EFFICIENT USE OF HYBRID ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM COMBINED WITH NONLINEAR DIMENSION REDUCTION METHOD IN PRODUCTION PROCESSES

Dhinaharan Nagamalai et al. (Eds) : ITCCMA, CSITY, AIFZ, NWCOM, SIGPRO - 2017
pp. 29– 43, 2017. © CS & IT-CSCP 2017 DOI : 10.5121/csit.2017.70703
EFFICIENT USE OF HYBRID ADAPTIVE
NEURO-FUZZY INFERENCE SYSTEM
COMBINED WITH NONLINEAR
DIMENSION REDUCTION METHOD IN
PRODUCTION PROCESSES
Saad Bashir Alvi, Robert Martin and Johannes Gottschling
Mathematics for Engineer, Duisburg-Essen University, Duisburg
ABSTRACT
This research study proposes a novel method for automatic fault prediction from foundry data
introducing the so-called Meta Prediction Function (MPF). Kernel Principal Component
Analysis (KPCA) is used for dimension reduction. Different algorithms are used for building the
MPF such as Multiple Linear Regression (MLR), Adaptive Neuro Fuzzy Inference System
(ANFIS), Support Vector Machine (SVM) and Neural Network (NN). We used classical
machine learning methods such as ANFIS, SVM and NN for comparison with our proposed
MPF. Our empirical results show that the MPF consistently outperform the classical methods.
KEYWORDS
Fuzzy Inference System, ANFIS, Neural Network, Support Vector Machine, KPCA
1. INTRODUCTION
Automatic fault prediction is an important topic of research in metal industry [1]. Since the
beginning of the first industrial revolution, industries are striving to produce fault-free products in
least possible amount of time. Customer expectations are ever-increasing in term of quality and
availability of products [2-4]. A fault in a product during the production process can be due to a
single cause or a combination of causes and industries are still using trial and error methods to
minimize them[5][6]. Faults in production can be reduced by analysis of the production process
data [7][8]. Data capturing of the production processes for analysis started manually but due to
human-errors, the quality of the data captured was compromised. Now, with the automation of
data capturing using sensors, the noisy data is getting reduced. With the advancement of the
technology and automation, more and more data is available for analysis to optimise the
production processes [9].
In the current globalized industry era with emphasis on automated smart industries, the analysis
of measured data plays an important role to improve the quality of the products [10]. Large

30 Computer Science & Information Technology (CS & IT)
number of parameters have an influence on the quality of a product. Deviations of process
parameters can have negative effect on the production performance. Measuring and evaluating all
the appropriate process parameters with a suitable method ensure a consistently high quality and
productivity in the automated production environment. The application of Machine Learning
(ML) methods in these processes is motivated mainly by two objectives: the prediction of quality
properties from the measured data and the identification of key process indicators, i.e. the process
parameters which have the strongest influence on the outcome. These objectives help to improve
the quality of the products and to understand the implicit relations among the parameters in the
production processes, which in turn result in reducing faulty products during production.
In this work, we will concentrate on the first objective, namely the prediction of quality
properties from the measured data by suggesting a novel methodology for automatic fault
prediction. The learnt methodology will help to reach target predictions, which will ensure
stability of the production process regulation and repeatability of the process conditions resulting
in quality products with minimal scrap. The potential of the proposed methodology is
demonstrated by using three actual production datasets.
In our approach, we use three stage process. In first stage, the input data is split into mutually
exclusive sets of training, validation and test data. In the second stage, Support Vector Machine
(SVM) [11][12] and Neural Network (NN) [13][14] are trained using the original data whereas
Adaptive Neuro Fuzzy Inference System (ANFIS) [15] is trained on the data transformed by
Kernel Principal Component Analysis (KPCA) [16][17]. In the last stage, we utilize a novel
fusion method to combine these different prediction functions and obtain a “Meta Prediction
Function" (MPF). Results during these stages are also collected and performance is measured.
Figure 1 shows major data processing units of proposed framework.
Figure 1 - Major data processing blocks
The performance of the MPF is compared with the performance of the classical machine learning
methods such as Neural Networks, SVM and ANFIS.
2. RELATED WORK
Lehaet al proposed a novel integrated method into a production plan realized in a physics-based
realistic simulator. Supervised machine learning techniques namely Model Trees and Neural
Networks were integrated. The online learning and on the fly control code modification were
allowed by integrating the learning capability into the control process. Averaging was used for

Computer Science & Information Technology (CS & IT) 31
measuring the produced optimization times through learning which outperform times of a
production process. [18]
Ashouret al proposed a method for automated identification for machined surfaces in
manufacturing.Image processing and computer vision technologies were used for automated
identification for reduction in inspection time and avoidance of human error due to inconsistency
and fatigue. SVM classifier with differentkernels were investigated for the categorization of
machined surfaces into six machining processes. The gray level histogram was used as
discriminating feature. Experimental results suggested that the SVM with the linear kernel
outperformed for a dataset consisting of seventy-two workpiece images. [19]
Yuan et al proposed a novel method for improving the machining quality of thin walled flexible
workpieces. Machining platform was established for thin-walled flexible workpieces. Sparse
Bayesian learning based method was used to predict the future deformation. The dual mode
predictive controller was developed to reduce the machining vibrations and quality of the
workpiece surface was improved. Experiments were performed for thin-walled flexible
workpieces and effectiveness of proposed method were demonstrated through machining
experiments. [20]
Rostamiet al proposed an efficient Equipment Condition Diagnosis (ECD) model for Fault
Detection and Classification (FDC). Different machine learning techniques such as SVM, K-
Means clustering and Self-Organizing Map (SOM) were used to develop ECD model. Principal
Component Analysis (PCA) was used to project the abnormal observations into normal models.
Experiments were conducted with industry data. The proposed Model showed effectiveness for
classifying the fault fingerprints to give evident guidelines in explaining the detected faults. [21]
Costa et al used ANFIS and CART (Classification and Regression Tree) for classifying the
defects that occur during the production process for packing of glass. The database project was
named as “Newglass”. Different operating variables of the furnace and the percentage defects in
the end products of the factory model were used in experiments. Finally, ANFIS, CART and glass
packing manufacturing results were compared. [22]
3. METHODOLOGY
3.1 Support Vector Machine
Support Vector Machine (SVM) is a supervised learning method[11][12] which computes a set of
hyperplanes in a higher dimensional feature space based on the given data for classification and
regression problems.
For each dataset, optimal settings for the SVM method were chosen with experimentation. Input
training and validation data was normalized with z-score normalization. Following kernel
functions were considered for learning:-
the Polynomial kernel,

where p is the power of the polynomial and r is a shifting parameter;
the Gaussian kernel,
where sigma (σ) is an adjustable parameter;
Sigmoid kernel as shown in equation 3 below,
where rho (γ) is the scaling parameter of the input data and r is the shifting parameter controlling
threshold of mapping.
The performance of the selected optimal settings was calculated on the test data and is shown in
the experiments section. The predictions from SVM were used as input in the MPFdescribed in
section 3.5.
3.2 Neural Networks
Neural Networks (NN) [13][14] are a non-linear supervised learning method based on a network
of so-called neurons which are interconnected by weighted links. NN learns by adjusting the
weights to optimal values based on the given training and validation data.
In this work, a NN is trained using one of the most popular backpropagation learning algorithm
with multilayer perceptron topology. The used feedforward NN consisted of three layers: an input
layer, one hidden layer and an output layer. The number of neurons in the input layer is equal to
the total number of independent variables and the number of neurons in the output layer is equal
to the number of dependent variables, whereas the hidden layer contains eight neurons for all
considered datasets. The input layer receives the input from the independent variables and
forwards it to all the neurons in the hidden layer. The neurons in the hidden layer apply their
activation function to the weighted sum of their inputs and compute an output. The output layer
then computes the predicted value for the dependent variable(s).
The back-propagation learning algorithm used here to train the multilayer network consists of two
passes. In the forward pass, with randomly selected weights and the input given by the training
data, the algorithm produces an output for the dependent variable(s). An error is then calculated
based on the difference between predicted and actual output. In the backward pass, this error is
propagated backwards through the network from the output layer to the input layer and weights in
the network are modified using the delta rule. The formula used to calculate the change in
weights of hidden and output layer neurons are shown in equation 4 and in equation 5
respectively:-
where the indices i, j, k refer to input, hidden and output layers,
is the learning rate,
is the Momentum with a value between 0 and 1,

, are the output of neurons i in the input layer and j in the hidden layer at iteration p,
, are the error gradients at the neurons j in hidden layer and k in the output layer at
iteration p.
During training, we used 1000 epochs. Weights of the NN were optimised using settings of
learning rate 0.1 and momentum 0.2. The Unipolar Sigmoid function was used as the activation
function.
During the learning phase, training and validation data is normalised. Then NN is trained and the
parameter settings which produce best results are preserved and the performance of these settings
is calculated on the test data.
The predictions from NN are also used as input in the MPF in section 3.5.
3.3 Kernel Principal Component Analysis
KernelPrincipal Component Analysis (KPCA) is a nonlinear dimension reduction method,
introduced by Sholkoptet al.[16][17], which maps data from the input space to a lower
dimensional feature space while retaining maximum possible variance in the data.
For each dataset, KPCA was applied using the Polynomial kernel and Gaussian kernel,which are
defined in equations(1) and (2) respectively. For the Polynomial kernel, settings used were p =
1,2,3; r = 0. For Gaussian kernel, the settings used were σ = 0.5, 1.0, 1.5.
The results are shown for the kernel settings which provided the most accurate prediction results.
3.4 Adaptive Neuro-Fuzzy Inference System
Adaptive Neuro-Fuzzy Inference System (ANFIS) [15] is a hybrid neuro-fuzzy model which
consists of 5 layers. In the first layer, for the inputs, degree of membership of the chosen
membership function is computed. In the second layer, firing strength of the rules is calculated
using t-norm (product for AND and maximum for OR) operators. In the third layer, computed
firing strengths are normalized. Forth layer is the adaptation layer in which rule consequent
parameters are computed. Fifth layer is the summation layer which computes sum of all the
computed consequents. In learning phase of ANFIS, membership function parameters and rules
consequent parameters are optimised.
In our approach, we use the ANFIS, which combines the advantages of fuzzy expert systems with
those of classical NN, in combination with KPCA. It has already been successfully employed for
data prediction in a variety of fields. KPCA was used to transform the data because for some of
the selected parameter settings like input partition method: ‘grid partitioning’, ANFIS was not
immediately applicable to the number of variables as large as 10. In addition, even in the case of
small number of variables, it was observed that ANFIS produced comparatively better results for
same parameter settings on the transformed data from KPCA in comparison to the original data as
input.
Before the training in ANFIS start, a Fuzzy Inference System (FIS) is initialized with information
about the number of input variables for the selected dataset, the selected number and type of
membership functions and the number of rules. The rules for the FIS were initialized based on the
Grid Partitioning (GP) method or Subtractive Clustering (SC) method. In case of GP method,

input membership function type as Gaussian was selected, output membership function type as
Linear was selected and in different independent runs for best parameter settings, number of
membership functions for input parameters were selected as 2, 3 or 4. For SC method, influence
radius was tested with inputs 0.1, 0.2, 0.3, 0.4 and 0.5 for different runs. It was trained using NN
backpropagation algorithm and least square methods.
In case of KPCA and ANFIS, the training and validation data was collectively normalised and the
corresponding symmetric kernel matrix was computed. The Eigen-values and Eigen-vectors were
computed for the computed kernel matrix. Then principal components of the training and
validation data were computed using the computed kernel matrix and Eigen-vectors. These
principal components along with the corresponding dependent variable values were used to train
ANFIS and parameter settings which produced the best result on the validation data were saved.
Finally, the performance was calculated on the principal components of the test data. These
predictions were also saved for later use in the MPF.
3.5 Meta Prediction Function
Our Meta Prediction Function(MPF) is inspired by the principle of ensemble methods [23]. The
ensemble method provides an optimized prediction by combining the results obtained from the
individual learning methods. The main component of our proposed MPF is the combination of
KPCA and the accumulator module which is implemented using machine learning methods
Multiple Linear Regression (MLR)[24], NN and SVM. The learning algorithm is given below for
MPF.
Step 1: Apply different ML methods such as SVM, NN and ANFIS as discussed in sections 3.1,
3.2, and 3.4 and save the prediction to be used as input.
Step 2: The predictions from the ANFIS, SVM and NN are pre-processed and multicollinearity
was removed from them by using KPCA as shown in equation 6
Step 3: The computed principal components obtained from KPCA are used as the independent
variables as shown in the equations 7, 8 and 9.These independent variables along with the
computed predictions are used to fit the MLR and to find the optimal settings for the NN and
SVM.
Then to compute the performance of the MPF on the test data, independent variables of the test
data are also normalized based on therespective mean and standard deviation of training and
validation data variables, its kernel matrix is computed and the principal components of the
computed kernel matrix are computed using the Eigen-vectors of the training and validation data.

Figure 2 – Proposed Meta Prediction Function (MPF)
To make a comparison with classical machine learning techniques, the test data results of these
three MPFs were compared with the best results of the SVM, NN and ANFIS obtained in sections
3.1, 3.2 and 3.4 respectively.
4. EXPERIMENTS
In order to evaluate our proposed MPF, we checked its performance on three different datasets
related to different fields of actual production data.
The first dataset is related to plastic deformation. It contains 4 variables and 1248 measurements.
The dependent variable is the measured flow stress, while the three independent variables are the
temperature, the strain and the strain rate during the plastic deformation.
The second dataset is related to the pressure strength of green molding sand. It consists of 5
variables and 1076 measurements. The dependent variable is the green compressive strength,
while the 4 independent variables are water content, carbon content, bentonite and clay content.
The third dataset is related to predicting the expected final cost of an ordered product. It consists
of 11 variables and 4878 measurements. The dependent variable is the final cost, while the
independent variables include delivery weight, material number, length, width, height, molding
process, primer, iron ratio and two time related parameters.
To perform the experiments. the input data was divided into three parts. 60% of the input data
was used for training and 20% was used for validation of the selected machine learning methods
to select their respective optimal parameter values and 20% of the data was used for testing. The
results of these methods for test data are shown in section 4.2.
During testing of MPF, the predictions from ANFIS, SVM and NN were transformed using
KPCA.
4.1 Performance Evaluation Measure
To check the performance of the selected machine learning methods, three different error
measures were computed namely Root Mean Square Error (RMSE), Relative Root Mean Square

Error (RRMSE) and Symmetric Mean Absolute Percentage Error (SMAPE), which are defined
as
where n is the total number of patterns. is the computed output and is the predicted output
from the used machine learning method.
4.2 Results and Discussion
4.2.1 Results for dataset Flow Stress
First dataset we used to observe our method performance contains actual measurements from
metal forming experiments.
For NN, same settings were used as described in the section 3.2 For SVM, Gaussian kernel with
sigma ( ) value of 0.5 produced best results.
Since the number of independent variables is smaller in this instance, after computing the
principal components, all three of them were selected, covering 100% of the variance in the data.
For ANFIS, Linear kernel along with subtractive clustering (radius: 0.2) produced best results as
shown in the Table 1 below.
Table 1: Flow stress - Machine learning tool results
Error type
Learning Methods
NN SVM ANFIS
RRMSE 0.0266 0.0336 0.0033
SMAPE 5.0818 9.4270 0.9259
RMSE 21.878 26.4546 2.6550
The following Table 2 shows that the results of the MPF as described in Section 3.5.

Table 2: Flow stress - MPF results
Error Type Kernel: Polynomial
NN SVM MR
RRMSE 0.0033 0.0033 0.0033
SMAPE 0.9569 0.9225 0.9278
RMSE 2.6397 2.6454 2.6533
For all the considered error measures, the results as shown in Figure 3 clearly show that the
combination of KPCA and ANFIS as well as the MPF with MLR outperformed NN as well as
SVM.
Figure 3: Results of all prediction methods for Flow Stress dataset
4.2.2 Results for dataset Pressure strength of green molding sand
Second dataset we used to compare the introduced methodology results is related to computing
the pressure strength of green molding sand.
For this dataset also, all four principal components were selected covering 100% of the variance.
For ANFIS, Linear kernel along with subtractive clustering (radius: 0.3) produced best results.
Computed results are shown in Table 3 below.

Table 3: Pressure shock resistance - Machine learning tools results
Error type
Learning Methods
NN SVM ANFIS
RRMSE 0.0425 0.0449 0.0426
SMAPE 4.4149 4.5869 4.3690
RMSE 1.1688 1.2348 1.1708
Again, for appropriate parameters, the results of the MPF as shown in the table 4 outperform all
of the separate prediction results on the test data.
Table 4: Pressure shock resistance - MPF results
NN SVM MR
RRMSE 0.0481 0.0434 0.0422
SMAPE 4.9242 4.4232 4.3410
RMSE 1.3227 1.1944 1.1615
As we can see in the Figure 4, our proposed MPF outperforms all the individual learning
methods.
Figure 4: Results of all prediction method for Pressure Strength dataset

4.2.3 Results for dataset Final cost
Final cost dataset contains data recorded from actual production environment and is used to
predict the expected final costs of the new ordered product.
In this case, seven out of ten principal components were selected covering 95.91% of the variance
in the original data. For ANFIS, Linear kernel along with subtractive clustering (radius: 0.2)
produced best results. Results are shown in Table 5 below.
Table 5: Final cost - Machine learning tool results
Error type
Learning Methods
NN SVM ANFIS
RRMSE 0.0143 0.0157 0.0233
SMAPE 9.9220 21.1273 23.9995
RMSE 347.7719 382.5654 565.8913
In this case, the prediction provided by the ANFIS in combination with KPCA was not able to
achieve the accuracy of the classical methods. However, the MPF performed best in comparison
with all the individual methods as shown in the Table 6 below. The ensemble method with MR
produced best predictions.
Table 6: Final costs - MPF results
NN SVM MR
RRMSE 0.0139 0.0218 0.0140
SMAPE 10.9912 66.8833 8.5959
RMSE 336.4586 529.2121 339.2884
Figure 5 given below shows the effectiveness of the MPF in comparison with the performance of
SVM and NN.

Figure 5: Results of all prediction methods for Final Costs dataset
For the test data, a random sample of prediction results from all the prediction methods used in
this work are shown in figure 6. Close observation reveals that the MPF is performing better in
comparison with other prediction methods. In Figure 7, which only shows the prediction by MPF,
clearly indicates that the MPF is able to predict the test data very accurately.
Figure 6: Predictions of sample of measurements for all prediction methods

Figure 7: Predictions of sample of measurements for Meta-Prediction Function
5. CONCLUSION AND FUTURE WORK
In this paper, we introduced a novel methodology which combines results of different learning
methods for quality prediction in production processes. Overall based on RMSE, SMAPE and
RRMSE measures, our proposed MPF outperformed the other individual learning methods.
Results show that the novel approach of transforming the results of different machine learning
methods using KPCA and again applying learning techniques to the obtained data returned very
good predictions. Also, it is shown that for the MPF, the combination of a linear kernel for KPCA
with MLR produced the best results. For future work, the proposed methodology can be
transferred for real time monitoring of production processes to identify parameter deviations
which produce faulty parts.
In the future, we will also explore the use of other machine learning methods such as Decision
Trees [25] and Bayesian Networks [26] and incorporate them in MPF. Finally, we note that usage
of our MPF is not only limited for fault prediction of foundry data, but can also be used for
forecasting cost, expenditure or sales prices of production process.
REFERENCES
[1] Ding, S. X., Yin, S., Peng, K., Hao, H., & Shen, B. (2013). A novel scheme for key performance
indicator prediction and diagnosis with application to an industrial hot strip mill. IEEE Transactions
on Industrial Informatics, 9(4), 2239-2247.
[2] Neely, A. (1999). The performance measurement revolution: why now and what next?. International
journal of operations & production management, 19(2), 205-228.
[3] Enkel, E., Perez‐Freije, J., & Gassmann, O. (2005). Minimizing market risks through customer
integration in new product development: learning from bad practice. Creativity and Innovation
Management, 14(4), 425-437.

[4] Kan, S. H., Basili, V. R., & Shapiro, L. N. (1994). Software quality: an overview from the perspective
of total quality management. IBM Systems Journal, 33(1), 4-19.
[5] Sosna, Marc, Rosa Nelly Trevinyo-Rodríguez, and S. Ramakrishna Velamuri. "Business model
innovation through trial-and-error learning: The Naturhouse case." Long range planning 43.2 (2010):
383-407.
[6] Shingo, S., & Dillon, A. P. (1989). A study of the Toyota production system: From an Industrial
Engineering Viewpoint. CRC Press.
[7] Dilts, D. M., Boyd, N. P., & Whorms, H. H. (1991). The evolution of control architectures for
automated manufacturing systems. Journal of manufacturing systems, 10(1), 79-93.
[8] Kang, S., Kim, E., Shim, J., Cho, S., Chang, W., & Kim, J. (2017). Mining the relationship between
production and customer service data for failure analysis of industrial products. Computers &
Industrial Engineering, 106, 137-146.
[9] Fayyad, U., Piatetsky-Shapiro, G., & Smyth, P. (1996). The KDD process for extracting useful
knowledge from volumes of data. Communications of the ACM, 39(11), 27-34.
[10] Ribeiro, L., & Barata, J. (2011). Re-thinking diagnosis for future automation systems: An analysis of
current diagnostic practices and their applicability in emerging IT based production paradigms.
Computers in Industry, 62(7), 639-659.
[11] Schölkopf, Bernhard; and Smola, Alexander J.;” Learning with Kernels”, MIT Press, Cambridge,
MA, 2002, ISBN 0-262-19475-9.
[12] Suykens, J. A., & Vandewalle, J. (1999). Least squares support vector machine classifiers. Neural
processing letters, 9(3), 293-300.
[13] Lawrence, Jeanette, “Introduction to Neural Networks”, California Scientific Software Press 1994,
ISBN 1-883157-00-5.
[14] Craven, M. W., & Shavlik, J. W. (1997). Using neural networks for data mining. Future generation
computer systems, 13(2-3), 211-229.
[15] Jang, J-SR. "ANFIS: Adaptive-Network-based Fuzzy Inference System." IEEE transactions on
systems, man, and cybernetics 23.3 (1993): 665-685.
[16] B. Scholkopf, A. Smola, and K. Muller. “Nonlinear component analysis as a kernel eigenvalue
problem”. Neural Computation, 10(5):1299–1319, 1998.
[17] Yang, J., Frangi, A. F., Yang, J. Y., Zhang, D., & Jin, Z. (2005). KPCA plus LDA: a complete kernel
Fisher discriminant framework for feature extraction and recognition. IEEE Transactions on pattern
analysis and machine intelligence, 27(2), 230-244.
[18] Leha, A., Pangercic, D., Rühr, T., & Beetz, M. (2009, September). “Optimization of simulated
production process performance using machine learning”. In Emerging Technologies & Factory
Automation, 2009. ETFA 2009. IEEE Conference on (pp. 1-5). IEEE.
[19] Ashour, M. W., Khalid, F., Halin, A. A., & Abdullah, L. N. (2015, October). “Machining process
classification using PCA reduced histogram features and the Support Vector Machine”. In Signal and
Image Processing Applications (ICSIPA), 2015 IEEE International Conference on (pp. 414-418).
IEEE.

[20] Yuan, Y., Zhang, H. T., Wu, Y., Zhu, T., & Ding, H. (2016). Bayesian learning-based model
predictive vibration control for thin-walled workpiece machining processes. IEEE/ASME
Transactions on Mechatronics.
[21] Rostami, H., Blue, J., & Yugma, C. (2016, December). Equipment Condition Diagnosis and Fault
Fingerprint Extraction in Semiconductor Manufacturing. In Machine Learning and Applications
(ICMLA), 2016 15th IEEE International Conference on (pp. 534-539). IEEE.
[22] H.R. d. N. Costa and A. La Neve, "A study on the application of regression trees and adaptive neuro-
fuzzy inference system in glass manufacturing process for packaging," 2016 Annual Conference of
the North American Fuzzy Information Processing Society (NAFIPS), El Paso, TX, USA, 2016, pp.
1-4.
[23] Dietterich, Thomas G. "Ensemble methods in machine learning." International workshop on multiple
classifier systems. Springer Berlin Heidelberg, 2000.
[24] Aiken, L. S., West, S. G., & Pitts, S. C. (2003). Multiple linear regression. Handbook of psychology.
[25] Safavian, S. R., & Landgrebe, D. (1991). A survey of decision tree classifier methodology. IEEE
transactions on systems, man, and cybernetics, 21(3), 660-674.
[26] Jansen, R., Yu, H., Greenbaum, D., Kluger, Y., Krogan, N. J., Chung, S., ... & Gerstein, M. (2003). A
Bayesian networks approach for predicting protein-protein interactions from genomic data. science,
302(5644), 449-453.
AUTHORS
Saad Bashir Alvi is currently a Ph.D. student and a research associate in the Institute for
Technologies of Metals at the University of Duisburg-Essen. He has previously worked
as a software engineer and received his M.Sc. from University of Bonn, Germany in
2013. His research interests include machine learning and Neural Networks with special
emphasis on comparative studies in predictive analytics and focus on dimension
reduction as well as neuro fuzzy modeling with applications to production processes.
Robert Martin received his Ph.D. from the University of Duisburg-Essen University,
Germany in 2016 and is currently working at the Institute for Technologies of Metals in
Duisburg.
Prof. Johannes Gottschling is working at the Institute for Technologies of Metals at the
University of Duisburg-Essen and is head of the chair “Mathematics for Engineers”.

EFFICIENT USE OF HYBRID ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM COMBINED WITH NONLINEAR DIMENSION REDUCTION METHOD IN PRODUCTION PROCESSES

More Related Content

What's hot (14)

Similar to EFFICIENT USE OF HYBRID ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM COMBINED WITH NONLINEAR DIMENSION REDUCTION METHOD IN PRODUCTION PROCESSES (20)

Recently uploaded (20)

EFFICIENT USE OF HYBRID ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM COMBINED WITH NONLINEAR DIMENSION REDUCTION METHOD IN PRODUCTION PROCESSES