This paper is focused particularly on application of fuzzy logic approach for solving fault isolation problem of some class of industrial actuators described in the benchmark actuator definition [1]. Particular attention was paid for searching of applicable and acceptable solutions in terms of industrial implementations. The rational solution of the problem of setting fuzzy partitions for residual evaluation was proposed. The industrial benchmark study was applied for evaluating of proposed approach by means of the real process data acquired in normal and abnormal process states. The chosen examples of achieved results concerning fault isolability issues are presented.
Keywords: fault detection and isolation, industrial actuators, benchmark study, modelling, diagnosis.
Bayesian belief networks represent and process probabilistic knowledge. This representation rigorously describes the knowledge of some domains and it is a human easy-use qualitative structure that facilitates communication between a user and a system incorporating the probabilistic model. Learning Bayesian network from data may be grouped into two modelling situations: qualitative learning and quantitative learning. The first one consists in establishing the structure of the network, whereas the second concerns determining parameters of the network (conditional probabilities). Both modelling methods were applied on exemplary data to show the possibilities and benefits of this methods. The results and conclusions are presented. It was necessary to preprocess the date first. The used method, described in detail in the paper, consists in discretization into linguistic states on the basis of evaluated signal derivative. Some remarks about adjusting the network, as a part of model identification, are also presented.
Keywords: Bayesian network, learning, diagnostic models.
The aim to the paper is to optimize 2-dimensional elastic structures subjected to cyclic load. The loading can result in crack forming, so the aim of the optimization is to reduce the possibility of crack growth. The number of loading cycles necessary to crack growth is maximized. To solve the optimization task the evolutionary algorithm is used. The boundary element method is applied to solve the crack problem. In order to reduce the number of design variables the parametrical NURSB curves are used to model the geometry of parts of the structural element boundary.
Keywords: boundary element method, optimization, evolutionary algorithm, crack, parametric curves.
System identification of a parametric "black box" model for the purpose of electrical motor diagnostics is discussed in this paper. The measured acoustic pressure signal is used for identification of a model which structure is considered as a transfer function. Poles of denominator are calculated and collected on a complex plane. Fuzzy, two-stage algorithm is used for clustering and classification of poles which are assumed as symptoms of the motor conditions. The statistical uncertainty and fuzzy imprecision of the poles placement is taken into account by the clasterization procedure. The aim of this procedure is a separation of classes regarding a priori information of their number. Classification was performed with the use of the faulty electrical motors.
Keywords: fuzzy classification algorithm, acoustic process control, parametric model.
This paper describes briefly the development and verification of a probabilistic system for the rapid diagnosis of plant status and radioactive releases during postulated severe accidents in a Boiling Water Reactor nuclear power plant. The probabilistic approach uses Bayesian belief network methodology, and was developed in the STERPS project in the European Union 5-th Euroatom Framework program.
Keywords: nuclear reactors, source term, Bayesian belief network, severe accidents, probabilistic safety assessment.
The paper describes shape optimisation of a turbine blade shank. The turbine blade shank zone with a compound fillet is a critical location where a high risk of failure exists. The APDL language operating in Ansys environment is used to write a parametric turbine blade shank FEM models generator, which is a basic part of evolutionary optimisation routine. The goal of the optimisation is the 1st principal stress reduction with maximum allowable mass constraint imposed. Parameterisation routine and optimisation results are presented and discussed.
Keywords: AI, evolutionary optimisation, APDL, turbine blade, shape optimisation, Finite Element Method (FEM).
During the manufacturing process of multilayered fibre-reinforced composites with variable fibre orientations, residual stresses build up due to the directional expansion of the single unidirectionally reinforced layers. Dependent on the laminate lay-up, these inhomogeneous residual stresses, which are caused by thermal effects, moisture absorption and chemical shrinkage, can lead to large multistable out-of-plane deformations. Instead of avoiding these laminate's curvatures, they can be advantageously used for technical applications following the near-net-shape technology. However, due to the effect that the laminate curvature depends on huge amount of different parameters such as anisotropic, hygroscopic and thermomechanical material properties, fibre orientations and ply thickness of each single layer as well as technological processing parameters, a search in a multi-dimensional search area is necessary. In order to solve such a task, Genetic Algorithms in combination with a fitness function based on a nonlinear semi-analytical calculation model for the laminate shape prediction have been applied and described in the paper. Using this approach, one can purposefully adapt the laminate lay-up dependent on the loading and process parameters.
The paper is devoted to computational grids applications in evolutionary optimization of structures. The two grid middleware are used, UNICORE and LCG2. The distributed evolutionary algorithm is used for optimizataion. The fitness function is computed using finite element method. Numerical examples are presented.
In the paper, Artificial Neural Network with hidden layers is used to approximate the functional dependence of the effective properties of a composite on the physical properties of its micro-components. Two numerical examples have been examined in order to demostrate this approach. The first one introduces geometrical parameters of the cell of periodicity into ANN training process. It proves the ability of ANN to catch the behaviour of the composite material based on the properties of the components and their spatial arrangement at micro level. The second example deals with a special case of the self-repetitive composite structure. It has been shown that, in the limit, the geometry and behaviour of such a composite is consistent with the fractal form known in the literature as the Sierpinski's carpet.
Keywords: neural network, homogenisation, hierarchical composite.
Fault diagnosis becomes more and more difficult and sophisticated task. This is so mainly due to growing complexity - contemporary technological systems are assembled from numerous components which cooperate and recursively include other components. The main goal of this paper consists in presentation of an approach which is able to reduce time of diagnosis and quantity of produced diagnoses by using hierarchical, logic-based approach. The reduction is achieved here due to two main factors. The first one is that a hierarchical model of systems is used. Such approach limits search space, because the system is considered at various levels of details and some diagnoses which are possible potential ones at more abstract levels can be verified to be impossible at more detailed levels. The second factor is that levels can be described with use of different kinds of a logic-based knowledge representation, what lets fit some best representation to a particular level.
Automatic knowledge base generation using techniques such as genetic algorithms tend to be highly dependent on the quality and size of the learning data. First of all, large data sets can lead to unnecessary time loss, when smaller data sets could describe the problem as well. Second of all, the presence of noise and outliers can cause the learning algorithm to degenerate. Clustering techniques allow compressing and filtering the data, thus making the generation of fuzzy knowledge bases faster and more accurate. Different clustering algorithms are compared and the validation of the results through a theoretical 3D surface, shows that when compressing the data to 5% of its original size, clustering algorithms accelerate the learning process by up to 94%. Moreover, when the learning data contains noise and/or a large amount of outliers, clustering algorithms can make the results more stable and improve the fitness of the obtained FKBs.
This paper focuses on features extraction based on cyclostationarity for diagnosis purpose. The objective is to derive new indicators for the diagnosis of rotating machinery. These indicators are based on cyclic higher order statistics and generalize some existing ones for the second order statistics. A comprehensive methodology is proposed for obtaining a diagnosis objective; a crucial example is presented, relating to vibration signals of a gearbox. Results demonstrate the effectiveness of these features to detect spalling in gearbox.
Standard for the Exchange of Product Model Data (STEP-ISO 10303) contains product information models covering most of the aspects of product lifecycle management (PLM). In designing ontology of knowledge base concerning any aspects of PLM it is necessary to use such product information models. Unfortunately, a model based on STEP is based on EXPRESS language which is not compatible with common technologies for ontology creation. The paper presents contemporary methods of ontology description and it describes the importance as well as current level of advancement in development of STEP standard. Attention has been paid to drawbacks in possibilities of already existing methods, included in STEP for ontology creation, which could form a foundation for knowledge base designing methodology founded on the defined ontologies. A method of ontology creation has been proposed, including STEP model and examples of this method application in knowledge base implementation have been given.
Keywords: knowledge base, ontology, object-oriented modeling, STEP, UML, XML.
The paper deals with the problem of new projects acceptance into the multi project environment, where constraints are limiting the number of projects that a company is able to carry out concurrently. The objective of this paper is to answer the question: Is it possible to execute new project on time in the multi-project environment? For answering the question combination of Theory of Constraints and conditions guaranteeing project due dates with constraint-based scheduling are proposed. As a result the decision of the project implementation and the schedule of project activities, which the company is able to implement concurrently are obtained.
Keywords: Theory of Constraints, multi-project scheduling, constraint programming.
The utility of Bayesian neural networks to predict concrete fatigue durability as a function of concrete mechanical parameters of a specimen and characteristics of the loading cycle is investigated. Bayesian approach to learning neural networks allows automatic control of the complexity of the non-linear model, calculation of error bars and automatic determination of the relevance of various input variables. Comparative results on experimental data set show that Bayesian neural network works well.
Keywords: Bayesian neural networks, concrete fatigue durability, prediction.
At present, inference about technical state of machinery or industrial processes on the basis of analysis of residual signals is one of the most developed fields of technical diagnostics. Results of analysis are usually huge sets of signal features, whose changes carry information about technical state of an object. Correct interpretation of such results is the most important problem of technical diagnostics. It is particularly important in the case of complex machinery and processes, whose operation parameters vary in time and additionally the environment of investigated objects, can not be assumed to be unchangeable. The approach described in the paper was based on the assumption that signal features can be presented in the form of a dynamic scene. Changes in the scene are determined by means of simple methods of image processing. The background of diagnostic inference is context-based reasoning (CxBR).
The paper deals with an application of the theory of optimum experimental design to the problem of selecting the data set for developing neural models. Another objective is to show how to design a robust fault detection scheme with neural networks and how to increase its fault sensitivity by decreasing model uncertainty. It is also shown that the optimum design is independent of the parameters that enter linearly into the neural network. The final part of this paper shows a comprehensive simulation study regarding modelling and fault detection with the proposed approach. In particular, the DAMADICS benchmark problem is utilized to verify the performance and reliability of the proposed technique.