Weria Khaksar

An agile FCM for real-time modeling of dynamic and real-life systems

Fuzzy cognitive map (FCM) is a well-established model of control and decision making based on neural network and fuzzy logic methodologies. It also serves as a powerful systematic way for analyzing real-life problems where tens of known, partially known, and even unknown factors contribute to complexity of a system. FCM-based inference requires a neural activation function much like other neural network systems. In modeling, in addition to an activation function, FCM involves with weight training to learn about relationships as they exist among contributing factors. Therefore, numerous contributing factors could be analyzed to understand the behaviors of factors within a real-life system and to represent it in form of tangible matrices of weights. This article discusses a new incremental FCM activation function, named cumulative activation, and introduces a new weight training technique using simulated annealing (SA) known as agile FCM. Smooth variation of FCM nodes that is due to cumulative nature of inference results into faster convergence, while a unique minimum cost solution is guaranteed using the SA training module that is entirely expert-independent. A combination of these two techniques suits time-related applications where inclusion of temporal features is necessary. The resulted system is examined through numerical example datasets where the candidate FCM shows sensitivity to dynamic variables over time. A real-life example case is included as well to further support the effectiveness of the developed FCM in modeling of natural and complex systems.

AN ALTERNATIVE APPROACH TO FCM ACTIVATION FOR MODELING DYNAMIC SYSTEMS

Recurrent neural models such as fuzzy cognitive maps (FCM) are well established in decision modeling through progressive variations of systems’ concepts. However, existing activation functions have shortcomings, such as a lack of sensitivity to weights of initial concepts, which is due to exaggerated focus on the training of networks’ causal links. Therefore, in most cases, decision outputs converge toward lower and higher extremes and do not represent gray scales. Another disadvantage is that current models require sufficient time delay for convergence toward results. This makes FCM unable to handle transient changes in input. A new technique has been examined in this article using a real-life example to improve FCM activation in terms of fast response to dynamic stimuli. A simple expert model of hexapod locomotion is developed without focus on weight training. The systems response to stimuli is evaluated through a complete six-phase stride to validate the effectiveness of the developed activation function.

A Low Dispersion Probabilistic Roadmaps (LD-PRM) Algorithm for Fast and Efficient Sampling-Based Motion Planning

In this paper, we propose a new learning strategy for a probabilistic roadmap (PRM) algorithm. The proposed strategy is based on reducing the dispersion of the generated set of samples. We defined a forbidden range around each selected sample and ignored this region in further sampling. The resultant planner, called low dispersion-PRM, is an effective multi-query sampling-based planner that is able to solve motion planning queries with smaller graphs. Simulation results indicated that the proposed planner improved the performance of the original PRM and other low-dispersion variants of PRM. Furthermore, the proposed planner is able to solve difficult motion planning instances, including narrow passages and bug traps, which represent particularly difficult tasks for classic sampling-based algorithms. For measuring the uniformity of the generated samples, a new algorithm was created to measure the dispersion of a set of samples based on a predetermined resolution.

Robotic motion planning in unknown dynamic environments: Existing approaches and challenges

Path planning with obstacles avoidance in dynamic environments is a crucial issue in robotics. Numerous approaches have been suggested for the navigation of mobile robots with moving obstacles. In this paper, about 50 articles have been reviewed and briefly described to offer an outline of the research progress in motion planning of mobile robot approaches in dynamic environments for the last five years. The benefits and drawbacks of each article are also explained. These papers are classified based on their issues into ten groups which are: stability, efficiency, smooth path, run time, path length, accuracy, safety, future prediction (uncertainties), control, and less computation cost. Finally, some scope and challenging topics are presented based on the papers mentioned.

Robotic Welding Technology

Since the first industrial robots were introduced in the early 1960s, the development of robotized welding has been truly remarkable and is today one of the major application areas for industrial robots. Robot welding is mainly concerned with the use of mechanized programmable tools, known as robots, which completely automate a welding process by both performing the weld and handling the part. Robots are quite versatile and hence have been used for a variety of welding types such as resistance welding and arc welding. This chapter describes the development and progress of robotization in welding over the years and also discusses many advantages and disadvantages of different robotic welding technologies.

FCM Relationship Modeling for Engineering Systems

Semantic graphs like fuzzy cognitive map (FCM) are known as powerful methodologies commonly used in control applications, as well as in relationship modeling. Besides, FCM is used as a systematic way for analyzing real-world problems with numerous known, partially known and unknown factors. This chapter discusses FCM application in relationship modeling context using some agile inference mechanisms. A sigmoid-based activation function is discussed with application in modeling hexapod locomotion gait. The activation algorithm is then added with a Hebbian weight training technique to enable automatic construction of FCMs. A numerical example case is included to show the performance of the developed model. The model is examined with perceptron learning rule as well. Finally a real-life example case is tested to evaluate the final model in terms of relationship modeling.

Application of Sampling-Based Motion Planning Algorithms in Autonomous Vehicle Navigation

With the development of the autonomous driving technology, the autonomous vehicle has become one of the key issues for supporting our daily life and economical activities. One of the challenging research areas in autonomous vehicle is the development of an intelligent motion planner, which is able to guide the vehicle in dynamic changing environments. In this chapter, a novel sampling-based navigation architecture is introduced, which employs the optimal properties of RRT* planner and the low running time property of low-dispersion sampling-based algorithms. Furthermore, a novel segmentation method is proposed, which divides the sampling domain into valid and tabu segments. The resulted navigation architecture is able to guide the autonomous vehicle in complex situations such as takeover or crowded environments. The perform‐ ance of the proposed method is tested through simulation in different scenarios and also by comparing the performances of RRT and RRT* algorithms. The proposed method provides near-optimal solutions with smaller trees and in lower running time.

A review on mobile robots motion path planning in unknown environments

Robotics sector have achieved enormous founds in recent years due to its high demands in factories to carry out high-precision jobs like riveting and welding. They are also often applied in special situations that would be hazardous for humans such as disposing toxic wastes or defusing bombs. Mobile robots alone however have gained much focus from researches relating optimization of their motion path planning. In this paper, a brief review on mobile robots motion path planning in unknown environment have been done based on recent founds. The paper categorizes motion path planning into two groups which is the Optimized Classic Approaches and Evolutionary and Hybrid Approaches. The optimized classic approaches represents the recent optimized motion path planning that implies the classic approaches such as A* search algorithm, Rapidly-exploring Random Trees (RRT), D* and D* Lite algorithm. The evolutionary and hybrid approaches are those adapts Artificial Intelligence (AI) such as neural networks (NN), genetic algorithms (GA), fuzzy systems and reinforced learning either acting alone or as hybrids together with other algorithms. Finally a comparison between these two categories are done differentiating their advantages and disadvantages.

A fuzzy-tabu real time controller for sampling-based motion planning in unknown environment

Sampling-based path planning methods for autonomous agents are one of the well-known classes of robotic navigation approaches with significant advantages including ease of implementation and efficiency in problems with high degrees of freedom. However, there are some serious drawbacks like inability to plan in unknown environments, failure in complex workspaces, instability of results in different runs, and generating non-optimal solutions; which make sampling-based planners less efficient in practice. In this paper, a fuzzy controller is proposed which utilizes the heuristic rules of Tabu search to improve the quality of generated samples. The main contribution of this work is the ability of the proposed sampling-based planner to work effectively in unknown environments and to plan efficiently in complex workspaces by letting the fuzzy-Tabu controller check the quality of the generated samples before any further processing. The efficiency of the proposed planner is tested in several workspaces and the comparison studies show significant improvement in runtime and failure rate. Furthermore, the decision variables of the proposed controller are discussed in detail to determine their effect on the performance of the algorithm.

Artificial Neural Network (ANN) Approach for Predicting Friction Coefficient of Roller Burnishing AL6061

Artificial Neural Network (ANN) approach is a fascinating mathematical tool, which can be used to simulate a wide variety of complex scientific and engineering problems. Due to its highly reliable prediction quality, the usage of it is growing rigorously and had already become an ultimate tool for various applications in the field of engineering. In this study an ANN technique was used to predict friction coefficient of roller burnishing AL6061 for two orientations which is parallel burnishing orientation (PB) and cross burnishing orientation (CB). The input parameters were defined by widths of roller curvature (7.5mm, 8mm and 8.5mm), burnishing speeds (110rpm, 230rpm, 330rpm and 490rpm), and burnishing forces (155.06N, 197.45N, 239.83N and 282.22N) while the output parameter was friction coefficient. 173 data was used for training the ANN and another 115 data was used to test the ANN. 60 different configurations of ANN was trained by using 6 different training algorithms. It was found that feed-forward back-propagation network with 15 neurons in hidden layer that was trained by Levenberg-Marquardt training algorithm gave the best result when compared to other training algorithms used. From the results it was found that the training performance and prediction performance was 0.000809 and 0.710 respectively. From this study, it became obvious that the selected ANN with the configuration and training algorithm proved to be the most suitable among the other ANN investigated for similar applications.