Development of Tribosensor System Using Neural Networks

本课题研究的是一种具有人类手指感受力的新型摩擦传感器的研制。传感器基本上是一个压电陶瓷声发射传感器。它被放置和移动在待识别的材料上。首先将来自传感器的电压信号存储在数字存储器中一段固定的时间。然后对其进行分析，得到最大信号强度、信号的凹凸总数以及与平均信号强度线相交的次数。得到的每一组参数与神经网络中的训练数据集进行比较，测量样本的材料是固定的。实验结果表明，该传感系统对6种不同材料的120个样品的正确率达到85%，通过两步分类，正确率提高到95%。

This research is a study on the development of a new tribosensor that has the sensibility of human finger. The sensor is basically a piezoceramic AE sensor. It is placed and moved on the material to be identified. The voltage signal from the sensor is first stored in the digital memory for a fixed period of time. It is then analyzed to obtain the maximum signal strength, the total number of concaves and convexes of the signal, and the number of crossings with the line of mean signal strength. Each set of the parameters obtained is compared with the training set of data in the neural networks, and the material of the measured sample is fixed. Experimental results obtained show that the present sensor system identifies the material 85% correctly for 120 samples of 6 different materials and the correct recognition is rised to 95% by two step assortment.