Skip to Main Content
 

Global Search Box

 
 
 

ETD Abstract Container

Abstract Header

Predicting the Deformation of 3D Printed ABS Plastic Using Machine Learning Regressions

Stellmar, Justin

Abstract Details

2020, Master of Science in Engineering, Youngstown State University, Department of Mechanical, Industrial and Manufacturing Engineering.
FDM printing is a relatively cheap and fast method to produce parts as compared to traditional machining, which makes it attractive for many applications including tooling. Contrary to the isotropic behavior of steel tooling, the anisotropic properties of the printed materials can be attributed to FDM printing parameters. It is possible that through a proper machine learning regression model, the behavior of these parts could be captured, modeled and predicted. Predicting the deformation and stress of FDM created tools and dies could be an excellent way to save on costs for low run manufacturing due to its lower lead time, and manufacturing cost. FDM material behavior is complex and unpredictable by current testing standards, so as an alternative strategy, machine learning is proposed to model deformation and stresses. ABS cylinders were printed and compressed to depths ranging from .025-.150 inches. Fujifilm Prescale paper was used during compression to record the pressure distribution across the surface of the sample interface. The Prescale was scanned, converted to grayscale, and processed using MATLAB. Random points on the affected area of the pressure papers were selected exported. Data points were split into two bodies; one to create regression models and one to test the models. The first file was imported into MATLABS’s regression learner to create regression models. Regression models were created with three different validation strategies and tested against holdout data to estimate predication accuracy. Residual model error for prediction of compression depth and stress distribution were compared. It was found that validation types had no impact on the RMSE of regression models for both the compression depth and grayscale value predictions. Compression depth for any of the samples could be predicted between 1 – 23 thousandths of an inch. Stress, which was represented as grayscale could be predicted within 293 thousandths of a grayscale. Grayscale is a unitless dimension which ranges from white (0) to black (1). Testing the model against samples used to build it led to an RMSE of 137 thousandths using a Bagged Trees regression. These numbers cannot be directly correlated to a range of stress due to a nonlinear relationship between grayscale values and stress, however the error is significant considering that the range of grayscale values measured ranged between 0.150 to 0.802. This method was excellent at predicting the deformation of samples that had dimensions used to create the regression models, and it was decent at predicting the deformation of samples that had at least one dimension not seen by the regression models. The method was poor at predicting grayscale values for any sample, regardless of whether it was seen by the model.
Darrell Wallace, PhD (Advisor)
Jason Walker, PhD (Committee Member)
Solomon Virgil, PhD (Committee Member)
84 p.

Recommended Citations

Citations

  • Stellmar, J. (2020). Predicting the Deformation of 3D Printed ABS Plastic Using Machine Learning Regressions [Master's thesis, Youngstown State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ysu1587462911261523

    APA Style (7th edition)

  • Stellmar, Justin. Predicting the Deformation of 3D Printed ABS Plastic Using Machine Learning Regressions. 2020. Youngstown State University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ysu1587462911261523.

    MLA Style (8th edition)

  • Stellmar, Justin. "Predicting the Deformation of 3D Printed ABS Plastic Using Machine Learning Regressions." Master's thesis, Youngstown State University, 2020. http://rave.ohiolink.edu/etdc/view?acc_num=ysu1587462911261523

    Chicago Manual of Style (17th edition)