基于智能技术的木材缺陷定量无损检测研究  

Study on quantitative nondestructive test of wood defects based on intelligent technology

作　　者：孙建平[1,2] 胡英成[1] 王逢瑚[1] 

机构地区：[1]东北林业大学生物质材料科学与技术教育部重点实验室,哈尔滨150040 [2]广西大学林学院,南宁530004

出　　处：《仪器仪表学报》2013年第9期1955-1960,共6页Chinese Journal of Scientific Instrument

基　　金：中央高校基本科研业务费专项资金(DL12EB03-03);国家自然科学基金(30571457);全国优秀博士学位论文作者专项基金(200764)资助项目

摘　　要：通过对直径不同孔洞缺陷振动信号进行处理分析,实现木材孔洞大小的无损检测。研究中首先采集孔洞面敲击和无孔洞面敲击的振动应力波信号,然后对直径不同孔径缺陷的振动信号的频谱特征进行分析,提取出频谱的特征向量作为训练样本;并利用获取的样本对构建好的体现信号特征与孔洞大小的非线性神经网络模型进行训练,然后利用训练好的网络对孔洞缺陷的大小进行无损检测。结果显示:随着孔洞直径的增大,振动信号频谱密度极大值所对应的频率逐渐减小;与无孔洞面敲击方式相比,孔洞面敲击所获得的信号频谱特征作为样本训练BP网络,网络仿真性能较好,仿真输出和目标值的相关系数都能达到0.98以上,对孔洞缺陷直径大小的识别准确率达到93.5%以上;孔洞缺陷大小检测的最佳模型为隐层节点6、传递函数为正切Sigmoid的单隐层网络模型。The vibration signals come from the wood with different diameter hole defects were processed and analyzed to realize the nondestructive test of the wood hole diameters. First, the vibration signals were collected by knocking the wood surfaces with and without holes, then the frequency spectrum analyses on the vibration signals were carried out, the feature vectors were extracted and taken as the training samples. The nonlinear neural network model that re- flects the relationship between the signal features and hole diameters was constructed. The model was trained with the obtained training samples and employed to carry out the nondestructive test of the wood hole diameters. The results show that with the increasing of the hole diameters, the frequency corresponding to the frequency spectrum maximum of the vibration signal decreases gradually ; and compared with that of the vibration signals for the wood surface with- out holes, when the signal frequency spectrum of the vibration signals for the wood surface with holes is employed to train the BP neural network model, the obtained simulation performance of the model is better. The correlation coeffi- cients between the simulation output and the target value are all above 0.98 ; the recognition accuracy of the hole di- ameters is above 93.5%. The optimum BP neural network for recognizing hole diameters is one-layer network model with the hidden nodes of 6 and the transfer function is tangent sigmoid.

关 键 词：频谱分析 神经网络 缺陷 无损检测 

分 类 号：TS67[轻工技术与工程]