Collaborative Research: Specific Energy-Based Prognosis for Machining Surface Integrity through Integration of Process Physics and Machine Learning

制造业每年提供超过1200万个就业岗位，为国内生产总值（GDP）贡献超过2万亿美元。与此同时，制造业占美国年能源消耗总量的28%左右，尤其是金属切削和加工工艺，这些工艺在价值创造、教育、劳动力发展和就业方面为国民经济做出了重要贡献。尽管传感和通信技术发展迅速，但对加工零件表面完整性的实时过程监测和预测仍然是节能、高质量加工的挑战。虽然将实时传感数据纳入基于物理的加工模型具有模型更新和校准的潜力，并且新兴的机器学习（ML）技术已经证明了制造数据分析的有效性，但ML模型的一般黑箱性质限制了ML结果的严格，基于物理的解释。该奖项通过引入物理指导的学习方法来解决这一现有差距，通过数据科学和过程物理的互补优势，提高加工表面完整性预测的准确性和透明度。该项目的成果影响了多个行业，从航空航天到汽车，能源和医疗保健。该项目的跨学科性质有助于通过扩大妇女和代表性不足的少数群体在研究和教育中的参与来培养下一代制造业劳动力。本研究调查了加工工艺参数对加工零件表面完整性的复合影响。研究方法是多方面的。(1)开发与加工表面完整性相关的特定能量的物理模型;（2）开发数据生成方法，通过自动表征合成刀具磨损和加工表面的图像;（3）将切削物理集成到递归神经网络（RNN）中，用于物理指导的表面完整性预测，以提高ML结果的可解释性和透明度;（4）在生产级机床上对所开发的方法进行了实验评估。该方法减少了加工后产品质量检测的时间和成本，并在三个方面创造了新的知识：（1）引入了一种新的以能量为中心的学习方法，该方法通过比能量表征加工表面完整性：（2）开发了一种新的数据合成方法，以解决模型构建和评估中表面完整性数据可用性的限制;以及（3）展示将机器学习与物理知识相结合的有效途径，以改进对网络结构及其预测逻辑的解释，该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的评估被认为值得支持。影响审查标准。

Manufacturing employs more than 12 million jobs and contributes over $2 trillion to the Gross Domestic Product (GDP) annually. At the same time, manufacturing accounts for about 28 percent of the annual total energy consumed in the U.S. This is particularly true for metal cutting and machining processes, which have been a major contributor to the national economy in value creation, education, workforce development and employment. Despite rapid advancement in sensing and communication technologies, real-time process monitoring and prediction of the surface integrity of machined parts have remained a challenge for energy efficient, high-quality machining. Although the incorporation of real-time sensing data into physics-based machining models has the potential for model updating and calibration, and emerging machine learning (ML) techniques have demonstrated the effectiveness in data analysis for manufacturing, the general black-box nature of ML models has limited rigorous, physics-based interpretations of ML outcomes. This award addresses this existing gap by introducing a physics-guided learning method for machining surface integrity prediction with improved accuracy and transparency, through the complementary strengths of data science and process physics. The outcome of this project impacts multiple industry sectors, from aerospace to automotive, energy, and healthcare. The project’s interdisciplinary nature helps train the next generation of manufacturing workforce by broadening participation of women and underrepresented minority groups in research and education.This research investigates the compounding effects of machining process parameters on the surface integrity of machined parts. The research approach is multifold. (1) Develop physical models for the specific energy associated with machining surface integrity; (2) Develop a data generative method to synthesize images of cutting tool wear and machined surfaces by automatic characterization; (3) Integrate cutting physics into a recurrent neural network (RNN) for physics-guided surface integrity prediction to improve the interpretability and transparency of the ML outcomes; and (4) Experimentally evaluate the developed methods on a production-grade machine. The resulting methodology reduces the time and cost for post-machining product quality inspection, and creates new knowledge in three areas: (1) Introducing a new, energy-centric learning method that characterizes the machining surface integrity by means of specific energy; (2) Developing a new data synthesis method to address limitations in surface integrity data availability for model construction and evaluation; and (3) Demonstrating an effective pathway to integrate machine learning with physical knowledge for improved interpretation of the network structure and its prediction logic, thereby enhancing the network’s transparency and acceptance by the manufacturing industry.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.