Gancho Vachkov

Intelligent Data Analysis for Performance Evaluation and Fault Diagnosis in Complex Systems

The paper proposes an efficient computational strategy for remote performance analysis and diagnosis of construction machines and other complex systems. A special information compression (IC) method is used to send the information obtained from various sensors to the maintenance center in a compact and economical way. The IC method uses the neural-gas unsupervised learning algorithm to locate a predefined number of neurons in the densest data areas of the parameter space. These neurons serve as a kind of information granules of the current machine operation that are later sent in a wireless way to the maintenance center for further information recovery (IR) and performance analysis. Here a special weighted moving window average (MWA) method is used, as well as an original fuzzy inference-based analysis for comparison of different operations and discovery of possible deteriorations. A knowledge-based fault diagnosis method is also proposed and analyzed in the paper. The whole IC/IR computational strategy is illustrated on real experimental data from a hydraulic excavator which demonstrate its merits and applicability.

Classification of images based on information compression and fuzzy rule based similarity analysis

This paper proposes a computational scheme for fuzzy similarity analysis and classification of images by comparison of the new (unknown) images with a predetermined number of known (core) images, contained in an image base. As a first step, an unsupervised competitive learning algorithm is used to create the so called compressed information model (CIM) which replaces the original ldquoraw datardquo (the RGB pixels) of the image with much smaller number of neurons. Then two specially introduced parameters of the CIM are computed, namely the center-of-gravity of the model and the generalized model size. These parameters are used as inputs of a special fuzzy inference procedure that computes numerically the similarity between a given pair if images as a difference degree between them. Finally, a sorting procedure with a predefined threshold is used to obtain the results from the classification. The flexibility and applicability of the whole proposed unsupervised classification scheme is illustrated on the example of classification of 18 different images by use of three different image bases containing, 3, 5 and 7 ldquocorerdquo images respectively.

Fuzzy emotion interpolation system for emotional autonomous agents

The paper presents a model that simulates emotional internal states for autonomous agents. The model connects simulated touch-sensor inputs to an emotion-labeling output and includes the perception, motivational and emotional systems. As part of the model, a Fuzzy Emotion Interpolation System (FEIS) is proposed and described in detail. The experimental results obtained using FEIS for emotional status simulation are presented.

Classification Of Machine Operations Based On Growing Neural Models And Fuzzy Decision

In this paper, a novel approach to analysis and classification of complex machine operations is presented. The available data sets from different machine operations are first compressed and saved in the form of neural models that are called compressed information models (CIM). Here an original algorithm for unsupervised learning is proposed. It creates the so called growing neural models in a sense that the number of neurons is gradually increasing (growing) during the learning process, until predetermined model accuracy (the “average minimum distance”) is satisfied. The proposed algorithm has much faster convergence compared with the classical neural-gas learning that uses preliminary fixed number of neurons. A special Knowledge Base classification scheme is also proposed in the paper. It uses a fuzzy decision block for computing the difference degree between each CIM in the Knowledge Base with the CIM of the current machine operation. The fuzzy inference procedure uses two parameters for comparison the CIMs, namely the decision the Center-of-Gravity and the General Size of the CIM. An example for classification of 45 specially generated operations from a diesel engine of a hydraulic excavator is used to demonstrate the whole proposed technology and its applicability. This fuzzy classification scheme is also able to discover new operations that significantly differ from all previously known operations.

On-Line Unsupervised Learning for Information Compression and Similarity Analysis of Large Data Sets

The growing huge amount of information from the operations of complex processes and systems requires suitable methods for information compression. Therefore in this paper three unsupervised learning algorithms for information compression are proposed and analysed, namely the fixed-model learning (FML), the growing-model learning (GML) and the on-line model learning (OML) algorithms. They convert the original large data set into a much smaller set of neurons in the same dimensional space. It is shown that the OML algorithm is the fastest one and the most suitable for large data compression. A procedure for similarity analysis of the compressed models is also presented and illustrated in the paper. It uses the preselected Key Points from the compressed model for comparison.

Vision based autonomous path tracking of a mobile robot using fuzzy logic

In this paper we present an algorithm for autonomous path tracking of a mobile robot to track straight and curved paths traced in the environment. The algorithm uses a fuzzy logic based approach for path tracking so that human driving behavior can be emulated in the mobile robot. The method combines a fuzzy steering controller, which controls the steering angle of the mobile robot for path tracking, and fuzzy velocity controller which controls the forward linear velocity of the mobile robot for safe path tracking. The inputs to the fuzzy system are given by the vision system of the mobile robot. A camera is used to capture images of the path ahead of the mobile robot and the vision system determines the lateral offset, heading error and the curvature of the path. We perform experiments using a mobile robot platform. In the first experiment the mobile robot is able to successfully track a straight path. This shows the effectiveness of the fuzzy steering controller. We also perform experiments on paths containing curved sections. The fuzzy velocity controller was able to command appropriate speed for safe tracking of the path ahead of the robot. The effectiveness of the fuzzy velocity controller is shown in this experiment.

Temporal and spatial Evolving Knowledge Base system with sequential clustering

This paper proposes a computational scheme of a novel Evolving Knowledge Base system that is able to gradually grow and update spatially and temporally. The main assumption is that the input information comes from the real environment in the form of chunks of data (not single data points). Therefore the whole system works in a quasi-real time. Each chunk of data is used for extraction of the so called knowledge items, which is done by a specially introduced sequential clustering algorithm. It is able to discover the separate knowledge items sequentially, in decreasing order of their size. Another important block of the proposed evolving knowledge base system is the updating algorithm, It is in charge of managing the Knowledge Base over time and performs (when necessary) one of the three types recursive computations, namely: learning, relearning and forgetting. The flexibility and the degree of generality of the proposed evolving system is illustrated on a specially constructed example that resembles a real case of data flow coming as a sequence of 20 chunks of data. These data exhibit evolving behavior during the sampling periods and the knowledge Base system is able to catch such behavior by properly updating its parameters. These results show the way of different possible practical applications.

Human-Assisted Fuzzy Image Similarity Analysis Based on Information Compression

The fuzzy similarity analysis we propose in this paper is used for unsupervised image classification. We introduce a special growing unsupervised learning algorithm for information compression (granulation) of the original “raw data” (the RGB pixels) of an image with a smaller number of neurons (information granules). Two important parameters are extracted from each image, namely the center of gravity (COG) and the model volume of the image, taken as the number of neurons obtained from information compression. These two features are used as inputs for special fuzzy inference for numerically calculating the degree of similarity between a pair of images. The fuzzy inference procedure can be tuned based on a predefined human preference (list of similar images), thus performing human-assisted similarity analysis. The choice of the optimization algorithm and the selection of the optimization criterion are questions open to the user to answer. The proposed computation scheme for similarity analysis is illustrated on a test example of 16 flower images and results are discussed.

Growing radial basis function network models

In this paper a learning algorithm for creating a Growing Radial Basis Function Network (RBFN) Model is presented and analyzed. The main concept of this algorithm is that the number of the Radial Basis Function (RBF) units is gradually increased at each learning step of the algorithm and the model is gradually improved, until a predetermined (desired) approximation error is achieved. The important point here is that at each step of increasing the number of the RBF units, an optimization algorithm is run to optimize the parameters of only this unit, while keeping the parameters of all the previously optimized RBF units. Such strategy, even if being suboptimal, leads to significant reduction in the number of the parameters that have to be optimized at each step. A modified constraint version of the particle swarm optimization (PSO) algorithm with inertia weight is develop and used in this paper. It allows for obtaining optimal solutions with clear practical meaning. A synthetic nonlinear test example is used in the paper to analyze the performance of the proposed learning algorithm for creating the Growing RBFN model. A comparison with the standard algorithm for simultaneous optimization of all parameters of the classical RBFN model with fixed number of units is also done. It shows that the learning of the Growing RBFN model leads to a more stable and in many cases more accurate solution.

Classification-based behavior model for detection of abnormal states in systems

Abstracf - In !his paper, a special model in the form of lookup Table is proposed for detecting different abnormal states in mechanical and other industrial systems. The model is generated by repetitive use of a classification procedure that uses a SelfOrganized Map and Ndural Gas algorithm for competitive unsupervised learning. This classification procedure makes a hind of mapping from the measured “Parameter-Space” to the preliminary defined “Operating Mode-Space”. The computed Modes Recognition Vectors show the relative weight of each operating mode in the entire set of operating modes. They represent the first part of the system behavior model, while the sewad part consists of the vectors of the Deviated (abnormal) Parameters. These are preliminary generated in a systematical, combinatorial way. Thus produced Classification-based Behavior Model (CBB Model) is further used for solving the general problem of the fault diagnosis. Here the most plausible deviated parameters vector has to be found which produces as close as possible the computed Modes Recognition Vector. Simulations with real dbta from a hydraulic excavator are used in order to prove the applicability and the merits of the proposed method for abnormality detection and fault diagnosis.