Soft Computing in Industrial Applications

Wireless sensor networks/data-logging devices are increasingly applied for distributed measurement and acquiring additional contextual data. These have been applied in large scale indoor and outdoor systems with solutions based on RF, light based and ultra sound based systems. Data-loggers in liquid filled containers pose new challenges for localization because of the high reflectivity of containers and high attenuation due to the liquids obstructing communication between wireless nodes. Magnetic localization techniques have been used in many places including military research [

14

]. This approach was adapted for use in liquid filled containers. In this project, two prototypes, a laboratory and an industrial installation have been conceived and served for acquisition of experimental data for localization. In our paper, we exploit the sparsity met in the particular magnetic MEMS sensor swarm localization concept by introducing NLMR which is a simplified form of Sammon’s mapping (NLM) and we combine it with different meta-heuristics and soft-computing techniques, e.g., gradient descent, Simulated Annealing and PSO. We compare this with Multilateration and conventional NLM localization technique. Our approach has improved the localization from a mean error of 20 cm in the first cut analysis for the industrial setup using conventional NLM down to 11 cm without and to 9 cm with apriori knowledge. Future improvements are to be expected from a thorough calibration of all system components. in [

5

]. The modified algorithm is capable of distributed localization producing mean localization error of 10 cm for the Warstein experiment data.

Contemporary medicine concentrates on providing high quality diagnostic services, yet it should not be forgotten that the comfort of the patient during the examination is also of high importance. Therefore non-invasive methods that allows to precisely predict the state of the disease are currently one of the key issues in the medical business. The paper presents a novel ensemble of neural networks applied to recognition of liver fibrosis stage from indirect examination method. Several neural network models are build on the basis of outputs of different feature selection algorithms. Then an ensemble pruning procedure with the usage of diversity measures is conducted in order to eliminate redundant predictors from the pool. Finally the weights of classifiers in the fusion process are assessed to establish their influence on the output of the whole ensemble. Proposed method is compared with several state-of-the-art ensemble methods. Extensive experimental investigations, carried out on a dataset collected by authors, show that the proposed method achieve a satisfactory level of the fibrosis level recognition, outperforming other machine learning algorithms and thus may be used as a real-time medical decision support system for this task.

In recent years, a plethora of multi-objective evolutionary algorithms (MOEAs) have been proposed which are able to effectively handle complex multi-objective problems. In this paper, we focus on Equilibrium Problems with Equilibrium Constraints. We show that one interpretation of the game can also be handled by MOEAs and then discuss a simple methodology to map the non-cooperative outcome to the cooperative outcome. We demonstrate our proposed methodology with examples sourced from the economics and transportation systems management literature. In doing so we suggest resulting policy implications which will be of importance to regulatory authorities.

In this chapter, a new GP-based algorithm is proposed. The algorithm, named SGP (Statistical GP), exploits statistical information, i.e. mean, variance and correlation-based operators, in order to improve the GP performance. SGP incorporates new genetic operators, i.e. Correlation Based Mutation, Correlation Based Crossover, and Variance Based Editing, to drive the search process towards fitter and shorter solutions. Furthermore, this work investigates the correlation between diversity and fitness in SGP, both in terms of phenotypic and genotypic diversity. First experiments conducted on four symbolic regression problems illustrate the goodness of the approach and permits to verify the different behavior of SGP in comparison with standard GP from the point of view of the diversity and its correlation with the fitness.

In this paper, a segmentation system with a modified automatic Seeded Region Growing (SRG) based on Particle Swarm Optimization (PSO) image clustering will be presented. The paper is focused on Magnetic Resonance Imaging (MRI) breast tumour segmentation. The PSO clusters’ intensities are involved in the proposed algorithms of the automated SRG initial seed and threshold value selection. Prior to that, some pre-processing methodologies are involved. And breast skin is detected and deleted using the integration of two algorithms, i.e. Level Set Active Contour and Morphological Thinning. The system is applied and tested on the RIDER breast MRI dataset, and the results are evaluated and presented in comparison to the Ground Truths of the dataset. The results show higher performance compared to the previous segmentation approaches that have been tested on the same dataset.

Many real life optimization problems are multimodal with multiple optima. Evolutionary Algorithms (EA) have successfully been used to solve these problems, but they have the disadvantage since that they converge to only one optimum, even though there are many optima. We proposed a hybrid algorithm combining differential evolution (DE) with tabu search (TS) to find multiple solutions of these problems. The proposed algorithm was tested on optimization problems with multiple optima and the results compared with those provided by the Particle Swarm Optimization (PSO) algorithm.

In this paper, a new variant of the Radial Basis Function Network with the Dynamic Decay Adjustment algorithm (i.e., RBFNDDA) is introduced for undertaking pattern classification problems with noisy data. The RBFNDDA network is integrated with the

k-

nearest neighbours algorithm to form the proposed RBFNDDA-KNN model. Given a set of labelled data samples, the RBFNDDA network undergoes a constructive learning algorithm that exhibits a greedy insertion behaviour. As a result, many prototypes (hidden neurons) that represent small (with respect to a threshold) clusters of labelled data are introduced in the hidden layer. This results in a large network size. Such small prototypes can be caused by noisy data, or they can be valid representatives of small clusters of labelled data. The KNN algorithm is used to identify small prototypes that exist in the vicinity (with respect to a distance metric) of the majority of large prototypes from different classes. These small prototypes are treated as noise, and are, therefore, pruned from the network. To evaluate the effectiveness of RBFNDDA-KNN, a series of experiments using pattern classification problems in the medical domain is conducted. Benchmark and real medical data sets are experimented, and the results are compared, analysed, and discussed. The outcomes show that RBFNDDA-KNN is able to learn information with a compact network structure and to produce fast and accurate classification results.

This chapter proposes an approach to fuzzy modeling of Anti-lock Braking Systems (ABSs). The local state-space models are derived by the linearization of the nonlinear ABS process model at ten operating points. The Takagi-Sugeno (T-S) fuzzy models are obtained by the modal equivalence principle, where the local state-space models are the rule consequents. The optimization problems are defined in order to minimize the objective functions expressed as the squared modeling errors, and the variables of these functions are a part of the parameters of input membership functions. Simulated Annealing algorithms are implemented to solve the optimization problems and to obtain optimal T-S fuzzy models. Real-time experimental results are included to validate the new optimal T-S fuzzy models for ABS laboratory equipment.

This paper describes the hybridization of an evolutionary algorithm with a greedy algorithm to solve a job-shop problem with recirculation. We model a real problem that arises within the domain of loads’ dispatch inside an automatic warehouse. The evolutionary algorithm is based on random key representation. It is very easy to implement and allows the use of conventional genetic operators for combinatorial optimization problems. A greedy algorithm is used to generate active schedules. This constructive algorithm reads the chromosome and decides which operation is scheduled next. This option increases the efficiency of the evolutionary algorithm. The algorithm was tested using some instances of the real problem and computational results are presented.

The present study deals with the fuzzy reliability analysis of washing unit in a paper plant utilizing available uncertain data which reflects their components’ failure and repair pattern. Paper computes different reliability parameters of the system in the form of fuzzy membership functions. Two soft-computing based hybridized techniques namely Genetic Algorithms Based Lambda-Tau (GABLT) and Neural Network and Genetic Algorithms Based Lambda-Tau (NGABLT) along with traditional Fuzzy Lambda-Tau (FLT) technique are used to evaluate the fuzzy reliability parameters of the system. In FLT, ordinary fuzzy arithmetic is utilized while in GABLT and NGABLT ordinary arithmetic and nonlinear programming approach are used. The computed results, as obtained by these techniques, are compared. Crisp and defuzzified results are also computed. Based on results some important suggestions are given for future course of action in maintenance planning.

This work treats the single machine scheduling problem in which the setup time depends on the sequence and the job family. The objective is to minimize the makespan and the total weighted tardiness. In order to solve the problem two multi-objective algorithms are analyzed: one based on Multi-objective Variable Neighborhood Search (

MOVNS

) and another on Pareto Iterated Local Search (

PILS

). Two literature algorithms based on

MOVNS

are adapted to solve the problem, resulting in the

MOVNS_Ottoni

and

MOVNS_Arroyo

variants. Also, a new perturbation procedure for the

PILS

is proposed, yielding the

PILS1

variant. Computational experiments done over randomly generated instances show that

PILS1

is statistically better than all other algorithms in relation to the cardinality, average distance, maximum distance, difference of hypervolume and epsilon metrics.

Advanced sensing systems, sophisticated algorithms and increasing computational resources continuously enhance active safety technology for vehicles. Driver status monitoring belongs to the key components of advanced driver assistance system which is capable of improving car and road safety without compromising driving experience. This paper presents a novel approach to driver status monitoring aimed at drowsiness detection based on depth camera, pulse rate sensor and steering angle sensor. Due to NIR active illumination depth camera can provide reliable head movement information in 3D alongside eye gaze estimation and blink detection in a non-intrusive manner. Multi-sensor data fusion on feature level and multilayer neural network facilitate the classification of driver drowsiness level based on which a warning can be issued to prevent traffic accidents. The presented approach is implemented on an integrated soft-computing system for driving simulation (DeCaDrive) with multi-sensing interfaces. The classification accuracy of

$$98.9\,\%$$

for up to three drowsiness levels has been achieved based on data sets of five test subjects with 588-min driving sequence.

The Self-Organizing Map (SOM) is a popular neural network model for clustering and visualization problems. However, it suffers from two major limitations,

viz.

, (1) it does not support online learning; and (2) the map size has to be pre-determined and this can potentially lead to many “trial-and-error” runs before arriving at an optimal map size. Thus, an evolving model, i.e., the Evolving Tree (ETree), is used as an alternative to the SOM for undertaking a text document clustering problem in this study. ETree forms a hierarchical (tree) structure in which nodes are allowed to grow, and each leaf node represents a cluster of documents. An experimental study using articles from a flagship conference of Universiti Malaysia Sarawak (UNIMAS), i.e., the

Engineering Conference

(ENCON), is conducted. The experimental results are analyzed and discussed, and the outcome shows a new application of ETree in text document clustering and visualization.

Examined are sources of brain activity, contributing to EEG patterns which correspond to motor imagery during training to control brain–computer interface (BCI). To identify individual source contribution into EEG recorded during the training, Independent Component Analysis (ICA) was employed. Those independent components, for which the BCI system classification accuracy was at maximum, were treated as relevant to performing the motor imagery tasks. Activities of the three most relevant components demonstrate well exposed event related desynchronization (ERD) and event related synchronization (ERS) of the mu-rhythm during imagining of contra- and ipsilateral hand and feet movements. To reveal neurophysiological nature of these components we solved the inverse EEG problem in order to localize the sources of brain activity causing these components to appear in EEG. Individual geometry of brain and its covers provided by anatomical MR images, was taken into account when localizing the sources. The sources were located in hand and feet representation areas of the primary somatosensory cortex (Brodmann areas 3a). Their positions were close to foci of BOLD activity obtained in fMRI study.

Despite of the popularity of the fuzzy Failure Mode and Effects Analysis (FMEA) methodology, there are several limitations in combining the Fuzzy Inference System (FIS) and the Risk Priority Number (RPN) model. Two main limitations are: (1) it is difficult and impractical to form a complete fuzzy rule base when the number of required rules is large; and (2) fulfillment of the monotonicity property is a difficult problem. In this paper, a new fuzzy FMEA methodology with a zero-order Single Input Rule Modules (SIRMs) connected FIS-based RPN model is proposed. An SIRMs connected FIS is adopted as an alternative to the traditional FIS to reduce the number of fuzzy rules required in the modeling process. To preserve the monotonicity property of the SIRMs-connected FIS-based RPN model, a number of theorems in the literature are simplified and adopted as the governing equations for the proposed fuzzy FMEA methodology. A case study relating to edible bird nest (EBN) processing in Sarawak (together with Sabah, known as the world’s number two source area of bird nest after Indonesia) is reported. In short, the findings in this paper contribute towards building a new fuzzy FMEA methodology using the SIRM s connected FIS-based RPN model. Besides that, the usefulness of the simplified theorems in a practical FMEA application is demonstrated.

Hierarchical architecture is an effective mechanism to make the Wireless Sensor Networks (WSNs) scalable and energy-efficient. Clustering the sensor nodes is a famous two-layered architecture which is suitable for WSNs and has been extensively explored for different purposes and applications. In this paper, a novel clustering approach called the Energy-Efficient Distance-based Clustering (EEDC) protocol is proposed for WSNs. Selecting the cluster heads in the proposed EEDC is performed based on a hybrid of residual energy and the distances among the cluster-heads. At first, the nodes with the most residual energy are elected and form an initial set of cluster-head candidates. Then the candidates with a suitable distance to other neighbour candidates are elected as the cluster-heads. The proposed algorithm is fast with a low time complexity. The proposed EEDC offers a long lifetime for the network, and at the same time, a proper level of fault tolerance. Different simulation experiments are done on different states and the algorithm is compared to some well-known clustering approaches. The experiments suggest that, in terms of longevity, the EEDC presents better performance than the existing protocols.

Tissue characterization of plaque in coronary arteries by using histogram-based frequency spectrum in window is proposed. Radio frequency (RF) signals, observed by the intravascular ultrasound catheter rotating in the coronary artery, are used for the tissue characterization. The conventional methods only use the frequency spectrum at the point of tissue of concern. However, in the proposed method the 2D histogram, concerning frequency and spectral band intensity, created from the window matrix of RF signals, is employed. The accuracy of the tissue characterization has been improved compared with the conventional methods which only use the statistical information of the frequency spectrum.

In this paper, a hybrid system is presented in which a convolutional neural network (CNN) and a Logistic regression classifier (LRC) are combined. A CNN is trained to detect and recognize face images, and a LRC is used to classify the features learned by the convolutional network. Applying feature extraction using CNN to normalized data causes the system to cope with faces subject to pose and lighting variations. LRC which is a discriminative classifier is used to classify the extracted features of face images. Discriminant analysis is more efficient when the normality assumptions are satisfied. The comprehensive experiments completed on Yale face database shows improved classification rates in smaller amount of time.

Network security is a growing issue, with the evolution of computer systems and expansion of attacks. Biological systems have been inspiring scientists and designs for new adaptive solutions, such as genetic algorithms. In this paper, an approach that uses the genetic algorithm to generate anomaly network intrusion detectors is used. An algorithm is proposed using a discretization method for the continuous features selection of intrusion detection, to create some homogeneity between values, which have different data types. Then, the intrusion detection system is tested against the NSL-KDD data set using different distance methods. A comparison is held amongst the results, and it is shown by the end that this proposed approach has good results, and recommendations are given for future experiments.

In practical data mining and process monitoring problems high-dimensional data has to be analyzed. In most of the cases it is very informative to map and visualize the hidden structure of complex data in a low-dimensional space. Industrial applications require easily implementable, interpretable and accurate projection. Nonlinear functions (aggregates) are useful for this purpose. A pair of these functions realise feature selection and transformation but finding the proper model structure is a complex nonlinear optimisation problem. We present a Genetic Programming (GP) based algorithm to generate aggregates represented in a tree structure. Results show that the developed tool can be effectively used to build an on-line spectroscopy based process monitoring system; the two-dimensional mapping of high dimensional spectral database can represent different operating ranges of the process.

In this work, Radial Basis (Exact Fit) and Radial Basis (Fewer Neurons) artificial neural network (ANN) models were developed to evaluate its capability in predicting the solubility index of roller dried goat whole milk powder. The ANN models were trained with a data file composed of variables: loose bulk density, packed bulk density, wettability and dispersibility, while solubility index was the output variable. The modeling results showed that there is an agreement between the experimental data and the predicted values, with coefficient of determination and Nash-Sutcliffe coefficient close to 1. Therefore, this method may be effective for rapid estimation of solubility index of roller dried goat whole milk powder.

This paper presents a computer model for online prediction of the wear contour of grooved rolls in the round-oval-round pass rolling process based on semi-analytical equations and artificial neural networks (ANN). This wear may adversely affect the shape quality of final product and is a result of complex interactions of many variables in the rolling process. The temperature of the material, amount of rolled material, water cooling system efficiency, diameters of the rolls, rolling speed and rolling load are some of these factors that play important role when assessing the wear of the rolls. A first ANN learns the average electrical current for thousands of hot rolled billets, and is done for ideal conditions, with new rolls. A second ANN calculates empirical coefficients in order to define the spread of the workpiece and then its contour is calculated accurately. This second ANN has inputs of differences on ideal and real electrical currents (and, thus, the rolling load variation) generated from the first ANN, temperature, water cooling pressure, speed of the rolls, diameters of the rolls, etc. Then the coefficients

$${\varvec{\gamma }}$$

and

$${\varvec{\kappa }}$$

(for wear profile) are calculated and input in semi-analytical equations to define the wear and its contour, as an online prediction. The works of Shinokura and Takai (1984) and Byon and Lee (2007) apply constant values for these two coefficients, limiting its application for other operational data variation during the rolling process. The model presented in this work uses an ANN to adapt

$${\varvec{\gamma }}$$

and

$${\varvec{\kappa }}$$

to cope with this variation. More than 50,000 billets were monitored and their operational data collected. The model was tested and the results agree well in real operational situations.

The growing use of cloud services, increased number of users, novel mobile operating systems and changes in network infrastructures that connect devices create novel challenges for cyber security. In order to counter arising threats, network security mechanisms and protection schemes also evolve and use sophisticated sensors and methods. The drawback is that the more sensors (probes) are applied and the more information they acquire, the volume of data to process grows significantly. In this paper, we present real-time network data analysis mechanism. We also show the results for SQL Injection Attacks detection.

Traditional recommendation techniques often rely on the user-item rating matrix, which explicitly represents a user’s preference among items. Recent studies on recommendations in the scenario of social networks still largely follow this principle. However, the challenge of recommending people to follow in social networks has yet to be studied thoroughly. In this paper, by using the utility instead of ratings and randomly sampling the negative cases in the recommendation log to create a balanced training dataset, we apply the popular matrix factorization techniques to predict whether a user will follow the person recommended or not. The asymmetric factor models are built with an extended item set incorporating the social graph information, which greatly improves the prediction accuracy. Other factors such as sequential patterns, CTR bias, and temporal dynamics are also exploited, which produce promising results on Task 1 of KDD Cup 2012.

This chapter describes the fuzzy logic approach to modelling of trip generation and trip distribution on country and country-pair levels. Different economic (GDP per capita of origin country, imports by destination countries) and social factors, as well as other ones (number of emigrants in destination country and destination country attractiveness) are considered. The case study of Serbia, illustrating possibilities of models, is given. Results of this research provide empirical evidence relating to successful use of fuzzy logic as a non-traditional technique.

Network design is one of the crucial activity in transportation engineering whose goal is to determine an optimal solution to traffic network layout with respect to given objectives and technical and/or economic constraints. In most of the practical problem the input data are not always precisely known as well as the information is not available regarding certain input parameters that are part of a mathematical model. Also constraints can be stated in approximate or ambiguous way. Thus, starting data and/or the problem constraints can be affected by uncertainty. Uncertain values can be represented using of fuzzy values/constraints and then handled in the framework of fuzzy optimization theory. In this paper we present a fuzzy linear programming method to solve the optimal signal timing problem on congested urban. The problem is formulated as a fixed point optimization subject to fuzzy constraints. The method has been applied to a test network for the case of priority corridors that are used for improve transit and emergency services. A deep sensitivity analysis of the signal setting parameters is then provided. The method is compared to classical linear programming approach with crisp constraints.

Modelling route choices is one of the most significant tasks in transportation models. Route choice models under Advanced Traveller Information Systems (ATIS) are often developed and calibrated by using, among other, Stated Preferences (SP) surveys. Different types of SP approaches can be adopted, alternatively based on Travel Simulators (TSs) or Driving Simulators (DSs). Here a pilot study is presented, aimed at setting up an SP-tool based on driving simulator developed at the Technical University of Bari. The obtained results are analysed in order to check the accordance with expectations in particular the results of application of data fusion technique are shown in order to explain how data collected by DSs, can be used to reduce the effect of choice of behaviour in unrealistic scenarios in TSs.

The widely debated concepts of sustainability and sustainable development represent nowadays an essential aspect in transportation studies, in particular for the analyses of interactions between transportation and land-use systems. In this paper the three-dimensional concept of sustainability (social, economic and environmental sustainability) is formalized by a Fuzzy-Based Evaluation Method, which has already been applied for evaluating the sustainability of alternative transportation policies. The method is tested as a tool to interpret the preferences expressed by the decision makers, to identify the most important characteristics of alternative transportation policies and to support the design of hypothetical transportation services, following a “

What to

” analysis.

This study proposed Artificial Bee Colony (ABC) algorithm for finding optimal setting of traffic signals in coordinated signalized networks for given fixed set of link flows. For optimizing traffic signal timings in coordinated signalized networks, ABC with TRANSYT-7F (ABCTRANS) model is developed. The ABC algorithm is a new population-based metaheuristic approach, and it is inspired by the foraging behavior of honeybee swarm. TRANSYT-7F traffic model is used to estimate total network performance index (PI). The ABCTRANS is tested on medium sized signalized road network. Results showed that the proposed model is slightly better in signal timing optimization in terms of final values of PI when it is compared with TRANSYT-7F in which Genetic Algorithm (GA) and Hill-climbing (HC) methods are exist. Results also showed that the ABCTRANS model improves the medium sized network’s PI by 2.4 and 2.7 % when it is compared with GA and HC methods.

Metering of merging traffic flows from on-ramp section of freeways is an important research issue for traffic engineers. Although metering signal is one of the recent applications for the subject, assignment of signal timing is problematic. The problem is based on dynamic structure of traffic flows and uncertainties coming up from driver behaviors. Because of variations in car following behavior and perception-reaction times of drivers, uncertainties are occurred. To handle these uncertainties, fuzzy logic approach is preferred in this research. A

Fu

zzy

L

ogic

C

ontrol based Dual Lane

R

amp

Me

tering (FuLCRMe) Model is proposed. The model considers following parameters as inputs; arrival headways of mainline, queue length at ramp and red time of ramp. Decision about red signal timing is made using these parameters. Based on this decision the final red time is assigned. The FuLCRMe model is tested by a simulation developed in Microsoft Excel program considering different cases. Results of the comparisons show that the FuLCRMe model provides significant decrease in delays, queue length, cycle time, CO

$$_{2}$$

emission, fuel consumption, travel time and total cost.

The containers drayage problem studied here arise in International Standards Organization (ISO) container distribution and collecting processes, in regions which are oriented to container sea ports or inland terminals. Containers of different sizes, but mostly 20 ft, and 40 ft empty and loaded should be delivered to, or collected from the customers. Therefore, the problem studied here is closely related to the vehicle routing problem with the time windows where an optimal set of routes is obtained. Both delivery and pickup demands can be satisfied in a single route. The specificity of the containers drayage problem analyzed here lies in the fact that a truck may simultaneously carry one 40 ft, or two 20 ft containers, using an appropriate trailer type. This means that in one route there can be one, two, three or four nodes, which is equivalent to the problem of matching nodes in single routes. This paper presents the Variable Neighborhood Search (VNS) heuristic for solving the Containers Drayage Problem with Time Windows (CDPTW). The results from the VNS heuristic are compared with the two optimal MIP mathematical formulations that were introduced in our previous research papers.

Nowadays, the collection of separated solid waste for recycling is still an expensive process, specially when performed in large-scale. One main problem resides in fleet-management, since the currently applied strategies usually have low efficiency. The waste collection process can be modelled as a vehicle routing problem, in particular as a Team Orienteering Problem (TOP). In the TOP, a vehicle fleet is assigned to visit a set of customers, while executing optimized routes that maximize total profit and minimize resources needed. The objective of this work is to optimize the waste collection process while addressing the specific issues around fleet-management. This should be achieved by developing a software tool that implements a genetic algorithm to solve the TOP. We were able to accomplish the proposed task, as our computational tests have produced some challenging results in comparison to previous work around this subject of study. Specifically, our results attained 60% of the best known scores in a selection of 24 TOP benchmark instances, with an average error of 18.7 in the remaining instances. The usage of a genetic algorithm to solve the TOP proved to be an efficient method by outputting good results in an acceptable time.

Determination of bus stop locations and bus stop frequencies are important issues in public transportation planning. This study analyzes the relationships among demand, travel time, bus stop locations, frequency, fleet size and passenger capacity parameters and develops models for bus stop locations and bus service frequency using fuzzy linear programming and linear goal programming approaches. The models are microscopic and applied to determine the bus stop locations and bus service frequency in the city of Izmir, Turkey, where 26 bus routes pass through two stops in the center city. The fuzzy optimization model minimizes the passenger access time and in-vehicle travel time. The reduction of the values of the bus service frequency and time parameters derived by the two proposed models are validated by a cost function. Encouraging results are obtained.