1. Iterative Numerical Method for Predicting Additive Manufacturing Surface Smoothing Processes
ME 535 Final Project Proposal
Eric Bol, Michael Daffon, Curtis Doyle, Andrew Gubel

2. Our Vision
Everything we do, we believe in more efficient transportation for the world, we believe in advancing the human experience. The way we make transportation more efficient is by additively manufacturing solutions to complex engineering problems. This is done utilizing our custom designed iterative numerical method which make our additive products everlasting and cost effective.

3. The Problem
Additively Manufactured (3D printed) metal components have poor surface finish when compared to traditional manufacturing methods
Surface smoothness is important to the long life of fatigue critical parts, especially for aerospace applications
With the potential for highly optimized complex geometry, a traditionally smooth machined surface finish is impractical and extremely costly
Surface smoothing operations remove unknown quantities of material requiring conservative (wasteful) estimates to be made in the preform design
Can we predict how much material will be removed by a smoothing method to give an engineer the preform surface offset dimension required to achieve the nominal surface after the smoothing operation?

4. Solution – Iterative Numerical Methods
Generate a metal 3D printed representative surface
Characterize Surface Roughness (Ra)
Calculate mean and maximum
Input desired smoothness
Scripted procedure that simulates a post process smoothing operation which can be calibrated for tumbling, chem-milling, grinding, grit blasting, etc.
Make use of FFT, Wavelets, and other numerical methods for code efficiency
Iterate to desired smoothness
Final Predicted Ra value
Reduced Thickness
Quantity of material reduced

5. Initial Surface Generate the representative surface dataset
Artificial Surface Generator
Extracted 2D Cross section
Maximum
Peak
Mean
Range
Ra = 861 μin
Valley
Minimum

6. FFT Smoothing Process Smooth Surface loop Apply local smoothing Input:
Desired Ra
Method of Smoothing
Measured Surface
Output:
Plot Initial and final Surface
Final Ra
Reduced Thickness
Reduced mass
Determine if first section is peak or valley
FFT surface data
Power Spectral Density (PSD) of FFT
Calculate:
New Mean
New Thickness
New Ra
Calculate:
Mean
Thickness Ra
Max Difference
FFT
Desired Ra met
Find large power frequencies
Zero others
Multiply by FFT
PSD
If Valley, smooth 1%
if Peak, smooth 5%
Forced mean reduction
iFFT
Last section smoothed
iFFT
Second derivative for inflection points
Go to next section

7. Results of FFT Smoothing Method
Sectioned smoothing (window method)
FFT of entire plot with high degree of smoothing
Second derivative to ID inflection points
Smooth Peaks and Valleys with different scalings

8. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 861
New = 695

9. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 695
New = 558

10. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 558
New = 446

11. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 446
New = 370

12. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 370
New = 302

13. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 302
New = 238

14. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 238
New = 205

15. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Previous = 205
New = 186

16. Results of FFT Smoothing Method
Ra (μin)
Initial = 861
Final = 186
Final Difference
2.403
Area Removed
2.432 x10-3 sq. in.

17. Results of FFT Smoothing Method
Too much smoothing of valleys
Can we try another method?

18. Weighted Smoothing Process
Defining weight factors
Peaks
Valleys
proximity to mean
Input:
Desired Ra
Method of Smoothing
Measured Surface
Output:
Plot Initial and final Surface
Final Ra
Reduced Thickness
Reduced mass
search for local peaks and valleys
apply weights
Calculate:
New Mean
New Thickness
New Ra
Calculate:
Mean
Thickness Ra
Max Difference
Desired Ra met
Determine distance to mean
apply weight
Forced mean reduction
Smooth with wavelet
bior 3.3 level 3

19. Weighted Smoothing Method
Ra (μin)
Initial = 861
Final = 192
Final Difference
1.704
Area Removed
1.831 x10-3 sq. in.

20. Weighted Smoothing Method

21. Comparison
Comparison of FFT Smoothing method to Weighted Smoothing

22. Actual Surface Smoothing
Before
After

23. Conclusion Parametric method developed to simulate surface smoothing
2nd Derivative of global FFT worked well for identifying peaks and valleys
FFT is not a great method for non-periodic data and localized smoothing
Wavelet performed better than FFT (knowing both time and space)
Follow on work:
Calibration & validation with actual smoothing process

24. Challenges and opportunities in Design for Additive Manufacturing
“To realize the full potential of AM, however, we must change the way we design things too. As design engineers, our first challenge is to break out of the conceptual barriers created by conventional fabrication techniques. Researchers in cognitive psychology and engineering design have demonstrated designers experience a powerful tendency to adhere to designs they have encountered previously. The difficulty is that most designers have primarily - and often exclusively - observed, reverse engineered, and designed conventionally fabricated parts. Those parts are subject to all of the design-for-manufacturing guidelines and restrictions that accompany injection molding, casting, machining, and other common manufacturing techniques. When presented with a clean slate and an AM machine, it is difficult for many of us to conceive of a marketable design that cannot be made any other way. Exemplars are important tools for changing perspectives, as are AM education initiatives that introduce new generations of engineers to these tools and techniques.”
Carolyn Conner Seepersad
Associate Professor, University of Texas
Department of Mechanical Engineering