PSO多决策融合刀具磨损评估

Multi-decision fusion assessment of tool wear based on PSO

针对刀具磨损状态监测中单一决策算法模型评估效果受工况参数影响较大,在不同工况条件下模型通用性不好的问题,提出一种基于粒子群优化(PSO)算法的多决策融合刀具磨损评估模型。首先对加工过程中的主轴电流信号和振动信号进行时域分析、频域分析和小波包分解提取有效特征,并利用主成分分析进行特征降维。之后在4种集成学习算法的基础上,构建PSO多决策融合模型,分别使用LDIWPSO和CFPSO两种PSO算法实现多集成学习算法的决策融合,进行刀具磨损状态评估。试验结果表明:PSO多决策融合模型相比于各种单一集成学习算法模型具有更好和更稳定的预测性能,且CFPSO决策融合的效果要优于LDIWPSO决策融合。

In order to deal with the problem that the performance of a single-decision tool wear monitoring model is greatly influenced by the machining conditions and the universality is not satisfying,a multi-decision fusion evaluation methodology of tool wear is proposed based on the PSO algorithm.The spindle current signal and vibration signal in the machining process are collected,and multiple valid signal features are extracted via time domain analysis,frequency domain analysis and wavelet packet decomposition.Then the principle components analysis is utilized to reduce the dimension of extracted features.After that,based on the results of four ensemble learning algorithms,the multi-decision fusion algorithm based on PSO is proposed,and LDIWPSO and CFPSO,two modified particle swarm optimization algorithms,are respectively applied to combine multiple ensemble learning algorithms to realize multi-decision fusion to evaluate the state of tool wear.The experimental results indicate that compared to each single ensemble learning algorithm,the multi-decision fusion algorithm has better and more stable predicting accuracy,and the performance of CFPSO is better than LDIWPSO.