Babak Forouraghi

A Genetic Algorithm for Multiobjective Robust Design

The goal of robust design is to develop stable products that exhibit minimum sensitivity to uncontrollable variations. The main drawback of many quality engineering approaches, including Taguchis ideology, is that they cannot efficiently handle presence of several often conflicting objectives and constraints that occur in various design environments.Classical vector optimization and multiobjective genetic algorithms offer numerous techniques for simultaneous optimization of multiple responses, but they have not addressed the central quality control activities of tolerance design and parameter optimization. Due to their ability to search populations of candidate designs in parallel without assumptions of continuity, unimodality or convexity of underlying objectives, genetic algorithms are an especially viable tool for off-line quality control.In this paper we introduce a new methodology which integrates key concepts from diverse fields of robust design, multiobjective optimization and genetic algorithms. The genetic algorithm developed in this work applies natural genetic operators of reproduction, crossover and mutation to evolve populations of hyper-rectangular design regions while simultaneously reducing the sensitivity of the generated designs to uncontrollable variations. The improvement in quality of successive generations of designs is achieved by conducting orthogonal array experiments as to increase the average signal-to-noise ratio of a pool of candidate designs from one generation to the next.

Worst-case tolerance design and quality assurance via genetic algorithms

In many engineering designs, several components are often placed together in a mechanical assembly. Due to manufacturing variations, there is a tolerance associated with the nominal dimension of each component in the assembly. The goal of worst-case tolerance analysis is to determine the effect of the smallest and largest assembly dimensions on the product performance. Furthermore, to achieve product quality and robustness, designers must ensure that the product performance variation is minimal.Recently, genetic algorithms (GAs) have gained a great deal of attention in the field of tolerance design. The main strength of GAs lies in their ability to effectively perform directed random search in a large space of design solutions and produce optimum results. However, simultaneous treatment of tolerance analysis and robust design for quality assurance via genetic algorithms has been marginal.In this paper, we introduce a new method based on GAs, which addresses both the worst-case tolerance analysis of mechanical assemblies and robust design. A novel formulation based on manufacturing capability indices allows the GA to rank candidate designs based on varying the tolerances around the nominal design parameter values. Standard genetic operators are then applied to ensure that the product performance measure exhibits minimal variation from the desired target value. The computational results in the design of a clutch assembly highlight the advantages of the proposed methodology.

A Particle Swarm Algorithm for Multiobjective Design Optimization

Many engineering design problems are characterized by presence of several conflicting objectives. This requires efficient search of the feasible design region for optimal solutions which simultaneously satisfy multiple design objectives. The search is further complicated in view of the fact that because of inherent manufacturing variations it is often necessary to allocate tolerances to design variables while guaranteeing low variances for product/process performance measures. Particle swarm optimization (PSO) is a powerful search technique with faster convergence rates than traditional evolutionary algorithms. This paper introduces a new PSO-based approach to multiobjective engineering design by incorporating the central quality-control notion of tolerance design. Unlike classical optimization techniques which rely on single-point representation of designs, the modified PSO algorithm allocates tolerances to design variables and flies a swarm of hypercubic particles through the feasible space. To demonstrate the utility of the proposed method, the multiobjective design of an I-beam is presented

Fuzzy multiobjective optimization with multivariate regression trees

We introduce a new methodology in which multiobjective optimization is formulated as unsupervised learning through induction of multivariate regression trees. In particular, it is shown that learning of Pareto-optimal solutions can be efficiently accomplished by using a number of fuzzy tree partitioning criteria. These include: a newly formulated fuzzy method based on Kendalls nonparametric measure of association (G. Simon, 1977), Bellman-Zadehs approach to multiobjective decision making utilized in an inductive framework (R.E. Bellman and L.A. Zadeh, 1970), and finally, multidimensional fuzzy entropy (B. Kosko, 1990). For purposes of comparison, the efficiency of learning with fuzzy partitioning criteria is compared with that of two conventional multivariate statistical techniques based on dispersion matrices. The widely used problem of design of a three bar truss is presented to highlight advantages of our new approach.<<ETX>>

An Interactive and Personalized Cloud-Based Virtual Learning System to Teach Computer Science

Virtual learning environments VLE provide up-to-date education and training for individuals, and in some cases they can generate personalized feedback based on the learners performance. Unlike other VLEs available to-date, the learning system developed in this work extends the basic instructional and assessment capabilities of a typical VLE by dynamically creating a cloud-based computer laboratory that is needed by Computer Science students. Specifically, the basic capabilities of Moodle are enhanced by developing two new modules: a virtual lab module VLM and a study progress module SPM. VLM utilizes Amazons EC2 cloud-based web services technology in order to create a personalized experimental environment especially suited for Computer Science students although the same concept can easily be applied to other disciplines. SPM, on the other hand, evaluates student progress in terms of specific learning objectives and offers personalized guidance using the Apriori data mining technique. An implemented cloud-based Internet Application Development laboratory highlights the advantages of the proposed approach.

On utility of inductive learning in multiobjective robust design

Most engineering design problems involve optimizing a number of often conflicting performance measures in the presence of multiple constraints. Traditional vector optimization techniques approach these problems by generating a set of Pareto-optimal solutions, where any specific objective can be further improved only at the cost of degrading one or more other objectives. The solutions obtained in this manner, however, are only single points within the space of all possible Pareto-optimal solutions and generally do not indicate to designers how small deviations from predicted design parameters settings affect the performance of the product or the process under study.In this paper we introduce a new approach to robust design based on the concept of inductive learning with regression trees. Given a set of training examples relating to a multiobjective design problem, we demonstrate how a multivariate regression tree can utilize an information-theoretic measure of covariance complexity to capture optimal, tradeoff design surfaces. The novelty of generating design surfaces as opposed to traditional points in the design space is that now designers are able to easily determine how the responses of a product or process vary as design parameters change. This ability is of paramount importance in situations where design parameter settings need to be modified during the lifetime of a product/process due to various economic or operational constraints. As a result, designers will be able to select optimal ranges for design parameters such that the products performance indices exhibit minimal or tolerable deviations from their target values. To highlight the advantages of our methodology, we present a multiobjective example that deals with optimum design of an electric discharge machining (EDM) process.

Multi-objective Particle Swarm Optimization with Gradient Descent Search

Particle swarm optimization (PSO) has been proven to be a reliable method to deal with many types of optimization problems. Specifically, when solving multi-objective PSO (MOPSO) optimization problems careful attention must be paid to parameter selection and implementation strategy in order to improve the performance of the optimizer. This paper proposes a novel MOPSO with enhanced local search ability. A new parameter-less sharing approach is introduced to estimate the density of particles’ neighborhood in the search space. Initially, the proposed method accurately determines the crowding factor of the solutions; in later stages, it effectively guides the entire swarm to converge closely to the true Pareto front. In addition, the algorithm utilizesthe local search method of gradient descent to better explore the Pareto-optimal region. The algorithm’s performance on several test functions and an engineering design problem is reported and compared with other approaches. The obtained results demonstrate that the proposed algorithm is capable of effectively searching along the Pareto-optimal front and identifying the trade-offsolutions.

Robust engineering design with genetic algorithms

One of the difficulties of using Artificial Neural Networks (ANNs) to estimate atmospheric temperature is the large number of potential input variables available. In this study, four different feature extraction methods were used to reduce the input vector to train four networks to estimate temperature at different atmospheric levels. The four techniques used were: genetic algorithms (GA), coefficient of determination (CoD), mutual information (MI) and simple neural analysis (SNA). The results demonstrate that of the four methods used for this data set, mutual information and simple neural analysis can generate networks that have a smaller input parameter set, while still maintaining a high degree of accuracy.

Concept Classification Using a Hybrid Data Mining Model

Apriori is a well-known algorithm which is used extensively in market-basket analysis and data mining. The algorithm is used for learning association rules from transactional data bases and is based on simple counting procedures. In this paper we propose enhancements to Apriori which allow it to perform concept classification similar to the way decision tree algorithms learn. Specifically, training examples are modified and treated as transactional data and the results are verified and further improved by C4.5 decision tree and k-means clustering algorithms, respectively. To demonstrate the novelty of the enhanced Apriori algorithm, we present a hybrid data mining model (HDMM) which identifies at-risk students based on their academic performance and other pertinent data.

On Scalability of Distributed Machine Learning with Big Data on Apache Spark

Performance of traditional machine learning systems does not scale up while working in the world of Big Data with training sets that can easily contain petabytes of data. Thus, new technologies and approaches are needed that can efficiently perform complex and time-consuming data analytics without having to rely on expensive super machines.