基于图像分割的深纹核桃挤压破壳裂纹规律研究

Investigation into Crack Propagation Laws in Compression Shell-breaking of Deeply Veined Walnuts Based on Image Segmentation

摘要:

对核桃挤压破壳裂纹规律的研究有利于提升核仁完整性，然而传统的破壳方法难以准确反映裂纹真实演化规律。针对上述问题，以云南泡核桃为研究对象，本文提出了一种基于图像分割的裂纹分析方法，研究在挤压载荷下核桃裂纹扩展规律。采用YOLO v8m-seg实例分割模型实现裂纹自动识别与量化表征，模型分割平均精度均值为82.3%，研究了不同等效直径与挤压方向条件下挤压力、变形量、做功等力学特性，并分析了裂纹总数、裂纹分布象限数、裂纹总面积占比等裂纹演化特征的影响规律及核仁完好率。搭建了融合双视角高速成像与力、位移传感技术的试验平台并进行试验，结果表明，等效直径与挤压方向对核桃的力学响应、裂纹特征及核仁完好率均具有显著影响;破壳最大挤压力为190~280 N，总功为620~1 000 J;裂纹数量先增加，达到峰值后逐渐贯通合并为2~3条，且分布在多个象限，挤压结束后裂纹总面积占比稳定在10%~20%;等效直径34~36 mm的中等尺寸核桃从横径方向挤压综合表现最优，最大挤压力约231 N，总功约760 J，挤压结束时裂纹总面积占比18%，核仁完好率为73%。研究结果为低损伤破壳机械装备设计与参数优化提供了理论依据，且能在一定程度上预测和提高核仁完好率。

Abstract:

It was designed to clarify the mechanical behaviour and crack propagation mechanisms of deep-ridged walnuts under compression shell-breaking, and improve kernel integrity rate by directly observing and quantifying crack evolution. A dual-view synchronous high-speed imaging platform integrating force-displacement acquisition was constructed, and a pixel-level crack segmentation pipeline based on YOLO v8m-seg was developed. In comparison with YOLO v8n-seg and YOLO 11n-seg, the proposed model was found to achieve a precision of 86.3% and an mAP50 of 82.3% for the crack class. An annotated dataset comprising 3,600 pairs of front- and back-view images was established, and an imaging magnification calibration coefficient was introduced to unify crack measurements between the two views;on this basis, total crack number, total crack area ratio, and crack quadrant count were defined as three reproducible crack descriptors to describe the spatiotemporal evolution of cracks. A two-factor full factorial design was adopted, with equivalent diameter (30~42 mm, six levels) and compression direction (transverse, ridge, longitudinal) as independent variables, and compression strokes were set to 6,7,8 mm according to walnut size. Two-way ANOVA results showed that equivalent diameter and compression direction had highly significant effects (p≤0.01) on mechanical response, crack features, and kernel integrity rate. Typical force-displacement curves were observed to be divided into three stages, consisting of an elastic rising stage, a stage in which primary cracks were initiated and were accompanied by a sharp force drop, and a subsequent stage of stable crack growth. For all test groups, the maximum compression force was found to range from 190 N to 280 N and the total work was found to range from 620 J to 1,000 J;transverse compression was associated with lower energy and more stable shell opening than the other directions. During mid-stage loading, the crack number was observed to reach 4~7 and then to merge into 2~3 cracks toward the end of compression, while the final crack area ratio was found to stabilize at 10%~20%. When the crack area ratio reached a peak of 29% and then was reduced to 18% as compression progressed, this range was shown to coincide with the best kernel integrity rate performance. Walnuts with an equivalent diameter of 34~36 mm compressed along the transverse direction were demonstrated to perform best, combining lower energy consumption, a stable shell-breaking mode, and a maximum kernel integrity rate of 73%. The proposed vision-based segmentation framework and crack descriptors were shown to provide an effective basis for in-situ fracture analysis of biological shells under load and to offer practical guidance for the design and optimization of low-damage, controllable shell-breaking equipment.