基于功率传感器的刀具磨损量预测方法

Tool Wear Prediction Approach Based on Power Sensor

摘要:

使用功率传感器监测机床加工功率，和切削力、声发射等传感器相比，功率传感器具有实用性强、对加工过程无影响等优点.针对采集到的功率信号，在分析信号特征相关性的基础上，提出了一个多目标优化RPSBL的刀具磨损量预测方法.对信号特征进行后处理(Reprocessing，RP)消除电网波动和切削中其他偶然因素的影响，进一步提高特征对刀具磨损敏感性.基于处理后的特征，运用稀疏贝叶斯学习(Sparse Bayesian Learning，SBL)方法建立刀具磨损量预测模型.此外，使用非支配排序遗传算法(Nondominated Sorting Genetic Algorithm II，NSGAII)对SBL模型相关参数进行优化以提高预测精度.实验研究表明，该方法能够实现刀具磨损量的准确预测.不同预测方法的对比表明，通过特征后处理提高信号特征对刀具磨损的敏感性，保证了刀具磨损量的准确预测，对SBL模型参数进行优化可进一步提高预测精度，减小预测误差的最大值.

Abstract:

The power sensor was used to monitor machine processing power which was more practical and of no influence on the cutting process in comparison with conventional sensors such as force and AE. For the collected power signal, based on the analysis of signal features, a Reprocessing Sparse Bayesian Learning(RPSBL) with Nondominated Sorting Genetic Algorithm Ⅱ(NSGAII) approach was proposed to achieve the tool wear prediction. First, the features reprocessing was applied to eliminating impacts caused by power fluctuation and other casual factors, and the sensitivity of tool wears enhanced. Then, the tool wear was predicted by Sparse Bayesian Learning based on the reprocessed features. Finally, the parameter of Sparse Bayesian Learning was also optimized by NSGAII to improve the prediction accuracy. The experimental results on a milling machine tool show the effectiveness in predicting the tools wear by the proposed approach. A comparative study of different methods shows feature sensitivity enhancement of the tool wear by feature reprocessing ensures its prediction accuracy; Prediction accuracy can be further improved and the maximum of the prediction error can be minimized through the optimization of SBL with NSGAII.