无先验知识下基于CHMM的刀具磨损监测技术

Tool Condition Monitoring without Priori Knowledge Based on Hidden Markov Model

切削刀具的状态直接影响工件加工质量、生产率和产品成本,因此在切削加工过程中监测刀具的状态显得尤为重要。针对实际监测系统通常无法获取刀具各磨损退化状态先验知识的情况,以切削力与切削振动为监测信号,提出无先验知识下基于小波包分析与连续隐马尔可夫模型的刀具磨损监测技术。应用小波包分析技术提取信号特征信息,采用S函数实现特征值归一化处理。利用监测过程中的刀具正常状态下归一化特征信息建立基于连续隐马尔可夫模型的监测模型;根据刀具未知状态特性向量与监测模型间的对数似然度获取刀具性能指标PV,实现刀具磨损状态评价。采用铣刀磨损全寿命数据来验证该方法的有效性,实验结果表明:该方法能在无先验知识的情况下对刀具的健康状态进行较为准确的评估,且所需样本数较少,训练速度快。该技术对实现无先验知识下的刀具智能化在线状态监测具有重要意义。

The condition of cutting tool has a direct effect on processing quality,productivity and produce cost.So,it is very important to monitor tool condition in cutting process.Aiming at the situation that the prior knowledge of all kinds of wear degradation mode cannot be attained usually,the cutting force and vibration signals were measured as monitoring signals by multi-sensors,a tool condition monitoring method without priori knowledge based on wavelet packet decomposition and continuous mixture hidden Markov model(CHMM) was presented.Features were extracted by wavelet package decomposition and normalized by sigmoid function.Fist,during the monitoring process,the normalization features which were attained in normal wear condition were inputted to CHMM to complete model training.Then,the trained model could be used to monitor tool condition through calculating the PV which was attained by calculating log-likelihood ratio of the unknown state and model.In order to validate the effectiveness of the proposed method,the whole life-cycle data of milling cutter wear were used.The experimental result shows that this method can be used to carry out an accurate assessment of the tool state when lack of priori knowledge.Also,it shows that the model has fast learning ability and needs few training samples.It has significant realistic meaning to tool intelligent on-line monitoring without priori knowledge.