Apr 062012

This work presents a complete fuzzy model for surface grinding of steel using aluminium oxide grinding wheels. Six independent variables—wheel grade, grit size, table speed, depth of cut, dressing, and coolant application—are input, and eight dependent variables—two force components, two residual stress components, two surface roughness parameters, change in surface hardness, and the onset of surface burns—are output model predictions. Thesis StatementThe entire model is automatically extracted from experimental data, without human intervention.

Chapter 1 is a brief introduction to the technological and economical importance of the grinding process. It provides a motivational foundation for the desirability of better understanding, predictability, and control of this important manufacturing process.

It is argued that our current portfolio of engineering sciences were not intended for describing behaviour under conditions prevailing during grinding. Before attempting to model the grinding process, analytically, there is a need for revamping engineering sciences and extending their reach to those extreme conditions. Moreover, there is a need for significant advances in material testing procedures, in order to acquire constitutive models and material properties under those extreme—not the idealized laboratory—conditions. As neither frontiers—fundamental theories or testing procedures—are likely to make quantum leaps in the near future, we seek alternative solutions, outside the classical analytical paradigm. This argument is promoted in Chapters 2 and 3.

Chapter 2 discusses the fundamental physical processes occurring during grinding. It is shown that several mechanisms take place during a typical grinding process, with varying degrees of intensity and interdependent nonlinear interference. Not only we do notunderstand the extent of each mechanism and its interplay with others, but also we do not have satisfactory basic knowledge of each individual mechanism, under the conditions prevailing in grinding. This lack of knowledge manifests itself in a number of conflicting reports by various research groups.

Chapter 3 is a road map of research in grinding over the last five decades, and a general review of various research efforts. It is shown that analytical research for grinding has had limited practical utility, because it (1) is based on simplified physical models, (2) applies many unrealistic assumptions, for the convenience of numerical solvability, and (3) requires input values that are not readily available to the user. It is estimated that grinding models include, typically on average, ± 30% uncertainty in their predictions, when tested over wide range of testing environments. At such a level of uncertainty, the computational demand by many analytical models is unjustifiable. Practicing engineers, often, find it easier to just rely on skills and experience of their grinding machine operators.

Fuzzy modelling is, therefore, offered as an interim pragmatic solution to modelling the grinding process. It may lack the intellectual superiority and mathematical robustness of an analytical model, but it offers meaningful answers to practical questions at the workshop level.

However, a major difficulty with fuzzy modelling is the acquisition of reliable fuzzy rules. Human experiences are known to be victim to bias, misconception, and misinformation, and are confined to a limited range of current and recurrent observations, with no possibility for extension beyond established operating conditions. It would be advantageous to fuzzy modelling, if we develop a methodology for fuzzy rules generation that is independent of human experts, and derive rules automatically from experimental data. Such fuzzy models would have the same—or better—credibility and extensibility as other empirical models, but would be much easier to use and comprehend.

Chapter 4 is concerned with the process of constructing objective fuzzy models that do not require any human input. This is achieved by extracting rules from experimental data directly, not from heuristic human experience. Further, no human contribution is required for defining membership functions, selecting inference mechanisms, number of rules, etc. A basic requirement for model parameters is that they should enable a model to predict data from which it was originally extracted, with minimum error. Hence, fuzzy modelling is reduced to a problem of non-linear optimization over a vast dimensional space, and innovative algorithms for solving such a problem are introduced. Chapter 4 describes a set of algorithms that, starting from a given empirical data set, produce a complete fuzzy model that is optimized to an arbitrary level of precision, determined by the user. The methodology and algorithms are applicable to any set of data, and any engineering system, including grinding.

Quality of a model is as good as the data used in its construction. Therefore, it is of utmost importance that experimental data are produced with greatest precision, reliability, and reproducibility, over the largest number of pertinent variables and the widest range of variation. Equally important is to have an estimate of model quality and the degree of confidence in its prediction, by measuring confidence in quality of the underlying experimental procedures.

Chapter 5 is an experimental investigation into surface grinding. An extensive set of grinding tests are performed on annealed A4140 steel, under a wide range of grinding conditions. A number of important product characteristics are measured and reported for each test point. Attention is given to detailed description of the experimental setup, calibrations, procedures, and standards followed. This is to help users of the fuzzy model identify the differences between their respective work environments, and the environment from which this model is extracted, and to estimate the likely effect of those differences on predictions made by the model.

Particular emphasis is placed on estimating the range of error, scatter, or uncertainty in the experimental data obtained. It is found that an average ±20% uncertainty in all grinding quantities can be expected. This uncertainty is irreducible by refinement or stricter control on the experimental environment, because it is an inherent characteristic of the grinding process itself. This provides further support for the fuzzy modelling paradigm.

Chapter 6 applies the algorithms in Chapter 4 to the experimental data from Chapter 5, and presents a complete, six dimensions eight variates, fuzzy model for the surface grinding process. Implementation of the model is demonstrated by two worked examples. This model is useful in the grinding environment from which it was extracted. Therefore, guidelines are offered on means of transporting, and optimizing the model for usage within other grinding environments.

The model developed in this work is original in (1) the way it was automatically extracted from data, (2) the large number of variables included, (3) the compact and efficient structure of the rule-base and its presentation, and (4) that the complete functional model is available for other users to impalement, e.g. within their machine control systems. Finally, Chapter 7 is a summary of main topics where further research may be needed. Brief discussions and proposals for conducting those future research tasks are also provided.

Sorry, the comment form is closed at this time.