CAREER: Unifying Sensing, Machine Perception and Control for High-precision Micromanufacturing

该学院早期职业发展（CAREER）资助研究基于机器学习的传感和控制方法，以提高可扩展和高精度减材和增材微制造工艺的效率和质量。 基于机器学习的方法利用从传感器和执行器收集的大量数据来发现模式，并使用数学算法和统计模型进行推理。 这代表了新兴微制造技术（例如，柔性电子器件的卷对卷印刷），其中先验知识有限但传感器数据丰富。通过利用由物理制造知识指导的先进机器学习方法，这项研究增加了对基于传感的控制的理解，从而显著提高了高精度和高速率制造所需的过程性能，稳定性和适应性。 最终，这项工作通过实现更精确、更可靠的下一代制造来造福社会，并以更少的材料浪费、更低的缺陷率和更高的效率生产更复杂的产品。从这项研究中获得的知识用于支持未来制造科学家和工程师的教育和培训，这些科学家和工程师来自一个多元化和充满活力的群体，包括该领域代表性不足的少数民族和妇女。该项目的目标是推进对数据驱动的机器学习的基础上的精密控制的微制造过程的基本理解。在这项研究中开发的新的统一框架和方法将最先进的过程控制从基于模型的标准转变为数据驱动的无模型范式，最终将复杂微制造系统的准确性和精度提升到新的水平。主要创新点包括一个新的感知-感知-学习-控制框架，该框架导致满足以下研究目标：1）通过概率深度学习方法，从丰富的多模态传感器数据中创建低维和低噪声的潜在状态表示，用于过程状态估计，并提供实现高质量监控所需的基础知识; 2）建立了一种新的基于感知的迭代学习控制（PILC）方法，以实现精密过程控制的前所未有的精度，以及3）实验性地展示和验证统一框架和一组基于传感器的深度推理和学习-基于两种特定的先进微制造工艺的控制算法，离子磨蚀刻（减法）和卷对卷凹版印刷（加法）。 这些基本的理解直接提高了具有更严格公差的高精度制造过程的实时控制能力，并指导了通过传感技术和先进数据分析增强制造能力的潜在路线。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估而被认为值得支持。

This Faculty Early Career Development (CAREER) grant investigates machine learning-based sensing and control methods to improve the efficiency and quality of scalable and high-precision subtractive and additive micromanufacturing processes. Machine learning-based methods take advantage of massive amounts of data collected from sensors and actuators to discover patterns and draw inference using mathematical algorithms and statistical models. This represents a new avenue for innovation in process monitoring and control for emerging micromanufacturing technologies (e.g., roll-to-roll printing of flexible electronics) where prior knowledge is limited but sensor data is rich. By leveraging advanced machine learning methods that are guided by physical manufacturing knowledge, this research increases understanding of sensing-based control, thus enabling significant enhancement in the process performance, stability and adaptiveness required for high-precision and high-rate manufacturing. Ultimately this work benefits society by enabling next-generation manufacturing that is more precise, more reliable and that produces more complex products with less material waste, lower defect rates, and higher efficiency. The knowledge obtained from this research is used to support the education and training of future manufacturing scientists and engineers recruited from a diverse and dynamic group that includes underrepresented minorities and women in this field.The goal of this project is to advance the fundamental understanding of data-driven machine learning-based precision control for micromanufacturing processes. The novel unified framework and methods developed in this research transform state-of-the-art process control from a model-based standard to a data-driven model-free paradigm, ultimately pushing new levels of accuracy and precision of complex micromanufacturing systems. The major innovation involves a novel sensing-perception-learning-control framework that leads to meeting the following research objectives: 1) create a low-dimensional and low-noise latent state representation from abundant multimodal sensor data for process state estimation through probabilistic deep learning methods, and provide fundamental knowledge required to realize high-quality monitoring; 2) establish a novel Perception-based Iterative Learning Control (PILC) method with controllers to achieve unprecedented accuracy in precision process control, and 3) experimentally demonstrate and validate the unified framework and a set of sensor-based deep inference and learning-based control algorithms on two specific advanced micromanufacturing processes, ion-mill etching (subtractive) and roll-to-roll gravure printing (additive). The fundamental understandings directly advance the real-time control capability of high-precision manufacturing processes with tighter tolerance, and guide potential routes for achieving manufacturing capabilities augmented by sensing technologies and advanced data analytics.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.