Skip to Main Content

Strategic Business Insights (SBI) logo

Novel Ceramic/Metallic Materials July 2016 Viewpoints

Technology Analyst: Cassandra Harris

Machine Learning Accelerates Materials Discovery

Why is this topic significant?

Researchers are developing machine-learning algorithms to identify materials with targeted properties. Machine learning could accelerate the discovery of novel materials in a range of applications.

Description

In May 2016, researchers at Los Alamos National Laboratory (New Mexico) developed a method of machine learning to predict compositions of nickel titanate shape-memory alloys (SMAs) with low thermal hysteresis. Machine learning is a branch of artificial intelligence that uses algorithms that can iteratively learn from past data and situations to identify patterns or make predictions without explicit programming for where to look.

Researchers designed the algorithm using data from experiments on several compositions of the alloy Ni50-x-y-zTi50CuxFeyPdz. The algorithm then determined concentrations (x, y, z) of the dopants copper (Cu), iron (Fe), and palladium (Pd), which are likely to give alloy compositions with the lowest thermal hysteresis. The researchers verified the thermal-hysteresis properties of the predicted SMAs using experimental feedback. Deformed SMAs remember their original shape—thermal hysteresis is an important property in determining the SMAs' level of recoverable strain (see the March 2016 Viewpoints).

This research—which appeared in the journal Nature Communications—is among the first to demonstrate how machine learning could facilitate the discovery of new materials.

Implications

Researchers often perform materials discovery and properties optimization by trial and error. These processes can be resource intensive and time consuming—often taking many years. Machine-learning techniques—although in their infancy—could accelerate the discovery and optimization of advanced materials and furthermore accelerate the development of a variety of material applications and technologies. For example, researchers could adapt machine-learning algorithms to aid the discovery of other materials such as ceramics or to target certain properties such as bandgaps and conductivity or to find application in the optimization of manufacturing methods.

The current application of machine learning in materials science has several limitations. For example, algorithms require prior experimental data and understanding of the system in question. However, the physical properties of many materials are still poorly characterized. The algorithm developed by the researchers at Los Alamos National Laboratory does not predict the ease by which researchers could synthesize or characterize targeted SMAs or the materials' stability under certain conditions. However, researchers could probably predict material stability by incorporating other parameters (such as thermodynamic or kinetic parameters) into algorithms.

Impacts/Disruptions

Although machine learning potentially holds great promise for materials discovery, several barriers will slow progress.

The application of machine learning in materials discovery requires algorithms that are specifically tailored to the system under investigation. At present, algorithm development is time intensive and necessitates access to supercomputers and intellectual input from people with both chemical and computer-science know-how. These resources are often restricted to universities or select institutions, and collaboration between research groups and industry are most likely. Clear intellectual-property rights will be important for the successful commercialization of algorithms for materials discovery.

However, machine learning is already finding increasing application in a variety of industry sectors such as finance, education, and health care. Breakthroughs in algorithm design and understanding will lead to application in materials science and discovery.

Scale of Impact

  • Low
  • Medium
  • High
The scale of impact for this topic is: Medium to High

Time of Impact

  • Now
  • 5 Years
  • 10 Years
  • 15 Years
The time of impact for this topic is: 5 Years to 15 Years

Opportunities in the following industry areas:

Materials science, materials discovery, properties optimization, aerospace, automobiles, robotics, civil engineering

Relevant to the following Explorer Technology Areas:

3D Printing Metal-Matrix Composites

Why is this topic significant?

Companies are using 3D-printing technology to manufacture increasingly complex materials such as metal-matrix composites, which find application in the aerospace and automobile industries. Start-up company Fabrisonic is developing a novel method of 3D printing metal-matrix composites.

Description

Fabrisonic (Columbus, Ohio) is developing a hybrid additive-subtractive 3D-printing process—ultrasonic additive manufacturing (UAM)—to 3D print metal-matrix composites (MMCs).

The UAM process uses ultrasound to bond layers of metal foils together in a brick-like network and a computer-numerical-control mill to create the final shape of the object. High-frequency ultrasonic sound waves remove metal-oxide layers that naturally form on the metal-foil surfaces, which enables the foil sheets to bond together through the process of solid-state welding. The UAM printing process can be interrupted at any point, and reinforcing materials can be incorporated into the object to form a composite. The layers of reinforcing material can orient in various directions to maximize the strength of the material.

Fabrisonic has manufactured MMCs composed of continuous alumina ceramic fibers in aluminum metal and silicon carbide fibers in aluminum laminate. The company is producing aerospace components for NASA (Washington, DC) and Boeing (Chicago, Illinois).

Implications

Researchers can prepare MMCs by other methods of 3D printing such as selective laser melting (SLM)—a technique similar to selective laser sintering (see the May 2015 Viewpoints)—or by using plastic inks (see the February 2016 Viewpoints). SLM and plastic-ink 3D-printing methods can be expensive, and both techniques rely on the application of high temperature, which can result in deformation, fracturing, or nonlinear shrinkage of the printed object.

The UAM technique does not use heat as the mechanism of material binding—Fabrisonic states that the UAM process does not exceed temperatures of 200°C). The company has incorporated thermocouples and sensors into UAM-printed objects by pausing the printing process and drilling holes into the object. This feature could be of significant interest to the aerospace and automobile industries, which demand both MMCs and precision-measurement technologies.

Fabrisonic states that the UAM process generates low-level waste and can print between 15 and 30 cubic inches of material per hour. Rapid printing in combination with low-temperature operation could give the UAM process competitive advantage over other 3D-printing techniques and deliver cost and energy savings to manufacturers of both composite and single-phase materials.

Impacts/Disruptions

Further research is necessary to determine whether the properties of UAM 3D-printed MMCs are compromised in any way. MMCs are in use in a range of structural applications where stress and fatigue resistance are essential. The UAM technique may not be suitable for the manufacture of MMCs with certain microstructures. For example, metal-matrix foams, which comprise a high level of porosity or intricacy, could be time consuming to manufacture by means of UAM, because the process must frequently pause. However, the technology could potentially find application in the 3D printing of MMCs and other composite materials with complex microstructures that are difficult to prepare by other techniques.

Scale of Impact

  • Low
  • Medium
  • High
The scale of impact for this topic is: Medium to High

Time of Impact

  • Now
  • 5 Years
  • 10 Years
  • 15 Years
The time of impact for this topic is: Now to 5 Years

Opportunities in the following industry areas:

Aerospace, automobiles, integrated systems, internet of things, structural engineering

Relevant to the following Explorer Technology Areas: