基于神经网络的表面粗糙度识别算法

提出一种新的表面粗糙度识别算法，该算法利用从标准样块上通过采样得到的散射光强度分布数据，把表征光强分布的数据和样块的标称值分别作为神经网络的输入和输出，采用改进的ＢＰ算法对神经网络进行训练。训练后，把某一工件的散射光强度分布数据输入给神经网络，则网络的输出就是该样块的表面粗糙度数值。该算法充分利用了神经网络的泛化能力和学习能力，可正确识别Ｒａ在０．８μｍ以下的被测表面，并可避免误识别

A Surface Roughness Recognizing Algorithm Based on Neural Networks

A new algorithm for surface roughness identification is introduced. A neural network (NN) is employed, which accepts inputs from distribution data of the scattering light reflected from standard sample-box, and use the declaration value of sample-box as the target outputs. The NN is trained by improved BP algorithm. When distribution data of some work piece are input to the NN after training, the output of the NN will be Ra of the piece. Better useing learning ability and other features of the NN, the correct identification could be achieved for work pieces that have Ra less than 0.8 μm. And possible miss-identification could be avoided.