3D printing, also known as additive manufacturing (Additive Manufacturing, AM), is currently one of the cutting-edge technologies of human manufacturing. It is expected to make it possible to manufacture products that were previously difficult to manufacture. Food, weapons and even artificial organs have broad application prospects.
Although there are many materials used for 3D printing, such as metal materials, non-metal materials, and medical biomaterials, most materials have performance trade-offs because many materials are designed with inefficient methods based on human intuition. Not the best material solution. Recently, a research team from the Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory (CSAIL) proposed a machine learning method that can accelerate the discovery of 3D printing materials with the best mechanical properties. Printing materials using data-driven multiobjective optimization” was published in the scientific journal Science Advances.
Without prior knowledge of the main formula, the method proposed in the paper automatically revealed 12 best formulas after only 30 experimental iterations, and expanded the performance space discovered by 288 times. This method is expected to be popularized Go to other material design systems to realize the automatic discovery of the best material.
Find the best compound formula
In recent years, glass, batteries, high-temperature ceramics, and artificial organs have successfully achieved 3D printing. Among various polymer printing methods, stereolithography and material jet 3D printing have shown good application prospects, such as robot components, Prostheses, biological stents and customized products (such as footwear, clothing, buildings, models, etc.). However, the development of new 3D printing materials currently relies on knowledge in the field of polymer chemistry and extensive experiments to discover, which limits the efficiency and scalability of material development. And nowadays, 3D printing materials are generally designed and optimized using one performance factor at a time. This method usually requires testing too many samples, resulting in a lot of waste and adverse environmental impact, but the best solution cannot be found. Therefore, if 3D printing technology wants to become more popular, it is very important to accelerate the development of materials with the best performance. Moreover, in order to meet the technical challenges of different application fields such as bioengineering and aerospace engineering in the future, 3D printing also needs to be able to optimize material properties for specific applications.
In the paper, the researchers proposed a semi-automated data-driven workflow, looking for a new type of light-curable ink for 3D printing technology, showing cost-effectiveness and efficiency. The purpose of the workflow is to find a set of optimal composites. In the experiment, the material scheme is composed of six main light-curable ink formulas to improve the mechanical performance and exceed the performance level of the main formula manually designed. These compound formulas can be automatically optimized for multiple performance goals. Conduct limited experiments.
Schematic diagram of the work flow of the accelerated material discovery system (Source: Science Advances)
The workflow is shown in the figure above. First, the researcher distributes the primary formula according to the specific ratio according to the needs (Figure A), and then mixes them thoroughly (Figure B) to prepare the compound formula, and then transfers each compound formula to the injection valve 3D Sample preparation is performed in the printer (Figure C), and then post-processing (Figure D) is performed to complete the sample preparation. Finally, the samples are tested to extract multiple quantitative mechanical performance parameters (ie toughness, compressive modulus and maximum compressive modulus, compressive strength) (Figure E).
In order to minimize the resources required to test different formulations and quickly find a better performance design, the researchers used a data-driven approach based on Bayesian optimization (Figure F).
A key insight in the entire decision-making process is to balance the use of the most promising formulas and explore the uncertain areas of the design space. The experimental results show the rapid performance improvement and the discovery of 12 3D printing materials. The best fusion scheme was achieved after only 30 algorithm iterations. The method can also be easily extended to other formulation design issues, such as toughness. Hydrogels, surgical sealants or nanocomposite coatings are being optimized.
Performance space volume increased by 288 times
Specifically, regarding the basic ingredients and material formulations, the researchers first generated a set of compatible light-curing primary formulations to mix and have different mechanical properties. Of course, they did not develop printing materials from scratch, but first determined Eight commercial formulation components (including one photoinitiator, three diluents, and four oligomers) were created. Then, six main formulations (A to F) were composed of eight main components in the library. To ensure that all possible combinations of formula ingredients can be 3D printed, and within the printable viscosity range, the researchers also added surfactants to adjust the surface tension of the material and increase compatibility with the printer.
The main formulas and main formula properties used in the system, covering a wide range of mechanical properties (Source: Science Advances)
After that, the researchers used 3D printing based on jet valve dispensing technology to conduct experiments. Compared with other types of 3D printing technology, jet valves can dispense ink materials with a variety of fluid characteristics and require less process parameter adjustments. Reliable printing process, these features increase the types of materials that can be tested, which can reduce the time for sample preparation and data collection.
Finally, in order to extract performance data from each formulation, the researchers used a universal tester to perform compression tests on the 3D printed and post-processed samples. The goal of the optimization algorithm proposed in the paper is to navigate in the 6D design space of the main formulas A to F, and quickly find the best performance design for three goals: toughness, compressive modulus, and maximum strength. These performance indicators are selected because these characteristics are important mechanical properties in engineering applications. Generally, the properties of these three materials need to be maximized.
However, these goals often conflict with each other, so there is no single optimal solution, but a set of optimal performance designs with different trade-offs. The machine learning method proposed in the paper learns to predict the performance of untested samples and guides the sampling of the design space to quickly find a design with better performance.
Overview of optimization algorithms used to find the best 3D printing material formulation (Source: Science Advances)
In order to test the material development workflow proposed in the experiment, the researchers performed a total of 30 algorithm iterations, because in addition to the initial data set, the budget was fixed at 120 samples. In each algorithm iteration, in order to reduce the time, four samples were tested in parallel, a total of 120 samples were tested in the optimization process, after a total of 150 samples (30 initial samples and 120 samples proposed by the algorithm) were tested , The system finally determined a set of 12 formulations, which have the best trade-off in terms of three mechanical properties of compression modulus, maximum compression strength and toughness.
The iterative algorithm encourages exploration of unknown areas in the performance space and discovers materials with large changes in performance.
When monitoring the compressive strength and compressive modulus performance of the main formula and all evaluation samples, the performance space will be expanded by 250%; the compressive strength and toughness will increase by 399%; in terms of compressive modulus and toughness, The performance space has increased by 584%. The convex shell is a measure of the volume of the enclosed performance space in all test samples. It is 288 times larger than the performance space volume of the first five main formulations. These improvements may be important for applications that require specific attribute ranges and cannot be easily found manually.
In the experiment, the researchers also found that the optimized data set can provide interesting results about the influence of the chemical composition on the final mechanical properties of the material. For example, polyurethane dimethacrylate (UDMA) is the main component in the base mixture F and is considered Great contribution to high modulus materials, this contribution may be due to its high conversion rate and the tendency to form hydrogen bonds. In addition, researchers have also seen that algorithm optimization engines tend to minimize the contribution of hexafunctional aliphatic urethane acrylate, a highly cross-linking agent that tends to fragile prints. Through the use of polyurethane modified acrylate oligomer (content 24% to 37%), aliphatic polyurethane diacrylate (content up to 26%) and UDMA (content up to 40%) formula, high toughness performance, high The toughness formula also contains the diluents acrylamide and acrylate in the range of 14% to 18% and 1% to 19%, respectively. The highest performance compressive strength compound formula includes oligomer, 34% polyurethane modified acrylate, 26% aliphatic polyurethane diacrylate and 6% UDMA. They also include diluent, 15% acrylamide and 19% acrylate.
Provide a new research foundation
The researchers concluded that the method proposed in this paper provides an automatic preparation “pipeline” for improving the performance characteristics of the hybrid polymer system. From mixing to sample processing, each step of the process can be fully automated, which provides a template for the automated process , The template can be adapted to various optimization needs, such as coating or molding, by changing the substrate used in the experiment.
However, this study also has some limitations. For example, when defining the design space, the basic components are limited to the selection of known printable inks or materials. Although this improves the efficiency of the experiment, some innovative combinations outside the basic ink material combination may be missed. The choice of jet valve distribution as the printing process allows a wide range of materials to be considered, but this also prevents the results from being directly applied to commercial printing processes to a certain extent.
What is worthy of recognition is that this kind of scientific research idea still opens a new door, and the material discovery system described in the paper is optimized3D printingThe photopolymer formulation provides a new method, using this system, the industry can find a set of3D printingThe new formulation of materials provides the best trade-off in terms of mechanical properties such as compressive modulus, compressive strength and toughness. This lays the foundation for material engineers and polymer chemists to find and optimize material formulations for various performance goals and applications.
(Editor in charge: admin)
0 Comments for “MIT’s latest AI algorithm expands the performance space of 3D printing materials by 288 times”