![A fully connected feedforward neural network with one hidden layer 2.1 Network architectures There are two general types of neural network architectures: feedforward and recurrent. A feedforward network is organized in the form of layers. It consists of the input layer, the hidden layers and the output layer. The input pattern is transmitted from the input layer to the next layer, the rst hidden layer. After that, the output signal of each layer is used as the input signal to the next layer, until the output signal of the output (last) layer is computed. The values of the output nodes constitute the overall response of the network to the input pattern applied. The number of hidden layers may be greater than or equal to zero. A network with no hidden layer is called single layer feedforward network. A feedforward network can be fully or partially connected depending on the existence or not of all the connections between the nodes of each layer and the forward adjacent layer. An example of a fully connected feedforward neural network with one hidden layer is shown in Fig. 2. An example of feedforward network is the multilayer perceptron (MLP), where the input signal propagates through the network in a feedforward direction, in a layer-by-layer fashion. The other general class of network architectures are recurrent architectures. A recurrent network diiers from the feedforward type in that it has one or more feedback loops, in the sense that each neuron feeds its output signal back to the inputs of all the other neurons. The most commonly used recurrent neural networks are the Hoppeld network and the Boltzmann machine. The Hoppeld network 43] is a recurrent network operating as a nonlinear associative memory, where a pattern stored in memory is retrieved given an incomplete or noisy form of that pattern. The Hoppeld neural network may be discrete (discrete input-discrete output) or analog (analog input-analog output). The discrete Hoppeld network performs](https://www.researchgate.net/profile/Konstantinos-Blekas/publication/2581388/figure/fig1/AS:669548076355588@1536644109323/A-fully-connected-feedforward-neural-network-with-one-hidden-layer-21-Network_Q320.jpg)
Content uploaded by Konstantinos Blekas
Author content
All content in this area was uploaded by Konstantinos Blekas on Jan 29, 2013
Content may be subject to copyright.
Hybrid Soft Computing Systems: A Critical Survey with Engineering Applications
Abstract and Figures
During the last decade the human behaviour and human imitating processing methods have become of central interest through the scientific community. The development of methods that mimic the human learning process being able to solve complex engineering problems which are difficult to deal with via conventional approaches, seems to be on an immediate emergency. Concepts such as nervous system, fuzziness and evolution come directly from human resources enclosing attractive properties and reach theory, and as a consequence lead to new scientific horizons. In this direction, soft computing indicates a new family of computing techniques that accommodate human computing resources and make them being utilized. Neural networks, fuzzy systems and genetic algorithms are mainly the three basic constituents that contribute to this juncture. Starting with the basic features in each one of these partners, this paper is focused on the examination of all the possible combined (hybrid) meth...
Figures - uploaded by Konstantinos Blekas
Author content
All figure content in this area was uploaded by Konstantinos Blekas
Content may be subject to copyright.
... By combining them, we can build Artificial neural systems can be considered as a massively parallel distributed model that has a natural propensity for storing experimental knowledge and making it available for use. They represent mathematical models of brain-like systems where knowledge is received through a learning process [83]. ...
... Fuzzy logic (FL) was developed by Zadeh in the mid 1960s to provide a mathematical basis for human thinking [80]. It uses fuzzy set theory, in which a variable is a member of one or more sets, with a specified degree of membership [83]. ...
... Several types of membership function can be used. The form of membership function is dependent on the structure of the corresponding fuzzy set [83]. In general, the shape of membership function depends on the application and can be monotonous, trapezoidal, triangular, Gaussian, etc., as shown in Figure ( and there are many ways in which the fuzzy implication may be defined [38]. ...
Transitional Bladder cell carcinoma is one of the most common
cancers of the urinary tract, accounting for approximately 90% of Bladder
cancers.
In this research a new computer-based system "Design Hybrid
Intelligent System for Transitional Bladder Cell Carcinoma Diagnosis"
(DHSTCCD) has been developed and implemented for the diagnosis of
transitional bladder cell carcinoma.
The proposed system is composed of two main phases, the first phase
is the "cell analysis phase " which consists of three main stages and the
second phase is the " patient data analysis phase ".
In the first stage of the cell analysis phase, a method named
"Genetic Optimization Based Fuzzy Image Segmentation Algorithm"
(GOFISA) has been proposed. This method has been used for the
segmentation of the nuclei of the selected cells images using hybrid
genetic algorithm and fuzzy system. These segmented nuclei are used in
the subsequent stage of features extraction.
In the second stage, for each segmented cell nucleus, morphometric
and photometric features are extracted from the segmented cell image.
In the third stage, these features were used as input to the
" Neuro-Fuzzy Classifier Model " (NFCM) that has been developed and
implemented to classify a set of normal and abnormal cells using hybrid
intelligent technique that combines the artificial neural network and fuzzy
logic to form NFCM.
The NFCM is inherently a fuzzy inference system with a capability
of learning fuzzy rules from data. The learning strategy consists of two
phases, a self-organizing clustering to establish the structure of the
network as well as the initial values of its parameters and a supervised
learning phase for optimal adjustment of these parameters.
In the patient data analysis phase of this work, a rule-based Fuzzy
Expert System has been proposed and implemented that uses the
laboratory and clinical data and simulates an expert doctor’s behavior.
The final diagnosis of the patients is determined from the results of the
fuzzy expert system and the NFCM.
The DHSTCCD has been tested on 80 microscope slides obtained
from 80 patients, where 100 images cells of the specimens under
microscopic examination have been extracted and analyzed. The system
has achieved diagnosis with accuracy rate (sensitivity) of the cell
classification of 88% .
The fact that this system would be a quick, less invasive and easy test,
assisting physician diagnosis has been taken into consideration.
سرطان الأنسجة الانتقالية للمثانة هو أحد اكثر أو ا رم المثانة شيوعا ويشكل ما يقرب ٩٠
من أو ا رم المثانة السرطانية.
يشمل هذا البحث تصميم وتنفيذ نظام حاسوبي جديد يتضمن بناء نظام ذكائي هجين
يتكون النظام المقترح من ."DHSTCCD" لتشخيص سرطان الأنسجة الانتقالية للمثانة سمي ب
والطور الثاني "cell analysis phase " طورين رئيسين: الطور الأول هو طور تحليل الخلايا
. "patient data analysis phase " هو طور تحليل المعلومات المتعلقة بالمريض
يتكون الطور الأول من ثلاث م ا رحل أساسية ، تشمل المرحلة الأولى بناء خوارزمية
لاستقطاع الخلايا باستخدام امثلية الخوارزميات الجينية للبيانات الصورية المضببة سميت ب
حيث يتضمن البرنامج تقطيع الخلايا المطلوبة وعزلها عن باقي خلايا الشريحة ، "GOFISA"
ومن ثم استقطاع انويه الخلايا المعزولة لغرض الاستفادة منها لاستخلاص المعلومات في
المرحلة التالية.
تتضمن المرحلة الثانية استخلاص المعلومات الصورية الضوئية والمعلومات التشكلية
والحجمية لكل من الانوية المستقطعة والتي تستخدم كمعلومات مغذية للشبكة العصبية المضببة.
أما المرحلة الثالثة فتتضمن بناء الشبكة العصبية المضببة عن طريق استخدام تقنيات
الشبكة العصبية و تهجينها مع تقنية المنطق المضبب للحصول على الشبكة المسماة
حيث تقوم الشبكة العصبية المضببة ." Neuro-Fuzzy Classifier Model (NFCM)"
بتصنيف الخلايا إلى نوعين، خلايا طبيعية وخلايا سرطانية اعتمادا على المعلومات المستخلصة
من صور انوية الخلايا المستقطعة.
إن نموذج الشبكة العصبية المضببة يقوم بتحديد وتدريب القواعد المضببة بالاعتماد على
البيانات التي تبث للشبكة، ان عملية تحديد عدد القواعد المضببة و تدريبها يتكون من طورين.
بناء هيكلية " Self-Organizing phase" حيث يتم في الطور الأول وهو طور التدريب الذاتي
الشبكة وتحديد القيم الأولية لعوامل الشبكة، أما في الطور الثاني وهو طور التدريب الموجه "
فيتم الحصول على القيم الامثلية لعوامل الشبكة. "Supervised phase
يتم في الطور الثاني من هذا النظام اقت ا رح و تنفيذ نظام خبير مضبب يستخدم المعلومات
المختبرية و السريرية لمحاكاة الأسلوب الطبي المتبع ويتم الحصول على التشخيص النهائي
اعتمادا على النتائج المستخلصة من هذا النظام و كذلك من نتائج الشبكة العصبية المضببة.
تم اختبار هذا النظام من خلال ٨٠ شريحة مجه رية أخذت من عينات ٨٠ مريض حيث تم
استخلاص و تحليل ١٠٠ صورة من الخلايا المأخوذة من العينات تحت الفحص المجهري، وقد
. % حقق النظام تشخيصا بدقة تصنيفية للخلايا بلغت ٨٨
يتميز هذا النظام بكونه سريع الإنجاز و سهل الاستخدام حيث يوفر المساعدة الطبية في
التشخيص
... Pour un problème complexe, la représentation indirecte est souvent utilisée avec une procédure de décodage pour convertir une représentation de solution indirecte en une solution réalisable. Une fois la solution décodée, safitness peut être évaluée.La population initiale peut être soit un échantillon aléatoire de l'espace des solutionspossiblessoit un ensemble construit avec des individus trouvés par des procédures de recherche locales simples, si celles-ci sont disponibles[10].En plus de la représentation de la solution, deux paramètres communs qui doivent être déterminés initialement sont la taille de la population et le nombre maximum d'itérations. Les choix de ces deux paramètres ont une influence majeure sur la qualité de lasolution et le temps d'exécution, et en pratique, ces valeurs sont presque toujours déterminées de manière empirique par des essais. ...
... Apprentissage paramétrique : effectué pour ajuster les paramètres antécédents et conséquents afin qu'une fonction objectif spécifiée soit minimisée. Il existe plusieurs méthodes d'apprentissage[10] pour mettre à jour ces paramètres, l'apprentissage hybride qui combine l'algorithme de rétro-propagation du gradient GD et la méthode des moindres carrés LSE est l'algorithme utilisé dans cette thèse.Cet algorithme est réalisé en deux étapes :-La passe avant : où les modèles d'entrée sont propagés vers la sortie et les paramètres conséquents optimaux sont estimés par le LSE, tandis que les paramètres ...
This thesis presents an evolutionary approach for the learning of the Adaptive Network Based Fuzzy Inference System (ANFIS). The previous works are based on the descent of the gradient (GD), this algorithm converges very slowly and stuck into bad local minima. In this thesis, we apply genetic algorithms (GA) and particle swarms (PSO) to optimize the antecedents and consequent parameters of ANFIS fuzzy rules. First, the subtractive clustering algorithm was used to determine the optimal structure of the ANFIS network, i.e., the best partitioning of the input space; then, adjusting the antecedent and consequent parameters of the fuzzy rules so that a specified objective function is minimized. The evolutionary process begins by randomly generating an initial population, each candidate solution is represented by a vector. The length of the latter is based on the number of antecedent and consequent parameters in the ANFIS model. Then, the entire population was made to improve gradually until the maximum number of iterations was reached. The proposed approach was applied for the recognition of phonemes
from TIMIT database and speaker recognition from the CHAINS database. The results obtained by the hybrid models AG-ANFIS and PSO-ANFIS showed an improvement in precision compared to a similar classic ANFIS based on the back-propagation of the gradient.
... This provides a completely different and unorthodox way to approach a modeling problem. Through FL is possible to find a simple, cheap and quick solution for the practical problem; instead of building a complex mathematical model (Tzafestas & Blekas, 1999). The method is easy to apply, and its results are often effective and useful. ...
The automotive industry uses Reliability Centered Maintenance (RCM) to establish efficient and cost-effective maintenance schedules of equipment (IATF 16949). The RCM methodology is based on equipment reliability which requires collection of failure times. However, in some cases it is not possible to record enough information. Sometimes in the field, only service times are available (these can be classified as “only censored data”). The most commonly found situation waiting until the component fails. In that sense, the RCM methodology mainly uses the classic Weibull model for estimating reliability indexes, which cannot be performed if there are only censored data. The maximum entropy can be used to propose an alternative reliability model that is able to work under the limited information conditions. Nevertheless, the uncertainty of only censored data and the limited information also affect the maximum entropy model. The information available, such as maintenance staff experience, must help provide accuracy to reliability estimation, which can be addressed through fuzzy numbers. Hence, a fuzzy maximum entropy approach is proposed to determine the maximum entropy reliability function considering the uncertainty and the maintenance staff knowledge; therefore, the replacement frequency. Finally, the proposed method was applied to an injection molding machine. Outcomes proved that the alternative model is a useful and reliable option if there are only censored data and personnel experience as available information. The RCM maintenance plan was scheduled using the proposed methodology and its results showed 40% savings in the annual maintenance costs, which represent an important benefit for company.
... Hybrid Soft Computing approaches incorporates all the features from individual fields and, moreover, has the ability to overcome difficulties and limitations that characterize each field. The use of intelligent hybrid systems is growing rapidly with successful applications in many areas including process control, robotics, manufacturing, medical diagnosis, etc. [8]. ...
Transitional Bladder cell carcinoma is one of the most common cancers of the urinary tract, accounting for approximately 90% of Bladder cancers. In this research a new computer-based system "Design of a Hybrid Intelligent System for Transitional Bladder Cell Carcinoma Diagnosis" (DHSTCCD) has been proposed and implemented. The proposed system is composed of two main phases, the first phase is the "cell analysis phase" which consists of three main stages including segmentation stage using "Genetic Optimization Based Fuzzy Image Segmentation Algorithm" (GOFISA), morphometric and photometric feature extraction stage and "Neuro-Fuzzy Classifier Model" (NFCM) stage that has been developed and implemented to classify a set of normal and abnormal cells using hybrid intelligent technique that combines the artificial neural network and fuzzy logic. The second phase is the "patient data analysis phase", in this phase a rule-based Fuzzy Expert System has been proposed and implemented that uses the laboratory and clinical data and simulates an expert doctor's behavior. The final diagnosis of the patients is determined from the results of the fuzzy expert system and the NFCM. ﺟﺎﻣﻌ ﻳﺎﺿﻳﺎت اﻟر و اﻟﺣﺎﺳوب ﻋﻠوم ﻛﻠﻳﺔ اﻟﻣوﺻﻝ/ ﺔ ﻗﺳم اﻟﺗﻌﻠﻳﻣﻲ اﻟﺟﻣﻬوري ﻣﺳﺗﺷﻔﻰ اﻟﻌﺎﻣﺔ/ اﺣﺔ اﻟﺟر ﺗ ﺎ اﻟﺒﺤﺚ اﺳﺘﻼم ﻳﺦ ر : ٢١ / ٩ / ٢٠٠٨ ﺗ ﺎ اﻟﺒﺤﺚ ﻗﺒﻮل ﻳﺦ ر : ٤ / ١٢ / ٢٠٠٨ اﻟﻣﻠﺧص ﻳﻘرب ﻣﺎ وﻳﺷﻛﻝ ﺷﻳوﻋﺎ اﻟﻣﺛﺎﻧﺔ ام أور اﻛﺛر أﺣد ﻫو ﻟﻠﻣﺛﺎﻧﺔ اﻻﻧﺗﻘﺎﻟﻳﺔ اﻷﻧﺳﺟﺔ ﺳرطﺎن 90 اﻟﻣﺛﺎﻧﺔ ام أور ﻣن % اﻟﺳ اﻷﻧﺳﺟﺔ ﺳرطﺎن ﻟﺗﺷﺧﻳص ﻫﺟﻳن ذﻛﺎﺋﻲ ﻧظﺎم ﺑﻧﺎء ﻳﺗﺿﻣن ﺟدﻳد ﺣﺎﺳوﺑﻲ ﻧظﺎم وﺗﻧﻔﻳذ ﺗﺻﻣﻳم اﻟﺑﺣث ﻫذا ﻳﺷﻣﻝ رطﺎﻧﻳﺔ. " ـ ﺑ ﺳﻣﻲ ﻟﻠﻣﺛﺎﻧﺔ اﻻﻧﺗﻘﺎﻟﻳﺔ DHSTCCD اﻟﺧﻼﻳﺎ ﺗﺣﻠﻳﻝ طور ﻫو اﻷوﻝ اﻟطور ﺋﻳﺳﻳن: ر ﻳن طور ﻣن اﻟﻣﻘﺗرح اﻟﻧظﺎم ﻳﺗﻛون ." " cell analysis phase ﺗﺣﻠﻳ طور ﻫو اﻟﺛﺎﻧﻲ اﻟطور و " ﻳض" ﺑﺎﻟﻣر اﻟﻣﺗﻌﻠﻘﺔ اﻟﻣﻌﻠوﻣﺎت ﻝ patient data analysis phase. " أﺳﺎﺳﻳﺔ اﺣﻝ ﻣر ﺛﻼث ﻣن اﻷوﻝ اﻟطور ﻳﺗﻛون ، اﻣﺛﻠﻳﺔ ﺑﺎﺳﺗﺧدام اﻟﺧﻼﻳﺎ ﻻﺳﺗﻘطﺎع ارزﻣﻳﺔ ﺧو ﺑﻧﺎء اﻷوﻟﻰ اﻟﻣرﺣﻠﺔ ﺗﺷﻣﻝ ﺳﻣﻳت اﻟﻣﺿﺑﺑﺔ ﻳﺔ اﻟﺻور ﻟﻠﺑﻳﺎﻧﺎت اﻟﺟﻳﻧﻳﺔ ارزﻣﻳﺎت اﻟﺧو " ـ ﺑ GOFISA ﺗﻘ ﻧﺎﻣﺞ اﻟﺑر ﻳﺗﺿﻣن ﺣﻳث ، " اﻟﻣطﻠوﺑﺔ اﻟﺧﻼﻳﺎ طﻳﻊ ﻓﻲ اﻟﻣﻌﻠوﻣﺎت ﻻﺳﺗﺧﻼص ﻣﻧﻬﺎ اﻻﺳﺗﻔﺎدة ﻟﻐرض اﻟﻣﻌزوﻟﺔ اﻟﺧﻼﻳﺎ اﻧوﻳﻪ اﺳﺗﻘطﺎع ﺛم وﻣن ﻳﺣﺔ اﻟﺷر ﺧﻼﻳﺎ ﺑﺎﻗﻲ ﻋن ﻟﻬﺎ وﻋز ﻣن ﻟﻛﻝ اﻟﺣﺟﻣﻳﺔ و اﻟﺗﺷﻛﻠﻳﺔ اﻟﻣﻌﻠوﻣﺎت و اﻟﺿوﺋﻳﺔ ﻳﺔ اﻟﺻور اﻟﻣﻌﻠوﻣﺎت اﺳﺗﺧﻼص اﻟﺛﺎﻧﻳﺔ اﻟﻣرﺣﻠﺔ ﺗﺗﺿﻣن اﻟﺗﺎﻟﻳﺔ. اﻟﻣرﺣﻠﺔ اﻟﻣﺳﺗﻘطﻌ اﻻﻧوﻳﺔ اﻟﺷﺑﻛﺔ ﺑﻧﺎء ﻓﺗﺗﺿﻣن اﻟﺛﺎﻟﺛﺔ اﻟﻣرﺣﻠﺔ أﻣﺎ اﻟﻣﺿﺑﺑﺔ. اﻟﻌﺻﺑﻳﺔ ﻟﻠﺷﺑﻛﺔ ﻣﻐذﻳﺔ ﻛﻣﻌﻠوﻣﺎت ﺗﺳﺗﺧدم اﻟﺗﻲ و ﺔ اﻟﺷﺑﻛﺔ ﻋﻠﻰ ﻟﻠﺣﺻوﻝ اﻟﻣﺿﺑب اﻟﻣﻧطق ﺗﻘﻧﻳﺔ ﻣﻊ ﺗﻬﺟﻳﻧﻬﺎ و اﻟﻌﺻﺑﻳﺔ اﻟﺷﺑﻛﺔ ﺗﻘﻧﻳﺎت اﺳﺗﺧدام ﻳق طر ﻋن اﻟﻣﺿﺑﺑﺔ اﻟﻌﺻﺑﻳﺔ Nada N. Saleem and Khalid N. Saleem 28 اﻟﻣﺳﻣﺎة Neuro-Fuzzy Classifier Model (NFCM) اﻟﺷ ﺗﻘوم ﺣﻳث. إﻟﻰ اﻟﺧﻼﻳﺎ ﺑﺗﺻﻧﻳف اﻟﻣﺿﺑﺑﺔ اﻟﻌﺻﺑﻳﺔ ﺑﻛﺔ اﻟﻣﺳﺗﻘطﻌﺔ. اﻟﺧﻼﻳﺎ اﻧوﻳﺔ ﺻور ﻣن اﻟﻣﺳﺗﺧﻠﺻﺔ اﻟﻣﻌﻠوﻣﺎت ﻋﻠﻰ اﻋﺗﻣﺎدا ﺳرطﺎﻧﻳﺔ وﺧﻼﻳﺎ طﺑﻳﻌﻳﺔ ﺧﻼﻳﺎ ﻧوﻋﻳن، ﺧﻼﻝ ﻣن اﻟﻧظﺎم ﻫذا اﺧﺗﺑﺎر ﺗم 80 ﻋﻳﻧﺎت ﻣن أﺧذت ﻳﺔ ﻣﺟﻬر ﻳﺣﺔ ﺷر 80 ﺗﺣﻠﻳﻝ و اﺳﺗﺧﻼص ﺗم ﺣﻳث ﻳض ﻣر 100 ة ﺻور اﻟﻣﺄﺧوذة اﻟﺧﻼﻳﺎ ﻣن ﺑﻠﻐت ﻟﻠﺧﻼﻳﺎ ﺗﺻﻧﻳﻔﻳﺔ ﺑدﻗﺔ ﺗﺷﺧﻳﺻﺎ اﻟﻧظﺎم ﺣﻘق وﻗد اﻟﻣﺟﻬري، اﻟﻔﺣص ﺗﺣت اﻟﻌﻳﻧﺎت ﻣن 88. % اﻟﺗﺷﺧﻳص. ﻓﻲ اﻟطﺑﻳﺔ اﻟﻣﺳﺎﻋدة ﻳوﻓر ﺣﻳث اﻻﺳﺗﺧدام ﺳﻬﻝ و اﻹﻧﺟﺎز ﻳﻊ ﺳر ﺑﻛوﻧﻪ اﻟﻧظﺎم ﻫذا ﻳﺗﻣﻳز اﻟﻣﻔﺗﺎﺣﻳﺔ: اﻟﻛﻠﻣﺎت ﻟﻠﻣﺛﺎﻧﺔ اﻻﻧﺗﻘﺎﻟﻳﺔ اﻷﻧﺳﺟﺔ ﺳرطﺎن ، ﻫﺟﻳن ذﻛﺎﺋﻲ ﻧظﺎم ، اﻟﻌ اﻟﺷﺑﻛﺔ اﻟﻣﺿﺑﺑﺔ ﺻﺑﻳﺔ اﻻﻧظﻣﺔ ، اﻟﻣﺿﺑﺑﺔ ة اﻟﺧﺑﻳر
... Hybrid Soft Computing approaches incorporates all the features from individual fields and, moreover, has the ability to overcome difficulties and limitations that characterize each field. The use of intelligent hybrid systems is growing rapidly with successful applications in many areas including process control, robotics, manufacturing, medical diagnosis, etc. [8]. ...
Transitional Bladder cell carcinoma is one of the most common cancers of the urinary tract, accounting for approximately 90% of Bladder cancers. In this research a new computer-based system "Design of a Hybrid Intelligent System for Transitional Bladder Cell Carcinoma Diagnosis" (DHSTCCD) has been proposed and implemented. The proposed system is composed of two main phases, the first phase is the "cell analysis phase" which consists of three main stages including segmentation stage using "Genetic Optimization Based Fuzzy Image Segmentation Algorithm" (GOFISA), morphometric and photometric feature extraction stage and "Neuro-Fuzzy Classifier Model" (NFCM) stage that has been developed and implemented to classify a set of normal and abnormal cells using hybrid intelligent technique that combines the artificial neural network and fuzzy logic.
The second phase is the "patient data analysis phase", in this phase a rule-based Fuzzy Expert System has been proposed and implemented that uses the laboratory and clinical data and simulates an expert doctor’s behavior. The final diagnosis of the patients is determined from the results of the fuzzy expert system and the NFCM.
... Atualmente, o uso das técnicas da inteligência computacional em conjunto, formando sistemas híbridos,é uma opção cada vez mais utilizada (Craenen & Eiben, 2002) (Tzafestas & Blekas, 2001). A interligação das várias técnicas e, mais importante, a maneira com que os pontos fortes de cada técnica são utilizados em conjunto, representa uma tendência da pesquisa naárea, e existem fortes expectativas com relação aos avanços que resultarão desta sinergia (Goldberg, 2000). ...
This work proposes the development of a neuro-fuzzy predictive control. A recur-
rent neural network is used to identify the process, providing predictions about its
behavior, based on control actions applied to the system. This information is used
by the fuzzy controller, in order to accomplish a better control of a process. Genetic
algorithms (GA’s) are used to automatically tune the controller.
The neural modelling is done through the Cascade Correlation algorithm. The
algorithm is modified to incrementally find the proper number of delayed inputs and
external recurrent connections that allow the neural modelling.
The combined use of these techniques make it possible to control multi-variable
processes using several fuzzy controllers, that have their membership functions evol-
ved by genetic algorithms. The coupling among controlled variables are modelled by
neural networks, and conflictive objectives can be inserted in the fitness function of
the GA.
The methodology was applied to three non-linear processes. These processes
were chosen due to their characteristics, such as inertia, large accommodation time,
and conflicting control objectives, that make these processes difficult to control with
traditional methods. The processes were: the steering control of a cargo ship, the
concentration control of a fermentation process, and the start-up of a continuous
distillation column.
... There are many several approaches proposed to solve the robot navigation problem including the model-based methods of artificial intelligence; the fuzzy logic, neural network and genetic algorithms [2][3] [4]. Soft-computing techniques have provided a powerful tool to deal with mobile robot navigation process, which exhibits incomplete and uncertain knowledge due to the inaccuracy and imprecision inherent from the sensory system [5] [6]. Fuzzy logic based decision-making [7] and neural networks [3] have been found to be the most attractive techniques that can be used for this purpose due to their properties. ...
This paper describes the use of soft computing approaches to the robot behavior design. In many applications, the robot's environment changes with time in a way that is not predictable by the designer in advance. In addition, the knowledge about the environment is often imprecise and incomplete due to the limited perceptual quality of sensors. In order to equip the robot by capacity of autonomy and intelligence in its environment, the control system must perform much complex information and processing tasks in real time, and it is well suited to use the soft-computing techniques. The objective of this paper is to elaborate and compare simple intelligent control systems for the path following behavior by an autonomous mobile robot using the most known approaches of the artificial intelligence science: fuzzy, neuro and neuro-fuzzy controllers. The proposed controllers are used for pursuing a moving target. The obtained simulation results show the effectiveness of the designed controllers. The results are discussed and compared.
... One can concentrate on solving the problem rather than trying to model the system mathematically, if that is even possible. This almost invariably leads to quicker, cheaper solutions [39]. Once understood, this technology is not difficult to apply and the results are usually quite surprising and pleasing. ...
Oftentimes, material components follow non-linear degradation because the material loss trough time is non-linear; the components present different levels of resistance during a degradation process. Constructing predictions of the component life span and performing reliability analyses regarding degradation is possible by non-linear regression. Generally, this method ignores the prediction error in the extrapolated time and it does not account measurement error in the observed sample paths. In this paper, it is proposed new means for predicting time to failure of the components, using a calibration regression method for measuring the error prediction in the extrapolation process. In order to consider the uncertainty, we proposed to perform the estimated times by a trapezoidal fuzzy numbers, providing a more reliable prediction without loss of information. Since the times to failure are shaped by fuzzy numbers, it was adapted the fuzzy probability theory to the classical reliability analysis for estimating the fuzzy reliability of a component. An accelerated test of a polymeric coating process was performed and also an application for a preventive maintenance frequency was proposed.
Abstract:
This thesis deals with the autonomous navigation problem of a mobile robot using hybrid neuro-fuzzy techniques. The objective of the presented work is to study and develop effective architectures
for a reactive navigation of an autonomous mobile robot in unknown environment, by using on the one hand the behavior based approach, and on the other hand the learning paradigm. The techniques
employed to tackle this problem are based on the fuzzy inference systems, artificial neural networks and the reinforcement learning. Firstly, we used fuzzy behavior based navigation approach for the
reactive and local planning, then for tuning the fuzzy behaviors parameters; we introduced the hybrid neuro-fuzzy models for autonomous navigation. The second type of learning method is particularly adapted to mobile robotic, which makes it possible to find by a process of tests and errors, the executed optimal action for each situation which the robot will perceive in order to maximize its
rewards. To combine the advantages of fuzzy logic and reinforcement learning, a control strategy with a learning capacity is used, it is an extension of Q-learning to the continuous spaces and considered as
and optimization method of fuzzy systems. The advantage of the fuzzy systems is the introduction of a priori knowledge in order to make the first behavior is acceptable. The effectiveness of the proposed and the studied architectures are demonstrated by various applications of the autonomous mobile robot navigation.
Refractory cement is one of the elementary materials for the thermocouples manufacture. It is important pointing out that viscosity greatly affects its quality and functionality. In this sense, there is a viscosity range in which refractory cement must be applied; this range is known as “pot life.” For this reason, the cement setting process (viscosity behavior) should be modeled in order to predict its life (useful time). Like this, some operational factors must be considered, among them: temperature and humidity as well as the fact that pot life behavior is nonlinear and must be performed by a growth model. At the modeling process, it is necessary considering the uncertainty which is not considered in the properties of the bodies studied in the rheological models for real materials. This work proposes an inverse prediction method for measuring the prediction error time. Furthermore, it is suggested performing the estimated times by a triangular fuzzy number with the purpose of considering all uncertainty information and providing a reliable prediction. Therefore, the fuzzy theory to the Weibull analysis was adapted in order to estimate some faculties: the fuzzy reliability, the fuzzy useful time with a desired reliability and the fuzzy mean pot life. Results show accuracy and useful pot life predictions.
Fuzzy set and fuzzy logic theory was initiated by Zadeh [1] and permits the treatment of vague, imprecise and ill defined knowledge and concepts in an exact mathematical way. Throughout the years this theory was fully studied and used for the analysis, modeling and control of technological and nontechnological systems. Actually, our life and world obey the principle of compatibility of Zadeh, according to which “the closer one looks at a ‘real’ world problem, the fuzzier becomes its solution”. Stated informally, the essence of this principle is that, as the complexity of a system increases, our ability to make precise and yet significant statements about its behavior diminishes until a threshold beyond which precision and significance (relevance) become almost exclusive characteristics.
Most training algorithms for radial basis function (RBF) neural networks start with a predetermined network structure which is chosen either by using a PRORI knowledge or based on previous experience. The resulting network is often insufficient or unnecessarily complicated and an appropriate network structure can only be obtained by trial and error. Training algorithms which incorporate structure selection mechanisms are usually based on local search methods and often suffer from a high probability of being posed to automatically configure RBF networks. The network configuration is formed as a subsequent selection problem. The task is then to find an optimal subset of nc terms from the Nt training data samples. Each network is coded as a variable length string with distinct integers and genetic operators are proposed to evolve a population of individuals. Criteria including single-objective and multi-objective functions are proposed to evaluate the fitness of individual networks. Training based on practical data set is used to demonstrate the performance of the new algorithms.
Section 1 (S1) describes specifically the problems to be discussed in succeeding chapters. In S2 a short analysis of the modeling process suggests that information and uncertainty will be key concepts in the development of new mathematical structures for pattern recognition. Fuzzy sets are introduced in the third section as a natural and tractable way to model physical situations which seem to require something more than is offered by deterministic or stochastic representations.
