Analytics for Additive Manufacturing

What Is Additive Manufacturing?

Conventional subtractive manufacturing involves starting with a piece of raw material and removing material, for example via drilling or milling, until the final desired shape is achieved. On the other hand, additively manufactured parts are built up through the deposition of thin layers of material. Most modern additively manufactured parts begin with a CAD model. Software is used to slice the model into the individual layers that will be deposited to form the final part. In one method, selective laser melting (SLM), a thin layer of metal powder is deposited on the build tray holding the part, and a laser is used to selectively melt the powder only in the locations specified by the sliced geometry data. As the melted powder cools, it solidifies together, the build tray is lowered, another layer of powder is deposited, and the process repeats for the next geometry slice.

Why Use Additive Manufacturing?

The value of additive manufacturing comes from greater flexibility in the manufacturing process, the possibility of customized parts, and the construction of complex geometric parts that would not be possible via conventional manufacturing techniques. Click on the bullet point for more information.

• Flexibility in the Manufacturing Process

Metal casting or injection molding requires expensive tooling to act as the mold for a desired part. The cost to machine a die for even a relatively simple part can be tens of thousands of dollars. Any changes to the part design will often require a completely new die or at a minimum additional machining work on the existing die. As the cost of tooling is fixed regardless of the quantity of parts manufactured, the manufacturing cost per part is greatly influenced by the number of parts produced. Thus, these conventional molding or casting processes work well for putting high volume parts into production. By contrast, the cost of additive manufacturing is not tied to the number of parts produced as no special tooling is required. This allows for the production of low volume parts such as custom pieces or quick prototypes.

• Construction of Complex Geometric Structures

Parts with internal cavities and very small, contorted, and/or intricate geometric features present a challenge to traditional manufacturing techniques. Directly machining or machining tooling for such parts may be extremely complicated, costly, or impossible. One solution is to break the desired part into individual components to be manufactured separately and assembled. This however adds complexity, generally making the part less reliable and incurs costs related to the design work, tooling, and assembly of the individual components. Additive manufacturing does not suffer from these limitations and is very flexible in handling different geometric features and levels of complexity. A common example of this is the acetabular shell (or cup) used in hip replacement prostheses. Additive manufacturing is used to give the part a surface with a fine, textured, lattice structure that would be impossible to create via conventional techniques. This structure assists in attaching the hip protheses to a patient’s bone, leading to higher success rates and more rapid recovery times.

• Paired with Conventional Manufacturing

Additive manufacturing can also be used in conjunction with conventional manufacturing such as manufacturing mold tooling that would otherwise be difficult or impossible to machine. Additive manufacturing enables engineers and manufacturers to reduce time and cost in developing a prototype, creating custom or one-off pieces, or constructing a complex part.

Additive Manufacturing Uncertainty and Analytics Challenges

As a new methodology, additive manufacturing is surrounded by uncertainty. Below are some examples of the uncertainties found in additive manufacturing and their effects. To combat these uncertainties, engineers will run highly computationally expensive simulation models that can take days to run and thus, limiting the value of its potential insight. Click on the bullet point for more information.

• Uncertainty in Mechanical Properties

Thanks to years of research, testing, and practical use, much is known about the mechanical properties, ultimately determined by the microstructure, of a wide range of materials. However, the additive manufacturing process can result in parts with drastically different microstructures compared to identical conventionally manufactured parts made of the same material. This uncertainty in the material properties makes existing knowledge of little use in predicting the additively manufactured part’s mechanical properties such as stiffness, elasticity, fatigue life, and yield strength.

• Uncertainty in Thermal Gradients

Enormous thermal gradients inherently resulting from the additive manufacturing process are a cause of the microstructure differences seen in additively manufactured parts as compared to conventionally manufactured parts. These thermal gradients are caused by the successive rapid heating and cooling, from near room temperature to the material’s melting point (for SLM) and back down, of extremely small regions of material. If the uncertainty in the thermal gradients are not properly understood and controlled, these gradients can lead to large residual stresses and cracking, or distortion of the part in excess of required tolerances.

• Uncertainty in Process Parameters

Laser power, scanning speed, scan pattern, and powder particle size are examples of process parameters which can influence the thermal response of an additively manufactured part and in turn its microstructure. A challenge of additive manufacturing is therefore to determine robust process parameters that produce a part both within tolerance and possessing of the desired mechanical properties.

In practice, this is often achieved using a trial and error approach, whereby the various process parameters are varied until the desired part is produced. Even applying basic Design of Experiments (DOE) techniques to systematically sample from the n-dimensional space of the process parameters can be a very time consuming and expensive exercise.

Even if the correct process parameters are determined, there is still uncertainty in the process, for example due to sources such as natural variation in the grain size of the material powder or fluctuation of the laser scan speed. Poorly understood or unaccounted for uncertainties can result in an inconsistent production process leading to high rates of scrap.

• Computational Burden of Simulation Models

Iterative computer simulations may also be used to help guide the selection of appropriate process parameters. However, additive manufacturing simulations are extremely computationally expensive. For example, the mesh size required to model an SLM process depends on the size of the laser, which is on the order of 100 microns. Further, the length of time steps required decreases in proportion to the laser’s scanning speed, which can range from hundreds to thousands of mm per second. So, to accurately model an SLM process, a very fine mesh with very short timesteps is required. This can result in terabytes of data for the engineer to analyze.

SmartUQ Solutions for Additive Manufacturing Analytics Challenges

SmartUQ has a variety of DOE tools for reducing sampling requirements for simulation and experimental modeling. These tools can be used to reduce the number of test pieces and/or simulations required to determine additive manufacturing process parameters. SmartUQ also features advanced DOEs specifically designed for iterative data collection in model improvement processes.

A key use of SmartUQ’s DOE tools is creating emulators, statistical models trained using machine learning algorithms. Emulators can be built from simulation or of physical data such as in-situ monitoring data. Once built, an emulator can be used to make rapid predictions in lieu of further full-fidelity simulation runs or physical data collection. For example, an emulator could be used to predict melt pool depth, important in SLM to the adhesion of material layers, from a set of process parameters. SmartUQ has a variety of emulation tools to handle different types of data sets including those with a spatial-temporal response such as the temperature distribution across a layer as a function of time for an additive manufacturing process.

The ability to perform rapid predictions with an emulator allows users of SmartUQ to apply analytics techniques whose sampling requirements would be too great for direct use. The analytics that can be performed within SmartUQ using an emulator include sensitivity analysis and uncertainty propagation.

A sensitivity analysis can be used to determine which process parameters have the greatest effect on a part or process’s desired properties such as melt pool depth, melt pool temperature, tensile strength, fatigue life, density, porosity, and surface roughness. Knowing which process parameters are insignificant to achieving a desired property allows that parameter to be eliminated from consideration in subsequent simulations or tests.

Sources of uncertainty can also be propagated through an emulator to determine the resulting uncertainty in output. For example, how do uncertainties in the scanning speed, laser beam power, or material powder particle radii contribute to the uncertainty in melt pool temperature. A sensitivity analysis can be used in conjunction with this process to identify the sources of input uncertainty most responsible for the resulting output uncertainty. Attention can then be efficiently focused on reducing the uncertainty in those factors most responsible for variability in the final part. Doing so helps in consistently producing parts within requirements and thus reducing scrap.

SmartUQ can also be used to perform statistical calibration, a means of utilizing physical or experimental data to tune model parameters to achieve improved accuracy. The tuned model parameters are ones that cannot be directly measured in the real world, for example a thermal conductivity value in a melting pool model. SmartUQ’s statistical calibration can also be used to produce uncertainty information for the calibration parameters and identify sources of model form uncertainty.

Often rather than wanting to predict a manufactured part’s properties given a set of process parameters, the opposite is desired. That is, given a set of desired properties, determine the process parameters required to produce a part with those properties. SmartUQ includes inverse analysis tools for this specific purpose.

Summary

Additive manufacturing will lead to new technologies and innovations with its flexibility and ability to handle complex structures. But without accounting for uncertainties in the changing material and mechanical properties, additive manufacturing is a long way off from reaching its full potential. By applying SmartUQ’s predictive analytics and Uncertainty Quantification techniques, engineers and manufacturers can combat the analytical challenges and continue to push the boundaries of additive manufacturing.