Researchers in Carnegie Mellon University's College of Engineering have developed a novel approach to optimizing soft-material 3D printing. The Expert-Guided Optimization (EGO) method combines expert judgment with an optimization algorithm that efficiently searches combinations of parameters relevant for 3D printing.
Tech Briefs: What was the problem you sought to solve with this technology?
Dr. Sara Abdollahi: I was looking to create soft material with 3D printing because I was trying to make a conformable cuff that I could integrate with small electronic sensors. I wanted to create a medical device that would measure things like heart rate and blood oxygen content, like the clunky oximeter they put on your finger when you go to the hospital. My goal was to make a wearable version of that.
I soon realized that there are a lot of parameters involved in 3D-printing soft materials, especially since the material starts as a liquid. For example, it requires a temporary sacrificial support bath. There's a lot of trial and error involved to align the parameters. I had taken some courses on policy analysis where I learned about the expert intervention that's used for policy decision-making. We decided to implement that approach for our engineering problem.
We were still able to benefit from the systematicity provided by an algorithm, but we could also utilize intervention from an expert to change the parameters and factor levels of the variables in order to escape when the model could not converge. We call this process Expert-Guided Optimization (EGO).
Tech Briefs: Where do you start when researching a new material that can be 3D-printed?
Dr. Abdollahi: There are three stages: the expert, the algorithm, and then again, the expert. The expert chooses the parameter search space and decides which factors are important; for example, bath concentration, pH, printer speed, layer height, acceleration, or nozzle diameter. We start with a random combination of the parameters and assess the print. We then take the parameters that gave the best print and start a hill climb — iteratively varying one factor at a time, while keeping the others constant. We look to see which direction we should move towards to get a higher score. Finally, it's up to the experts to judge the results. They can choose to change the factors, the factor levels, or the entire parameter.
Tech Briefs: How do you analyze the quality at each iteration?
Dr. Abdollahi: We used two different calibration structures: a hollow cylinder and a five-sided cube. We then defined variables we thought were important; for example, for the cylinder, it was layer adhesion, or stringiness, and if it was well fused with the rest of the print. For the cylinder, an important parameter is “infill,” or whether the cylinder is completely hollow or if some material protrudes into the interior. I then created a scoring system. For layer fusion, zero means the print is a big mess and 10 is when the layers are fully fused. I anticipate that at some point, we could integrate some form of physical evaluation.
Tech Briefs: What are the next steps?
Dr. Abdollahi: I would like to see this technique used with other objects and other additive manufacturing processes. Ideally, it would be nice if it could be standardized, scaled, and used in industry as a tool to get a desired product without the random trials that often lead to a lot of product waste.