationship. Based on data from 12 development specimens, the SNN achieved a nested-LOOCV RMSE of 0.722 kN and a MAPE of 2.21%. For two additional independent off-design specimens excluded from model development and selection, the absolute percentage errors were 2.33% and 2.24%, respectively. Because one specimen was tested per configuration, these values do not account for experimental scatter. The SNN was trained and validated only for hollow HTs under quasi-static compression and was not used for core-filled MCTs. Experimental and numerical results showed that small perforations (Φ6) preserved structural integrity, whereas large, dense perforations (Φ14 with six holes per side) promoted localized collapse and reduced load capacity. Relocating perforations from mid-height to the upper region increased MaCL by approximately 3.5–5.5% and specific energy absorption (SEA) by 2–3%. For the investigated MCTs under axial impact at 15.6 m/s, simultaneously increasing the tube and ABS-core wall thicknesses from 1.2 to 1.8 mm increased the mean crushing load by 75–85%, the crushing load ratio by 25–29%, and SEA by approximately 22%. Within the examined 6–14 mm perforation-diameter range, varying the diameter changed the first and second MaCL by less than 2% and 5%, respectively, indicating a secondary effect relative to the coupled tube/core wall-thickness level under the investigated design and loading conditions.
|
