For each measurement, a beef fillet sample of 5 g was introduced inside a 100 ml volume glass jar and left at room temperature (20°C ±2°C) for 15 min to enhance desorption of volatile compounds from the meat into the headspace. Pre- processing of the data obtained from enose sensors is required to obtain the “olfactory image” of the sample. This process involves extracting certain significant characteristics from the sensor response curves in order to produce a set of data that can be processed by the recognition system of the enose. Different features can be extracted and used depending on the characteristics of the enose used such as the type of sensors adopted, and the stability of the responses of the latter to the reference gas, to variations in humidity and temperature levels. The responses of all sensor signals classes for meat samples stored at 4°C are shown in Fig. 3. Considering that each measurement can be represented as a point in an 8-dimensional space, a dimensionality reduction algorithm has been applied on those enose data used for training purposes. The robust PCA (RPCA) scheme has been utilized to obtain principal components that are not influenced much by outliers. RPCA scheme was implemented in MATLAB, with the aid of PLS_Toolbox (ver. 8.0 Eigenvector.com).
Abstract—Freshness and safety of muscle foods are generally considered as the most important parameters for the food industry. To address the rapid detection of meatspoilage microorganisms during aerobic or modified atmosphere storage, an electronicnose with the aid of fuzzy wavelet network has been considered in this research. The proposed model incorporates a clustering pre-processing stage for the definition of fuzzy rules. The dual purpose of the proposed modelling approach is not only to classify beef samples in the respective quality class (i.e. fresh, semi-fresh and spoiled), but also to predict their associated microbiological population directly from volatile compounds fingerprints. Comparison results against neural networks and neurofuzzy systems indicated that the proposed modelling scheme could be considered as a valuable detection methodology in food microbiology.
A prototype of electronicnose (e-nose) with the gas sensor system for evaluation of fresh chicken meat freshness was developed. In this paper a rapid, simple and not expensive system for fresh chicken meatspoilage detection was investigated that provides objective and reliable results. Quality changes in fresh chicken meat during storage were monitored by the metal oxide sensor (MOS) system and compared with the results of traditional chemical measure- ments. Gas sensor selection was tested for evaluation of volatile fatty acids (VFA) mainly representing meatspoilage. The study demonstrated that a correlation coefficient (R 2 = 0.89) between e-nose signals and traditional chemical
their potential in assessing meat quality. In recent years, spectral imaging (i.e., hyperspectral and multispectral) has been also considered as an alternative tool for safety and quality inspection of various agricultural products . This technique integrates the conventional imaging and spectroscopy technique to attain simultaneously both spatial and spectral information from the target product. The “mechanism” of these approaches is based on the assumption that the metabolic activity of micro-organisms on meat results in biochemical changes, with the simultaneous formation of metabolic by-products, which may contribute to the spoilage phenomenon. The quantification of these metabolic activities corresponds to a unique “signature”, providing thus information about the type and rate of spoilage . The huge amount of information provided by analytical sensors/devices requires an advanced data analysis approach. This has been achieved through the integration of modern analytical platforms with computational and chemometric techniques . Multivariate statistical analyses (e.g., partial least square (PLS) regression, discriminant function analysis (DFA), cluster analysis) have resulted in the development of decision support systems for timely determination of safety/quality of meat products . Considering that microbial meatspoilage is a complex process, which involves growth of microorganisms during storage, their spectra contain highly non-linear characteristics. Hence, linear-based techniques might not provide a complete solution to such complex identification/classification problem . Neural networks (NNs) have gained much interest in predictive engineering and quantitative modelling due to their flexibility and high accuracy as compared to other modelling techniques (e.g., statistical models). In comparison to other NN-based application areas, the field of food science is still in an early development stage. Recently, advanced NN algorithms have shown promising results in applications such as growth parameter estimation of microorganisms . NNs usually require a large number of neurons for solving the majority of approximation problems and are prone to dimensionality problems, as each single neuron- node cannot define a multi-dimensional hyper-sphere of the input domain. Although fuzzy logic systems, provide such input space mapping, they do not have learning ability, thus it is difficult to analyse complex systems without prior and accurate knowledge on the system being analysed .
The idea of the learning algorithm, see , is to create a rule base first and then to refine it by modifying the initially given membership functions (usually fuzzy partitions of in- put and output variables). The rule base is created by finding for each pattern in the training set a rule that best classifies it. If a rule with an identical antecedent is not already in the rule base, it will be added. The learning algorithm of the mem- bership functions uses the output error that tells whether the degree of fulfillment of a rule has to be higher or lower. This information is used to change the input fuzzy sets by shifting the membership functions and making their supports larger or smaller. There are defined diﬀerent shapes for fuzzy mem- berships (triangles, bells, etc.), and all of them can be easily modified with parameters. Constraints are defined here for the learning procedure (fuzzy sets must not pass each other or that they must intersect at 0 . 5, etc.), in order to obtain an interpretable rule base. After the rule learning algorithm has terminated, the predefined fuzzy partitioning on both input variables defines a partitioning of the input space created by overlapping hyperboxes where each hyper-box is formed by the Cartesian product of the supports of the defined fuzzy sets (the number of sets is predefined). Each hyper-box rep- resents the support of an n-dimensional fuzzy set which is the antecedent of a fuzzy rule.
This means the sniffing process continued sequentially and the pump in jected the vapour from e ither the s ample vessel or reference vessel alternately one at a time. We had found experimentally that pumping from re ference vessel vapour for 600 s duration is sufficient to a llo w the sensors to return to their baseline between two consecutive cycles of tea aro ma sniffing. This ensures that the EN system only responds to the tea aroma rather than any residual smell fro m the surroundings. Figure 3 shows a response curve obtained using Lab VIEW software (National Instruments Inc.) during EN sniffing of tea samples. It can be observed that the curve increases exponentially during the sniffing of the tea samples and then decreases to a level commensurate with the level of response determined by the odour in the reference vessel. The sensor responses were continuously monitored and stored in data files for the purpose of subsequent data processing off line. The resulting dataset comprised 260 record ings.
Due to the emerging requirement for a reliable, real-time, low-cost, and portable device to measure, detect, and classify volatile compounds and odors, the research community has developed instruments to mimic the human olfactory system . At the present, these instruments, usually called electronic noses (e-noses), have been gaining favor in several industrial applications , . Typically, the e-nose comprises three main functional components: a sampling unit, a signal processing unit, and an odor classification unit. The function of the odor classification unit is to correctly identify or assign an unknown odor sample to a true class. Although the traditional odor classification algorithms work well for simple classification tasks, their performance can be highly degraded for applications that have a large number of odor classes and for those with highly correlated datasets. In some e- nose machines, a specialized signal-processing unit is utilized to improve the classification performance . Unfortunately, this improved performance is still inadequate in some practical applications which require near perfect precision such as in medical diagnosis. In addition, we know that there is no single method or algorithm that is superior to others in all applications. Consequently, other techniques are needed to solve that kind of problem.
This paper presents the development and evaluation of different versions of neuro-fuzzy model for prediction of spike discharge patterns. The author aims to predict the spike discharge variation using first spike latency and frequency-following interval. In order to study the spike discharge dynamics, the author analyzed the cerebral cortex data of the cat. Adaptive neuro-fuzzy inference systems (ANFIS), Wang and Mendel, dynamic evolving neural-fuzzy inference system, hybrid neural fuzzy inference system, genetic for lateral tuning and rule selection of linguistic fuzzy system (GFS.LT.RS) and subtractive clustering and fuzzy c-means algorithms are applied for data. Among these algorithms, ANFIS and GFS.LT.RS models have better performance. On the other hand, ANFIS and GFS.LT.RS algorithms can be used to predict the spike discharge dynamics as a function of first spike latency and frequency with a higher accuracy compared to other algorithms. Key words:
University of Jammu Campus, India Abstract- Data clustering is a well known technique for fuzzy model identification or fuzzy modelling for apprehending the system behavior in the form of fuzzy if-then rules based on experimental data. Fuzzy c- Means (FCM) clustering and subtractive clustering (SC) are efficient techniques for fuzzy rule extraction in fuzzy modeling of Adaptive Neuro-fuzzy Inference System (ANFIS). In this paper we have employed a novel technique to build the rule base of ANFIS based on the kernel based variants of these two clustering techniques which have shown better clustering accuracy. In kernel based clustering approach, the kernel functions are used to calculate the distance measure between the data points during clustering which enables to map the data to a higher dimensional space. This generalization makes data set more distinctly separable which results in more accurate cluster centers and therefore a more precise rule base for the ANFIS can be constructed which increases the prediction performance of the system. The performance analysis of ANFIS models built using kernel based FCM and kernel based SC has been done on three business prediction problems viz. sales forecasting, stock price prediction and qualitative bankruptcy prediction. A performance comparison with the ANFIS models based on conventional SC and FCM clustering for each of these forecasting problems has been provided and discussed.
The first step in the training phase is feature extraction. Where the GM and CM are sued to extract 18 features form each single train image. Then these features stored in the database of the system to be used in the training stage. The NN designed according to the set of input data and the size of the desired outputs. After that, the designed NN will be training to recognize the image stored in the database of the system. In the testing stage, the sets of test image that used to test the accuracy of the designed NN. The first step in the testing phase is the feuture extraction by using GM and CM. Then the extracted features used in testing stage to find in they match with feature that stored in database of the system.
Takagi-Sugeno-Kang fuzzy inference method also known as Sugeno fuzzy inference method has been used. Sugeno method works same as the Mamdani method but generates either linear or constant outputs. The Sugeno type Fuzzy Inference System is designed with 4 different inputs, an output and rules based on the classification-regression decision tree analysis output structure. The Gaussian curve member function is used for inputs and constant membership function is used for the output. Gaussian membership functions are used for input values because Gaussian membership functions are most adequate in representing uncertainty in measurements . A Gaussian membership function is a characteristic symmetric bell curve produced by bell function, which tend to fall towards zero. From , Gaussian function can be represented in the following form:
The difficulty that arises when using ANFIS as a classifier is due that, in fact, it is a Function Approximator with one unique output, and in order to use it for other applications (e.g. prediction, classification, control, etc.) requires certain subterfuges . In , they have to use five ANFIS systems, four of them trained with the back propagation gradient descent method in combination with the least squares method. In reference  it is proposed a Multiple Instance ANFIS for realizing the described applications. In  it is used the same 5-ANFIS technique as in . In reference  the method employed is based on Independent Component Analysis (ICA), Power Spectrum and ANFIS; finally, in , the structure developed consists of six binary ANFIS classifiers.
ABSTRACT: This paper presents an approach of direct integration scheme for wind energy conversion systems using a capacitor-clamped three-level inverter-based supercapacitor. The main idea of this paper is to increase the capacitance of the clamping capacitors with the use of supercapacitors. The supercapacitor voltage is varied within a defined range. This variable voltage changes brings about many challenges. The uneven distribution of the space vector is the first challenge. To overcome this, space vector modulation technique is proposed. This generates undistorted currents even in the presence of dynamic changes in supercapacitor voltages. The harmonics are reduced by replacing conventional PI controllers with the Neuro-Fuzzy controllers which give better performance. The control strategies of the proposed system are discussed in detail. Simulation results are presented to study the efficacy of the proposed system in reducing wind power fluctuations.
FPGA basedFuzzy Logic Control for Single Phase PWM Multilevel Inverter (MLI) produces AC output voltage of desired magnitude and frequency . Here, the objective of reducing the THD of output under steady-state as well as set point tracking with fast transient response is approached from control point of view. A FLC is developed using Matlab-Simulink and implemented using FPGA. In this paper, direct torque control (DTC) of induction motor (IM) using intelligent techniques: ANN and FLC is purposed .The results of Matlab-Simulink simulations are presented which shows that the direct torque control usingfuzzy logic scheme is best for control of induction motor. The model used consists of an adaptive neuro-fuzzy inference system modified for efficient HW/SW implementation . The designs of two different on-chip approaches are presented. The device contains an embedded-processor core and a large FPGA.
automatically learn and adapt. Hybrid systems have been used by researchers for modelling and predictions in various engineering systems. The basic idea behind these neuro-adaptive learning techniques is to provide a method for the fuzzy modelling procedure to learn information about a data set, in order to automatically compute the membership function parameters that best allow the associated FIS to track the given input/output data. The membership function parameters are tuned using a combination of least squares estimation and back- propagation algorithm for membership function parameter estimation. These parameters associated with the membership functions will change through the learning process similar to that of a neural network. Their adjustment is facilitated by a gradient vector, which provides a measure of how well the FIS is modelling the input/output data for a given set of parameters. Once the gradient vector is obtained, any of several optimization routines could be applied in order to adjust the parameters so as to reduce error between the actual and desired outputs. This allows the fuzzy system to learn from the data it is modelling. The approach has the advantage over the pure fuzzy paradigm that the need for the human operator to tune the system by adjusting the bounds of the membership functions is removed.
ANFIS was first developed by J.S Roger in the year 1993 by combining fuzzy logic system and neural network . The ANFIS is a form of neural network that functions like the Sugueno-type “ÍF….THEN” fuzzy inference system rule being a network structure and is considered to be more efficient than the individual neural network or fuzzy logic system, it provides more optimal solution than any of the two system . A typical ANFIS structure is presented in Figure 3 with two inputs x and y and one output f, it also consist of five layers with each layer having different function. The ANFIS used for this study comprises of four inputs and a single output. Each of the five layers consist of nodes, the nodes on each layer perform the same functions.
lectrical power systems in modern era are characterized by extensive system interconnections and increasing dependence on control for optimum utilization of existing resources. The supply of reliable and economic electric energy is a major determinant of industrial progress and consequent rise in the standard of living . The growth of the power systems in the future will rely on increasing the capability of existing power transmission systems rather than building the new transmission lines and the power stations for an economical and an environmental reasons. The requirement of the new power flow controllers, which is capable of increasing the transmission reliability and controlling the power flow through the predefined corridors, will certainly increase due to the deregulation of the electricity markets. Additionally, these new controllers must be regulate the voltage levels and the flow of the real and reactive power in the transmission line to use full capability of the system in some cases with no reduction in the system stability and security margins . Flexible Alternating Current Transmission Systems (FACTS) is an evolving technology based solution to help electric utilities fully utilize their transmission assets. The technology was presented in the late of 1980s . FACTS devices enhance the stability of the power system with its fast control characteristics and continuous compensating capability.
The HT12E Encoder IC is used at the end of the sending arduino for Remote Control of the system. They are capable of Encoding 12 bit of information which consists of N address bits and 12-N data bits. Each address/data input is externally binary programmable if bonded out. The HT12D decoder IC is placed before the receiver arduino to decode encoded the signals. For proper operation a pair of encoder/decoder with the same number of address and data format should be selected. The Decoder receive the serial address and data from its corresponding decoder, transmitted by a carrier using an RF transmission medium and gives output to the output pins after processing the data.
In [3, 4] a modified Kohonen clustering network ar- chitecture was introduced in which the neurons were replaced by fuzzy sets and fuzzy rules. Although this network works quite well in many problems of pattern recognition, its learning algorithm is computationally ineffective. In  a new Kohonen network with fuzzy inference based on cosine neighborhood-membership functions and combined learning algorithm based on Kohonen and Grossberg rules was proposed. The main disadvantage of this network is the dependence of clus- tering results on the choice of free parameters of the learning procedure. In , the so-called fuzzy Kohonen clustering network (FKCN) was introduced and later improved in [7, 8, 9]. The learning procedure for this network is based on Bezdek’s fuzzy c-means cluster- ing algorithm , which operates in a batch mode and requires all training data available a priori. Thus, the FKCN cannot be effectively applied to the problems where on-line learning is preferable and the training data must be processed sequentially.