In this paper,a trajtory **tracking** **method** **based** on the combination of **fuzzy** control and **neural** **network** control is proposed.We use the **gauss** **function** as **fuzzy** membership **function** and realize the **fuzzy** reasoning by **neural** net- work.Then a robotic controller using the **method** is designed for **welding** **manipulator** **trajectory** tracking.Simulation results illustrate the performace of the proposed controller.

Vehicles as means of transportation are considered a necessity in today’s world. However, the use of vehicles has resulted in detrimental outcomes, as well. Road accidents involving automotive have taken the lives of 1.3 million individuals around the world in 2015 and about 1.4 million in 2016. According to the World Health Organization, fatal driving accidents rank as the 10th cause of death in the world [1]. Driver error is the cause of 72% of car accidents [2]. With the rapid advancement of artificial intelligence and car technology, autonomous vehicles are expected to take the pressure and stress of driving off the drivers and thus improve their workload and safety. Here, we have focused on the control system of an autonomous vehicle, and tried to present a controller for a basic model of such a car. Many of the previous works have introduced **tracking** controllers for driver-less automotive. Researchers initially designed motion controllers for these cars **based** on their kinematic model [3] and then started designing **tracking** controllers using the dynamic model. A simple **fuzzy** proportionalintegralderivative (PID) controller for the kinematic model of a car-like robot has been presented in [4]. Some researchers have designed kinematic model-**based** controllers while considering the skidding and slipping of car tires [5]. Using the virtual vehicle **method**, several articles have presented controllers for car-like robots **based** on their dynamic models [6]. The **trajectory** **tracking** of autonomous vehicles using the predictive control **method** has been discussed in [7]. In this study, the front wheel angle has been used in the control scheme. A path-following scheme for an autonomous four-wheeler using the combination of sliding mode and feedback control, along with the control of yaw moment, has been presented in [8] for the direct control of the car wheel. Numerous research works have been conducted on the lateral motion control of autonomous vehicles at the Stanford

Show more
14 Read more

Over the past few decades, the sliding mode control (SMC) technique for nonlinear mechanical systems has been studied extensively by many researchers [1-15]. The main idea of using the SMC approach is to cope with the parametric uncertainties and external disturbances under matching conditions for the complex nonlinear systems exist in practical applications such as robotics **manipulator**, **welding** mobile robot,...[1,2]. Recently, many robust control algorithms using SMC [8-15] or combined with **fuzzy** logic [7], **neural** **network** [4,6], adaptive [3-4] have been proposed to deal with the **trajectory** **tracking** problem including system dynamics

Show more
In this paper, a synergistic combination of Radial Basis **Function** **Neural** **Network** (RBFNN) with **fuzzy** sliding mode control methodology is proposed. The slope of the sliding surface is used to adjust with **Fuzzy** logic. The weights of the RBFNN are adjusted according to an adaptive algorithm for the purpose of controlling the system states to hit the sliding surface and then slide along it. The proposed **method** and PID control are implemented on an industrial robot (Manu- tec-r15) and the results obtained from the applications are presented.

Then for the **trajectory** **tracking** error of a robot ma- nipulator is uniformly ultimately bounded along the tra- jectory i.e. the **tracking** errors and its derivatives converge to zero with application of additional control law (15) in presence of parametric uncertainities.

order to minimize the cost **function** under certain con- anipulator, the external force exerted by the fluid drag needs to be added to the robot **manipulator** dynamics. The fluid drag on an object is proportional to the square of the object’s speed [16]. The fluid drag always has a positive value in the calculation of the robot **manipulator** dynamics. With respect to the motion direction of the **manipulator**, the fluid drag acts in the opposite direction in a real environment. Therefore, the sign of the fluid drag direction has to be determined according to the mo- tion of the **manipulator**. The motion direction of the ma- nipulator can be identified by the sign of the translational velocities.

Show more
14 Read more

Assume that the **network** input layer and output layer are respectively composed of same M neurons, and neuron number K of the middle layer is smaller than M Provide the same learning mode in the input layer and output layer(that is, the teacher mode is the input mode). After **network** studying, its underlying layer shall be able to give different encoding expression for each input modes among M input modes. Its basic idea is to make the original data pass the waist type **network** bottleneck, and expect to gain relatively compact data expression at the **network** bottleneck, in order to achieve the purpose of compression. In the process of **network** learning, adjust the **network** weights through training and make the reconstruction image similar with the training image possibly in mean error sense. The trained **network** can be used to perform the data compression task, and the weighted value between **network** input layer and the middle layer is equivalent to encoder. The original image data transmitted from the input end is processed by the **fuzzy** **neural** **network** to gain the output data in the middle layer, and the output data is the compression code of the original image, and the vector of the output layer is the reconstructed image data after the compression [11].

Show more
In this section we derive the equations of motion for an individual link **based** on the direct **method** has been derived, i.e. Newton-Euler Formulation. The motion of a rigid body can be decomposed into the translational motion with respect to an arbitrary point fixed to the rigid body, and the rotational motion of the rigid body about that point. The dynamic equations of a rigid body can also be represented by two equations: one describes the translational motion of the centroid (or center of mass), while the other describes the rotational motion about the centroid. The former is Newton's equation of motion for a mass particle, and the latter is called Euler's equation of motion.

Show more
The objective type **fuzzy** demonstrating has incredible learning capacities and requires less computational exertion. A normal **fuzzy** model has fundamentally 4 stages: fuzzification with input constraints to the **fuzzy** territory, (ii) **fuzzy** rules generation help of membership **function** (iii) **fuzzy** inference system which processes **fuzzy** variable inputs to acquire **fuzzy** outputs, and (iv) defuzzification strategy to change over the **fuzzy** output back to the general input imperatives. In **fuzzy** rationale, estimations of various criteria are mapped into linguistic values that describe the level of fulfillment with the numerical estimation of the targets [28]. As indicated by three sorts of trust value: friend, acquaintance, and stranger, characterize three **fuzzy** sets: high, medium and low, individually.

Show more
Abstract: **Neural** networks with their inherent learning ability have been widely ap- plied to solve the robot **manipulator** inverse kinematics problems. However, there are still two open problems: (1) without knowing inverse kinematic expressions, these solutions have the di ﬃ culty of how to collect training sets, and (2) the gradient-**based** learning algorithms can cause a very slow training process, especially for a complex configuration, or a large set of training data. Unlike these traditional implementa- tions, the proposed metho trains **neural** **network** in joint subspace which can be easily calculated with electromagnetism-like **method**. The kinematics equation and its in- verse are one-to-one mapping within the subspace. Thus the constrained training sets can be easily collected by forward kinematics relations. For issue 2, this paper uses a novel learning algorithm called extreme learning machine (ELM) which randomly choose the input weights and analytically determines the output weights of the single hidden layer feedforward **neural** networks (SLFNs). In theory, this algorithm tends to provide the best generalization performance at extremely fast learning speed. The results show that the proposed approach has not only greatly reduced the computation time but also improved the precision.

Show more
14 Read more

Abstrat : The **manipulator** provides a gain in terms of production rate compared to conventional serial manipulators. The **trajectory** planning made it possible to optimize the performance of the parallel 3RRR **manipulator** in its workspace. This planning was done on the octave software, **based** on the modeling of the **manipulator** using the inverse geometry **method**. Which allowed to determine the workspace of the robot. The working space of a parallel robot therefore depends on the different independent kinematic chains that connect the base to the effector (mobile platform).

Show more
10 Read more

K-means clustering **method**, as a common and effec- tive clustering **method**, can confirm the extension constant of hidden nodes on the basis of the distance among various clustering centers. The main idea of this **method** is that the K value and K initial class clus- ter centers should be set firstly, then each center point should be allocated in the class cluster where there is the class cluster center which is closest to the center. After this, the mean of class clusters should be recal- culated according to the data in each class cluster. It is a process of allocating center points and updating the class cluster centers by iteration **method**. The al- gorithm will not stop until the change of class cluster center is very small (convergence) or the specified it- eration number are achieved. Rational determination of K value and K initial class cluster centers greatly influences of the clustering effect. Therefore, K val- ue and K initial class cluster centers shall be selected through calculation.

Show more
11 Read more

physical systems, and their valid range of operation is small. A better choice to tackle **tracking** control prob- lems, while satisfying the input and state constraints, is nonlinear optimal control. Nonlinear optimal control satises any of the desirable constraints, and is also suitable for nonlinear systems [7]. In an optimal control problem, the goal is determination of the states and controls that minimize a cost functional subject to nonlinear dynamic constraints, boundary conditions and inequality path constraints. There are two meth- ods for resolving optimal control problems: direct and indirect [8]. In an indirect **method**, rst-order necessary conditions for optimality are derived from the optimal control problem via the calculus of variations and Pontryagin's minimum principle [9]. These neces- sary conditions form a Hamiltonian Boundary-Value Problem (HBVP), which is then solved numerically for extremal trajectories. The optimal solution is then found by choosing the extremal **trajectory** with the lowest cost. The primary advantages of indirect methods are high accuracy in the solution and the assurance that the solution satises the rst-order optimality conditions. However, indirect methods have several disadvantages. First, the solution of HBVP must be usually derived analytically, which can be often non-trivial. Second, if one wishes to obtain the solution numerically, as numerical techniques used in indirect methods typically have small radii of convergence, an extremely good initial guess of the unknown solution or boundary conditions is generally required. Finally, for problems with path constraints, it is necessary to have a priori knowledge about the constrained and unconstrained arcs or switching structure [10]. Cimen and Banks (2004) used the indirect **method** to solve nonlinear optimal **tracking** control of an oil tanker [7].

Show more
Abstract—In this paper, a reliability evaluation model **based** on **fuzzy** **neural** **network** is proposed to evaluate the reliability of wireless sensor networks with- out a unified standard. Firstly, the reliability is analyzed from the point of view of topology structure, protocol stack structure and reliability mechanism of wire- less sensor **network**, and the performance indexes that affect the reliability are extracted. Secondly, some performance indexes are screened out, and the stand- ard value matrix of reliability evaluation for index **fuzzy** quantization is estab- lished. The sample data is generated by interpolation, and the reliability evalua- tion model **based** on **fuzzy** **neural** **network** is established. The **neural** **network** model takes the selected index values as input, and outputs are the reliability of the wireless sensor **network**. The simulation results show that the evaluation model is basically consistent with the actual situation, and it can evaluate the wireless sensor **network** from the system level.

Show more
13 Read more

the filtering phase. The ones club capabilities are used for **fuzzy** set impulse noise depiction. The proposed novel **fuzzy** technique is specifically evolved for suppressing impulse noise from coloration Image s whilst stopping other image records and texture. Ibrahim [3] gave a simple adaptive median filter out for the removal of impulse noise from highly corrupted snap shots. This creator proposed a easy, but green technique to suppress impulse noise from noise affected Image s. This new technique composed of stages.

Abstract: Robot manipulators are subject to different types of uncertainties which may degrade the **tracking** control performance or even make the system unstable. In this paper, a **neural** **network** **tracking** controller with disturbance observer is developed to deal with both the external disturbance and the dynamic parametric uncertainties. First, RBF **neural** **network** is introduced to learn and approximate the uncertain robot dynamic by using adaptive algorithm. Next, a nonlinear disturbance observer is designed to estimate and compensate for external disturbances and remove the effect of the disturbance. Simulation results show that the proposed control scheme has good **tracking** performance, which can effectively suppress the uncertain dynamics and external disturbances of the **manipulator** systems.

Show more
intelligence models are very useful in robot controller design. In better words, the controllers need immediate feedback for appropriate functioning, but normal operation of the robot is not fast enough to satisfy these limits. The computer models like the two **neural** **network** models explained in this paper can be used to eliminate such limitations. Accordingly, analogous approaches with help of **neural** **network** methods can be useful to help many other robotic and intelligent systems to perform more effective.

Abstract. The previous motion simulator is almost all built on the basis of parallel mechanism, and the most used and most mature is **based** on the six degree of freedom steward platform, and the simulator of the series mechanism has not been widely studied. Because the body organs can not accurately distinguish the degree of motion, the theory of **fuzzy** control is introduced to improve the classical wash out algorithm, and the motion error of the sensory organ of the vestibule of the human vestibule and the space of the motion platform are combined, and the fidelity of the motion simulation can be raised as much as possible under the limit of the non super motion platform.

Show more
This chapter presents a brief introduction to the notions that are used to achieve the aim of this thesis. The aim is to use granular computing for problems solving. The introduced notions are granular computing, granular **neural** networks, **fuzzy** artificial **neural** networks, **fuzzy** information granulation, generic algorithms and **fuzzy** sets theory. The reason for this study is to investigate the aim of artificial computations that is to solve a problem with the least amount of cost and the best accuracy. Problems become more difficult to be solved when their corresponding data sets are large or contain uncertain information. In the literature of artificial computations, there are many nature inspired computations such as evolutionary computations, artificial **neural** networks, artificial immune systems, swarm intelligence, etc. (Zomaya, 2006) (Kari, et al., 2008). However, there are two issues behind the proposed algorithms; which are the ability of each algorithm to solve only a particular type of problem; and their performance in solving the problem. To overcome these two issues, this thesis uses granular computing with the aid of learning and optimization mechanisms for an optimal learning from granular rules. Therefore, granular **neural** networks have been used for learning mechanism; meanwhile **fuzzy** artificial **neural** networks are centered in the granular **neural** networks. Also, the genetic algorithms (GA) are used to increase the performance of **fuzzy** artificial **neural** networks. Therefore, GA-**based** **fuzzy** artificial **neural** networks are used in the main part for granular **neural** networks.

Show more
50 Read more

support them in a proper location which is capable of preventing distortions in workpieces during **welding**. For this the locating elements need to be placed carefully, clamping has to be light but firm, placement of clamping elements has to be clear of the **welding** area and the fixture has to be quite stable and rigid to withstand the **welding** stresses. With the aid of **manipulator** the **welding** fixture on which the rooftop will be placed is rotated for **welding** purpose. After necessary **welding** operations being performed, the fixture is rotated back to its original position. Then the rooftop is unclamped and unscrewed from its fixture in order to get lifted by the crane to be placed on the train top. For carrying out these operations appropriate design and functioning of this mechanism is of prime concern.As a result of complex alignment and positioning equipment are important as they are required in nearly all research andmanufacturing processes

Show more