基于极限学习机的铁谱磨粒模式识别

Wear Particle Recognition of Ferrography Based on ELM Based on ELM Ferrography Pattern Recognition

将极限学习机(ELM)应用于铁谱磨粒模式识别中,从磨粒彩色图像中提取出磨粒的形状尺寸、颜色、纹理3个方面的特征参数作为ELM的输入,以正常滑动磨粒、严重滑动磨粒、球状磨粒、切削磨粒、氧化物磨粒这5种类型磨粒作为ELM的输出,建立基于ELM的磨粒分类器;将3个方面的17个特征参数进行排列组合建立不同的模型,通过对比实验及分析,确定出最优的模型和磨粒分类器;通过实验比较基于ELM与基于BP神经网络的磨粒分类器性能。结果表明:基于ELM神经网络的磨粒分类器的识别速度平均为150 ms,准确率最高为96%,基于BP神经网络的磨粒分类器的识别速度平均为250 ms,准确率最高为90%。因此,基于ELM的磨粒分类器识别速度更快、准确率更高。

The extreme learning machine(ELM)was used for wear particle recognition of ferrography.With the characteristic parameters of the shape size,color and texture extracted from the abrasive color image as the input of the ELM,with five types of abrasives,the normal sliding abrasive,severe sliding abrasive, spherical abrasive,cutting abrasive,oxide abrasive,as the output of the ELM,an ELM-based abrasives classifier was established.Seventeen feature parameters from three aspects were arranged and combined to establish different models.Through comparative experiments and analysis,the optimal model and abrasive classifier were determined.The performance of the wear particle classifier based on ELM and BP neural network was compared through experiments.The results show that the wear particle classifier based on ELM has an average recognition speed of 150 ms and a maximum recognition accuracy of 96%,the wear particle classifier based on BP neural network has an average recognition speed of 250 ms and a maximum recognition accuracy of 90%,which indicates that the wear particle classifier based on ELM has faster recognition speed and higher accuracy.