Cutting-edge technologies for interdisciplinary materials research
by Amber King, Louisiana EPSCoR
Bringing together different disciplines of science is at the core of the advanced manufacturing and materials research effort being conducted by the Louisiana Materials Design Alliance (LAMDA), consisting of researchers from five universities: Louisiana State University (LSU), Louisiana Tech University, Southern University, Tulane University, and University of Louisiana at Lafayette. The alliance is funded by a $20 million cooperative agreement with the National Science Foundation and support from the Louisiana Board of Regents.
This is an exciting era in advanced manufacturing research, bringing together Louisiana’s lead researchers from several disciplines and partnering with industry and national labs. An important part of this research team is graduate and undergraduate students. Louisiana’s bright minds perform critical research under the mentorship of LAMDA’s primary investigators. One of LAMDA’s key student researchers is Mr. Saber Nemati, a PhD student at LSU.
Photo: LAMDA PhD student Saber Nemati working with Avizo for 3D reconstruction of tomography data at LSU.
Nemati researches how to detect materials defects inside of advanced manufactured parts. “Currently, I am working on the applications of machine learning in materials engineering. Particularly, we are trying to apply novel intelligent methods for real-time monitoring of additive manufacturing parts using synchrotron/neutron beamline tomography for material characterization and microstructure analysis. In other words, we want to create an ‘artificial materials specialist’ who is as smart as an experienced engineer and as fast as a computer, and of course does not face the common safety issues that humans do,” said Nemati.
“This is not only a hot field of study, but also an exceptional opportunity to deal with today’s technological trends in research and development,” added Nemati.
This research is truly interdisciplinary, combining theoretical, experimental and numerical aspects of materials science, engineering, computer science, and physics fundamentals. Nemati belongs to a team who designs different polymer and alloy materials, algorithms, testbeds and specimens and then tests their strength with supercomputers. After the virtual testing, the experimental results are physically tested in the lab to verify the results.
Nemati works with Dr. Shengmin Guo, an expert in Additive Manufacturing, Dr. Xin Li, who is known for his expertise in Image Processing, and Dr. Les Butler whose research is focused on Tomography. “It is a great honor and a big opportunity for me to work under the supervision of these acknowledged experts,” said Nemati.
Photo: Saber Nemati performing high frequency ultrasonic fatigue testing of small key-shaped specimens for evaluating fatigue life of additive manufactured parts.
“Is there any way to do this better?”
This was the recurring question that Nemati would ask himself, as a young student who loved puzzle magazines and, later as an engineer, in ordinary situations, such as riding a public bus and thinking about improving the paper ticket process. This mindset led him to study mathematical sciences in high school and mechanical engineering in college in Iran where he grew up.
After acquiring the basic knowledge earning his undergraduate degree and a masters degrees in Mechanical Engineering, he broadened his experience by working in industry and facing the challenges of real-world problems. He later felt the need for a broader perspective, and earned his MBA with a focus on Strategic Management, which helps him have a bigger picture of the whole process, from identifying customer needs to production process to sales and marketing.
Ultimately, he wants to be an entrepreneur who can develop a sustainable relationship between science and industry. He has persistently paved the way to achieving this goal during his education and career.
He is currently working toward his goal of completing his PhD in Engineering Science at LSU. He decided on attending LSU when he visited Dr. Guo’s webpage, and saw that under the research goal it was written, “Provide challenges and opportunities for future engineers and high-tech leaders.”
“I believe with his attitude, no other mentor can contribute more to my decision than him,” said Nemati.
“The most challenging and exciting skill that I’m learning in this project, is having the ability to communicate with various scientists and experts from different backgrounds.
Normally, computer scientists and materials engineers do not have so much in common. But with LAMDA and its interdisciplinary nature, it is absolutely crucial to learn how to maintain a scientific communication among different groups,” concluded Nemati.
Accelerating deep learning discovery of new thermoset shape memory polymers
Cheng Yan, Xiaming Feng, Louisiana State University; Collin Wick, Andrew Peters, Louisiana Tech University; Guoqiang Li, Southern University
Researchers with the Louisiana Materials Design Alliance (LAMDA) established a machine learning framework to predict the recovery stress of thermoset shape memory polymers (TSMPs) and to discover new TSMPs with superior recovery stress. The team leveraged a new linear notation computer language for the digital representation of polymers, called BigSMILES, to fingerprint complex TSMP structures and establish structure-property correlations using a small training dataset. This information helped to identify two new TSMPs predicted to have high recovery stress, which was synthesized in the lab to validate the model predictions. Finally, they explored a chemical space with 4,459 possible TSMPs and screened 14 mostly unknown TSMPs with higher recovery stress than any TSMPs in the training dataset. One of 14 TSMPs was modeled by molecular dynamics simulation and found to have calculated recovery stress in agreement with predicted values.
Mr. Cheng Yan, a Ph.D. student in the Department of Mechanical & Industrial Engineering at Louisiana State University, is designing and encoding the machine learning pipeline.
To our knowledge, this study is the first to discover TSMPs with high recovery stress by leveraging machine learning. Using a small dataset of about 100 molecules, this method has the potential to greatly increase the ability to explore chemical space and bring remarkable advancements over previous materials discovery methods.
a) A bottleneck for current TSMPs persists in their low recovery stress in their rubbery state, limiting their applications as actuators or as crack closing devices in self-healing applications.
b) Due to the time it takes to synthesize new TSMPs, the traditional trial-and-error method for materials discovery needs a long period of time, deep domain knowledge and skills, and some luck. This new method overcame these limitations and quickly discovered 14 new TSMPs.
c) The machine learning approach is at least hundreds of times faster than the traditional molecular or atomistic computational approaches, such as molecular dynamics simulation and density functional theory based electronic calculation.
Traditionally, predictions of thermomechanical behaviors of TSMPs, such as recovery stress, rely on multi-parameter constitutive models, and most of the parameters need to be determined by curve-fitting. This machine learning model, on the other hand, only depends on the basic chemical structures and so can be applied nearly universally.
TSMPs are a new class of smart polymers, which, after deformation, can maintain their deformation nearly permanently until a stimulus, such as heat or an electric field, is applied. This stimulus causes the deformed TSMP to restore its original shape. TSMPs have found many applications, such as crack self-healing in lightweight structures, stents in medical applications, and artificial muscles in soft robots. However, a persisting critical limitation of existing TSMPs is their very low recovery stress in their rubbery state, usually less than a few Newtons per square millimeter. This recovery stress is too low for some critical applications, such as the crack closing. It is also too low to compete with shape memory alloys, which may have tens to hundreds of Newtons per square millimeters in recovery stress. Unfortunately, the chemical space constituting TSMPs is very large, and using trial and error approaches is not sufficient to identify new TSMPs with higher recovery stress. Therefore, machine learning is a natural choice. As compared to conventional polymers—which have already had a large database in Materials Genome Initiative—only a limited number of TSMPs are available for training and fingerprinting. In this study, we successfully overcame this limitation and discovered 14 new TSMPs. This machine learning approach can be expanded to discover other types of materials.
Pipeline for new TSMPs discovery: first, collecting monomer set in the two datasets (c1), and then automatically generating random combinations of monomers and crosslinkers, which produce new TSMPs (c2). Next, fingerprinting is performed for these new TSMPs (c3) and input into the glass transition temperature model (c4) to predict the corresponding glass transition temperature (c5). Then, programming temperature Ttr, recovery temperature (c6) (approximated from glass transition temperature, i.e., Ttr = Tg +20 °C), presumptive uniform strain (c7), and fingerprints of the newly formed TSMP structures (c3) are input into recovery stress model (c8) to predict the corresponding recovery stress σs (c9), which is validated by experiments (c10). Finally, two-step screening processes were conducted. First, if the predicted recovery stress is greater than the maximum recovery stress in the training data, then the corresponding TSMPs are recorded (c11); second, by employing the prior knowledge, promising TSMPs with higher recovery stress than that in the training dataset are be further screened (c12).
Phosphate esters dynamic chemistry enables flame retardant vitrimers
Xiaming Feng, Louisiana State University and Guoquiang Li, Southern University
Researchers with the Louisiana Materials Design Alliance (LAMDA) report the first phosphate esters based thermoset polymers that are mechanically strong, completely malleable and recyclable, and significantly safe under fire. These promising properties rely on the unique behaviors of β-hydroxy phosphate esters at low, medium, and high temperatures, respectively. At room temperature, the abundant hydrogen bonds in the network contribute to outstanding toughness (5.44 MJ/m3). Around 100 oC, the catalyst-free rapid exchange reaction between phosphate esters and neighboring β-hydroxyls endows the polymer with almost 100% recycling efficiency. Above 250 oC, a cellular layer of charred phosphoric acid generated from β-hydroxy phosphate esters could separate/insulate the heat effectively, providing fire protection. In addition, by combining phosphate diesters and acrylates, a new polymer integrated with ultraviolet (UV) curability, recyclability, and flame retardancy are also developed. This highly crosslinked network exhibited attractive recyclability even at the temperature lower than glass transition temperature. The fast exchange reactions via catalyst-free mixed transesterification between phosphate diesters and carboxylate esters of acylate structures are validated.
(Top row) The broken thermoset vitrimer pieces after service can be reshaped into a new transparent sheet by a simple hot pressing for reuse. (Bottom row) The vitrimer sheet under the cotton ball generates an expanded char layer upon external fire to protect the cotton ball from temperature rising and burning for a couple of minutes.
The phosphate dynamic chemistry proposed here is a fantastic drop-in technology that can be easily used to develop a broad range of high-performance vitrimers while possessing intrinsic flame retardancy. Coupled with the high transparency, these self-healing fire-safe vitrimers can serve as multifunctional coatings for metallic structures or components with a high risk of fire and corrosion, such as in construction fields and electronics. The UV curability of the combination of phosphate diesters and acrylates enables the printing of customized and complicated structures in these advanced fields, such as robotics and aerospace, using digital light processing (DLP) technology. We believe that this work could expand the scope of dynamic covalent chemistry and create new directions in developing multifunctional thermoset polymers.
Fire hazards are a well-known limitation for polymers. Thermoset polymers, while they have high mechanical strength and thermal stability, are usually not recyclable, which causes a significant waste disposal issue. Therefore, sustainability and safety have been one of the key issues in polymer science and engineering due to the shortage of natural resources, the crisis in waste disposal, and fire hazards caused by flammable polymers. Recyclable thermosets or vitrimers and fire-retardant polymers have been developed separately for years to address these challenges partially. To our knowledge, no other vitrimer has demonstrated fire-retardant capability without adding extra flame-retardant structures. Therefore, integrating robust mechanical performance, recyclability, and flame-retardancy into one polymer using the new facile dynamic covalent chemistry initiated in this study is of significant value to both academia and industry.
Recyclable Composites: The future of lightweight materials
John Konlan, Louisiana State University; Sam Ibekwe, Patrick Mensah, Karen Crosby, Guoquiang Li, Southern University
Researchers with the Louisiana Materials Design Alliance (LAMDA) developed a new self-healable and recyclable fiber reinforced thermoset shape memory vitrimer composite laminate. In this study, they used pre-tensioned shape memory alloy (SMA) reinforcing fibers called “z-pins” to control delamination, which is cracking at the interface between neighboring layers, and used the thermoset shape memory vitrimer for molecular-scale healing of the delamination. The team found significant enhancement in low velocity impact tolerance, and high healing efficiency even under repeated impact damage and healing cycles for the laminated composite.
Mr. John Konlan, a Ph.D. student in the Department of Mechanical & Industrial Engineering at Louisiana State University, is examining the self-healing composite laminate.
In recent years, self-healing and recycling of fiber-reinforced polymer composites have become a popular topic of research. However, most of the studies are either focused on damage healing of pure polymers or limited to damage healing of polymer composites with microscale cracks. In this study, by a novel combination of SMA z-pins and thermoset shape memory vitrimers, we are able to repeatedly heal wide-opened delamination in glass fiber reinforced vitrimer composites. The combination of SMA, shape memory and thermoset vitrimer opens up a new opportunity to develop the next generation of laminated composites for lightweight structural applications.
Delamination induced by low velocity impact: (left) with SMA z-pins and (right) without SMA z-pins.
Continuous fiber-reinforced polymer composite laminates have been widely used in lightweight load-bearing structures. It is well known that laminated composite is vulnerable to impact damage. For example, even a drop of a hammer during a routine inspection of laminated composite structures represents a low-velocity impact event, which may induce delamination, matrix cracking, matrix/fiber interfacial debonding, and fiber fracture. Among them, delamination can reduce the compressive load carrying capacity of the composite by over 50%. On the other hand, recycling of the end-of-service laminated composites poses a solid waste disposal issue. Therefore, how to make a laminate composite that is self-healable, recyclable, and impact tolerant is highly desired. In this study, we used a combination of SMA z-pins and thermoset shape memory vitrimer to prepare SMA z-pinned, continuous glass fiber reinforced vitrimer laminated composite. The results show that SMA-pins have significantly reduced the delamination opening (from > 0.2 mm for those without SMA z-pins to < 0.03 mm for those with SMA z-pins), and the vitrimer matrix can be healed and recycled repeatedly under multiple low-velocity impact cycles.
Inverse machine learning technique optimizes discovery of materials for sandwich structures
Adithya Challapalli, Louisiana State University and Guoqiang Li, Southern University
Researchers with the Louisiana Materials Design Alliance (LAMDA) have developed an inverse machine learning approach using the generative adversarial network (GAN) and regression analysis. Using this technique, novel lightweight lattice and cellular structures with optimized biomimetic elements are discovered, which are used as sandwich core. Thermoset shape memory polymer (TSMP) is then used to 3D print these designs, which adds smartness to the sandwich structures.
Mr. Adithya Challapalli, a Ph.D. student in the Louisiana State University Department of Mechanical & Industrial Engineering, is using a digital light processing (DLP) printer to print the machine learning optimized lattice core using a TSMP ink.
Inverse machine learning techniques have been previously used to discover new small molecules, polymers, and drugs. This helped in speeding up the process of discovering new materials with superior properties. Here, we used GAN and regression analysis to learn the correlation between the microstructures and mechanical properties of biomimetic rods and then further optimized the biomimetic rods into lattice and cellular structures as a sandwich core. 3D printing of the newly discovered lattice core by TSMP makes the new core with robust structural properties and shape memory effect. This technique can be applied to any other structural designs targeting specific features such as strength, bucking load, vibration damping, acoustic insulation, and many other mechanical, thermal, and electrical features required for a particular application. By further generalizing this inverse design technique, it can be incorporated into existing computational and simulation software as a futuristic optimization tool or as an individual application that can predict optimal designs.
Step-by-step optimization by machine learning for discovering lattice core and thin-walled cellular core for lightweight sandwich structures.
Sandwich structures have been widely used in lightweight engineering structures, such as aircraft fuselages, ship hulls, car bodies, wind turbine blades, pressure vessels, pipelines, and bridge decks. The mechanical and functional performance of sandwich structures depends on the core. While many core materials have been used, such as foam and basal wood, further advancement depends on the design and manufacturing of new core structures. Here, we use a combination of GAN and regression analysis and optimize lattice and cellular cores for sandwich structure with biomimetic rods, which is infeasible using the classical design approach. Using 3D printing and TSMP ink, these lattice and cellular cores are manufactured, which is a great challenge using the classical reduction manufacturing approach. Therefore, the combination of inverse design using machine learning and 3D printing using TSMP ink opens up many opportunities for developing extremely lightweight, robust, and smart sandwich structures for engineering applications.
New specimen design improves rapid fatigue testing
Hamed Ghadimi, Louisiana State University
A new small-sized test specimen has been designed by researchers with the Louisiana Materials Design Alliance (LAMDA) to accelerate bending-fatigue testing. Small-sized test pieces are more adaptable, economical, and timesaving in fabrication processes and contain quantifiable imperfections and defects, which makes them highly desirable for rapid qualification of new alloys and new fabrication processes.
Louisiana State University graduate student Hamed Ghadimi working with SHIMADZU USF-2000 ultrasonic fatigue testing system.
This new design of a small-sized test specimen enables fatigue property studies of new 3D printed alloys to be carried out rapidly and with significantly reduced cost. Rapid fatigue testing capability helps scientists to explore new alloys and 3D printing techniques for reliable and durable products.
Bending-fatigue testing setup and designed small-sized test specimen.
Fabrication of test pieces and doing experiments on them is important, particularly for additive manufacturing (AM) studies, like developing new AM alloys, investigating proper AM design, and studying the effects of AM build parameters on the mechanical properties. Fatigue testing is an essential part of such studies, and the high-cycle life regimes are time-consuming and costly, and a large number of test specimens with a suitable design are needed. For AM research, scientists also encounter severe restraints such as limited building-capacity, small chamber size, high costs, and unpredictable microstructure defects. Ultrasonic high-frequency fatigue testing systems overcome the time- and cost-related limitations, and in addition, reducing the size of the test specimens is a way to tackle the above-mentioned obstacles. Small-sized test specimens are also beneficial from the affordability point of view as they make the controlling of the manufacturing processes more feasible, and thus, they facilitate the production of more qualified specimens under a variety of different processing conditions.
Developing artificial intelligence for defect detection
Saber Nemati, Hao Wen, Brian P. Tsai, Louisiana State University
Supervised Machine Learning (ML)-based algorithms are being developed by researchers with the Louisiana Materials Design Alliance (LAMDA) for crack detection in X-ray tomography images. Using this method, a large number of images can be segmented with accuracy after training with just a limited number of manually segmented images.
Louisiana State University graduate students, Saber Nemati, Hao Wen and Brian P. Tsai evaluating different network architectures.
The ability to efficiently detect the cracks and different types of inclusion in tomography images is a key component in evaluating and optimizing additively manufactured (AM) parts in the testing stage. This paves the way for developing faster and more intelligent algorithms with minimal supervision, which consequently leads to a more optimized way of designing stronger, premium alloys.
Sample algorithm segmented image. (a) original image (b) Fe-Inclusions (c) Mg-inclusions.
Creating suitable additively manufactured alloys for different applications is a challenging task, as it depends on many parameters during the designing, manufacturing, and testing processes. One of the most challenging phases of this process is the evaluation of the parts since defects in AM parts are hard to detect and cumbersome in nature. Tomography is a well-known nondestructive method for observing the defects inside 3D metal parts on micro scale. Due to recent advances in machine learning, there are a lot of efficient algorithms that can identify these defects and categorize them into different classes. With the implementation and validation of such algorithms, several days of manual work can be completed in a couple of seconds. Because of LAMDA’s interdisciplinary nature, the collaboration of experts with different scientific backgrounds is inevitable. One of the most valuable yet challenging outcomes of LAMDA is building a synergetic framework that makes use of each groups’ capabilities.
Virtual Training Modules for Additive Manufacturing Bridges Knowledge Gaps During COVID Closures
Mohammad Khondoker, Southern University
A group of 72 attendees, including faculty members, graduate/undergraduate students, participated in the workshop to learn about all seven categories of additive manufacturing technologies and their underlying process physics.
Additive manufacturing development opportunities serve as a crucial early component for the workforce development initiatives aimed to benefit the U.S. manufacturing industry. In addition, faculty members and researchers from other disciplines have also gained the necessary knowledge to advance their research. More than 10 LAMDA-affiliated faculty members earned in-depth knowledge on additive manufacturing which will be useful in performing LAMDA’s research activities (both SD-1 and SD-2), as well as in developing new grant proposals in that field.
Coursework that covers all seven categories of additive manufacturing is not available at universities in Louisiana. Therefore there was a need for a workshop training module to train graduate and undergraduate students in Louisiana. Through this workshop, more than 15 graduate students and 35 undergraduate students received the necessary knowledge to prepare them for the manufacturing industry in Louisiana.
Additive manufacturing (AM) is one of the core components of the fourth industrial revolution, called Industry 4.0. Therefore, it is essential that the U.S. manufacturing industry is supplied with a trained workforce that can utilize the true capabilities of AM technologies. There are seven different types of AM technologies with unique advantages/disadvantages. A manufacturing engineer needs to know the principles of these technologies and their underlying process physics, which will help in improving the performance and lowering the cost of manufactured parts. Being a relatively new technology, AM is not well understood by students because most programs in higher education do not offer AM-related courses. Hence, arranging workshops to educate graduate/undergraduate engineering students on AM technologies plays a vital role in workforce development for the U.S. advanced manufacturing industry. Such initiatives will help the U.S. to remain the global leader in this field.
2020 and earlier
LAMDA is built upon a foundation of decades of research and leverages current and previous awards and infrastructure investments to continue the growth of materials science research and industries in Louisiana. This foundation was boosted by two previous NSF Track-1 RII awards, the Consortium for Innovation in Manufacturing and Materials (CIMM), and Louisiana Alliance for Simulation-Guided Materials Applications (LA-SiGMA).