Freeform surface parts, such as blades, exhibit complex structures and excellent aerodynamic performance, making them widely utilized in aerospace propulsion systems. However, monitoring and ensuring surface quality during the milling process of such components is challenging, leading to high scrap rates and unguaranteed processing efficiency. To address these issues, this paper investigated the milling process monitoring and position-dependent surface roughness prediction for the thin-walled blade with the material of Ti-6Al-4 V. The monitored blade-root acceleration signal was utilized to develop a discrete spatial vibration model based on the machining characteristics of the blade. This involved using Fourier transform and inverse techniques to combine the frequency response functions and cutting force for acceleration calculation, which was then compared to measured values to validate the model’s monitorability. Aiming at the surface roughness prediction, a predictive method for the entire machined surface was proposed, consisting of signal pre-processing, feature extraction and selection, and construction of extreme learning machine (ELM) model. Time-domain, frequency-domain, and time–frequency-domain methods were adopted for feature extraction. To enhance the generalization ability and accuracy of the predictive model, the maximal information coefficient was employed for correlation analysis, resulting in the selection of 12 features as input for the ELM-based surface roughness prediction system. Comparison of the measured and predicted surface roughness results revealed that all errors were less than 14%, with an average error of only 6.70%, demonstrating the validity and reliability of the prediction method. Notably, the proposed monitoring method does not interfere with the milling process and enables prediction of surface roughness at arbitrary positions on the entire surface during variable-parameter processing of freeform surface parts, thereby improving the quality and precision of the machined surface. The potential application of this paper lies in the final inspection of defects in industrial products and reducing the service risk of non-conforming parts.

叶片等自由曲面零件具有复杂的结构和优异的气动性能，使其在航空航天推进系统中得到广泛应用。然而，在此类部件的铣削过程中监控和确保表面质量具有挑战性，从而导致高废品率和无法保证的加工效率。为了解决这些问题，本文研究了采用 Ti-6Al-4 V 材料的薄壁叶片的铣削过程监测和位置相关的表面粗糙度预测。利用监测到的叶片根部加速度信号开发离散空间基于叶片加工特性的振动模型。这涉及使用傅立叶变换和逆向技术将频率响应函数和切削力结合起来进行加速度计算，然后将其与测量值进行比较以验证模型的可监控性。针对表面粗糙度预测，提出了一种整体加工表面的预测方法，包括信号预处理、特征提取与选择、极限学习机（ELM）模型的构建。采用时域、频域和时频域方法进行特征提取。为了增强预测模型的泛化能力和准确性，采用最大信息系数进行相关性分析，最终选择了12个特征作为基于ELM的表面粗糙度预测系统的输入。测量和预测表面粗糙度结果的比较表明，所有误差均小于 14%，平均误差仅为 6.70%，证明预测方法的有效性和可靠性。值得注意的是，所提出的监测方法不会干扰铣削过程，并且可以在自由曲面零件的变参数加工过程中预测整个表面任意位置的表面粗糙度，从而提高加工表面的质量和精度。本文的潜在应用在于对工业产品缺陷进行最终检测，降低不合格件的服务风险。从而提高加工表面的质量和精度。本文的潜在应用在于对工业产品缺陷进行最终检测，降低不合格件的服务风险。从而提高加工表面的质量和精度。本文的潜在应用在于对工业产品缺陷进行最终检测，降低不合格件的服务风险。