摘要
针对目前智能制造动态生产过程质量异常诊断中,实时质量数据呈现出高维非线性等复杂特征导致模型过拟合以及诊断准确率低等问题,提出一种基于t-SNE-LSTM的质量异常诊断方法。该方法首先利用t分布随机近邻嵌入算法(t-SNE)对控制图数据进行降维得到具有聚类效果的低维特征;进而将其输入到长短期记忆神经网络(LSTM)进行训练学习捕捉其特征,从而达到对生产过程质量进行异常诊断的目的。最后进行了仿真实验,结果表明:基于t-SNE-LSTM的质量异常诊断方法能够有效提高质量异常模式的识别精度,验证了方法的有效性。
In the current quality anomaly diagnosis of intelligent manufacturing dynamic production process,real-time quality data presents complex characteristics such as high-dimensional nonlinearity,which leads to model overfitting and low diagnostic accuracy.To solve the problems,a quality anomaly diagnosis method based on t-SNE-LSTM was proposed.Firstly,the t-distributed random Neighbor embedding algorithm(t-SNE)was used to reduce the dimension of the control chart data to obtain the low-dimensional features with clustering effect.Then,it was input into the Long Short-Term Memory neural Network(LSTM)for training and learning to capture its characteristics,so as to achieve the purpose of quality diagnosis of production process quality.Finally,the simulation results show that the quality anomaly diagnosis method based on t-SNE-LSTM can effectively improve the recognition accuracy of quality anomaly patterns,which verifies the effectiveness of the method.
作者
罗强
石宇强
LUO Qiang;SHI Yuqiang(School of Manufacturing Science and Engineering,Southwest University of Science and Technology,Mianyang 621010,China)
出处
《机械工程与自动化》
2026年第1期148-151,156,共5页
Mechanical Engineering & Automation
