飞机蒙皮激光除漆过程的在线监测, 是实现分层可控除漆、 满足适航维修要求的重要手段, 也是推进激光除漆工程应用、 飞机维修自动化的核心技术。 激光诱导击穿等离子体光谱（LIBS）技术可通过激光材料作用过程中产生的等离子体发射光谱快速分析材料表面元素变化, 实现激光清洗表面状态的在线监测。 基于搭建的高频纳秒红外脉冲激光除漆LIBS在线监测平台, 分别采集了不同激光功率下, 面漆、 底漆、 铝合金基体去除过程中的3类LIBS光谱（各100幅）。 分析了不同激光功率下, 各类光谱示踪元素特征谱线的变化情况, 初步筛选了12条特征谱线作为光谱识别的特征。 进一步对这12个特征进行主成分分析（PCA）, 并将前3个主成分（PC1、 PC2、 PC3）构成的数据集作为支持向量机（SVM）识别模型的输入量, 建立了3类光谱的识别模型。 形成了多漆层结构激光分层可控清除过程的LIBS在线监测判定规则, 并对该规则的有效性进行了实验验证。 结果表明, 与低频脉冲激光单点作用采集的针状LIBS光谱相比, 基于该平台采集的LIBS光谱普遍存在较强的连续背景（大于5 000 a.u.）以及1.5 nm左右的半峰全宽; 针对此类光谱设计了改进均值平滑滤波算法, 在去除背景光谱的同时有效避免了特征谱线强度失真; 示踪元素的特征谱线存在不稳定性; 主成分分析中前3个主成分PC1、 PC2、 PC3对光谱的解释率达95%, 在其构成的三维空间中, 同类光谱呈区域性聚集; PCA-SVM模型对训练集、 测试集的识别准确率分别为99.44%、 100%; 验证实验结果表明3类光谱的识别模型与在线监测判定规则有效。 所建立的识别模型与判定规则, 可为飞机蒙皮激光分层除漆过程监测及自动化解决方案提供核心技术支撑。

Online monitoring of the aircraft skin laser paint removal process is an important means to achieve layered and controllable paint removal and meet airworthiness maintenance requirements. It is also the key technology to promote the industrial application of laser paint removal and aircraft maintenance automation. Currently, the main monitoring methods include surface imaging and process performance parameter measurement methods. However, these methods have inherent limitations, making it difficult to be online and real-time. Laser-induced plasma breakdown spectroscopy (LIBS) technology has the advantages of equipment simplicity, flexibility, quickness and sensitivity, which has been widely used in online monitoring and research of laser cleaning of artworks and cultural relics. Based on the established high-frequency nanosecond infrared pulsed laser paint removal LIBS online monitoring platform, three LIBS spectra (100 frames each) were collected during the removal of topcoat, primer and aluminum alloy substrate under different laser powers. The changes of characteristic spectral lines of various spectral tracer elements under different laser powers were analyzed, and 12 characteristic spectral lines were preliminarily screened as the characteristics of spectral identification. Principal component analysis (PCA) was further performed on these 12 characteristics. The data set composed of the first three principal components (PC1, PC2 and PC3) was used as the input of the support vector machines (SVM) identification model, and the identification model of three types of spectral data was established. A LIBS online monitoring and judgment rule for the controllable removal process of laser layering of multi-paint-layer structure was formed, and the rules validity was experimentally verified. It can be seen from the results that, compared with the needle-like LIBS spectra collected based on low-frequency pulsed laser single-point action, in general, the LIBS spectra collected based on this platform show a strong continuous background (greater than 5 000 a.u.) and a full width at half maximum of about 1.5 nm; an improved mean smoothing filtering algorithm was designed for this type of spectrum, which effectively avoids the intensity distortion of the characteristic spectral line while removing the background spectrum; under different laser powers, the characteristic spectral line of the tracer element is unstable; the contribution of the first three principal components, i.e., PC1, PC2, and PC3 in the principal component analysis to the explanation of the spectral data reaches 95%. The same type of spectra is clustered regionally in the three-dimensional space formed by them. The recognition accuracy of the PCA-SVM model on the training set and test set is 99.44% and 100%, respectively; the verification experimental results show that the identification models of the three types of spectra and the online monitoring and judgment rules are effective. The established identification model and judgment rules can provide key technical support for the monitoring and automation solutions of the aircraft skin laser layered paint removal process.