基于随机失活层优化深度置信网络算法的刀具磨损预测  

Tool Wear Prediction Based on DBN Algorithm Optimized by Dropout

作　　者：陈莉[1] Chen Li

机构地区：[1]新乡职业技术学院现代设计与工程学院,河南新乡453006

出　　处：《装备机械》2023年第3期77-80,共4页The Magazine on Equipment Machinery

摘　　要：为了提高数控刀具的加工精度,应用随机失活层优化深度置信网络算法,进行数控刀具磨损状态预测,并开展试验分析。通过受限玻尔兹曼机训练优化网络权值与偏置量,利用误差反向算法实现网络的微调,并使由特征提取获得的高级特征能够良好匹配预测需求。分析结果表明,各层受限玻尔兹曼机都需要在初期训练阶段重新学习,此时会引起重构误差方式突变,最终逐渐达到稳定状态。相比深度反向传播网络和支持向量机,加入随机失活层后深度置信网络有助于提升模型的学习性能,得到更稳定与更准确的预测结果。In order to improve the machining accuracy of CNC tool,the Dropout was applied to optimize deep belief network algorithm to predict the wear state of CNC tool and carry out experimental analysis.The network weight and bias were trained and optimized by restricted Boltzmann machine,and the error inverse algorithm was used to realize the fine-tuning of the network,and the advanced feature obtained by feature extraction can be well matched with the prediction requirement.The analysis results show that restricted Boltzmann machine in each layer needs to re-learn in the initial training stage,which causes the sudden change of reconstruction error mode,and finally gradually reaches a stable state.Compared with the deep back propagation network and the support vector machine,the deep belief network after adding the Dropout can help improve the learning performance of the model and obtain more stable and accurate prediction result.

关 键 词：刀具 磨损 预测 随机失活层 深度置信网络 

分 类 号：TG707[金属学及工艺—刀具与模具]