基于太赫兹时域光谱技术的CFRP薄涂层厚度检测方法

Thin Coatings Thickness Detection of CFRP Based on Terahertz Time-Domain Spectroscopy

针对碳纤维增强复合材料(CFRP)上的薄涂层测厚存在的基底各向异性、多重反射、回波重叠、环境噪声等问题,提出基于太赫兹时域光谱技术与稀疏分解的涂层测厚方法,并在仿真波形和实际样品上成功验证。CFRP通常由碳纤维与环氧基体牢固结合并多层复合而成,具有各向异性特点,在光学性质上表现为CFRP上同一位置不同入射方向、同一入射方向不同位置的反射波形均存在差别。这致使基于理论模型的优化算法用于其上涂层测厚时精度有待提高。另外,在涂层较薄时,受噪声与多重反射等因素的干扰,无法直接在时域根据反射波形区分涂层界面,即直接使用飞行时间法(TOF)难以获得涂层厚度。为此,使用稀疏分解法来定位薄涂层的分界面,再结合TOF法计算CFRP上薄涂层的厚度。首先,分析了多层结构上的太赫兹反射信号与信号稀疏分解模型之间的联系,并利用参考信号构成稀疏字典。然后,鉴于太赫兹信号的传播路径只在分界面发生改变这一先验知识,即涂层层数决定了信号稀疏分解后非零脉冲个数,使用前向正交匹配追踪算法(LAOMP)求解太赫兹反射信号的稀疏表示。LAOMP算法可直接指定稀疏度,便于选择稀疏脉冲定位涂层分界面。获得对应于各涂层界面的脉冲后,使用TOF法求得涂层的厚度。最后,在总厚度分别为102和66μm的双涂层仿真数据与实际样品上验证方法的性能,并与经典基于L1范数的谱投影梯度算法(SPGL1)进行了对比。结果表明:在相邻涂层折射率差别较小的情况下,LAOMP算法可以有效检测CFRP上35μm的薄涂层。在总厚度为66μm的双涂层样件上,检测误差在11%以内,具有调参简单、结果稳定的优点。

To address the challenges of substrate anisotropy,multiple reflections,echo overlap,and environmental noise in measuring thin coatings on CFRP,a coating thickness measurement method based on terahertz time-domain spectroscopy and sparse decomposition is proposed.This method has been successfully validated on simulated waveforms and real samples.CFRP composites are multilayer;each layer is typically composed of carbon fibers firmly bound with an epoxy matrix.CFRP composites are anisotropic,with variable reflection at the same location in different incidence directions and locations in the same incidence direction.As a result,the measurement accuracy of coating thickness needs to be improved when the optimization algorithm based on the theoretical model is used.In the case of thin coatings,direct differentiation of coating interfaces in the time domain is challenging due to noise and multiple reflections.Therefore,the conventional time-of-flight method(TOF)alone can't work well.To address this,sparse decomposition is utilized to locate the interfaces of thin coatings,and then the TOF is applied to calculate the thickness.Initially,the relationship between terahertz reflection signals on multilayer structures and sparse decomposition models is analyzed,and a sparse dictionary is constructed using reference signals.Then,with the prior knowledge that the propagation path of terahertz signals only changes at coating interfaces,the number of nonzero pulses in the sparse decomposition of the terahertz reflected in the number of coating layers determines the signal.Based on the above facts,the LAOMP algorithm is selected to solve the sparse representation of the terahertz-reflected signal.The LAOMP algorithm specifies the sparsity directly,facilitating the selection of sparse pulses to locate the layer interface.After obtaining the pulses corresponding to each coating interface,the thickness can be solved by TOF.Finally,this proposed method is validated for both simulated data and the samples with two coating layers.The total thickness of CFRP samples is 102 and 66μm,respectively.The method also is compared with the classical spectral projection gradient-based L1(SPGL1).The results show that the LAOMP algorithm can effectively detect the thin coating of 35μm on CFRP with a small refractive index difference of adjacent coatings,and the total error of the double coating sample is less than 11%when the total thickness is 66μm.Compared with SPGL1,the parameter tuning of the LAOMP is easier,and the decomposed sparse impulses are more stable.