Accurate porosity characterization is of great significance to evaluate abradable seal coating comprehensive performance. In this paper, a support vector regression model optimized by particle swarm optimization algorithm, termed PSO-SVR, is proposed to predict coating porosity based on multi-scale ultrasonic attenuation coefficient. To decouple the ‘multi-scale scattering effect’ of ultrasonic propagation in abradable seal coating, the echo signals are decomposed using Continuous Wavelet Transform (CWT). The ultrasonic responses in different frequency bands could be sufficiently extracted through the multi-scale ultrasonic attenuation coefficient obtained by CWT. Subsequently, taking the coefficients as input vectors, the SVR model is established. The parameters of SVR, including the penalty factor C, kernel function parameter γ, and insensitive loss ε, are optimized through PSO algorithm. Finally, the optimized SVR model is applied to predict the porosity of the AlSi-polyester abradable seal coating prepared by plasma spraying. The normalized mean squared error MSE of the validation set is 0.067 with a determination coefficient R² of 0.947. The prediction results show that the PSO-SVR model has higher accuracy, better generalization ability, and stronger robustness in the case with limited experimental data, compared with classical Artificial Neural Network (ANN) models, e.g. Back Propagation (BP), Radial Basis Function (RBF), and General Regression Neural Network (GRNN).

准确的孔隙率表征对评估耐磨密封涂层的综合性能具有重要意义。本文提出了一种基于粒子群优化算法的支持向量回归模型，称为PSO-SVR，用于基于多尺度超声衰减系数预测涂层孔隙度。为了消除可磨损密封涂层中超声波传播的“多尺度散射效应”，使用连续小波变换（CWT）对回波信号进行分解。通过CWT获得的多尺度超声衰减系数可以充分提取不同频段的超声响应。随后，将系数作为输入向量，建立SVR模型。SVR的参数，包括惩罚因子C通过PSO算法对，核函数参数γ和不敏感损失ε进行了优化。最后，将优化的SVR模型应用于预测通过等离子喷涂制备的AlSi聚酯耐磨密封涂层的孔隙率。验证集的归一化均方误差MSE为0.067，确定系数R ²为0.947。预测结果表明，与经典人工神经网络（ANN）模型（例如，反向传播（BP），径向基函数）相比，PSO-SVR模型在实验数据有限的情况下具有更高的准确性，更好的泛化能力和更强的鲁棒性。 （RBF）和通用回归神经网络（GRNN）。