KINEMATIC-BASED STRUCTURAL OPTIMIZATION OF ROBOTS

: In this paper a detailed description of a method is presented to estimate the minimum structural dimensions of the robot arms. A comparative study is conducted between the harmony search and artificial bee colony algorithms in this scientific application. The comparison process was done through the kinematic equations of the serial robot manipulator to find the optimum lengths of links of the robot. A novel design for a seven-degrees-of-freedom robot arm was presented to conduct the comparative study on the presented optimization algorithms. This novel robot mimics the functionality of the SANDVIK robot arm for tunnelling works, but the presented type synthesis was designed to overcome the restrictions on the original SANDVIK arm.


Introduction
In the kinematic combination of a multi-linkage robot arm, topology optimization is performed to obtain the best type synthesis [1] for a single task or multitask robot manipulator. Next, the minimum required dimensions should be calculated during the dimensional synthesis process [2]. Both the type and dimensional synthesis can be done in two separated operations or one single operation. A genetic algorithm has been employed [3] to find the optimum topology and dimensions of a planar linkage depending on graph theory definition. A graph theory-based approach has been used [4]  to estimate the type synthesis while the best dimension parameters are calculated by the precision position method. On the dimensional level, shape optimization of the link of the robot using level set techniques have been dealt with in [5]. On the type level, the critical point in most works is the graph theory [6], where possible topologies are enumerated to get the best type. Various optimization techniques can be employed to solve different problems like traveling salesman [7], stiffened plates [8] or heat exchangers [9].
In this work, harmony search and artificial bee colony optimization algorithms were compared to investigate their efficiency to solve the structural optimization problem of robot manipulators. A specific topology of robot arm was used to find the dimensional synthesis depending on the kinematic equations. A robot arm with novel topology and 7 degrees of freedom (7 DOF) was designed and employed to estimate the best dimensionality for multitasking purposes. A comparative study also has been done between the artificial bee colony and harmony search algorithm. The test function for this comparison is the kinematic equations of the manipulator mentioned above, and this test function of the ten-dimensional length is hard enough to be a challenge to the optimization algorithms to reach to the global minimum. The new type of robot arm which is proposed in this paper is designed to overcome restrictions on the arm SANDVIK DD421 [10].

Harmony search
The behavior of the musicians in a jazz ensemble during improvisation of a new harmony first inspired the Harmony Search (HS) algorithm [11]. HS is a robust heuristic optimization algorithm [12]. The following procedure can be used to solve a problem:

Artificial bee colony
This is a metaheuristic algorithm belonging to swarm intelligence and was inspired by the foraging behavior of honey bees [13]. The artificial bee colony ABC is efficient and returns perfect solutions for many optimization problems. The procedure to solve a problem is given in detail in the following: 1. The random population of candidate solutions has to be initialized randomly; 2. Employed bee section: For each solution in the population calculate the neighborhood structure using where sol denotes a particular solution, i is the current iteration number, φ is a 3. Onlooker bee section: The probability of each solution has to be calculated according to its fitness value where i O is the objective value for any solution; f denotes fitness value; and P is the probability of the current solution in the iteration. Neighborhood structure should be applied to the solutions as mentioned in section 2. 4. Scout bee section: If any solution does not get better after a moderated time or number of iterations, it should be replaced by a random solution within the search space. The following pseudo-code can demonstrated the ABC algorithm.

Initialization:
Initialize the random population of solutions; Calculate the initial cost function value, ( ) Apply the neighborhood structure on each solution in the population Sort the solutions in ascending order based on the penalty cost; Sol * ← select the solution that has the best value; else A i ++ end if Determine the probability for each solution, based on the following: OnlookerBee Sol * ← select the solution, which has the higher probability; Evaluate random number Apply the neighborhood structure on each solution in the population Sort the solutions in ascending order based on the penalty cost; Sol * ← select the solution that has the best value; Set random solution instead of the current solution; The main parameters of the Artificial bee colony algorithm are given in Table II. The parameters of the HS and ABC algorithm have been chosen unfairly, the function evaluation FE for HS is less than FE for ABC. However, harmony search gives the best results, as it will be explained later.

Structural optimization of 7 DOF type
In this paper, a massive hydraulic tunneling manipulator SANDVIK DD421 (Fig. 1) was taken into consideration to inspire a new design of robot manipulator that overcomes the restrictions on this model of SANDVIK. The proposed topology of the robot should enable the arm to work in a confined environment which is a real challenge for designers. The most prominent examples of a confined space are found in tunneling and mining works, especially when the requirements are digging a narrow or low overhead tunnel. Also, the problem gets more complicated in case of manipulation of the rubble as well as drilling the hole or pit. Fig. 2 and Table III illustrate the type synthesis and Denavit-Hartenberg parameter respectively of the new robot, which is designed to meet the requirements mentioned above; all of the variables are dimensionless. It is assumed that 2 a , 3 a , 4 a , 5 a are equal, and for this problem, ten variables represent a single task point. Fig. 3 presents a given trajectory that should be followed by the robot end-effector with its optimum dimensions. For each point in the trajectory, there should be optimal structure, and finally, the enumeration should be applied to all the proposed structures to choose the best option for reaching all of the task points in the working space.
The homogeneous transformation matrices will be estimated from the Denavit matrix and Denavit parameters in Table III as follows: where the subscript t denotes task point coordinates and the penalty is a high number like 100 or more; it is added to the objective function if and only if any point belonging to the robot arm touch the boundaries of the workspace, which are denoted as constraints in Fig. 3. A random 100 task points are continuously distributed in the workspace, and during the enumeration process, the best structure -one that can achieve all points -should be chosen. For any point among the random 100, there should be a robot configuration that can be represented by ten kinematic variables. The ten variables are seven for joint angles ( 1 θ , 2 θ , 3 θ , 4 θ , 5 θ , 6 θ , 7 θ ) plus three for the length of the link 1 d , a , and 7 d . These ten variables are substituted in equations (4)- (11) to get x, y, and z coordinates to be replaced in the objective function (12    The expected result is that the optimization algorithm creates a configuration that reaches the task point. In other words, the optimized joint angles and lengths of links should converge the objective function to zero. As it is shown in Table IV, harmony search performed the job well and converged to zero, while Artificial bee colony could not go to zero and stuck at an objective value equal to 4.7094. Though the artificial bee colony is sufficient for most optimization problems, it starts showing a reduction in its efficiency when there are many decision variables, as it is confirmed by this problem (ten variables).
The novel 7 DOF robot manipulator with its optimized dimensions is simulated in a 3D environment as it is shown in Fig. 5; this was created using the 3Ds Max software package. The constrained workspace matches the space in Fig. 3, and the simulated model worked well in its virtual environment.

Conclusion
By following the rules of the structural optimization and dimensional synthesis, the cost of the components can be reduced to the minimum value, which is important for economic and environmental reasons. Kinematic equations of robot manipulators are one of the quantities that can be optimized to reach the optimum structure of the robot arm. While there are many optimization algorithms, a comparative study has been done between harmony search and artificial bee colony because they are classified as effective optimization methods. The comparison process was made by optimizing the structural components of a novel 7 DOF robot manipulator. According to many tests, harmony search regularly drives the objective value to zero, which is the optimum value, but artificial bee colony algorithm failed to converge to the global of the function even with more function evaluation than HS. Another aspect of this study is that it presents a novel design for a multitask robot arm for tunneling and mining works.