|
| 电阻点焊质量Hopfield神经网络在线评估方法 |
| Online Evaluation Method of Resistance Spot Welding Quality Based on Hopfield Neural Network |
| |
| DOI：10.3969/j.issn.1674-6457.2023.03.021 |
| 中文关键词: 电阻点焊 焊接质量 动态功率 Hopfield神经网络 低碳钢 质量评估 |
| 英文关键词: resistance spot welding welding quality dynamic power Hopfield neural network mild steel quality evaluate |
| 基金项目:国家自然科学基金(52275317);广东省自然科学基金(2023A1515012172) |
|
| Hits: 1936 |
| Download times: 788 |
| 中文摘要: |
| 目的 电阻点焊广泛应用于汽车、家电等领域，但目前少有准确的无损质量评价方法。为此，研究一种基于低碳钢板焊接功率信号的焊接质量在线评估方法，并探索利用该信号来评价电阻点焊焊点质量的可能性。方法 对焊接电流、电压信号进行测量和分析，研究功率信号表征焊接质量的可靠性，提出一种有效的模式特征提取方法，将动态功率信号转换为二值图像并用二值矩阵表征，该方法避免特征提取和选择，且尽可能保留焊点质量信息。通过拉剪试验将焊接样本分为6种不同的焊接等级，利用Hopfield关联记忆神经网络建立焊接质量分类器，将具有不同焊接质量水平的焊接样本模式特征矩阵记忆为稳定状态。结果 将焊接样本的模式特征矩阵输入分类器，通过Hopfield网络关联记忆将其收敛到最相似的稳定状态，最终锁定了稳定状态对应的焊接质量。60个测试样本中59个样本都可以被正确分类，该分类器的分类准确率达到98%。结论 分类性能试验结果表明，所提出的模式特征提取方法快速、有效，并能可靠地在线评估低碳钢板的焊接质量。 |
| 英文摘要: |
| Resistance spot welding is widely used in automotive and appliance applications, but there are few accurate non-destructive quality evaluation methods available. The work aims to develop a method for online evaluation of weld quality based on the weld power signal of mild steel plates, and discuss the possibility of evaluating the quality of resistance spot welded joints with this signal. The welding current and voltage signals were measured and analyzed to study the reliability of characterize the weld quality with power signal, and an effective pattern feature extraction method was proposed to convert the dynamic power signal into a binary image and characterize it with a binary matrix, which avoided feature extraction and selection and retained as much information as possible about the quality of the welded joint. The weld samples were classified into six different weld levels through a tensile-shear test, and the Hopfield associative memory neural network was used to build a weld quality classifier to memorize the pattern feature matrix of weld samples with different weld quality levels into a steady state. The pattern feature matrices of the welding samples were fed into the classifier, which converged to the most similar stable state through the Hopfield network associative memory, and finally locked in the welding quality corresponding to the stable state. As a result, 59 out of the 60 test samples could be correctly classified and the classifier achieved a classification accuracy of 98%. The classification performance test results show that the proposed pattern feature extraction method is fast, effective and can reliably evaluate the weld quality of mild steel plates online. |
|
View Full Text
View/Add Comment Download reader |
| Close |
渝公网安备 50010702501719号