Vous êtes sur la page 1sur 1

Executive Summary

Optimal Path Following of Planar Manipulator using Novel Hybrid Optimization


Scheme – Suren Kumar
Problem Summary: This project introduces a novel optimization scheme derived by using the elements
of Powell’s Conjugate direction method and artificial potential field method to transform constrained
optimization problem into unconstrained optimization problem. This hybrid approach is used to solve
dimensional synthesis problem of a four bar for following a closed loop trajectory. Classic approach for
trajectory generation in to approximate trajectory by a polynomial or use spline interpolation which fail
in the case of four-bar because its output trajectory is intricately non-linear. Previous literature in
mechanism design refers to the point that must lie on output trajectory as precision point. Number of
such precision point achievable by mechanism is restricted by number of design variables of mechanism
which is limited in mechanism. Trajectory such an ellipse tracking would require infinite number of
precision point synthesis which is not possible by classical approaches. Using optimization for synthesis
can produce results which can closely track a desired trajectory within acceptable tolerances.

Solution Approach: Direct search is the only method to solve optimization problems when taking
gradient of objective function and constraints is either computationally expensive or not available.

Four- bar involves a logical grashof constraint which takes values true or false and hence its derivative
can’t be found. If grashof constraint fails, no solution can be found which allows four-bar to follow a
closed loop trajectory. Powell’s conjugate direction method is the best direct search algorithm till date
but it can only solve unconstrained problem.

This project proposes treating violation of constraints as high potential peaks which create a steep wall
whenever a constraint is violated and hence powell’s method is forced to retreat and search in other
directions for finding optimal solution.

Outcomes of the Optimization: Efficacy of this approach is showed by results wherein search vector
dimensions are achieved within tolerance < 0.01 and function value is achieved with tolerance < 0.01
and the solutions of search vector never violated the constraint in the entire solution process. Within 5
complete loops of powell’s search, this method decrease the objective function value from 355 to 0.5.
The hybrid optimization scheme also posses’ super linear rate of convergence.

Recommendations for Next Steps: Traditional optimization techniques converge to a local minimum
near the initial design point. However there are algorithms like Genetic Algorithms for optimization can
help in exploring global design space. Currently only 6 parameters are varied due to lack of computing
power, varying base location parameters effect on performance of optimization needs to be
investigated. Stricter convergence criteria and small step size can also in achieving a better solution. Also
the effectiveness of this hybrid optimization scheme on other optimization problems needs to be seen
which currently might suffer from numerically ill-conditioning when taking gradient by finite difference
method.

Vous aimerez peut-être aussi