Smart Manufacturing Seminar Series

Recently, various advanced 3D scanners have been widely used in manufacturing industries to collect 3D point cloud data of fabricated artifacts. The extra dimension of 3D point cloud data can provide more detailed descriptions of anomalies in artifact surfaces than 2D image data. 3D point cloud data can be categorized into structured and unstructured point clouds. Compared with structured 3D point cloud data, unstructured point cloud data can capture the surface geometry more completely. However, anomaly detection and classification by using unstructured 3D point cloud data is more challenging due to unstructured data representation, inconsistent point sizes, and high dimensionality. To deal with these challenges, this talk will present some recent advances in machine learning for anomaly detection and classification by using unstructured 3D point cloud data. The accuracy and robustness of the proposed method are validated by simulation studies and case studies.

The ongoing reliability test (ORT) process is a standard industry practice for manufacturing reliability. It is designed to provide assurance to customers that weekly shipments from volume production are free from issues that could cause them major material interruptions (e.g., Line Purges, Line Rework, or Field Rework). The ORT provides an early indication about product reliability in the field operation. It assures that the demonstrated long-term reliability during validation stage can be maintained through a robust manufacturing process and a closed loop of failure detection, failure analysis (FA), and corrective action (CA) processes. In classical reliability engineering, ORT is focused on true failures that manifest themselves as “physical” loss of function or wrong output outside of the desired target range. In real world, both true failures and degradation-based virtual failures can co-exist within the same product. Scientifically integrating these two categories of failures provides holistic view of product reliability. Introducing parametric-degradation-based virtual failure concept adds a micro-dimension to the traditional failure domain. Incorporating virtual failures into reliability assessment provides more insights about underlying product reliability and enhances the detection sensitivity of various reliability monitoring mechanisms, such as ORT during production and therefore can surface the poor vintage with potential high future failure rate due to “invisible” high degradation of critical parameters. To stimulate discussion and participation from both academia and fellow industry practitioners, a couple of potential philosophies of describing the holistic failure and algorithms of quantifying the holistic failure counts, including complex number (vector) approach, Yin-Yang theory, etc. An example is given to illustrate the potential value to manufacturing reliability monitoring. It is hoped that this work will be beneficial to a wide range of audience including reliability practitioners, theorists, and management.

The increased digitization in manufacturing and Industry 4.0 have shown promise for improving manufacturing operations and safety. However, it is still challenging to effectively integrate physical engineering models and high dimensional process sensing data for improving decision-making in process monitoring, diagnosis, and control. In this talk, I will present our group’s work funded by the NSF Future Manufacturing program, which aims to improve the inkjet printing processes in additive manufacturing. In particular, I will first present a general digital twin framework for automatically monitoring the multistage inkjet printing processes (e.g. jetting, evolution, and solidification). The detailed data analytics methods will be discussed using two examples: one is how to use deep learning models to analyze and monitor the in-situ streaming strobing videos and physics simulations to capture the complex spatiotemporal evolving characteristics of droplets. The other is how to build an efficient emulator using joint tensor factorization and Nearest Neighbor Gaussian Process (NNGP) to emulate the high-dimensional physical model’s outputs at the solidification stage. Finally, I will share some of our data analytics research in other applications such as how to use wearable sensors to improve workers’ safety in manufacturing.

Batteries and semiconductors are two of the most strategically important manufacturing industries, with planned US investments exceeding $200 billion in semiconductor foundries and $100 billion in battery manufacturing. In this talk, we explore common material concepts for Li-ion batteries and nonvolatile memory, one of the key component of microelectronics. We first show how ion migration and phase transformations, key principles behind the Li-ion battery, are needed to not only understand but also improve new types of nonvolatile memory technology. We then show how microfabrication can be applied to understand new mechanisms underlying the Li-ion battery. Overall, we emphasize that a combined integrated understanding of these two fields provide excellent opportunities for research, development, and workforce training.

Would you feel comfortable flying on a 3D printed airplane? Initially used for rapid prototyping in the 1980s, additive manufacturing (AM) has developed into a proven manufacturing technology. For demanding industrial applications, laser powder bed fusion (LPBF) is now the most established AM technology, and it is also qualified in highly regulated industries like aerospace or medical. The presentation will give an overview of the development of metal LPBF from prototyping to a mature manufacturing technology. Process capability, reliable and repeatable part properties, will be demonstrated with a machine capability study. Furthermore, the feasibility of distributed manufacturing, one of the big promises of AM, will be proven with a multi-machine capability study including different production sites. Applications from different industries will be presented and their requirements will be discussed. Tooling for automotive, space and aviation applications, medical implants, gas turbine applications from the energy sector and heat exchangers in electronics are dominant use cases for industrial AM in 2023.

Industry already accounts for approximately one-third of global greenhouse gas emissions and these emissions are growing quickly as the developing world industrializes and emissions-intensive materials are used to deliver better performing technologies. We need sustainable materials processing solutions that fit the scale and urgency of the challenge. However, the relevant data for informed decision making on emerging process technologies and supply chains is often sparse and noisy. We develop Bayesian techniques for rapid quantification and updating of environmental model uncertainties, demonstrated for the case of laser powder bed fusion and the U.S. steel supply chain. Such modeling reveals the potential for material efficiency (providing engineering services with less material production from natural resources) as a carbon abatement strategy. We highlight its critical role in the decarbonization of key sectors, particularly before midcentury while low-carbon electricity grids and fuels are still being developed. We identify opportunities for material efficiency across the product life cycle and then focus on manufacturing process innovations for reduction and reuse of light metal scrap and increased end-of-life metal recycling in the face of increasing scrap contamination and changing demand. Finally, we discuss some of the most critical developments needed for material efficiency to become widespread and make significant contributions to the decarbonization of global industry.

This presentation is about a 3D-printing based method to manufacture environment-friendly products using biomass (from agricultural wastes such as wheat straw and switchgrass) and fungi. The biomass serves as a nutrition source for fungi, and the fungi grow through the biomass particles and bind the biomass particles together. Products manufactured using this method can substitute those currently made from petroleum-based plastics. Initial targeted applications of these manufactured products will be in packaging, furniture, and construction. The presentation will cover three experimental studies on feasibility of this new method, the relationship between the composition of the mixture (prepared for 3D printing) and print quality, and the relationship between waiting time (from the time when the mixture is prepared till the time 3D printing is performed) and properties of the prepared mixture. The presentation will conclude by discussing research challenges and future research topics for this new manufacturing method.

The CHIPS and Science Act was signed into law in August 2022. In this case “CHIPS” stands for Creating Helpful Incentives to Produce Semiconductors. In this once in a lifetime event, the U.S. Congress has appropriated $11 billion for research and development in all aspects of the semiconductor and microelectronics world. The nation has a unique opportunity to bolster the U.S. integrated circuit and packaging production supply chain, and with this effort, we can prepare for long-term U.S. leadership in microelectronics. NIST has been tasked with managing the CHIPS program, and this talk will present the opportunities described in the legislation and the approach that NIST and the Department of Commerce are taking for implementation. In addition, the capabilities of the NIST research labs in metrology and standards support of the microelectronics industry will be described.

Metal Additive Manufacturing (AM) technologies can produce complex geometries beyond conventional processes, and also allow the design and processing of new alloys. A grand challenge for the metal AM technologies is to accurately control the alloy composition and microstructure throughout entire builds, which are highly sensitive to the local materials chemistry/physics and processing conditions. To tackle this challenge, a physics-based modeling framework has been developed at the University of Michigan to predict the process-microstructure relationship in metal AM processes. A 3D computational fluid/powder dynamics model is first used to simulate the complex multi-physics (e.g., laser-matter interaction, multi-phase fluid flow, and fluid-particle interaction) in the processes. The model has predicted several critical events for defect formation (e.g., keyhole collapse and powder spattering) as well as the thermal conditions for metal solidification. Then a computational materials model is used to predict the 3D grain growth during metal solidification. The model has quantitatively revealed the effects of nucleation mechanisms and heat source scanning patterns on grain structure development in AM metal builds. The collaboration of these two models will enable Integrated Computational Materials Engineering (ICME) for holistic strategies to design materials and process simultaneously for the metal AM technologies.

I view manufacturing as an integration platform that translates ideas and resources into products used by societies. Current research efforts at our manufacturing group are rooted in advancing new flexible manufacturing processes using the combination of the mechanics-driven and data-driven approaches. In this talk, I will post the manufacturing challenges that we are facing and use two flexible processes, i.e., metal powder-based additive manufacturing and rapid dieless forming for producing three-dimensional parts without geometry-specific tooling, as demonstration cases. Specifically, I will show how the integration of the fundamental process mechanics, process control, and techniques including machine learning to achieve effective and efficient predictions of material’s mechanical behavior due to or during a manufacturing process. Our solutions particularly target three notoriously challenging aspects of the process, i.e., long history-dependent properties, complex geometric features, and the high dimensionality of their design space. Finally, I will state the research needs towards paving the foundation for better connecting designers and manufacturers.

Digital Twin development is driving the transformation of organizations today and beyond. Enabled by the rapid advances in information technology, Digital Twins are the implementation of moving work from the physical world into the virtual world. Digital Twins are intended to both replace and mirror the events of physical world and provide a probabilistic window into the future. By substituting information for wasted physical resources, Digital Twins drive more effective and efficient decisions. Digital Twins are being proposed and considered for a wide variety of industries and many aspects of human activity. Critical to the adoption and success of Digital Twin is understanding what are Digital Twins and their virtual environment, their use over the entire lifecycle of their physical counterparts, and their potential and challenges. If Digital Twins continue on their current trajectory, they promise to transform not only industries but many aspects of society at large.

One of the basic principles of manufacturing is control of the unit processes that make the core components of any product. The foundations of product quality begin at this stage. But with a continual flow of material through the process, and increasingly rapid changes in product demands, significant variation is inevitable. Process control is about responding to that variation and, of course, seeking to minimize it. In this presentation, the distinction between controlling the process equipment and the process output itself is first made and then a classical feedback approach and examples are presented. The limitations to this approach lead to embracing a more classical statistical control method, and then turn to a hybrid, combining both statistical methods and feedback. Finally, the “holy grail” of continuously controlling the output of a process is presented as the goal of Smart Manufacturing. Speculation on how new tools for data collection and supervised learning methods are presented as the next generation of process control.