Novel Applications of Intelligent Systems

The paper presents the approaches, methods and tools for assessment of main quality features of grain samples which are based on color image and spectra analyses. Visible features like grain color, shape, and dimensions are extracted from the object images. Information about object color and surface texture is obtained from the object spectral characteristics. The categorization of the grain sample elements in three quality groups is accomplished using two data fusion approaches. The first approach is based on the fusion of the results about object color and shape characteristics obtained using image analysis only. The second approach fuses the shape data obtained by image analysis and the color and surface texture data obtained by spectra analysis. The results obtained by the two data fusion approaches are compared.

The mill fans (MF) are centrifugal fans of the simplest type with flat radial blades adapted for simultaneous operation both like fans and also like mills. The key variable that could be used for diagnostic purposes is vibration amplitude of MF corpse. However its mode values include a great deal of randomness. Therefore the application of deterministic dependencies with correcting coefficients is non-effective for MF predictive modeling. Standard statistical and probabilistic (Bayesian) approaches are also inapplicable to estimate MF vibration state due to non-stationarity, non-ergodicity and the significant noise level of the monitored vibrations. Adequate for the case methods of computational intelligence [fuzzy logic, neural networks and more general AI techniques—the precedents’ method or machine learning (ML)] must be used. The present paper describes promising initial results on applying the Case-Based Reasoning (CBR) approach for intelligent diagnostic of the mill fan working capacity using its vibration state.

We focused on the autonomous control of a three-dimensional snake-like robot that moves on rubble. To realize an autonomous controller, we employed reinforcement learning. However, applying reinforcement learning in a conventional framework to a robot with many degrees of freedom and moving in a complex environment is difficult. There are three problems: state explosion, lack of reproducibility, and lack of generality. To solve these problems, we previously proposed abstracting the state-action space by utilizing the universal properties of the body and environment. The effectiveness of the proposed framework was demonstrated experimentally. Unfortunately, analysis of the obtained policy was lacking. In the present study, we analyzed the obtained policy (i.e., Q-values of Q-learning) to determine the mechanism for abstraction of the state-action space and confirmed that the three problems were solved.

Friction and disturbances have an important influence on the robot manipulator dynamics. They are highly nonlinear terms that cannot be easily modeled. In this investigation an incrementally tuned parametric type-2 fuzzy cerebellar model articulation controller (P-T2FCMAC) neural network is proposed for compensation of friction and disturbance effects during the trajectory tracking control of rigid robot manipulators. CMAC networks have been widely applied to problems involving modeling and control of complex dynamical systems because of their computational simplicity, fast learning and good generalization capability. The integration of fuzzy logic systems and CMAC networks into fuzzy CMAC structures helps to improve their function approximation accuracy in terms of the CMAC weighting coefficients. Type-2 fuzzy logic systems are an area of growing interest over the last years since they are able to model uncertainties and to perform under noisy conditions in a better way than type-1 fuzzy systems. The proposed intelligent compensator makes use of a newly developed stable variable structure systems theory-based learning algorithm that can tune on-line the parameters of the membership functions and the weights in the fourth and fifth layer of the P-T2FCMAC network. Simulation results from the trajectory tracking control of two degrees of freedom RR planar robot manipulator using feedback linearization techniques and the proposed adaptive P-T2FCMAC neural compensator have shown that the joint positions are well controlled under wide variation of operation conditions and existing uncertainties.

In this study, OWA (Ordered Weighted Averaging) distance based C × K-nearest neighbor algorithm (C × K-NN) is considered. In this approach, from each class, where the number of classes is C, K-nearest neighbors are taken. The distance between the new sample and its K-nearest set is determined based on the OWA operator. It is shown that by adjusting the weights of the OWA operator, it is possible to obtain the results of various clustering strategies like single-linkage, complete-linkage, average-linkage, etc.

A new approach to autonomously detect moving objects in a video captured by a moving camera is proposed in this chapter. The proposed method is separated in two modules. In the first part, the well-known scale invariant feature transformation (SIFT) and the RANSAC algorithm are used to estimate the camera movement. In the second part, recursive density estimation (RDE) is used to build a model of the background and detect moving objects in a scene. The results are presented for both indoor and outdoor video sequences taken from a UAV for outdoor scenario and handheld camera for indoor experiment.

Different intelligent techniques have been proposed to solve the problem of downlink resource allocation in orthogonal frequency division multiple access (OFDMA)-based networks. These include mathematical optimization, game theory and heuristic algorithms. In an attempt to improve the performance of traditional genetic algorithm (GA) and its heuristics, we propose an improved GA (IGA) that optimizes the search space and GA iterations. Using concepts from ordinal optimization (OO) to determine the stopping criteria and sub-sampling alternatives to generate the initial population, IGA shows faster convergence when applied to downlink carrier allocation in an OFDMA system. IGA workflow also includes a new “swap if better”‘ mutation operator that replaces the random mutation and a novel fitness function that seeks to maximize the total throughput while minimizing the under-allocation in an attempt to meet the quality of service (QoS) requirements for different types of users. Comparing performance of IGA with different fitness functions published in literature shows improved fairness, comparable throughput and standard deviation. Most importantly IGA is able to better meet the QoS requirements for the different types of users (real time and non real-time) and this, within few milliseconds, making it attractive for real time implementation. Future work plans a parallel implementation of IGA to further improve its computational time.

Focusing on only one type of structural component in the process of clustering XML documents may produce clusters with a certain extent of inner structural inhomogeneity, due either to uncaught differences in the overall logical structures of the available XML documents or to inappropriate choices of the targeted structural component. To overcome these limitations, two approaches to clustering XML documents by multiple heterogeneous structures are proposed. An approach looks at the simultaneous occurrences of such structures across the individual XML documents. The other approach instead combines multiple clusterings of the XML documents, separately performed with respect to the individual types of structures in isolation. A comparative evaluation over both real and synthetic XML data proved that the effectiveness of the devised approaches is at least on a par and even superior with respect to the effectiveness of state-of-the-art competitors. Additionally, the empirical evidence also reveals that the proposed approaches outperform such competitors in terms of time efficiency.

New types of constraints are considered. Reasons are described that lead to applications of intelligent, logic-based methods aiming at reduction of risk factors to ATMs. Special attention is paid to applications of Puzzle method in ATMs. To make a more independently functioning ATM, the proposed methods should be applied to data/knowledge/metaknowledge elicitation, knowledge refinement, analysis of different logical connections aiming at information checks.

This chapter describes two methods of semiautomatic position and combined telecontrol by multi-link manipulators using special setting devices (SD), which kinematic schemes are differed from the kinematic schemes of manipulators. For the survey of working space the mobile television cameras are used. Optical axes of these cameras can change its spatial orientations during performance of operations. They can rotate around two mutually perpendicular axes by operator commands. The algorithms of work of computing systems which form the setting signals for drives of all degrees of freedom of manipulators are represented and researched. The results of executed experiments and mathematical simulation of the systems work confirm effectiveness of these methods.

This work focuses on developing efficient environment representation and localization for mobile robots. A solution-approach is proposed for hybrid map-building and localization, which suits operating environments with unstructuredness and moderate dynamics. The solution-approach is vision-based and includes two phases. In the first phase, the map-building reduces the domain of extracted point features from local places through an information-theoretic analysis. The analysis simultaneously selects the most distinctive features only. The selected features are further compressed into codewords. The uncompressed features are also tagged with their metric position. In such a way, a unified map is created with hybrid data representation. In the second phase, the map is used to localize the robot. For fast topological localization, features extracted from the local place are compared to the codewords. To extend the localization into a metric pose, triangulation is executed hierarchically for the identified topological place with the use of the positional metric data of features. To ensure accurate position estimate, the dynamics of the environment are detected through the spatial layout of features and are isolated at the metric localization level. The proposed map-building and localization solution enables for a fast hybrid localization without degenerating the accuracy of localization.

This chapter is largely based on the paper “Pusher Reheating Furnace Control via Fuzzy-Neural Model Predictive Control Synthesis” presented at IEEE IS 2012 in Sofia, Bulgaria. A design of innovative fuzzy model-based predictive control for industrial furnaces has been derived and applied to the model of three-zone 25 MW RZS pusher furnace at Skopje Steelworks. The fuzzy-neural variant of Sugeno fuzzy model, as an adaptive neuro-fuzzy implementation, is employed as a predictor in a predictive controller. In order to build the predictive controller the adaptation of the fuzzy model using dynamic process information is carried out. Optimization procedure employing a simplified gradient technique is used to calculate predictions of the future control actions.

The paper presents the system-“Exactus Expert”—search and analytical engine. The system aims to provide comprehensive tools for analysis of large-scale collections of scientific documents for experts and researchers. The system challenges many tasks, among them full-text search, search for similar documents, automatic quality assessment, term and definition extraction, results extraction and comparison, detection of scientific directions and analysis of references. These features help to aggregate information about different sides of scientific activity and can be useful for evaluation of research projects and groups. The paper discusses general architecture of the system, implemented methods of scientific publication analysis and some experimental results.

Appliance-specific Load Monitoring (LM) provides a possible solution to the problem of energy conservation which is becoming increasingly challenging, due to growing energy demands within offices and residential spaces. It is essential to perform automatic appliance recognition and monitoring for optimal resource utilization. In this paper, we study the use of non-intrusive LM methods that rely on steady-state appliance signatures for classifying most commonly used office appliances, while demonstrating their limitation in terms of accurately discerning the low-power devices due to overlapping load signatures. We propose a multi-layer decision architecture that makes use of audio features derived from device sounds and fuse it with load signatures acquired from energy meter. For the recognition of device sounds, we perform feature set selection by evaluating the combination of time-domain and FFT-based audio features on the state of the art machine learning algorithms. Further, we demonstrate that our proposed feature set which is a concatenation of device audio feature and load signature significantly improves the device recognition accuracy in comparison to the use of steady-state load signatures only.