基于机器学习算法模型的焊接接头疲劳寿命预测

Machine-Learning-Based Fatigue Life Prediction Method for Welded Joints

中文关键词:  机器学习  随机森林算法  轻梯度提升机算法  焊接接头  疲劳寿命  几何形状  预测模型

英文关键词:machine learning  random forest algorithm  LightGBM (Light Gradient Boosting Machine) algorithm  welded joints  fatigue life  geometry  prediction model

中文摘要:

      焊接接头具有非均匀的微观组织和梯度过渡的力学性能及随机分布的焊接缺陷等特征，相较于其他结构更容易产生疲 劳断裂，特别是焊接接头的疲劳载荷下的强度和寿命问题已成为工程界和学术界的研究热点。为了研究焊接接头的疲劳行为，开 展了基于随机森林（RF）模型的焊接接头疲劳寿命预测模型的全新研究。通过采用RF和轻梯度提升机（LightGBM）2种不同的机 器学习算法模型对焊接接头的疲劳数据集进行分析和预测，从中选择预测性能更优的机器学习模型；通过比较在不同几何形状下 疲劳寿命的预测结果，评估几何形状对机器学习模型预测性能的影响；利用RF算法对输入条件进行重要度排序，分析焊接接头疲 劳寿命的影响因素；通过计算模型在不同材料下的疲劳寿命结果验证机器学习模型的泛化能力。结果表明：机器学习模型对不同 几何形状的焊接接头疲劳寿命的预测效果较好，且可用于预测在不同材料下的焊接接头疲劳寿命。研究结果对焊接结构的强度 设计与焊接工艺参数优化等具有重要意义。

英文摘要:

      Welded joints are characterized by non-uniform microstructure, gradient transitions in mechanical properties, and randomly distributed welding defects, which are more prone to fatigue fracture than other structures. Therefore, studying the strength and life of welded joints (especially under fatigue loading) has become a hot research topic in engineering and academia. A new study of a fatigue life prediction model for welded joints based on a random forest model was carried out to study the fatigue behavior of welded joints. To select a machine learning model with better prediction performance, the fatigue data set of welded joints was analyzed and predicted using two different machine learning algorithm models, the Random Forest Model, and LightGBM. The random forest algorithm was used to rank the importance of the input conditions to analyze the factors influencing the fatigue life of welded joints; The fatigue life results of the model were calculated with different materials to verify the generalization ability of the machine learning model. The results show that the machine learning model performs well in predicting the fatigue life of welded joints with different geometries and can be used to predict the fatigue life of welded joints with different materials. The results are of great importance for the strength design of welded structures and the optimization of welding process parameters.