Marcin Perzyk

Prediction of ductile cast iron quality by artificial neural networks

Abstract The prediction of ductile cast iron properties by an analysis of physical and chemical phenomena occurring during melting process is discussed. Characterisation of the problem from the standpoint of artificial neural network (ANN) modelling is presented and the network input parameters are defined, based on their significance and availability in industrial practice. A large number of different networks have been constructed and trained using about 700 melts results recorded in a typical foundry. A comprehensive analysis of the prediction errors of tensile strength, elongation and hardness made by the networks is presented. It is concluded that ANNs can be used as a valuable tool for the on-line production control in the melt shops.

Comparison of data mining tools for significance analysis of process parameters in applications to process fault diagnosis

This paper presents an evaluation of various methodologies used to determine relative significances of input variables in data-driven models. Significance analysis applied to manufacturing process parameters can be a useful tool in fault diagnosis for various types of manufacturing processes. It can also be applied to building models that are used in process control. The relative significances of input variables can be determined by various data mining methods, including relatively simple statistical procedures as well as more advanced machine learning systems. Several methodologies suitable for carrying out classification tasks which are characteristic of fault diagnosis were evaluated and compared from the viewpoint of their accuracy, robustness of results and applicability. Two types of testing data were used: synthetic data with assumed dependencies and real data obtained from the foundry industry. The simple statistical method based on contingency tables revealed the best overall performance, whereas advanced machine learning models, such as ANNs and SVMs, appeared to be of less value.

A hybrid system with regression trees in steel-making process

The paper presents a hybrid regresseion model with the main emphasis put on the regression tree unit. It discusses input and output variable transformation, determining the final decision of hybrid models and node split optimization of regression trees. Because of the ability to generate logical rules, a regression tree maybe the preferred module if it produces comparable results to other modules, therefore the optimization of node split in regression trees is discussed in more detail. A set of split criteria based on different forms of variance reduction is analyzed and guidelines for the choice of the criterion are discussed, including the trade-off between the accuracy of the tree, its size and balance between minimizing the node variance and keeping a symmetric structure of the tree. The presented approach found practical applications in the metallurgical industry.

Knowledge in Imperfect Data

Data bases collecting a huge amount of information pertaining to real-world processes, for example industrial ones, contain a significant number of data which are imprecise, mutually incoherent, and frequently even contradictory. It is often the case that data bases of this kind often lack important information. All available means and resources may and should be used to eliminate or at least minimize such problems at the stage of data collection. It should be emphasized, however, that the character of industrial data bases, as well as the ways in which such bases are created and the data are collected, preclude the elimination of all errors. It is, therefore, a necessity to find and develop methods for eliminating errors from already-existing data bases or for reducing their influence on the accuracy of analyses or hypotheses proposed with the application of these data bases. There are at least three main reasons for data preparation: (a) the possibility of using the data for modeling, (b) modeling acceleration, and (c) an increase in the accuracy of the model. An additional motivation for data preparation is that it offers a possibility of arriving at a deeper understanding of the process under modeling, including the understanding of the significance of its most important parameters.

Applications of Data Mining to Diagnosis and Control of Manufacturing Processes

In the majority of manufacturing companies large amounts of data are collected and stored, related to designs, products, equipment, materials, manufacturing processes etc. Utilization of that data for the improvement of product quality and lowering manufacturing costs requires extraction of knowledge from the data, in the form of conclusions, rules, relationships and procedures. Consequently, a rapidly growing interest in DM applications in manufacturing organizations, including the development of complex DM systems, can be observed in recent years (Chen et al. 2004; Chen et al. 2005; Dagli & Lee, 2001; Hur et al., 2006; Malh & Krikler, 2007; Tsang et al., 2007). A comprehensive and insightful characterization of the problems in manufacturing enterprises, as well as the potential benefits from the application of data mining (DM) in this area was presented in (Shahbaz et al., 2006). Examples and general characteristics of problems related to the usage of data mining techniques and systems in a manufacturing environment can be found in several review papers (Harding et al., 2006; Kusiak, 2006; Wang, 2007). Application of DM techniques can bring valuable information, both for designing new processes and for control of currently running ones. Designing the processes and tooling can be assisted by varied computer tools, including simulation software, expert systems based on knowledge acquired from human experts, as well as the knowledge extracted semi automatically by DM methods. The proper choice of the manufacturing process version and its parameters allows to reduce the number of necessary corrections resulting from simulation and/or floor tests. The knowledge obtained by DM methods can significantly contribute to the right decision making and optimum settings of the process parameters. In the design phase two main forms of knowledge may be particularly useful: the decision logic rules in the form: ‘IF (conditions) THEN (decision class)’ and the regression–type relationships. Although the latter have been widely utilized before the emergence of DM methods (e.g. in the form of empirical formulas) and the rules created by the human experts were also in use, the computational intelligence (CI) methods (learning systems) have remarkably enhanced possibilities of the knowledge extraction and its quality. For the manufacturing process control many varied methods are used, ranging from paper Statistical Process Control (SPC) charts to automated closed loop systems. In spite of the

A Micromechanical Prediction of Elastic Properties of Composites with Spherical Particles

A corrected solution of the stress and strain distribution problem around a single spherical elastic inclusion in an elastic body is presented. Subsequent ly, by an appropriate selection of boundary conditions and elastic energy balance equations, the stress and strain distribution in a composite material, consisting of a matrix and regularly arranged spherical particles, is approx imated. In the result the Youngs modulus and the Poissons ratio of the composite considered as a continuous medium is obtained. These values are presented graphically as a function of the elastic properties and concentra tions of both components. A partial experimental verification of the theoretical calculations, based on some published experimental results of other authors as well as experiments performed especially for this purpose, has been carried out. A good agreement with the presented theory has been achieved.

Generalization of the Kocks-Mecking Type Constitutive Model for Cyclic Deformation Description of Metals

An elasto-viscoplastic, history dependent, one-dimensional constitutive model for metallic materials with the reference stress as an internal parameter has been extended for description of the phenomena related to strain reversals. At any instant of arbitrarily complex thermomechanical process, one of the two different directions of the viscoplastic deformation can be active: ‘primary’ or ‘secondary’. The reference stress is split into two variables and the evolutionary equations for both of them as well as the switch condition for both flow directions are presented. Varied numerical simulations of cyclic deformation are demonstrated showing appreciable ‘flexibility’ of the plots as well as their similarity to commonly observed experimental results. It is concluded that the predictive capabilities, the attainability of the material data as well as the relative simplicity of the model are encouraging for its practical applications.

The Role of Deformation History Effects in Generation of Residual Thermal Stresses

Possible sources of the deformation history effects in generation of residual thermal stresses in metallic materials are briefly discussed. The application of a history dependent model to calculation of unidirectional residual stresses in a simple three-rods rectangular symmetrical grid structure is demonstrated. The obtained values are compared with those calculated using the history independent model, i.e. the mechanical equation of state proposed by Ludwik. Various cooling conditions and material properties were tried. The obtained results show that the temperature and strain-rate history exhibits a vital influence on the residual thermal stresses while the Bauschinger’ s effect seems to be a less important factor.