论文部分内容阅读
为实现红外热波检测对缺陷的定量识别,应用BP神经网络,拟合函数关系来实现定量识别。在脉冲热像中,以最佳检测时间和表面最大温差为输入量,以缺陷的深度和直径为输出量,利用BP神经网络拟合输入量与输出量之间的关系。借助数值计算的方法,提供样本训练神经网络,并进行了30次随机计算。通过结果分析,发现使用BP神经网络进行计算具备以下特点:网络收敛速度并不决定计算的精度;网络训练过程中,是否达到计算目标误差不会对计算精度带来较大影响;该方法具有较好的计算稳定性。针对计算结果分布特点,设计计算方法,对数据中的较大误差点进行剔除,最后使用取均值的方法减小获得较大误差的风险,提高计算精度。计算结果表明,在4个参数的计算中,最大误差为3.32%,最小误差为0.1%,证明BP神经网络方法可以用于实现缺陷的定量识别计算。
In order to realize the quantitative identification of defects in infrared thermal wave detection, BP neural network is applied to fit the relationship between functions to achieve quantitative identification. In the pulse thermal imaging, the best detection time and the maximum temperature difference of the surface are taken as the input, and the depth and the diameter of the defect are taken as the output. The BP neural network is used to fit the relationship between the input and the output. By means of numerical calculation, a sample training neural network is provided and 30 random calculations are performed. Through the analysis of the results, it is found that the calculation using BP neural network has the following characteristics: the convergence speed of the network does not determine the accuracy of the calculation; whether the calculation of the target error in the network training process will not have a significant impact on the calculation accuracy; Good computational stability. According to the characteristics of the distribution of the calculation results, the calculation method is designed to eliminate the larger error points in the data. At last, the average value is used to reduce the risk of obtaining larger errors and improve the calculation accuracy. The calculation results show that the maximum error is 3.32% and the minimum error is 0.1% in the calculation of four parameters, which proves that the BP neural network method can be used to realize the quantitative identification calculation of defects.