Gradient Enhanced Vibration Control

1
Gradient Enhanced Vibration Control

• Alex Forrester Andy Keane
• Computational Engineering and Design Research
  Group
• University of Southampton

2
Damper

• Triangular truss concertinaed to provide
  vibration control mounting
• Leads to problems with
• acute joint angles
• beam intersections

3
Acute joint angles
4
New joint design
5
Beam intersection problems

• Ignore infeasible designs?
• Gradient descent geometry repair?

6
Geometry repair using FD/AD

• (one function evaluation 15s)

• Three bay geometry
• FD 9s
• AD 25s

initial
repaired
7
Hybrid GA/gradient descent

• 3 bay structure
• simultaneously optimise positions of 6 joints (18
  variables)
• 5 bay
• 12 joints (36 variables)

8
Genetic Algorithm

• 3 bay
• population 50
• generations 10
• 19 minutes
• opt 4.836x10-4
• 5 bay
• population 100
• generations 10
• 56 minutes
• opt 1.719x10-4

9
GA Convergence
10
Clustering
11
Gradients
12
Gradient descent - 3 bay

• FD
• 244 evaluations
• 59 minutes
• best opt 3.528x10-4
• avg. opt 4.067x10-4
• AD
• evaluations
• 63 minutes
• best opt 3.565x10-4
• avg. opt 4.506x10-4

13
Gradient descent - 5 bay

• FD
• 445 evaluations
• 69 minutes
• best opt 1.018x10-4
• avg. opt 1.409x10-4
• AD
• 77 evaluations
• 45 minutes
• best opt 1.001x10-4
• avg. opt 1.883x10-4

14
(No Transcript)
15
NSGA2 energy / gradients

• s estimated from AD gradients
• GA based algorithm produces Pareto front of non
  dominated designs
• Time saved compared to local DoE techniques

16
Possible future work with AD

• Extend geometry repair to new joint design
• Geometry repair in gradient based search?
• AD of active vibration control code