Lists
Lists
Davim, J. Paulo (2013) Intelligent systems applied in manufacturing technology: statement of the nature and significance of the work. Doctoral thesis, London Metropolitan University.
Manufacturing industry has a great economic importance not only in industrialized and developed countries (G7) but also in countries with emerging economies (BRICS). Nowadays, there has been an increased interest in the development of intelligent systems (computational and statistical methods) in order to be applied in the modelling, simulation and optimization of the manufacturing technology. Computational and statistical methods have achieved several applications, namely, monitoring and controlling, process parameters modeling, processes optimization and computer-aided process planning (CAPP) [1-4]. Manufacturing processes optimization is a key aspect, too relevant in what concerns the design of competitive, efficient and modern manufacturing technology. Today, the modeling of physical phenomena involved in manufacturing processes has been recognized as one of the most important tasks in manufacturing technology research. Due to their complex nature, manufacturing processes are very difficult to be understood, modeled or simulated. However, in the present days, and in order to implement this task, researches can appeal to the classical statistical techniques, the finite element method (FEM) as well as sophisticated artificial intelligence (soft computational) techniques, such as artificial neural networks (ANNs), fuzzy logic and neuro-fuzzy systems. Also, the evolutionary computation, simulating annealing (SA) and ant-colony optimization (ACO), and particle swarm optimization (PSO) allow them carrying out the optimization of complex systems, where classical techniques have problems [5].
Classical statistical techniques are based in design of experiments (DOE). This methodology was proposed by Sir Ronald A. Fisher in the famous book The Design of experiments (1935). The DOE takes levels of several input parameters to formulate the different combinations at which the output parameters are to be computed [6]. There are several types of DOE available, which are based on statistical theory, namely, Full factorial design, Fractional factorial design, Taguchi orthogonal arrays, Central composite design and Box-Behnken design. The response surface methodology (RSM) is a collection of statistical and mathematical techniques useful for modeling and optimizing the processes that provide an overall perspective of the system response within the design space [6]. The classical experimental design methods are too complex and not easy to use. If there are more parameters, a large number of experiments need to be carried out. This problem is overcome in Taguchi method using special design of orthogonal arrays [6].
Finite element method (FEM) has applied to a wide field of engineering problems in the last years as a powerful and versatile numeric method for finding good approach solutions for systems of partial differential equations [?]. The domain is discretized into a number of small elements and the primary dependent variables of the partial differential equations are approached by suitable functions over an element [3]. It can be used in order to obtain numerical solutions from complex phenomenological models. FEM has become the main instrument to modeling and simulate manufacturing processes. Experimental approaches to the study of manufacturing processes are important but they can be replaced by FEM analysis with the advantage of saving time and money required to undertake experimental procedures in laboratory. However, the experimental validation of FEM analysis in manufacturing processes is also very important [3].
Artificial intelligence (soft computing) techniques are more based in biological systems than in formal logic and include, but is not limited to, artificial neural networks (ANNs), fuzzy logic, neuro-fuzzy systems, evolutionary computation, simulating annealing (SA), ant-colony optimization (AGO), and particle swarm optimization (PSO). ANNs are connectionist structures that model the operations of biological neural systems. Any ANN is composed of a number of neurons (or nodes). These nodes are interconnected, so the output of one (or some) of them is the input of another one(s) [7]. Fuzzy logic, proposed by Zadeh (1960), is the generalization of Boolean (0 or 1) logic that deals with fuzzy sets and logical connectives for modeling problems. In fuzzy logic propositions can have different values that are known as degree-of-true ranging between O and 1 [8]. By combining the artificial neural networks and fuzzy logic, a class of powerful tools, called neuro fuzzy systems, arose. Evolutionary computation study computational systems by using ideas inspired from natural evolution, for example, genetic algorithms (GA) developed by Holland and evolution strategies (ES) proposed by Rechenberg and Schwefel [5]. Simulated annealing (SA) resembles the cooling process of molten metals through annealing. The SA procedure simulates this process of annealing to achieve the minimization function value in the problem [7]. Ant colony optimization (AGO), introduced by Dorigo, is a metaheuristic algorithm that tries to provide solutions to combinatorial optimization tasks [9]. Particle swarm optimization (PSO), introduced by Kennedy and Eberhat, is an optimization strategy that explores the search space in order to maximize a specific quality target. The population-based algorithm is inspired by social behavior of bird flocking or fishing schooling. The PSO presents similarities with evolutionary computation [9].
More recently, the combination of the FEM with artificial intelligence (soft computing) techniques is used with two different approaches [5-7]. The first is based on using FEM to simulate the manufacturing process under different conditions and the output of these simulations is employed as a dataset for training soft computing models. The other approach is based in the use of soft computing models as input for the FEM. Here, the experimental data is processed by some soft computing techniques, such as artificial neural networks or fuzzy logic, in order to obtain the empirical model. Finally this empirical model is introduced to FEM model to predicting some variables of manufacturing processes [5].
Considering the importance of this subject, the main aim of this work is concerned with the modeling, simulation and optimization of manufacturing processes with special emphasis in machining technology. Conscious of the importance of this scientific area, along the last 15 years, these topics have been developed and coordinated by the applicant, along with his systematic scientific research, aiming to promote the modeling, simulation and optimization of machining processes. The firsts scientific articles in international journals indexed in JCR-ISI Web of Knowledge (Thomson Reuters©) were published in 2001 [10-12]. In a first step they were studied statistical models based on experimental results which were, subsequently, extended to computational models. The experimental results were obtained using systematic laboratorial research in machining technology (or in some cases Laser technology). Later, based on experimental results, it was possible to apply to statistical and computational models for optimizing manufacturing parameters. Research allowed to modeling and optimizing manufacturing parameters using design of experiments (DOE) [13-14], Taguchi techniques [13-16], genetic algorithms (GA) [16-19], artificial neural networks (ANNs) [20-22], fuzzy logic [17, 23], response surface methodology (RSM) [21], particle swarm optimization (PSO) [24-26] among others.
The research is also focused on FEM analysis to modeling the material thermo mechanical behavior during machining process and to determine the influence of the friction coefficient in the tool-chip interface, the cutting and feed forces, cutting temperature, plastic strain, plastic strain rate, maximum shear stress and residual stresses [27-30]. It is important to note that the material behavior modeling has great influence on the design of process, tools and the final product [31]. In general, an experimental validation of the machining process was conducted in order to verify the numerical simulated results.
The innovation of this research is due not only to the original experimental results modeled, but also to the application of computational and statistical techniques for optimizing machining process modeled. Modeling based in response surface methodology (RSM) using DOE (Design of experiments) and artificial neural networks (ANN) are two good options used. The development of the model by RSM requires a minimum number of experiments to be conducted, although restricted to only small range of input variables and hence not suitable for complex and highly nonlinear processes. ANN is a powerful modeling tool for application in complex and nonlinear problems. However, the main novelty of this work is the combination of FEM analysis with the flexibility and adaptability of artificial intelligent techniques for solving the complex problems related to the study of manufacturing processes.
Taking into account these considerations we can say that this undertaken research is of interest for direct application in modern industry and for the large number of researchers working in this important field in the laboratories of companies and universities in different parts of the world. In future, with the development, for example, of the combination of artificial intelligence techniques with FEM analysis, this research line continues with a great potential of evolution as well as with application of sophisticated heuristic search algorithms, namely, simulated annealing (SA), ant colony algorithm (ACO), particle swarm optimization (PSO), for optimizing machining processes.
The original results of a considerable part of this research, developed and under the leadership of the applicant, with the collaboration of numerous researchers from different countries, are published in a set of works presented in this application and that you can see above, namely: 4 edited books of international circulation [1-4], 2 books as co-author [5, 32], 10 book chapters [6-9; 16, 19-20, 23-25], articles published in journals indexed in JCR-ISI Web of Knowledge (Thomson Reuters©) [10-15, 17-18, 21-22, 26-31] and in 2 journal "special issues" of international circulation indexed in SCOPUS© [33-34].
 |
View Item |