Single-layer Perceptron Network Research Articles

Facile estimation of viscosity of natural amino acid salt solutions: Empirical models vs artificial intelligence

Natural amino acid salt solutions (NAASs) are paving the way for greener carbon capture. This study developed simple and precise methods for the viscosity modeling of NAASs. Two approaches, namely, empirical correlations and artificial intelligence, were assessed using a large databank (16 NAAs, 3 alkaline compounds, 25 NAASs, and 1582 data points). Two general correlations and a global equation were suggested. Benefitting from the input of single reference-point data, the modified global equation yielded the best results with a 2.28% deviation. The other empirical models represented viscosities with less than a 7.20% error. The second approach, employing artificial neural networks (ANNs) with different algorithms, was also proposed. The best ANNs were a single-layer perceptron network with tansig + trainlm functions, a double-layer perceptron network with logsig + tansig + trainlm functions, and a radial basis function network with the maximum neurons. They managed to calculate the viscosities with errors of 2.82%, 1.82%, and 0.47%, respectively.

Performance analysis and artificial intelligence modeling for enhanced hydrogen production by catalytic bio-alcohol reforming in a membrane-assisted reactor

Due to the significance of green processing and artificial intelligence (AI) modeling, herein, we proposed a mathematical and an AI model to enhance green hydrogen production from bio-alcohols in a membrane-assisted reactor. A sensitivity analysis for the effective variables was conducted in terms of conversion, thermal behavior, pressure drop, and hydrogen distribution. A single-layer perceptron network with tansig + trainlm functions and eight neurons yielded a 0.72 % error (mean squared error (MSE) = 2.69, R2 = 0.99994) in the bio-methanol reformer, while the same functions with nine neurons presented a 0.22 % error (MSE = 0.32, R2 = ∼1.00000) in the bio-ethanol reformer. Lastly, a multi-objective optimization was performed to determine the optimum operating conditions, which enhanced the hydrogen production in the bio-methanol (yMeOH = 0.4 and 0.204 mol/h at 517 K and 6 bar) and bio-ethanol (yEtOH = 0.3 and 1.8 mol/h at 823 K and 6 bar) reforming.

Combined mathematical and artificial intelligence modeling of catalytic bio-methanol conversion to dimethyl ether

This study is devoted to developing a combined mathematical and artificial intelligence modeling for simulating the conversion of bio-methanol into dimethyl ether. This methodology couples a precise mathematical model with single-layer perceptron artificial neural networks formulated by several combinations of training and transfer functions, to constitute a machine learning that overcomes the computational tools with the capability of learning without being explicitly programmed, so that it may be used to teach these tools to handle the data with high efficiency in order to predict the upcoming unseen ones precisely. Hence, the novelty of this study deals with the development of a machine learning modeling combined with artificial neural networks, benefitting from an extended mathematical-model-driven dataset, applied for the first time in the conversion of bio-methanol into dimethyl ether in large-scale catalytic dehydration reactors, assessing two process schemes, namely adiabatic and water-cooled operations. To this scope, datasets have been generated by an accurate mathematical model validated experimentally by reference pilot data, adopting the best equilibrium and kinetic expressions, and the simulated numerical data have been then employed as the source of the machine learning models. Extended databanks for the sensitivity analyses of the two process schemes have been then generated by the model (2688 data points in each case), 70% of which has been employed for the artificial neural networks training with mean squared error as the performance function. A single-layer perceptron network yielded the best performances in the adiabatic and water-cooled process schemes, revealing less than 0.5% deviations for the input variables (feed temperature, pressure, flow rate, and concentration) and outputs (temperature, pressure, and flow rates of methanol, dimethyl ether, and water). The simulations revealed that, as the best result of this work, methanol may be converted into dimethyl ether up to ∼80% at 498 K, depending on the validated equilibrium constant adopted, with a maximum error of 0.11%. As a further relevant result, this study demonstrated that analyzing the methanol conversion into dimethyl ether reaction as a function of several variables such as feed flow rate, feed methanol concentration, and input temperature, the accuracy of the neural network with a small number of neurons (5) did not provide accurate output results, whereas 15 neurons in the hidden layer intensify their precision, with the interesting feature that a single-layer perceptron network may detect even small changes to the operating variables.

MODENN: A Shallow Broad Neural Network Model Based on Multi-Order Descartes Expansion.

Deep neural networks have achieved great success in almost every field of artificial intelligence. However, several weaknesses keep bothering researchers due to its hierarchical structure, particularly when large-scale parallelism, faster learning, better performance, and high reliability are required. Inspired by the parallel and large-scale information processing structures in the human brain, a shallow broad neural network model is proposed on a specially designed multi-order Descartes expansion operation. Such Descartes expansion acts as an efficient feature extraction method for the network, improve the separability of the original pattern by transforming the raw data pattern into a high-dimensional feature space, the multi-order Descartes expansion space. As a result, a single-layer perceptron network will be able to accomplish the classification task. The multi-order Descartes expansion neural network (MODENN) is thus created by combining the multi-order Descartes expansion operation and the single-layer perceptron together, and its capacity is proved equivalent to the traditional multi-layer perceptron and the deep neural networks. Three kinds of experiments were implemented, the results showed that the proposed MODENN model retains great potentiality in many aspects, including implementability, parallelizability, performance, robustness, and interpretability, indicating MODENN would be an excellent alternative to mainstream neural networks.

Enhanced Error Reduction of Signal Power Loss During Electromagnetic Propagation: Architectural Composition and Learning Rate Selection

This research work analyses the effect of the architectural composition of Multi-Layer Perceptron (MLP) Artificial Neural Network (ANN) combined with the effect of the learning rate for effective prediction of signal power loss during electromagnetic signal propagation. A single hidden layer and two hidden layers of MLP ANN have been considered. Different configurations of the neural network architecture ranging from 4 to 100 for both MLP networks have been analyzed. The required hidden layer neurons for optimal training of a single layer multi-layer network were 40 neurons with 0.99670 coefficient of correlation and 1.28020 standard deviations, while [68 72] trained two hidden layers multi-layer perceptron with 0.98880 coefficient of correlation and standard deviation of 1.42820. Different learning rates were also adopted for the network training. The results further validate better MLP neural network training for signal power loss prediction using single-layer perceptron network compared to two hidden layers perceptron network with the coefficient of correlation of 0.99670 for single-layer network and 0.9888 for two hidden layers network. Furthermore, the learning rate of 0.003 shows the best training capability with lower mean squared error and higher training regression compared to other values of learning rate used for both single layer and two hidden layers perceptron MLP networks.

Evaporate facies recognition using an unsupervised artificial neural network in the northern Arabia Plate

The carbonate–evaporite depositional combination of Late Jurassic age, incorporates the most prolific oil-producing intervals in the world and forms many giant fields. The succession is the top member of four upwards-shoaling carbonate–anhydrite cycles of Upper Kimmeridgian age, and is overlain by the impermeable anhydrite in northern Arabia. The weak depositional contrasts in carbonate ramp settings make the lateral seal configurations subtle and tough to recognize. Multiple attribute analyses based on an artificial neural network (ANN) can delineate the internal character of the reservoir and seal in a consistent way. In order to recognize the sedimentary facies and characterize stratigraphic traps within this reservoir interval, multiple seismic attributes were input to an unsupervised ANN. Unsupervised ANN offers a powerful means of classification, implemented here using a single-layer perceptron network. The network is trained by comparing the neurons to the input vectors using competitive-learning techniques. Once a neuron migrates to the centre of a class, the network stabilizes, training is finished and the neuron is assigned to a representative class. Without prior information, the unlabelled class is calibrated and analysed by lithofacies generated from log and core data. Further sedimentary facies are recognized by integrating local geological knowledge. The depositional environments in the study area are well characterized by the unsupervised ANN, and the recognized sedimentary facies are consistent with the drilled wells and the resulting geological model. Lagoonal deposits of the inner-ramp, ramp-crest shoal and proximal deposits of the middle ramp are recognized within the study area. The widespread ramp crest with peloid and oolitic grainstones provides good reservoirs, whereas the lagoonal deposits distributed between the shoals have a greater abundance of tight limestone with low porosity and permeability, thereby forming a good lateral seal. The selected study area, covering the Rimthan Arch is considered a favorable area for the presence of stratigraphic traps. The sedimentary facies recognition helps to define potential areas for favourable prospect definition and hence prospect ranking.

Evaluation of scoliosis using baropodometer and artificial neural network

Abstract Introduction: One of the most recurrent pathologies in the spine is scoliosis. It occurs in the frontal plane and is formed by one or more curves in the spinal column. The scoliosis causes global postural misalignment in an individual. One of the modifications produced by postural misalignment is the way in which an individual distributes weight to the feet. We aimed to implement an electronic system for separating patients with Degree I scoliosis (i.e., 1° to 19° scoliosis according to the Ricard classification) into two groups: C1 (1°-9°) and C2 (10°-9°). The highest percentage of patients with scoliosis is in this range: those who do not need to wear vests or undergo surgery and whose treatment is performed via special physical exercise and frequent evaluations by healthcare professionals. Methods The electronic system consists of a baropodometer and artificial neural networks (ANNs). The classification of patients in the scoliosis groups was performed with MATLAB software and a Single Layer Perceptron network using the backpropagation training algorithm. Evaluations were performed on 63 volunteers. Results The mean classification sensitivity was 93.7% in the C1 group and 94.5% in the C2 group. The classification accuracy was 83.3% in the C1 group and 96.0% in the C2 group. Conclusion The implemented system can contribute to the treatment of patients with scoliosis grades ranging from 1° to 19°, which represents the highest incidence of this pathology, for which the monitoring of the clinical condition using noninvasive techniques is of fundamental importance.

Classification of reflected signals from cavitated tooth surfaces using an artificial intelligence technique incorporating a fiber optic displacement sensor.

An enhanced dental cavity diameter measurement mechanism using an intensity-modulated fiber optic displacement sensor (FODS) scanning and imaging system, fuzzy logic as well as a single-layer perceptron (SLP) neural network, is presented. The SLP network was employed for the classification of the reflected signals, which were obtained from the surfaces of teeth samples and captured using FODS. Two features were used for the classification of the reflected signals with one of them being the output of a fuzzy logic. The test results showed that the combined fuzzy logic and SLP network methodology contributed to a 100% classification accuracy of the network. The high-classification accuracy significantly demonstrates the suitability of the proposed features and classification using SLP networks for classifying the reflected signals from teeth surfaces, enabling the sensor to accurately measure small diameters of tooth cavity of up to 0.6 mm. The method remains simple enough to allow its easy integration in existing dental restoration support systems.

Pattern classification by memristive crossbar circuits using ex situ and in situ training

Memristors are memory resistors that promise the efficient implementation of synaptic weights in artificial neural networks. Whereas demonstrations of the synaptic operation of memristors already exist, the implementation of even simple networks is more challenging and has yet to be reported. Here we demonstrate pattern classification using a single-layer perceptron network implemented with a memrisitive crossbar circuit and trained using the perceptron learning rule by ex situ and in situ methods. In the first case, synaptic weights, which are realized as conductances of titanium dioxide memristors, are calculated on a precursor software-based network and then imported sequentially into the crossbar circuit. In the second case, training is implemented in situ, so the weights are adjusted in parallel. Both methods work satisfactorily despite significant variations in the switching behaviour of the memristors. These results give hope for the anticipated efficient implementation of artificial neuromorphic networks and pave the way for dense, high-performance information processing systems.

Computer Aided Identification of Inter-Layer Faults in Gas Insulated Capacitively Graded Bushing during Switching

In a Gas Insulated Substation (GIS), Very Fast Transients (VFTs) are generated mainly due to switching operations. These transients may cause internal faults, i.e., layer-to-layer faults in a capacitively graded bushing as it is one of the most important terminal equipment for GIS. The healthiness of the bushing is generally verified by measuring its leakage current. However, the change in current magnitude/pattern is only marginal for different types of fault conditions. Leakage current monitoring (LCM) systems generate large amounts of data and computer aided interpretation of defects may be of great assistance when analyzing this data. In view of the above, ANN techniques have been used in this study for identification of these minor faults. A single layer perceptron network, a two layer feed-forward back propagation network and cascade correlation (CC) network models are used to identify interlayer faults in the bushing. The effectiveness of the CC network over perceptron and back propagation networks in identification of a fault has been analysed as part of the paper.

USING VIBRATION MONITORING FOR LOCAL FAULT DETECTION ON GEARS OPERATING UNDER FLUCTUATING LOAD CONDITIONS

Gearboxes often operate under fluctuating load conditions during service. Conventional techniques for monitoring vibration are based on the assumption that changes in the measured structural response are caused by deterioration in the condition of the gearbox. However, this assumption is not valid for fluctuating load conditions. To find a methodology that could deal with such conditions, experiments were conducted on a gearbox test rig with different levels of tooth damage severity and the capability of applying fluctuating loads to the gear system. Different levels of fluctuation in constant loads as well as in sinusoidal, step and chirp loads were considered. The test data were order tracked and time synchronously averaged with the rotation of the shaft in order to compensate for the variation in rotational speed induced by the fluctuating loads. A pseudo-Wigner–Ville distribution was then applied to the test data, in order to identify the influence of the fluctuating load conditions. In this work, a vibration waveform normalisation approach is presented, which enables the use of the pseudo-Wigner–Ville distribution to indicate deteriorating fault conditions under fluctuating load conditions. Statistical parameters and various other features were extracted from the distribution in order to indicate the linear separation of the values for various fault conditions, after applying the vibration waveform normalisation approach. Feature vectors were compiled for the various fault and load conditions. Mahalanobis distances were calculated between the various feature vectors and an average feature vector was compiled from data measured on the undamaged gearbox. It was proved that the Mahalanobis distance could be used as a single parameter, which can readily be monotonically trended to indicate the progression of a fault condition under fluctuating load conditions. It was shown that a single layer perceptron network could be trained with the perceptron learning rule within a finite number of iterations.