Application of Neuro-Fuzzy Estimation for Rotor Flux in Speed Vector Control of Bearingles
自适应神经模糊系统及其应用研究
自适应神经模糊系统及其应用研究人工智能技术的发展，为科学家们开辟了一片全新的研究领域。
神经网络、模糊控制等技术的不断发展带来了自适应神经模糊系统的出现。
自适应神经模糊系统，又称为ANFIS（Adaptive Neuro Fuzzy Inference System），是一种基于神经网络与模糊逻辑综合的自适应智能系统。
本文将从它的概念、结构及应用等几个方面进行探讨。
一、概念自适应神经模糊系统是一种结合神经网络和模糊控制的新型智能系统。
它能够利用神经网络来自动完成输入与输出间的映射，同时利用模糊控制来实现自适应和推理功能，从而实现对系统的智能化控制。
ANFIS的核心部分是模糊推理机，它通过“如果……那么”的形式进行推理，将输入的模糊信号通过规则的运算，转化为输出信号。
在推理的过程中，ANFIS通过神经网络进行学习，并根据学习的结果来优化推理机的结构和参数，从而提高其推理的精度与效率。
二、结构ANFIS的结构是由输入层、隐含层、输出层和反向传播算法组成。
其中，输入层是将系统的输入变量进行接受和处理的部分；隐含层是神经网络部分，它利用了Takagi-Sugeno-Kang（TSK）模型作为模糊推理的核心，并通过反向传播算法对其进行训练；输出层则是将隐含层的结果进行处理并转化为系统输出的部分。
此外，ANFIS还包括规则库、模糊化和去模糊化等部分，用来处理系统中的模糊数据，使系统具有推理、记忆和自适应等能力。
三、应用自适应神经模糊系统在工业控制、模式识别、信号处理等多个领域拥有广泛的应用。
其中，应用最为广泛的是控制领域。
ANFIS通过有效的模糊推理机制和自适应能力，可以实现对复杂系统的精准控制。
例如，在工业生产过程中，ANFIS可以通过学习数据的变化趋势，自动调节系统中各部分的运行状态，达到节省能源、提高产量等效果。
在车辆控制方面，ANFIS可以通过对车辆行驶数据分析，对车辆的驾驶状态进行自适应控制，从而达到提高驾驶安全性和车辆性能的效果。
neuro-fuzzy
Combining classifiers of pesticides toxicity through aneuro-fuzzy approachEmilio Benfenati1, Paolo Mazzatorta1, Daniel Neagu2, and Giuseppina Gini21 Istituto di Ricerche Farmacologiche "Mario Negri" Milano,Via Eritrea, 62, 20157 Milano, Italy{Benfenati, Mazzatorta}@ marionegri.it2 Dipartimento di Elettronica e Informazione, Politecnico di Milano,Piazza L. da Vinci 32, 20133 Milano, ItalyNeagu@fusberta.elet.polimi.it, Gini@elet.polimi.ithttp://airlab.elet.polimi.it/imagetoxAbstract. The increasing amount and complexity of data in toxicity predictioncalls for new approaches based on hybrid intelligent methods for mining thedata. This focus is required even more in the context of increasing number ofdifferent classifiers applied in toxicity prediction. Consequently, there exist aneed to develop tools to integrate various approaches. The goal of this researchis to apply neuro-fuzzy networks to provide an improvement in combining theresults of five classifiers applied in toxicity of pesticides. Nevertheless, fuzzyrules extracted from the trained developed networks can be used to performuseful comparisons between the performances of the involved classifiers. Ourresults suggest that the neuro-fuzzy approach of combining classifiers has thepotential to significantly improve common classification methods for the use intoxicity of pesticides characterization, and knowledge discovery.1 IntroductionQuantitative structure–activity relationships (QSARs) correlate chemical structure to a wide variety of physical, chemical, biological (including biomedical, toxicological, ecotoxicological) and technological (glass transition temperatures of polymers, critical micelle concentrations of surfactants, rubber vulcanization rates) properties. Suitable correlations, once established and validated, can be used to predict properties for compounds as yet unmeasured or even unknown.Classification systems for QSAR studies are quite usual for carcinogenicity [9], because in this case carcinogenicity classes are defined by regulatory bodies such as IARC and EPA. For ecotoxicity, most of the QSAR models are regressions, referring to the dose giving the toxic effect in 50% of the animals (for instance LC50: lethal concentration for 50% of the test animals). This dose is a continuous value and regression seems the most appropriate algorithm. However, classification affords some advantages. Indeed, i) the regulatory values are indicated as toxicity classes and ii) classification can allow a better management of noisy data. For this reason we investigated classification in the past [7], [8], [9] and also in this study. No generalrule exists to define an approach suitable to solve a specific classification problem. In several cases, a selection of descriptors is the only essential condition to develop a general system. The next step consists in defining the best computational method to develop robust structure–activity models.Artificial neural networks (ANNs) represent an excellent tool that have been used to develop a wide range of real-world applications, especially when traditional solving methods fail [3]. They exhibit advantages such as ideal learning ability from data, classification capabilities and generalization, computationally fastness once trained due to parallel processing, and noise tolerance. The major shortcoming of neural networks is represented by their low degree of human comprehensibility. More transparency is offered by fuzzy neural networks FNN [14], [16], [18], which represent a paradigm combining the comprehensibility and capabilities of fuzzy reasoning to handle uncertainty, and the capabilities to learn from examples.The paper is organized as follows. Section 2 briefly presents the aspects of data preparation, based on chemical descriptors, some of the most common classification techniques and shows how they behave for toxicology modeling, with a emphasis to pesticides task. Section 3 proposes the neuro-fuzzy approach in order to manage the integration of all the studied classifiers, based on the structure developed as FNN Implicit Knowledge Module (IKM) of the hybrid intelligent system NIKE (Neural explicit&Implicit Knowledge inference system [17]). Preliminary results indicate that combination of several classifiers may lead to the improved performance [5], [11], [12]. The extracted fuzzy rules give new insights about the applicability domain of the implied classifiers. Conclusions of the paper are summarized in the last section.2 Materials and Methods2.1 Data setFor this paper a data set constituted of 57 common organophosphorous compounds has been investigated. The main objective is to propose a good benchmark for the classification studies developed in this area. The toxicity values are the result of a wide bibliographic research mainly from “the Pesticide Manual”, ECOTOX database system, RTECS and HSDB [1]. An important problem that we faced is connected with the variability that the toxicity data presents [2]. Indeed, it is possible to find different fonts showing for the same compound and the same end–point LC50 different for about two orders of magnitude. Such variability is due to different factors, as the different individual reactions of organisms tested, the different laboratory procedures, or is due to different experimental conditions or accidental errors.The toxicity value was expressed using the form Log10 (1/LC50). Then the values were scaled in the interval [-1..1]. Four classes were defined: Class 1 [-1..-0.5), Class 2 [-0.5..0), Class 3 [0..0.5), Class 4 [0.5..1] (Table 2).2.2 DescriptorsA set of about 150 descriptors were calculated by different software: Hyperchem 5.01, CODESSA 2.2.12, Pallas 2.13. They are split into six categories: Constitutional (34 descriptors), Geometrical (14), Topological (38), Electrostatic (57), Quantum–chemicals (6), and Physico–chemical (4). In order to obtain a good model, a selection of the variables, which better describe the molecules, is necessary. There is the risk that some descriptors does not add information, and increase the noise, making more complex the result analysis. Furthermore, using a relatively low number of variables, the risk of overfitting is reduced. The descriptors selection (table 1) was obtained by Principal Components Analysis (PCA), using SCAN4:Table 1. Names of the chemical descriptors involved in the classification task.Cat. Cod.Moment of inertia A G D1Relative number of N atoms C D2Binding energy (Kcal/mol) Q D3DPSA-3 Difference in CPSAs (PPSA3-PNSA3) [Zefirov’s PC] E D4Max partial charge (Qmax) [Zefirov’s PC] E D5ZX Shadow / ZX Rectangle G D6Number of atoms C D7Moment of inertia C G D8PNSA-3 Atomic charge weighted PNSA [Zefirov’s PC] E D9HOMO (eV) E D10LUMO (eV) Q D11Kier&Hall index (order 3) T D122.3 Classification algorithmsThe classification algorithms used for this work are five: LDA (Linear Discriminant Analysis), RDA (Regularized Discriminant Analysis), SIMCA (Soft Independent Modeling of Class Analogy), KNN (K Nearest Neighbors classification), CART (Classification And Regression Tree). The first four are parametric statistical systems based on the Fisher’s discriminant analysis, the fifth and sixth are not parametrical statistical methods, the last one is a classification tree.LDA: the Fischer’s linear discrimination is an empirical method based on p–dimensional vectors of attributes. Thus the separation between classes occurs by an hyperplane, which divides the p–dimensional space of attributes.RDA: The variations introduced in this model have the aim to obviate the principal problems that afflict both the linear and quadratic discrimination. The regulation more efficient was carried out by Friedman, who proposed a compromise between the two previous techniques using a biparametrical method for the estimation (λ and γ).1 Hypercube Inc., Gainsville, Florida, USA2 SemiChem Inc., Shawnee, Kansas, USA3 CompuDrug; Budapest, Hungary4 SCAN (Software for Chemometric Analysis) v.1.1, from Minitab: SIMCA: the model is one of the first used in chemometry for modeling classes and, contrarily to the techniques before described, is not parametrical. The idea is to consider separately each class and to look for a representation using the principal components. An object is assigned to a class on the basis of the residual distance, rsd 2, that it has from the model which represent the class itself:()22ˆigj igj igj x xr −=, )(22j j igjM p r rsd −=∑ (1)where x igj = co –ordinates of the object’s projections on the inner space of the mathematical model for the class, x igj = object’s co –ordinates, p=number of variables, M j = number of the principal components significant for the j class.KNN: this technique classifies each record in a data set based on a combination of the classes of the k record(s) most similar to it in a historical data set (where k = 1). CART is a tree –shaped structure that represents sets of decisions. These decisions generate rules for the classification of a data set. CART provides a set of rules that can be applied to a new (unclassified) data set to predict which records will have a given outcome. It segments a data set by creating two –way splits.The classification obtained using these algorithms is shown in Table 2.2.4 ValidationThe more common methods for validation are: i) Leave –one –out (LOO); ii) Leave –more –out (LMO); iii) Train & Test; iv) Bootstrap. We used LOO, since it is considered the best working on data set of small dimension [10]. According to LOO, given n objects, n models are computed. For each model, the training set consists of n –1 objects and the evaluation set consists of the object left. To estimate the predictive ability, we considered the gap between the experimental (fitting) and the predicted value (cross –validation) for the n objects left, one by one, out from the model.Table 2. True class and class assigned by the algorithms for each compound 5.True Class CART LDA KNN SIMCARDAAnilofos 2 2 2 1 2 2 Chlorpyrifos1 2 2 1 2 2 Chlorpyryfos-methyl 2 2 2 1 2 2 Isazofos 1 1 1 2 1 1 Phosalone 2 2 2 2 2 2 Profenofos 1 2 2 1 2 2 Prothiofos 2 2 2 2 2 2 Azamethiphos 2 2 2 1 4 2 Azinphos methyl 1 1 1 2 1 1 Diazinon 3 3 1 1 4 1 Phosmet2 2 2 1 2 2 Pirimiphos ethyl 1 1 1 1 1 1 Pirimiphos methyl2312115 The 40 molecules with a blank background were used to train the neuro-fuzzy classifier.Pyrazophos 2 2 1 4 2 1Quinalphos 1 1 1 2 1 1Azinphos-ethyl 1 1 1 1 2 1Etrimfos 1 1 1 3 3 1Fosthiazate 4 2 2 2 4 2Methidathion 1 1 1 1 1 1Piperophos 3 3 3 2 2 3Tebupirimfos 4 1 1 3 4 1Triazophos 1 1 1 2 1 1Dichlorvos 2 4 2 2 2 2Disulfoton 3 3 3 1 3 3Ethephon 4 4 4 4 4 4Fenamiphos 1 1 3 2 1 1Fenthion 2 2 3 2 2 3Fonofos 1 1 3 2 1 3Glyphosate 4 4 4 4 4 4Isofenphos (isophenphos) 3 3 3 1 3 3Methamidophos 4 4 4 3 4 4Omethoate 3 3 3 3 3 3Oxydemeton-methyl 3 3 3 3 3 3Parathion ethyl (parathion) 2 2 2 3 1 3Parathion methyl 3 3 3 3 3 3Phoxim 2 2 1 1 1 1Sulfotep 1 1 3 2 2 2Tribufos 2 2 2 2 2 2Trichlorfon 2 2 2 1 2 4Acephate 4 4 1 3 4 4Cadusafos 2 2 3 3 2 2Chlorethoxyfos 2 2 2 3 2 2Demeton-S-methyl 3 3 3 3 3 3Dimethoate 3 3 1 1 3 3Edifenphos 2 2 3 1 2 2EPN 2 2 2 2 2 2Ethion 2 2 2 2 2 2Ethoprophos 3 3 3 2 2 3Fenitrothion 3 2 3 3 3 3Formothion 3 3 2 3 3 3Methacrifos 2 2 2 2 2 3Phorate 1 1 3 2 1 3Propetamphos 3 3 3 4 2 3Sulprofos 3 3 3 2 3 3Temephos 3 3 2 1 3 2Terbufos 1 1 3 2 3 3Thiometon 3 3 3 3 3 33.1 The neuro-fuzzy combination of the classifiers3.2 Motivations and architectureCombining multiple classifiers could be considered as a direction for the development of highly reliable pattern recognition systems, coming from the hybrid intelligent systems approach. Combination of several classifiers may result in improved performances [4], [5]. The necessity of combining multiple classifiers is arising from the main demand of increasing quality and reliability of the final models. There are different classification algorithms in almost all the current pattern recognition application areas, each one having certain degrees of success, but none of them beingas good as expected in applications. The combination technique we propose for the toxicity classification is a neuro-fuzzy gating of the implied classifiers, trained against the correct data. This approach allows multiple classifiers to work together.For this task, the hybrid intelligent system NIKE was used, in order to automate the processes involved, from the data representation for toxicity measurements, to the prediction of toxicity for given new input. It also suggests how the fuzzy inference produced the result, when required [17], based on the effect measure method to combine the weights between the layers of the network in order to select the strongest input-output dependencies [6]. Consequently, for NIKE, we defined the implicit knowledge as the knowledge acquired by neural/neuro-fuzzy nets.Fig. 1. Implicit Knowledge Module implemented as FNN2.The IKM-FNN is implemented as a multilayered neural structure with an input layer, establishing the inputs to perform the membership degrees of the current values, a fully connected three-layered FNN2 [16], and a defuzzification layer [17] (fig.1). The weights of the connections between layer 1 and layer 2 are set to one. A linguistic variable X i is described by m i fuzzy sets, A ij, having the degrees of membership performed by the functions µij(x i), j=1,2,...,m i, i=1,2,..,p., (in our case, p=5, all m i=4, on the classes of the prediction result of the classifiers, as inputs, and on the classes of the toxicity values, as the output y defuz). The layers 1 and 5 are used in the fuzzification process in the training and prediction steps, and the layers 2-4 are organized as a feedforward network to represent the implicit rules through FNN training [15][19]. 3.2 ResultsSince NIKE modules process only data scaled into the interval [0..1], every class was represented by the centroid of each of the four classes in which the available domain was split: 0.135 (class 1), 0.375 (class 2), 0.625 (class 3), and 0.875 (class 4). The inputs and the output followed a trapezoidal (de)fuzzification (fig. 2): VeryLow (0-0.25), Low (0.25-0.5), Medium (0.5-0.75), High (0.75-1).The neuro-fuzzy network was trained on a training set of 40 objects (70% of the entire set, as depicted in Table 2). The training set was used for the adjustment of the connections of the neural and neuro-fuzzy networks with backpropagation (traingdx) algorithm; traingdx is a network training function that updates weight and bias values according to gradient descent momentum and an adaptive learning rate. The neuro-fuzzy network was a multi-layered structure with the 5x4 above described fuzzy inputs and 4 fuzzy output neurons, the toxicity class linguistic variable (fig. 2.a). The number of hidden neurons parameterized the FNN. After different models (5 to 50 hidden units), a medium number of hidden units is desirable and had the same best results: IKM-FNN with 10, 12 and 19 neurons (fig. 3).(a) (b)Fig. 2. NIKE: (a)The fuzzy terms of the generic linguistic variable Class; (b) the FNN model. Table 3. Performances of the classification algorithms computed.NER% fitting NER%validation DescriptorsLDA 64.91 61.40 D1,D2, D3, D4RDA 84.21 71.93 D1, D2, D3, D4, D6, D7, D8, D11, D12, D13 SIMCA 92.98 77.19 D1, D2, D3, D4, D5, D6, D7, D8, D10, D11, D12 KNN - 61.40 D1, D12CART 85.96 77.19 D1, D2, D3, D4, D5, D9Table 4. Confusion matrix of the neuro-fuzzy combination of classifiers.N° of objectsAssigned Class1 2 3 41 132 15True Class2 20 203 1 15 164 6 6Table 5. True class and class assigned by all the classifiers for each compound wrong predicted by the neuro-fuzzy combination of classifiers.True Class CART LDA KNN SIMCA RDA FNN Chlorpyrifos 1 2 2 1 2 2 2Profenofos 1 2 2 1 2 2 2Fenitrothion 3 2 3 3 3 3 2(a)(b)(c)(d)(e)(f)Fig. 3. The results of training FNNs: (a) 3-5 errors, the best are FNN10H, FNN12H and FNN19H; (b) the chosen model, FNN10H, against the SIMCA results and the real ones; (c) the bad fuzzy inference prediction for 2 cases in class 1 (Chlorpyrifos and Profenofos); (d) the bad fuzzy inference prediction for the case in class 3 (Fenitrothion); two samples of good prediction for test cases: (e) a class 1 sample (Phorate); (f) a class 2 sample (Edinfenphos).A momentum term of 0.95 was used (to prevent too many oscillations of the error function). The nets were trained up to 5000 epochs, giving an error about 0.015. The recognition error for the above models is 5.26% (table 4, 5, fig. 3).The confusion matrix shows the ability in prediction of our approach. Looking of Table 3, we notice that the best performance was obtained by SIMCA, which could correctly classify almost 93% of the molecules. This encouraging result was obtained with whole data set involved in developing the model. If we take a look to the NER% validated with LOO, we can notice that we loss a lot of the reliability of the model when we predict the toxicity of an external object. Such a behavior proves the ability in modeling of these algorithms, but shows also their incapacity in generalization. The neuro-fuzzy approach seems to overcome this problem, succeeding in voting for the best opinion and underling all the considered classification algorithms (fig. 3).3.3 Interpreting the results of the neuro-fuzzy combination of the classifiers The most relevant fuzzy rules were extracted from the IKM-FNN structures using Effect Measure Method [6][13]. Finally, after deleting the contradictory rules, the next list of the most trusty fuzzy rules were considered for the chosen net IKM-FNN10H: IF CarFit1 is:VeryLow THEN class is:High (39.22%)IF CarFit1 is:Low THEN class is:High (82.30%)IF CarFit1 is:Medium THEN class is:High (48.74%)IF CarFit1 is:High THEN class is:High (39.04%)IF SimFit1 is:VeryLow THEN class is:Medium (61.25%)IF SimFit1 is:Low THEN class is:Medium (36.04%)IF SimFit1 is:High THEN class is:Medium (43.72%)IF RdaFit1 is:VeryLow THEN class is:Low (75.65%)IF RdaFit1 is:Low THEN class is:Low (100.00%)IF RdaFit1 is:High THEN class is:High (76.39%)Three types of fuzzy rules were obtained: some could be grouped by the same output, or by having the same fuzzy term in the premise and conclusion, and, finally, rules with mixed terms in premises and conclusion parts. From the first two groups of fuzzy rules (italics), we could conclude that, the opinion of the entry classifier is not important for the given output. More precisely, CART prediction for High values of toxicity (class 4) is better to not be taken in consideration.IF (CarFit1 is:VeryLow) OR (CarFit1 is:Low) OR (CarFit1 is:Medium) OR (CarFit1 is:High) THEN class is:HighSimilarly, SIMCA outputs are not so important for predicting class 3 (Medium toxicity: the second group of fuzzy rules). From the second last group of rules, we could find which is the best classifier from the involved systems. In our case, in order to predict class 2 (Low toxicity) is better to consider the opinion coming from RDA. The same opinion is very important for predicting the class 4 (High toxicity) cases too.ConclusionsClassification of the toxicity requires a high degree of experience from computational chemistry experts. Several approaches were described to generate suitable computer-based classifiers for the considered patterns. We investigated five different classifiers and a neuro-fuzzy correlation of them, to organize and classify toxicity data sets. Our approach shown an improved behaviour as a combination of classifiers. Some results viewing fuzzy rules extraction, as well as the possibility to interpret particular inferences suggest that the Neuro-Fuzzy approach has the potential to significantly improve common classification methods for the use in toxicity characterization. AcknowledgmentThis work is partially funded by the E.U. under the contract HPRN-CT-1999-00015. References1. Benfenati, E., Pelagatti, S., Grasso, P., Gini, G.: COMET: the approach of a project in evaluatingtoxicity. In: Gini, G. C.; Katritzky, A. R. (eds.): Predictive Toxicology of Chemicals: Experiences and Impact of AI Tools. AAAI 1999 Spring Symposium Series. AAAI Press, Menlo Park, CA (1999) 40-43 2. Benfenati, E., Piclin, N., Roncaglioni,A., Varì, M.R.: Factors Influencing Predictive Models ForToxicology. SAR and QSAR in environmental research, 12 (2001) 593-603.3. Bishop, C.M.: Neural networks for pattern recognition. Clarendon Press, Oxford (1995)4. Chen, K., Chi, H.: A method of combining multiple probabilistic classifiers through soft competition ondifferent feature sets. Neurocomputing 20 (1998) 227-2525. Duin, R.P.W., Tax, D.M.J.: Experiments with Classifier Combining Rules. Lecture Notes in ComputerScience, 1857, Springer-Verlag, Berlin (2000) 16-296. Enbutsu, I., Baba, K., Hara, N.: Fuzzy Rule Extraction from a Multilayered Network, in Procs. ofIJCNN'91, Seattle (1991) 461-4657. Gini, G., Benfenati, E., Boley, D.: Clustering and Classification Techniques to Assess Aquatic Toxicity.Procs. of the Fourth Int'l Conf. KES2000, Brighton, UK, Vol. 1 (2000) 166-1728. Gini, G., Giumelli, M., Benfenati, E., Lorenzini, P., Boger, Z.: Classification methods to predict activityof toxic compounds, V Seminar on Molecular Similarity, Girona, Spain (2001 forthcoming)9. Gini, G., Lorenzini, M., Benfenati, E., Brambilla, R., Malvé, L.: Mixing a Symbolic and a SubsymbolicExpert to Improve Carcinogenicity Prediction of Aromatic Compounds. In: Kittler,J.,Roli,F.(eds.):Multiple Classifier Systems, Springler-Verlag, Berlin (2001)126-135.10. Helma, C., Gottmann, E., Kramer, S.: Knowledge discovery and data mining in toxicology. Statisticalmethods in medical research, 9 (2000) 131-13511. Ho, T., Hull, J., Srihari, S.: Decision combination in multiple classifier systems. IEEE Trans. PatternAnal. Mach. Intell. 16/1 (1994) 66-7512. Jacobs, R.A.: Methods for combining experts' probability assessments, Neur. Comp. 7/5(1995)867-88813. Jagielska, I., Matthews, C., Whitfort, T.: An investigation into the application of ANN, FL, GA, andrough sets to automated knowledge acquisition for classification problems. Neurocomp,24(1999)37-5414. Kosko, B.: Neural Networks and Fuzzy System. Prentice-Hall, Englewood Cliffs (1992)15. Lin, C.T., George Lee, C.S.: Neural - Network Based Fuzzy Logic Control and Decision System. IEEETransactions on Computers, 40/12 (1991) 1320-133616. Nauck, D., Kruse, R.: NEFCLASS-X: A Neuro-Fuzzy Tool to Build Readable Fuzzy Classifiers. BTTech. J. 16/3 (1998) 180-192.17. Neagu, C.-D., Avouris, N.M., Kalapanidas, E., Palade, V.: Neural and Neuro-fuzzy Integration in aKnowledge-based System for Air Quality Prediction. App Intell. J. (2001 accepted)18. Palade, V., Neagu, C.-D., Patton, R.J.: Interpretation of Trained Neural Networks by Rule Extraction,Procs. of Int'l Conf. 7th Fuzzy Days in Dortmund (2001) 152-161.19. Rumelhart, D.E., McClelland, J.L.: Parallel Distributed Processing, Explanations in the Microstructureof Cognition. MIT Press (1986)。
R.Fullér, A neuro-fuzzy approach to medical diagnostics, in P.Eklund and J.Mattila eds
A Neuro-Fuzzy Approach to MedicalDiagnostics∗†Patrik EklundDepartment of Computer Science,˚A bo Akademi University, Lemmink¨a inengatan14,SF-20520˚A bo,FinlandRobert Full´e rDepartment of Computer Science,E¨o tv¨o s Lor´a nd University, P.O.Box157,H-1502Budapest112,HungaryAbstractThere has been growing interest and activity in the area of medical decision making,especially in the last20years.As it has been pointed out by manyauthors[Adl86,Hay90,San92,Wan91],fuzzy set theory and neural nets havea number of properties that make them suitable for formalizing the uncertaininformation upon which medical diagnosis and treatment is usually based.Generalizing the earlier results of thefirst author[Ekl92],we provide a formal model of this process using the neuro-fuzzy theory,and illustrate ourapproach on a two-symptom artificial disease.1IntroductionIn[Ekl92a]we proposed an architecture(Diagai D)for a generic tool which supports data analysis and developement of diagnostic modules in clinical medicine.This architecture contains three main modules,each of which constitute different software packages.Thefirst one extracts related data(symptoms and signs)from hospital databases.The prepocessing module transforms(by the help of sigmoid functions)the ex-tracted raw data into a corresponding logical form.Finally,the third one is a single ∗Supported by GeDeMeDeS-project at˚A bo Akademi University funded by the Technology Development Centre,Helsinki.†in:P.Eklund and J.Mattila eds.,Proceedings of Fuzziness in Finland’93Workshop,˚A bo Akademis Tryckeri,˚A bo,199319-22;also in:Proceedings of EUFIT’93Conference,September 7-10,1993,Aachen,Germany,Verlag der Augustinus Buchhandlung,Aachen,1993810-813;also in:Fuzzy Systems&A.I.,3(1994),53-56.1layer backpropagation network,where the weights and the parameters of prepocess-ings functions are tuned in order to optimize the diagnostic performance.In[Ekl92]we demonstrated how preprocessed(fuzzified)data from single layer net-works can provide faster convergence and better diagnostic performance than raw data from multi-layered networks in backpropagation networks.In this paper we show that by adding a fourth module(in which we tune the exponents of a polynomial combination of the modified inputs)to Eklund’s method even better diagnostic performance can be reached without enlarging the network structure.We illustrate our ideas on a two-class discriminant problem.2Improving the diagnostic performanceA full description of the three modules is outside of the scope of this short paper and we refer to[Ekl92,Ekl92a]for futher reading.Suppose that after the backpropaga-tion learning algorithm we obtained the weights,w1,...w n,where n is the number of symptoms.Then in the additional fourth module(which is a single layer backpropagation network),instead of the weighted sum of the inputs we use their polynomial combi-nation as followsnsign(w i)(|w i|x pi)λii=1where x pi is the(preprocessed)value of the i-th symptom of the p-th patient,and we tune onlyλi,i=1,...,n,from the initial valuesλ1=1,...,λn=1.It is clear(because it is easier to separate by a continuous curve than by a line)that in many cases the new module provides a better diagnostic performance.However,as it was pointed out by many authors,it is impossible to provide the correct diagnosis in every case,because two persons can have approximately the same symptoms,and one suffers the disease and the other is healthy.We would need more information(additional significant symptoms)to separate them.3IllustrationFor simplicity,we illustrate our approach on a two-symptom artificial disesase.Sup-pose that a linear transformation of the raw data into the unit interval resulted in the following picture2: healthy : illIt is clear that there does not exist a line separating the ill persons from the healthy ones(due to the presense of many XOR structures).The meaning of the prepocession is to move as much as possible the ill persons to the direction of the top-right corner and the healthy ones to the direction of the bottom-left corner.Then we shall have more freedom to separate them.Sup-pose that after the backpropagation learning algorithm(third module)we got the following pictureIt is easy to see that by the help of fuzzification we could separate more persons than by a simple linear transformation.However,in certain cases we are still not able to separate them via a line.The fourth module provides a moreflexible curve for separation,but as we can see from the following picture,we could not provide the correct answer in every case.34SummaryWe have proposed to add a new module to Diagai D architecture,which can improve the diagnostic performance in many cases.The network structure is still kept small and explainable,which allows to spare time at computerized implementation and provides an easy way to communicate with the users of the software package. References[Adl86]K.P.Adlassing,Fuzzy set theory in medical diagnostics,IEEE Trans.on Systems,Man,and Cybernetics,Vol.SMC-16(1986)260-264.[Ekl92]P.Eklund,Network size versus propocessing,in:R.R.Yager and L.A.Zadeh eds.,Fuzzy Sets,Neural Networks and Soft Computing,(to appear). [Ekl92a]P.Eklund,Mrten Fogstrm and Jari Forsstrm,A generic neuro-fuzzy tool for developing medical decision support,in:P.Eklund ed.,Proceedings ofthe MEPP’92,˚A bo Akademi Press,˚A bo,199211-27.[Hay90]Y.Hayashi and A.Imura,Fuzzy neural expert system and its application to medical diagnosis,in:C.N.Manikopoulos,ed.,Proceedings of the8thInternational Congress of Cybernetics and Systems,New Jersey Instituteof Technology Press,Newark,NJ,199054-61.[San92] E.Sanchez,Fuzzy logic knowledge systems and artificial neural networks in medicine and biology,in:R.R.Yager and L.A.Zadeh,eds.,An In-troduction to Fuzzy Logic Applications in Intelligent Systems,Kluwer,Dordrecht-Boston,1992235-252.[Wan91]Y.Wang,The fuzzy neural network system for diagnosing silicosis,in: T.Terano,M.Sugeno,M.Mukaidono and K.Shigemasu eds.,Proceedingsof the International Fuzzy Engineering Symposium’91,I.O.S.Press,1992546-549.4。
车用锂离子电池的SOC估算方法研究现状
AUTO TIME161AUTO PARTS | 汽车零部件时代汽车 车用锂离子电池的SOC 估算方法研究现状陆张浩 潘正军 许祥进金肯职业技术学院 江苏省南京市 211106摘 要： S OC(State of charge)，即电池的荷电状态，它描述的是电池的剩余容量，其数值上表示为电池剩余的荷电量占电池总电量的比值，常用百分数表示。
它是电池状态的一个关键指标，SOC 的准确估算可以有效的提高电池使用效率，延长电池的使用寿命。
荷电状态不能通过直接测量获得，而是需要其它方式来估算。
本文对车用锂离子电池SOC 估算方法进行了简单的描述，分析了不同方法的优缺点，最后进行了总结。
关键词：荷电状态 电池 估算目前世界上各汽车生产厂家纷纷开发并推广使用电动汽车,电动汽车有着广阔的发展前景。
电动汽车的蓬勃发展,促进了动力电池技术的发展,世界各大汽车公司纷纷投巨资并采取结盟的方式研究各种类型的电动车用动力电池。
电池的荷电状态(SOC)在使用过程中是一个非常重要的参数，它直接影响电池的电压、电流以及内阻等物理量，并且和电池的使用寿命、效率息息相关。
所以，SOC 估计是锂电池管理系统的一个核心技术。
为了保证电池具有良好的性能并且拥有较长的使用寿命，必须要对电池进行一定的管理和控制，从而对电池组进行均衡充电延长电池使用寿命，因此准确有效地预测估计电池荷电状态(state of charge，SOC)是BMS 中最基本和最核心的目标[1-4]。
本文对目前经常使用的SOC 估算方法进行了研究综述。
1 SOC 的定义电池的荷电状态SOC 反映电池的剩余电量的情况，也就是在一定的放电电流下，当前电池的剩余电量与总的可用电量的比值。
它的数学表达式如公式1所示[5]：%100SOC 0×=Q Q t（1）式中：Q t 为电池在计算时间的剩余电量；Q 0为蓄电池的总容量。
2 SOC 的估算方法2.1 安时积分法安时积分法是最常用的SOC 估算方法。
基于实时工况的装载机智能换挡规律
71
2
油门开度、 车速和油泵压力三参数换挡规律设计
b T
在牵引特性曲线图中，不同油门开度下各挡位牵引 力曲线的交点就是理论上低挡换高挡的升挡点。因此， 在牵引力计算的基础上，求出不同油门开度下各挡交点 的车速值，以此车速值为横坐标，油门开度值作为纵坐 标，即可得到自动换挡的升挡规律。 装载机由于其牵引力随工况的不同变化较大，因此， 根据工作泵在满载工况、半载工况、空载工况的状态， 分3种情况来考虑。3种工况下换挡规律曲线如图4所示。
TT K Tb
（4）
根据工作泵的3种典型工作状态，按照式（5） 、 （6） 分别计算不同油门开度下装载机的牵引特性，图3为工作 泵满载工作时不同油门开度下的牵引特性曲线。 同理可以计算工作泵空载和工作泵半载时不同油门 开度下的牵引力曲线，限于篇幅，本文不再列出。
式中 nT ——液力变矩器输出转速； TT ——液力变矩器 输出转矩；K——变矩系数。 根据合作企业提供的试验数据，得到ZL50装载机输 出特性如图2所示。
第 25 卷 第 3 期 2009 年 3月
农 业 工 程 学 报 Transactions of the CSAE
Vol.25 No.3 Mar. 2009
69
基于实时工况的装载机智能换挡规律
常 绿
（淮阴工学院交通工程系，淮安 223003） 摘 要：把装载机工作泵分为空载工作、满载工作、半载工作 3 种情况，分别计算 ZL50 装载机的输入特性和输出特性； 完成 ZL50 装载机的牵引特性的计算。在此基础上，提出根据装载机油门开度、车速和油泵的压力为控制参数的 3 参数 换挡规律。建立装载机传动系统数学模型，进行了计算机仿真。从仿真结果看，3 参数换挡规律符合工程车辆的实际工 作状态，能更好地满足工程车辆的动力性要求。 关键词：装载机，自动变速，数学模型，仿真，换挡规律 中图分类号：U463.2 文献标识码：A 文章编号：1002-6819(2009)-3-0069-05 常 绿. 基于实时工况的装载机智能换挡规律[J]. 农业工程学报，2009，25(3)：69－73. Chang Lü . Intelligent shift schedule based on working conditions of loader[J]. Transactions of the CSAE, 2009,25(3)：69－73.(in Chinese with English abstract)
模糊神经网络外文翻译文献
模糊神经网络外文翻译文献(文档含中英文对照即英文原文和中文翻译)原文：Neuro-fuzzy generalized predictive control ofboiler steam temperatureXiangjie LIU, Jizhen LIU, Ping GUANABSTRACTPower plants are nonlinear and uncertain complex systems. Reliablecontrol of superheated steam temperature is necessary to ensure high efficiency and high load-following capability in the operation of modern power plant. A nonlinear generalized predictive controller based on neuro-fuzzy network (NFGPC) is proposed in this paper. The proposed nonlinear controller is applied to control the superheated steam temperature of a 200MW power plant. From the experiments on the plant and the simulation of the plant, much better performance than the traditional controller is obtained.Keywords:Neuro-fuzzy networks; Generalized predictive control; Superheated steam temperature1. IntroductionContinuous process in power plant and power station are complex systems characterized by nonlinearity, uncertainty and load disturbance. The superheater is an important part of the steam generation process in the boiler-turbine system, where steam is superheated before entering the turbine that drives the generator. Controlling superheated steam temperature is not only technically challenging, but also economically important.From Fig.1,the steam generated from the boiler drum passes through the low-temperature superheater before it enters the radiant-type platen superheater. Water is sprayed onto the steam to control the superheated steam temperature in both the low and high temperature superheaters. Proper control of the superheated steam temperature is extremely important to ensure theoverall efficiency and safety of the power plant. It is undesirable that the steam temperature is too high, as it can damage the superheater and the high pressure turbine, or too low, as it will lower the efficiency of the power plant. It is also important to reduce the temperature fluctuations inside the superheater, as it helps to minimize mechanical stress that causes micro-cracks in the unit, in order to prolong the life of the unit and to reduce maintenance costs. As the GPC is derived by minimizing these fluctuations, it is amongst the controllers that are most suitable for achieving this goal.The multivariable multi-step adaptive regulator has been applied to control the superheated steam temperature in a 150 t/h boiler, and generalized predictive control was proposed to control the steam temperature. A nonlinear long-range predictive controller based on neural networks is developed into control the main steam temperature and pressure, and the reheated steam temperature at several operating levels. The control of the main steam pressure and temperature based on a nonlinear model that consists of nonlinear static constants and linear dynamics is presented in that.Fig.1 The boiler and superheater steam generation processFuzzy logic is capable of incorporating human experiences via the fuzzy rules. Nevertheless, the design of fuzzy logic controllers is somehow time consuming, as the fuzzy rules are often obtained by trials and errors. In contrast, neural networks not only have the ability to approximate non-linear functions with arbitrary accuracy, they can also be trained from experimental data. The neuro-fuzzy networks developed recently have the advantages of model transparency of fuzzy logic and learning capability of neural networks. The NFN is have been used to develop self-tuning control, and is therefore a useful tool for developing nonlinear predictive control. Since NFN is can be considered as a network that consists of several local re-gions, each of which contains a local linear model, nonlinear predictive control based on NFN can be devised with the network incorporating all the local generalized predictive controllers (GPC) designed using the respective local linear models. Following this approach, the nonlinear generalized predictive controllers based on the NFN, or simply, the neuro-fuzzy generalized predictive controllers (NFG-PCs)are derived here. The proposed controller is then applied to control the superheated steam temperature of the 200MW power unit. Experimental data obtained from the plant are used to train the NFN model, and from which local GPC that form part of the NFGPC is then designed. The proposed controller is tested first on the simulation of the process, before applying it to control the power plant.2. Neuro-fuzzy network modellingConsider the following general single-input single-output nonlinear dynamic system:),1(),...,(),(),...,1([)(''+-----=u y n d t u d t u n t y t y f t y∆+--/)()](),...,1('t e n t e t e e (1)where f[.]is a smooth nonlinear function such that a Taylor series expansion exists, e(t)is a zero mean white noise and Δis the differencing operator,''',,e u y n n n and d are respectively the known orders and time delay ofthe system. Let the local linear model of the nonlinear system (1) at the operating point )(t o be given by the following Controlled Auto-Regressive Integrated Moving Average (CARIMA) model:)()()()()()(111t e z C t u z B z t y z A d ----+∆= (2) Where)()(),()(1111----∆=z andC z B z A z A are polynomials in 1-z , the backward shift operator. Note that the coefficients of these polynomials are a function of the operating point )(t o .The nonlinear system (1) is partitioned into several operating regions, such that each region can be approximated by a local linear model. Since NFN is a class of associative memory networks with knowledge stored locally, they can be applied to model this class of nonlinear systems. A schematic diagram of the NFN is shown in Fig.2.B-spline functions are used as the membership functions in the NFN for the following reasons. First,B-spline functions can be readily specified by the order of the basis function and the number of inner knots. Second, they are defined on a bounded support, andthe output of the basis function is always positive, i.e.,],[,0)(j k j j k x x λλμ-∉=and ],[,0)(j k j j k x x λλμ-∈>.Third, the basis functions form a partition of unity, i.e.,.][,1)(min,∑∈≡jmam j k x x x x μ (3)And fourth, the output of the basis functions can be obtained by a recurrence equation.Fig. 2 neuro-fuzzy network The membership functions of the fuzzy variables derived from the fuzzy rules can be obtained by the tensor product of the univariate basis functions. As an example, consider the NFN shown in Fig.2, which consists of the following fuzzy rules:IF operating condition i (1x is positive small, ... , and n x is negative large),THEN the output is given by the local CARIMA model i:...)()(ˆ...)1(ˆ)(ˆ01+-∆+-++-=d t u b n t y a t y a t y i i a i in i i i a )(...)()(c i in i b i in n t e c t e n d t u b cb -+++--∆+ (4)or)()()()()(ˆ)(111t e z C t u z B z t yz A i i i i d i i ----+∆= (5) Where )()(),(111---z andC z B z A i i i are polynomials in the backward shift operator 1-z , and d is the dead time of the plant,)(t u i is the control, and )(t e i isa zero mean independent random variable with a variance of 2δ. Themultivariate basis function )(k i x a is obtained by the tensor products of the univariate basis functions,p i x A a nk k i k i ,...,2,1,)(1==∏=μ(6)where n is the dimension of the input vector x, and p, the total number of weights in the NFN, is given by,∏=+=nk i i k R p 1)((7)Where i k and i R are the order of the basis function and the number of inner knots respectively. The properties of the univariate B-spline basis functions described previously also apply to the multivariate basis function, which is defined on the hyper-rectangles. The output of the NFN is,∑∑∑=====p i i i p i ip i i i a y aa y y 111ˆˆˆ (8)译文：锅炉蒸汽温度模糊神经网络的广义预测控制Xiangjie LIU, Jizhen LIU, Ping GUAN摘要发电厂是非线性和不确定性的复杂系统。
A Neuro-Fuzzy Based Adaptive Set-Point Heat Exchan
Nov. 2012, Volume 6, No. 11 (Serial No. 60), pp. 1584–1588Journal of Civil Engineering and Architecture, ISSN 1934-7359, USAA Neuro-Fuzzy Based Adaptive Set-Point Heat Exchanger Control Scheme in District Heating SystemLiang Huang1, Zaiyi Liao2 and Zhao Lian11. Department of Electrical and Computer Engineering, Ryerson University, Toronto M5B2K3, Canada2. College of Hydraulic and Environmental Engineering, China Three Gorges University, Yichang 443003, ChinaAbstract: The control of heat exchange stations in district heating system is critical for the overall energy efficiency and can be very difficult due to high level of complexity. A conventional method is to control the equipment such that the temperature of hot water supply is maintained at a set-point that may be a fixed value or be compensated against the external temperature. This paper presents a novel scheme that can determine the optimal set-point of hot water supply that maximizes the energy efficiency whilst providing sufficient heating capacity to the load. This scheme is based on Adaptive Neuro-Fuzzy Inferential System. The aim of this study is to improve the overall performance of district heating systems.Key words: District heating system, neuro-fuzzy, inferential sensor, energy efficiency, control.1. IntroductionDistrict heating systems are considered energy efficient and widely used in Canada. Hot water from CHP (combined heat and power) carries heat to the heat exchangers, in which heat is transferred to the water in secondary loops. At individual buildings, appropriate operation of the heat exchangers is essential for harnessing the benefits made possible by district heating systems. The temperature of hot water supply in the secondary loop is conventionally controlled to fluctuate around a set-point, which may be constant for certain period of time or compensated against the external air temperature. Previous studies have shown that these two methods are likely to cause energy waste and/or discomfort [1]. A new approach is to change the set-point according to a measurement of thermal comfort at the buildings using temperature sensors [1]. However, using a lot of temperature sensors in a building can be practically infeasible and unstable. Liao and Dexter [1] proposed a simplifiedCorresponding author: Liang Huang, master, research fields: neuro-fuzzy network, artificial intelligence, building automation system, and control system. E-mail: **********************.physical model for estimating the average indoor air temperature by using measurable variables, such as outdoor temperature, solar radiation, and the power supplied to terminal. This model makes it possible to estimate heating load based on the outputs of simple sensors that are easily available to the controller in practice. In recent years, fuzzy logic [2] and neural networks have been proposed as alternatives to traditional statistical ones in building technology, in terms of improvement of indoor comfort and energy conservation. Researchers extensively applied fuzzy logic to the built environment to improve the performance and to reduce energy consumption [2–6], while neural networks are used for improving performance of built environment [7, 8] and estimate the operative temperature in a building [9, 10] designed an ANFIS based inferential sensor model, which estimates the average air temperature in the buildings that heated by a hydraulic heating system.In this paper, we present a neuro-fuzzy based control scheme that can estimate the heating load and according determine optimal value for the set-point of hot water supply in the secondary loop. When thesystem is operated with such set-point, the energyAll Rights Reserved.A Neuro-Fuzzy Based Adaptive Set-Point Heat Exchanger Control Scheme in District Heating System 1585efficiency can be maximized whilst desired indoor thermal environment is maintained.2. Research MethodsIn the conventional heat exchanger, the heat from the heat source is transferred to the water in secondary loops (Fig. 1) and the required flow rate of hot water from the heat source depends on required heating load, water temperature, and heating transfer rate.sr sssr ( - )*m ( - )*T T T T M η∙∙=(1)where, η is the heating transfer rate of the heatexchanger, m and T s are the water flow rate and temperature at hot-fluid outlet, M and T ss are the hot water flow rate and temperature at hot-fluid inlet, and T sr and T r are the water temperatures at hot fluid inlet and cold-fluid inlet.In this paper, two parameters are used to define the performance of a heating system: overall performance of the heating system and a measure of the thermal comfort in the zone [11]. A comfort range is defined as Φref = [T min , T max ]. The total energy consumption (E) in secondary loop is normalized to the total energy supplied to heat exchanger when the set-point is constant.100%*/o e E E = (2)A measure of the overall performance of the heating system is given by(1)1e e w w γγγγξ-+=+ (3)where, W γ is a weighting constant, which determines the importance of thermal comfort in assessing the overall performance. It should be noted that the largerFig. 1 Shell-and-tube heating exchanger [12]. the value of overall performance, the higher is the overall performance of the heating system [11].The impact of heat exchanger control on the overall performance of heating systems has been studied in simulation. Two types of heat exchanger controllers are studied:• Type I: the constant set-point controller. The supply water temperature set-point is fixed at a constant level specified during commissioning. This is a most commonly used heat exchanger controller because of its simplicity;• Type II: the adaptive set-point controller. The supply water temperature set-point in secondary loop is varied in inverse proportion to a moving average of the external environment in a certain time interval. During the test period, the temperature set-point of Type I is a constant, however the set-point changing of Type II varies based on the required heating load and the capacity of supplied heating load. The adaptive set-point in Type II cannot be varied frequently, since the profile of the control system. The temperature set-point changing time point is decided by estimating the time of instantaneous indoor air temperature equals to the average indoor temperature in one day. To look for a suitable set-point of supplied water temperature in every interval in the test period, indoor temperature comfort is considered firstly, and then, energy efficient. Liao’s simply physical model and Jassar’s [10] model is used in calculating optimal required energy. This optimal set-point need satisfy the system has a lowest energy cost when the indoor temperature in comfortable range during a certain period. The optimal required heating load is⎰1t t d Q Min (4)S.T. 0>sol Qmax min a a a T T T <<max min o o o T T T <<Once the required heating load is decided, the temperature set-point can be calculated byAll Rights Reserved.A Neuro-Fuzzy Based Adaptive Set-Point Heat Exchanger Control Scheme in District Heating System1586)()(t m Q T T dr s =-∙(5)Therefore, the temperature set-pointT QT r ds mt +=)((6)Then, an adaptive neuro-fuzzy inferential heat exchanger control scheme (illustrated in Fig. 2) is proposed and its control process is simulated. The impact of adaptive set-point heat exchanger control scheme on the overall performance of energy efficiency is studied in simulation. The experimental data used to estimate set-point temperature is obtained from a laboratory heating system monitored in an EU CRAFT project [13].3. ResultsIn the proposed control scheme, the temperature set-point estimator estimates the optimal set-point temperature of the hot water in the secondary loop and optimal set-point changing time. The thermalcomfortable range in test period is between 18︒C and 21︒C in our simulation and indoor air temperature is estimated by using adaptive neuro-fuzzy based inferential sensor model [10]. The supplied hot water temperature in the secondary loop is also sensed and the corresponding control signals is generated in heat exchanger operation module, which includes a PID (proportion integration differentiation ) controller, is sent to heat exchanger. In this case, the supplied hot temperature can follow the set-point temperature by controlling the flow rate of the hot water from CHP.In this scheme, the set-point changes twice a day at the 7.58th hour and the 18.67th hour that the indoor air temperature equals to average air temperature of the day.Fig. 3 shows a good performance of adaptive set-point in controlling indoor air temperature in thermal comfortable range. Comparing to constant set-point control heating, the indoor air temperature controlled by adaptive set-point satisfies the desired comfortable temperature range which is between 18︒C and 21︒C.Not only the adaptive has a good performance in keep indoor thermal comfortable, but also it has a good energy cost performance. Fig. 4 shows the adaptive set-point temperature control fulfill the indoor thermal comfort requirement. At the same time, the energy efficiency is also higher than constant set-point control.4. DiscussionThe neuro-fuzzy based adaptive set-point heat exchanger control scheme has a very good performance in maximizing the energy efficiency whilst providing sufficient heating capacity to the load. Jassar’s neuro-fuzzy based inferential sensor model is based on three inputs, power supplied to terminals Q in (derived from temperature difference between hot-fluid inlet and hot-fluid outlet), solar RadiationFig. 2 All Rights Reserved.A Neuro-Fuzzy Based Adaptive Set-Point Heat Exchanger Control Scheme in District Heating System1587Fig. 3 Thermal comfort performances of two types of control.Fig. 4 Impacts of heat exchanger control on the performance of heating systems.Q sol, and external temperature T O. Also, Liao’s simplified physical model for estimation of air temperature is based on the same variables. Therefore, the estimated average air temperature by Jassar’s model is possible to be used in deduction of the optimal set-point of supplied water estimation in secondary loop by using Liao’s model. Although the heating source of the proposed scheme in this paper is heat exchanger not a boiler, they are both hot-water space heating systems.In Fig. 4, the performance of Type I is far below that of the Tpye II, the reasons for the poor performance are as follows:Once commissioned the set-point is fixed for the entire test period.All Rights Reserved.A Neuro-Fuzzy Based Adaptive Set-Point Heat Exchanger Control Scheme in District Heating System 1588•If too high a value of the set-point is selected, more energy will be consumed and the room temperature is more frequently above the upper level of the desired range, resulting in lower overall performance;•If too low a value for the set-point is selected, the benefit of lower energy consumption is at the cost of significant discomfort because the room temperature is more frequently below the lower level of the desired range. Consequently the overall performance remains low.The performance of the Type II controllers is such better than Type I controller. Less energy is consumed and the room temperature is more frequently in the desired comfortable range when the controller is commissioned, such that too high a set-point is used at high external temperatures. As a result the overall performance improved in both energy consumption and comfort ratio.5. ConclusionA neuro-fuzzy based adaptive control scheme is developed to control the heat exchangers in district heating systems for maximize the energy efficiency whilst providing sufficient heating capacity to the load that the indoor temperature is controlled in a thermal comfortable range.In the future, an estimation model which can keep high robustness and high accuracy of the prediction in the indoor temperature estimation will be researched, so that the set-point estimator will have better performance and the robustness of the heat exchanger will be further improved.References[1]Z. Liao and A. L. Dexter, A simplified physical model forestimating the average air temperature in multi-zoneheating systems, Building and Environment 39 (9) (2004)1013–1022.[2]L. Zadeh, Outline of a new approach to the analysis ofcomplex systems and decision processes, in: IEEETransactions on System, Man, and Cybernetics, BrowseJournals & Magazines 3 (1) (1973) 28–44.[3] A. L. Dexter and D. W. Trewhella, Building controlsystems: fuzzy rule-based approach to performanceassessment, Building Services Research and Technology11 (4) (1990) 115–124.[4] A. I. Dounis, M. J. Santamouris and C. C. Lefas, Buildingvisual comfort control with fuzzy reasoning, EnergyConservation and Management 34 (1) (1993) 17–28.[5] A. I. Dounis, M. Bruant, M. Santamouris, G. Guaraccinoand P. Michel, Comparison of conventional and fuzzycontrol of indoor air quality in buildings, Journal ofIntelligent and Fuzzy Systems 4 (1996) 131–140.[6]P. Angelov, A fuzzy approach to building thermalsystems optimization, Vol. 2, in: Proceedings of theeighth IFSA World congress, Taipai, Taiwan, 1999, pp.528–531.[7]J. F. Kreider, Neural networks applied to building energystudies, in: H. Bloem (Ed.), Workshop on ParameterIdentification, Joint Research Center, Ispra, 1995, pp.233–251.[8]S. J. Hepeworth and A. L. Arthur, Adaptive neuralcontrol with stable learning, Mathematics and Computersin Simulation 41 (2000) 39–51.[9]M. S. Moheseni, B. Thomas and P. Fahlen, Estimation ofoperative temperature in buildings using artificial neuralnetworks, Energy and Buildings 38 (2006) 635–640. [10]S. Jassar, Z. Liao and L. Zhao, Adaptive neuro-fuzzybased inferential sensor model for estimating the averageair temperature in space heating systems, Building andEnvironment 44 (8) (2009) 1609–1616.[11]Z. Liao and A. L. Dexter, An inferential control schemefor optimizing the operation of boilers in multi-zoneheating systems, Building Service Engineering Researchand Technology 24 (4) (2003) 245–266.[12]R. K. Shah and D. P. Sekulic, Fundamental of HeatExchanger Design, John Wiley & Sons, Inc., 2003.[13]BRE (Building Research Establishment), ICITE,Controller efficiency improvement for commercial andindustrial gas and oil fired boilers, A CRAFT project,Brittech Controls Europe Ltd., 1999–2001.All Rights Reserved.。
信息检索与利用报告
机电一体化又称机械电子学，英语称为Mechatronics，它是由英文机械学Mechanics的前半部分与电子学Electronics的后半部分组合而成。机电一体化最早出现在1971年日本杂志《机械设计》的副刊上，随着机电一体化技术的快速发展，机电一体化的概念被我们广泛接受和普遍应用。随着计算机技术的迅猛发展和广泛应用，机电一体化技术获得前所未有的发展。现在的机电一体化技术，是机械和微电子技术紧密集合的一门技术，他的发展使冷冰冰的机器有了人性化，智能化。
检索结果（存取号、篇名、作者、文献出处，外文文献篇名翻译为中文）
SCI
Electromechanical integration ; automobile
Electromechanical integration* automobile
主题
all
4
篇名：造型与工程设计问题
入藏号:WOS:000072197400003
电力汽车
Invertor(s):HE HONGWENLIN[CN]
Applicant(s):BEIJING INST TECHNOLOGY [CN]
Application number:CN20061064955 20060320
（四）外文摘数据库（15%）
数据库名
检索词
检索式
检索字段
检索年限
检中量(篇)
二、文献检索范围、策略及结果：
（一）中外期刊全文数据库（40%）
数据库名
检索词
检索式
检索字段
检索
年限
检中量(篇)
检索结果（篇名、作者、文献出处，外文文献篇名翻译为中文）
维普
基于自适应神经-模糊推理系统的边坡变形预测模型
基于自适应神经-模糊推理系统的边坡变形预测模型摘要：利用自适应神经模糊推理系统(anfis)处理非线性关系的强大能力，将其应用于边坡变形预测。
发现建立的anfis模型预测精度远高于灰色模型。
最后用工程实例与灰色模型进行了对比，结果表明，anfis模型优于灰色模型，特别是在模拟多输入变量、高维数下边坡变形预测问题时有着独特的优势，具有一定的推广应用价值。
abstract: to compensate the defect of the growth model and the artificial neural network model which are the most commonly used method to settlement prediction of soft clay roadbed, proposed to adaptive neuro-fuzzy inference system (anfis) applied to soft clay roadbed settlement prediction. anfis put the fuzzy reasoning process of the expert inherent in the neural network structure, so, the neural network nodes and the weights have a clear physical meaning, and avoid the “black box”through the working process of neural network. at the same time that the system can used the least square method and gradient descent algorithm for the combination of mixed, both with the self-adaptive and learning ability of a neural network, but also overcome its disadvantages such as the local minimum, the prediction accuracy is also much higher than the growth curve model. they are applied to thecalculation for examples, the result showed that the anfis model has great theoretic significance and practical value in preventing roadbed sink and guaranteeing the road behavior in practice.关键词：边坡工程; anfis模型; 变形预测key words: road engineering;anfis model;soft clay roadbed;settlement prediction中图分类号：k826.16 文献标识码：a 文章编号中图分类号：文献标识码：文章编号：0 引言现代建（构）筑物建设中，经常会碰到许多高填深挖的边坡问题，边坡失效往往会给工程建设造成难以弥补的损失。
注塑模具英文文献
Employing current design approaches for plastic parts will fail to produce the true minimum manufacturing cost in these cases.
Minimizing manufacturing costs for thin injection
molded plastic components
1. Introduction
In most industrial applications, the manufacturing cost of a plastic part is mainly governed by the amount of material used in the molding procend the part deformation after molding [12], analyzing the effects of wall thickness and the flow length of the part [13], and analyzing the internal structure of the plastic part design and filling materials flows of the mold design [14]. Reifschneider [15] has compared three types of mold filling simulation programs, including Part Adviser, Fusion, and Insight, with actual experimental testing. All these approaches have established methods that can save a lot of time and cost. However, they just tackled the design parameters of the plastic part and mold individually during the design stage. In addition, they did not provide the design parameters with minimum manufacturing cost. Studies applying various artificial intelligence methods and techniques have been found that mainly focus on optimization analysis of injection molding parameters [16,17]. For in-stance He et al. [3] introduced a fuzzy- neuro approach for automatic resetting of molding process parameters. By contrast , Helps et al. [18,19] adopted artificial neural networks to predict the setting of molding conditions and plastic part quality control in molding. Clearly, the development of comprehensive molding process models and computer-aided manufacturing provides a basis for realizing molding parameter optimization [3 , 16,17]. Mok et al. [20] propose a hybrid neural network and genetic algorithm approach incorporating Case-Based Reasoning (CBR) to derive initial settings for molding parameters for parts with similar design features quickly and with acceptable accuracy. Mok’s approach was based on past product processing data, and was limited to designs that are similar to previous product data. However, no real R&D effort has been found that considers minimizing manufacturing costs for thin plastic components. Generally, the current practical approach for minimizing the manufacturing cost of plastic components is to minimize the thickness and the dimensions of the part at the product design stage, and then to calculate the costs of the mold design and molding process for the part accordingly, as shown in Fig. 1. The current approach may not be able to obtain the real minimum manufacturing cost when handling thin plastic components. 1.2Manufacturing requirements for a typical thin plastic component As a test example, the typical manufacturing requirements for a thin square plastic part with a center hole, as shown in Fig. 2, are given in Table 1.
Neuro-fuzzy Systems(A. Tettamanzi, M. Tomassini)
Fuzzy Neurons
y = g(w.x) y = g(A(w,x))
Instead of weighted sum of inputs, more general aggregation function is used Fuzzy union, fuzzy intersection and, more generally, s-norms and t-norms can be used as an aggregation function for the weighted input to an artificial neuron
Off-line – adaptation On-line – algorithms are used to adapt as the
system operates
Concurrent
– where the two techniques are applied after one another as pre- or post-processing
w ij
An Example: NEFPROX
NEuro Fuzzy function apPROXimator Three-layer feedforward network (no cycles in the network and no connections exist between layer n and layer n+j, with j>1 input variables / hidden layer - fuzzy rules / output variables Hidden and output units use t-norms and tconorms as aggregation functions The fuzzy sets are encoded as fuzzy connection weights and fuzzy inputs
围岩位移特性曲线预测的自适应神经模糊推理方法在软岩巷道中的应用
价值工程0引言近年来，随着矿产开采越来越向深部发展，软岩巷道问题便越来越突出。
据统计，我国目前矿产开采每年巷道掘进上万千米，软岩巷道约占其中的1/10。
由于软岩支护的问题，大约有1/6的软岩巷道需要返修、维护，所以能否解决好软岩巷道的支护等问题，是我国矿开采向纵深发展和安全生产的关键问题之一。
软岩巷道稳定既是一个强度稳定性，又是一个变形稳定性，而且往往表现在因变形过大而引起巷道稳定问题。
目前，软岩巷道主要采用新奥法施工技术，分两次支护。
初次喷锚支护加强围岩自承能力，然后通过对变形的观测确定最佳二次支护时间，最佳支护时机的确定有利于稳定围岩和节约支护材料，因此寻找和确定最佳支护时机，成为巷道工程支护技术的核心。
所谓最佳支护时机，就是最大限度的调动围岩变形释放能量，最大程度的发挥围岩自承载能力，充分调动支护体系的作用，使二次支护系统的抗力降为最低同时保证支护材料的使用最为合理。
因此，建立一种科学的方法能够准确确定围岩的位移特性曲线，并据此选择合适的支护时间和衬砌刚度是确保围岩和巷道衬砌稳定的关键。
研究表明，自适应神经模糊推理系统具有拟合能力强、预测精度高、收敛速度快、训练结果唯一等特点。
基于此，作者研究和建立了自适应神经模糊推理系统。
虽然该方法经初步检验具有收敛速度快、解的稳定性好、优化性能好等优点，但其工程适用性仍有待验证。
因此，本文应用自适应神经模糊推理系统建立了围岩位移特性曲线预测的自适应神经模糊推理方法，对水口山矿务局康家湾铅锌矿巷道围岩位移特征曲线进行了预测，并据以对巷道施工过程中的支护时间，支护刚度进行了分析。
分析结果表明，该方法具有良好的工程适用性，可在工程实践中推广应用。
1巷道围岩位移特性曲线预测问题的数学描述在平面应变状态下，已知巷道围岩的力学参数向量{b}={σx ，σy ，E ，μ，c ，准}来求巷道顶帮的竖向位移向量{δ}={δ1，δ2，δ3，δ4，δ5}，其中δ1，δ2，δ3，δ4，δ5是巷道围岩位移特性曲线上的5个点，这些点所对应的支护反力为p 1，p 2，p 3，p 4，p 5，它们分别是p 1=47σy ，p 2=37σy ，p 3=27σy ，p 4=17σy ，p 5=114σy ，求解该问题的思路是：给定n 个力学参数向量{b i }={σxi ，σyi ，E i ，μi ，c i ，准i }（i=1,2,…n ），可以通过数值计算，计算p 1i =47σyi ，p 2i =3σyi ，p 3i =2σyi ，p 4i =1σyi ，p 5i =1σyi 所对应的巷道顶帮的竖向位移δ1i （47σyi ），δ2i （37σyi ），δ3i （27σyi ），δ4i （17σyi ），δ5i（114σyi ），这样就得到了n 个模拟位移向量{δi s }={δ1i s ，δ2i s ，…，δ5i s }（i=1，2，…，n ）。
fuzzy estimation method
文章标题：深度剖析模糊估计方法1. 引言在现代科学和工程领域中，面对复杂系统和模糊问题的时候，我们常常需要借助于模糊估计方法来处理。
模糊估计方法是一种能够处理不确定性和模糊性的数学工具，其在决策分析、控制系统、模式识别等领域有着广泛的应用。
2. 模糊估计方法的基本原理模糊估计方法是基于模糊集理论的，模糊集是一种能够描述不确定性和模糊性的数学工具。
在模糊估计方法中，我们首先需要对问题领域进行模糊化，然后利用模糊集的运算规则进行模糊推理，最终得到模糊的估计结果。
模糊估计方法的基本原理是将模糊集的概念应用于估计过程，以应对现实世界中的模糊和不确定性。
3. 模糊估计方法的应用领域模糊估计方法在实际应用中有着广泛的应用，例如在工程领域中，我们可以利用模糊估计方法来处理复杂系统的控制问题；在经济领域中，模糊估计方法可以用来进行风险评估和投资决策；在医学领域中，模糊估计方法可以用来处理疾病诊断和治疗方案设计等。
4. 模糊估计方法的优缺点模糊估计方法的优点在于能够处理不确定性和模糊性，对于复杂系统和问题有着良好的适用性；但其缺点在于需要准确地建立模糊集的隶属函数和运算规则，且在实际应用中需要大量的专业知识和经验。
5. 个人观点和总结作为一种处理模糊问题的数学工具，模糊估计方法在现实世界中有着重要的应用意义。
但在实际应用中，我们需要充分了解问题的领域特点，结合专业知识和经验来准确地建立模糊集和运算规则，以获得可靠的估计结果。
随着人工智能和大数据技术的发展，模糊估计方法也在不断得到改进和拓展，将为解决更多复杂问题提供更加有效的数学工具。
以上是我对模糊估计方法的个人理解和观点，希望能够对您有所帮助。
6. 模糊估计方法的发展历程模糊估计方法作为一种处理模糊和不确定性问题的数学工具，其发展历程可以追溯至上世纪50年代。
最早提出模糊集的概念是由日本学者石川明于1965年提出的，其将不确定性因素引入了数学模型中。
随后，模糊估计方法被广泛应用于控制理论、人工智能、决策分析等领域，且在不断得到改进和拓展的过程中，逐渐成为一种重要的工程工具。
基于神经网络的智能刀具状态检测系统（英文）
AN INTELLIGENT TOOL CONDITION MONITORING SYSTEMUSING FUZZY NEURAL NETWORKSZhao DongbiaoDepa rtme nt of Mechanica l Engineering,N UAA 29Yuda o Str eet,Nanjing 210016,P.R.ChinaKesheng Wang Oliver K rimmelDepa rtment of Pr oduction and Quality Engineering,Norwegian University of Scie nce and Technology,7034Trondheim ,Nor wa yABSTRACTReliable on-line cutting tool conditioning monitoring is an essentia l featur e of a utomatic machine tool and flexi-ble manufac turing system (FM S )and computer integra ted manufacturing system (CIM S).Recently a r tificial neura l networ ks (ANNs)ar e used for this pur pose in conjunc-tion with suita ble sensory systems.The pr esent wor k in Nor wegia nUniversityofScienceandTechnology(N TN U)uses back-pr opaga tion neur al networ ks (BP )and fuzzy neura l ne tworks (FNN )to pr ocess the cut ting tool state data m ea sur ed with for ce and acoustic emission (AE )sensors ,and implements a valuable on -line tool condition monitoring system using the ANNs.Different ANN st ructures ar e designed and investigated to estimate the tool wea r sta te based on the fusion of acoustic emis-sion and force signals .Finally ,four ca se studies a re in-tr oduced for the sensing a nd ANN processing of the tool wear sta tes and th e failures of the tool with practical ex-perime nt exa mples .The r esults indicate tha t a tool -wea r identifica tion system can be achieved using the sensors in-tegration with ANNs,and that ANNs provide a ve ry ef-fective method of implementing se nsor integra tion for on-line monitoring of tool wear states and abnormalities .K y ;;fuzzy logic;acoustic emission;force sensor;fuzzy neural ne tworksINTRODUCTIONRecently ,FMS and CIM S ,which replace human operators with a robotic counterpart in manufacturing or assembly cells,have been con-cerned in industrial field .The absence of huma n operators requires that the process-monitoring function controlled by sensors associated with a decision-making system is able to analyze the sensory inputs a nd make an a ppropria te control action.Reliable on-line cutting tool conditioning monitoring is an essential feature of automatic machine tool a nd of FMS and CIM S to gua rantee that the tool is in good condition while in use ;therefore,automatic and rapid detection of tool breakage must be integral to FMS and CIM S .Recently ANNs a re used for this purpose in con-junction with suitable sensory systems.A large number of research papers have been published on the topic of tool wear monitoring o ver the last 20years and have concentratedS y S F f (5)R ;5S e wor ds :tool condition monitoring neural networks u pport ed b Aeronauti c ci ence o u ndation o C hi na 97H 2072.eceived 7J un.2000revisi on recei ved ep .2000mainly on turning,milling and drilling[1,2].Ref.[3]observed AE signals with large amplitudes when tool cracking,chipping or fracture took place during interrupted cutting.An approach was developed for in-process monitoring of tool wear in milling using frequency signatures of the cutting force[4].Ref.[5]used AE signals to monitor tool wear in face milling.It was found that both AE root mean square(RMS)a nd a special function of cutting force a nd cutting coef-ficients shown an increasing trend with the growth of flank wear.Advanced signal process-ing techniques and ar tificial neura l networks have been widely used in the development of tool con-dition monitoring systems.Ref.[6]demonstrat-ed the feasibility of using neural networks for sensor integration in tool wea r monitoring tasks. The networks are used as learning and pattern recogn ition devices,and found to be suitable for integrating acoustic emission and cutting force signals to assess and predict tool wear in machin-ing operations.Ref.[7]developed a fuzzy driven neural network based pa ttern recognition algo-rithm to accomplish multi-sensor information in-tegration and tool wea r sta te classifica tion.Ref.[8]developed a fuzzy logic inference by monitor-ing the cutting force signal for achieving the ob-jective of in-process detecting tool breakage sys-tem on an uneven surface in end milling opera-tions.Ref.[9]used cutting forces signals and a-coustic emission as inputs to a feedforward back-propagation neural network a nd evaluated the ef-fectiveness of some training techniques.The present work in N TNU uses back-prop-aga tion neural networks and fuzz y neural net-works to process the cutting tool state data mea-sured with force sensor and acoustic emission sensor,and implements a valuable intelligentyN N D ff NNf f f signals.Finally,four ca se studies a re introduced for the sensing and ANN processing of the tool wear states a nd tool failures with practical exper-iment examples.The results indicate tha t a tool-wear identification system can be achieved using the sensors integration with ANNs,a nd tha t ANNs provide a very effective method of imple-menting sensor integration for on-line monitoring of tool wea r states and abnormalities.1 EXPER IMENTAL CONDI-TIONS AND SAM PLE DATAThe experiments are conducted on a conven-tional CNC machine tool.The samples for the collection of the force and AE signals are cylin-ders of free-cutting steel9SMn28K(DIN). The tools used in the experiments are sta nda rd twist drills according to DIN338with a soft MoS2coating a nd a TiN coating respectively. The diameters of the drills a re4mm a nd10mm, as shown in Table1.The cutting pa ra meters for the samples a re shown in the Table2.Higher cutting speeds and lower feeds a re used in order Table1 Types of dr ills used in the exper imentsDia meter MoS2TiN4mm10mmTable2 Cutting speed and f eed f or the9SM n28K samples4mm10mm MoS2-andTiN-coa teddrillsRotationalspeed2400/(r·min-1)1000/(r·min-1)Feed0.15/(mm·r-1)0.3/(mm·r-1).T yy j y fFtool condition monitoring s stem using the A s.i erent A structures are designed and investigated to estimate the tool wear state based on the usion o acoustic emission and orce to accelerate tool wear he have been continu-ousl ad usted to avo id too earl breakage o the tool.or the same drill diameter the same cut-ting parameters were used in order to derive con-clusions of the influence of the coa ting alone .The tools in the same diameter used are absolute-ly identical apa rt from the coating.The tests are run without cooling fluid .The AE sensor used is a Kistler 8154B 1sen-sor with a frequency range of 50Hz ～400k Hz .The force sensor is a standa rd one -dimensiona l force sensor Kistler dynamometer measuring the drilling force in z -direction.The cylindrical test samples are in diameter of 20mm.The depths of drilled holes are 20mm for 4mm drills a nd 50mm for 10mm drills.The surface is cold drawn to assure repea table contact conditions between the jaws and the workpiece .The samples are centred prior to drilling to avoid wandering of the drill.The AE-sensor is applied on a steel ba r that was clamped between the sample and the clamping jaw of the vice .A v -groo ve is milled in-to the steel bar to assure a consta nt distance be-tween AE-sensor a nd sample.The vice is mount-ed on a Kistler dynamometer that was used to log the feed force.The feed force and the acoustic e-mission a re registered in a da ta logger at the rate of 5ms .This leads to 650groups of data per hole for the 4mm drills and 2600groups of data for the 10mm drills.In addition to the root-mean-square of acoustic emission AERMS signa l stored in the data base together with the cutting force signal,the raw signal is monitored on an online frequency analyzer.The mea sured force and AERMS data are shown in Figs.1,2,respectively,only for 4mm MoS 2-coa ted in this paper.Other three cases of 4mm TiN-coated,10mm MoS 2-coated,and10mm TiN-coated drills a re similar to those inFigs .1and 2,not given in the paper .These cha rts give surfaces over time a nd the depth of the hole.The Fig.3shows the averages of the f f RMS f 4M S T f y ,y x,Fig.1 Cutting for ce for 4mm Mo S 2-coa teddrillFig.2 AE RMS signal sur faces for 4mmMoS 2-coa teddrillFig .3 Average of cutting force and AE RMS for4mm Mo S 2-coa ted drillgiven in the Figure.The stage 1represents tha t the drill is a fresh one or a little -worn one .The f y T 3f f y T No 171.2Zha o D ongbiao,et a l.An Intelligent Tool Condition Monitoring System Using ……eed orces and the AE values or the mm o 2-coated drills.he tool wear stages oreach t pe drill evaluated b the e perimenters according to their worn characteristics are alsostage 2is or the drill worn moderatel .he stage is or the drill worn signi icantl and can be still used.he stage 4means that the drill isworn out thoroughly.2 AN INTELLIGENT CUT CON-DITION MONITOR ING SYS-TEMFuzzy logic is a technology that mimics the human decision-making process on the very high abstraction level of na tural language.On the contrary,neural networks try to copy th e way a human brain works on the lowest level,the `hardware′level.The key benefit of fuzzy logic is tha t it lets you define the desired system be-haviour with simple`IF-THEN′relations.In many applica tions,this gives you a simpler solu-tion in less design time.In addition,you can use all available engineering know-how to optimise the performance directly.Each neuron in a neural network transforms the incoming inputs into an output.The output is then linked to other neurons.Some of the neu-rons form the interface of the neural network. The informa tion enters the neural network a t the input layer.All la yers of the neural network pro-cess these signals th rough the network until they reach the output.The objective of a neural net-work is to process the information in the way in which it has been trained.Training involves ei-ther sample da ta sets of inputs and corresponding outputs or a teacher who ra tes the performance of the neura l network.For this training,neural networks use′lea rning algorithms′.The lea rn ing algorithms modify the individual neurons of the network a nd the weight of their connections in such a way tha t the behaviour of the network re-flects the desired behaviour.Both neural networks and fuzzy logic are powerful design tech n iques that have their strengths and weaknesses.Neural networks can lea rn from da ta sets while fuzz y logic solutions are easy to verify and optimise.If you look at these properties in a comparison Table(Table 3),it becomes obvious that a clever combination of the two techniques delivers the best of both worlds.A combina tion of the explicit knowledge representa tion of fuzzy logic with the lea rning power of neural networks is implemented,as shown in Fig.4.Table3 Str engths a nd weaknesses of f uzzy logic and neur al networ ksFuzzy logic Neural networ ksKnowledge r epresentationLinguistic repr esenta tion Bla ck box representationExper t knowledge requiredExample da ta or per formancefunction required Explicit,ve rifica tion and optimisa tionar e ea sy and efficient(★★★)Implicit,the syst em ca nnot be easyinterpr eted or modified(×)Tr ainability None,you have to defineeverything explicitly(×)Trains itself by lea rning f rom datasets(★★★)Adapta tionSome adapta tion A daptation mecha nisms availableFa ult tolerant Fault tolera ntMultiple description possible Multiple description possible172Transac tions of Nanjing University of Aeronautics&Ast ronautics Vol.17Fig.4 St ructure model of fuzzy neural ne twork for cutting tool condition monitoring The intelligent fuzzy neural network for manufacturing process monitoring uses the fea-ture selection and extraction from acoustic emis-sion and force signals sampled on the drilling processing to estima te the tool wea r states a nd the failures of the tool,shown as in Figs.1,2. The inputs for the training of the FNN are the parameters of the processing:the drill′s diame-ter,coating type,feed speed,spindle speed, dry/lubrica tion,material of the workp iece;the parameters of cutting state:No.of the hole,the drilling depth of the hole;samples measured on-line:cutting force,acoustic emission signal;a nd pre-processing da ta of the measured signals:the average of forces and AE RMS for every drilled seg-ment of a hole.The output is the wear stage of drills.The process of training in fuzz y neural net-works consists of modifying their parameters by presenting them with examples of their past ex-perience,typically by adjusting the weigh ts of the networks so that a certain performa nce indexT q ff qf x x network maps inputs into the corresponding tar-get values of the output.Synaptic connections in a crisp neuron are linear gains multiplying in-puts.Any adaptation or lea rn ing occurring with-in an individual neuron involves modifying the value of these gains by adjusting weight.For a fuzzy neuron,synaptic connections are repre-sented as a two-dimensional fuzz y relation be-tween synaptic inputs a nd outputs.Hence, learning in fuzz y neuron,in the most general case,involves changing a two-dimensional rela-tion surface at each synaptic.The determination of the membership func-tions of all the features for each model a nd the construction of FNN for classification mark the end of the lea rning stage.After training the fi-nal structure and associated weigh ts of the FNN can reflect the distinct importance of every input for each model under specific cutting conditions. So the future tool wear classification results ca n be reliable and accurate.3 R SULTSf f FNNNo173 .2Zha o D ongbiao,et a l.An Intelligent Tool Condition Monitoring System Using……is optimised.his re uires that a collection o input-output pairs be speci ied and also re uires a per ormance inde that e presses how well theEA ter constructing and training o thefor drill condition monitoring ,we ca n use FNN to evaluate (classify )the wea r stages with the sample data of four type drills.The figures from 5to 8show the results computed with the FNNfor the four drills:4mm MoS 2-coa ted ,4mm TiN-coa ted,10mm MoS 2-coa ted,and 10mm TiN-coa ted.The results evaluated by a back-propagation (BP )network a re also shown in thefigures for comparison with those by FNN a nd experts.We ca n figure out the differences be-tween the wear stages computed by FNN and ac-cessed by experts .The wea r stages computed by FNN go nea r to those by exper ts though the wear steps by ex-perts a re not reflect the real wear sta tes .In con-trast,the results by the BP network are much smooth and actual than those by FNN in despite of bigger er ror between the results and the wea r steps given by experts .The FNN can give more accurate assessment if the more actual wea r stages given by experts as train ing targets for the FNN are available.Fig .5 Compa rison of wear stages computed a nd setby manua l for 4mm MoS 2-coa ted drill4 CONCLUSION SAn intelligent tool condition monitoringsystem has been established a nd evalua ted based on fuzzy neural networks.The system utiliz ed (f )f zzyT Fig.6 Com parison of wea r stages computed and setby ma nual for 4mm TiN-coa teddrillFig.7 Com parison of wea r stages computed and setby ma nual for 10mm MoS 2-coa teddrillFig .8 Com parison of wea r stages computed and set by ma nual for 10mm TiN-coated drillshown as follows :(1)AE and cutting force sensors are appli-cable for monitoring tool wea r in metal cutting process .The healthy signals picked up by these sensors describe tool condition comprehensively .(2)T f NN f zzy f y f f f x y f x 174Transac tions of Nanjing University of Aeronautics &Ast ronautics Vol.17multiple sensors acoustic emission and orce and a u neural network as the learning and decision -making component .he results arehe combination o A and u logic integrates the strong learning and class i ication abilit o the ormer and superb le ibilit o the latter to e press the distribution characteristicsof signal features with vague bounda ries.(3)The intelligent tool condition monitoring system has the advantages of being suitable for different machining conditions.It is robust to noise and tolerant of fault.(4)It can reach a high success rate for re-cognizing tool wear under a range of process con-ditions.REFERENCES1 Byr ne G,D or nfeld D,Inasaki I,et a l.Tool condition monitoring(TCM)——The sta tus of r esea rch a nd in-dustrial a pplication.Annals of the CIRP,1995,44(2):541～5672 Leopold J,Gunther H.Quality control of the cutting process with neural nets.Ma nufacturing Systems, 1996,25(1):81～863 Moriwaki T.Application of a coustic emission mea-sur ement to sensing of wear and breakage of cutting tool.Bull Japan Soc of Prec Eng,1983,17(3):154～1604 Elbestawi M A,Papa zafiriou T A,Du R X.In-pr o-cess monitoring of tool wea r in milling using cutting force signatur e.Int J Mach Tools Manuf,1991,31(1):55～735 D iei E N,Dornfeld D A.Acoustic emission se nsing of tool wea r in face milling.Trans ASM E J of Engg Ind,1987,109/235:234～2406 Rangwala S,Dornfeld D.Sensor integration using neural ne tworks for intelligent tool condition monitor-ing.Tra nsactions of the ASME:J ournal of Engineer-ing for Industr y,1990,112:219～2287 Fu P,H ope A D,King G A.Intelligent tool wear monitoring of machine tools.In:COM ADEM Pr o-ceedings,1998,279～2858 Huang P T,Chen J C.Fuzzy logic-base tool breakage detecting system in end milling opera put-er s Ind Engng,1998,35(1-2):37～409 Jemielniak K,Kwiatkowski L,Wr zosek P.D iagnosis of tool wear based on cut ting forces and acoustic e-mission measur es as inputs to a neural network.J our-nal of Intelligent Ma nufacturing,1998,(9):447～455基于神经网络的智能刀具状态检测系统赵东标(南京航空航天大学机电工程学院 南京,210016)Kesheng Wa ng Oliver Krimmel(挪威科学技术大学生产与质量工程系 挪威特隆赫姆,7034)摘要 可靠的在线刀具磨损状态检测是柔性制造系统、计算机集成制造系统以及自动化机床必不可少的一个环节。
巷道围岩位移特性曲线预测的自适应神经模糊推理方法及应用
巷道围岩位移特性曲线预测的自适应神经模糊推理方法及应用李 明,刘 永(南华大学建筑工程与资源环境学院,湖南衡阳421001)摘 要:应用自适应神经模糊推理系统的原理,建立了巷道开挖围岩位移特性曲线预测的自适应神经模糊推理方法。
应用该方法,预测了水口山矿务局康家湾铅锌矿十二中段排泥巷道围岩的位移特性曲线,并据此对实际施工过程中出现的衬砌破坏、合理的支护时间和衬砌刚度进行了分析,结果表明,该方法具有良好的适用性。
关键词:巷道开挖;围岩位移特性曲线;自适应神经模糊推理系统中图分类号:T D353 文献标识码:A 文章编号:1008-4495(2005)03-0049-03 收稿日期:2004-11-11作者简介:李 明(1971-),男,江西吉水人,硕士,讲师,主要从事采矿工程灾害预测与控制研究。
巷道支护一直是地下矿山生产及建设的重大难题之一。
巷道掘进时,在岩土体中形成了新的空间,致使巷道周边岩土体失去原有的支撑。
在应力释放和应力重新分布过程中,围岩向着巷道硐内产生变形,并可能发生围岩的破坏。
为防止发生过度变形而导致围岩发生严重松弛甚至破坏,需要对围岩进行及时支护。
国内外大量的巷道支护失效的经验教训表明,对巷道围岩的位移特性曲线不能准确把握,因而对支护时间和支护刚度未准确把握是造成巷道失稳的主要原因之一。
因此,建立一种科学的方法能够准确确定围岩的位移特性曲线,并据此选择合适的支护时间和衬砌刚度是确保围岩和巷道衬砌稳定的关键。
最新研究表明,将人工神经网络和模糊逻辑推理相结合构成的自适应神经模糊推理系统ANFIS (Adaptive Neuro -Fuzzy In ference System ),具有收敛速度快、拟合能力强、预测精度高、网络训练结果具有唯一性等特点。
这些正是研究和建立预测巷道开挖围岩位移特性曲线分析方法所需要的。
因此,作者应用自适应神经模糊推理系统的原理,建立了巷道开挖围岩位移特性曲线预测的自适应神经模糊推理方法,并应用此方法对工程实例进行了预测。