Search results

Modeling Wastewater Treatment Plant (WWTP) constitutes an important tool for controlling the operation of the process and for predicting its performance with substantial influent fluctuations. The purpose of this paper is to apply an artificial neural network (ANN) approach with a feed-forward back-propagation in order to predict the ten-month performance of Touggourt WWTP in terms of effluent Chemical Oxygen Demand (CODeff).

The influent variables such as (pHinf), temperature (TEinf), suspended solid (SSinf), Kjeldahl Nitrogen (KNinf), biochemical oxygen demand (BODinf) and chemical oxygen demand (CODinf) were used as input variables of neural networks. To determine the appropriate architecture of the neural network models, several steps of training were conducted, namely the validation and testing of the models by varying the number of neurons and activation functions in the hidden layer, the activation function in output layer as well as the learning algorithms.

The better results were achieved with an architecture network [6-50-1], hyperbolic tangent sigmoid activation functions at hidden layer, linear activation functions at output layer and a Levenberg – Marquardt method as learning algorithm. The results showed that the ANN model could predict the experimental results with high correlation coefficient 0.89, 0.96 and 0.87 during learning, validation and testing phases, respectively. The overall results indicated that the ANN modeling approach can provide an effective tool for simulating, controlling and predicting the performance of WWTP.

This work is the first of its kind in this region due to the remarkable development in terms of population and agricultural activity in the region, which drove to the increase of water pollutants, so it is necessary to use the modern technologies to modeling and controlling of WWTP.

This paper aims to propose an efficient artificial neural network (ANN) model using multi-layer perceptron philosophy to predict the fireside corrosion rate of superheater tubes in coal fire boiler assembly using operational data of an Indian typical thermal power plant.

An efficient gradient-based network training algorithm has been used to minimize the network training errors. The input parameters comprise of coal chemistry, namely, coal ash and sulfur contents, flue gas temperature, SOX concentrations in flue gas, fly ash chemistry (Wt.% Na₂O and K₂O).

Effects of coal ash and sulfur contents, Wt.% of Na₂O and K₂O in fly ash and operating variables such as flue gas temperature and percentage excess air intake for coal combustion on the fireside corrosion behavior of superheater boiler tubes have been computationally investigated and parametric sensitivity analysis has been undertaken.

Quite good agreement between ANN model predictions and the measured values of fireside corrosion rate has been observed which is corroborated by the regression fit between these values.

This paper aims to discuss the utilization of artificial neural networks (ANNs) and multiple regression method for estimating surface roughness in milling medium density fiberboard (MDF) material with a parallel robot.

In ANN modeling, performance parameters such as root mean square error, mean error percentage, mean square error and correlation coefficients (R²) for the experimental data were determined based on conjugate gradient back propagation, Levenberg–Marquardt (LM), resilient back propagation, scaled conjugate gradient and quasi-Newton back propagation feed forward back propagation training algorithm with logistic transfer function.

In the ANN architecture established for the surface roughness (Ra), three neurons [cutting speed (V), feed rate (f) and depth of cut (a)] were contained in the input layer, five neurons were included in its hidden layer and one neuron was contained in the output layer (3-5-1).Trials showed that LM learning algorithm was the best learning algorithm for the surface roughness. The ANN model obtained with the LM learning algorithm yielded estimation training values R² (97.5 per cent) and testing values R² (99 per cent). The R² for multiple regressions was obtained as 96.1 per cent.

The result of the surface roughness estimation model showed that the equation obtained from the multiple regressions with quadratic model had an acceptable estimation capacity. The ANN model showed a more dependable estimation when compared with the multiple regression models. Hereby, these models can be used to effectively control the milling process to reach a satisfactory surface quality.

This paper presents the results from two supervised Artificial Neural Networks (ANN) developed for the spalling classification and failure prediction of high strength concrete columns (HSCC) subjected to fire. The experimental test data used for the ANN are based on the HSCC tests undertaken at the Fire Research Laboratories at the University of Ulster. 80% of the chosen experimental test data was used to train the network with the remaining 20% used for testing. In the spalling classification example the key ANN input parameters were; furnace temperature, restraint, loading level, force, spalling degree, failure time and spalling type. This was also the case for the failure prediction example except for spalling type. The networks were trained using the resilient propagation algorithm. A 6-10-3 and 5-10-1 ANN architecture gave the best results for the classification and failure prediction times respectively. The results demonstrate that HSCC can be assessed using ANN.

To analyse a self‐acting parallel surface thrust bearing using a proposed feedforward neural network.

Firstly, a one‐piece hydrodynamic thrust bearing with an initially flat surface is analysed, designed and tested. Analysis of the configuration used is particularly simple and gives good agreement with experimental results. Secondly, some artificial neural network types are designed to analyse minimum film thickness for specified load of thrust bearing system.

A more efficient film shape might result if the length of the cantilever did not increase with radius, since with the configuration used, the deflection of the outer corner was almost three times greater than the deflection of the inner corner, although this effect only becomes acute with regard to film thickness at fairly high loads. The design analysis of an asymmetric cantilever would be more lengthy and less easy to apply. Extrapolation of results for the plain bearing shows that high loads could be carried, but under severe conditions of temperature and clearance.

Owing to finance problems, it was not easy to setup system in real time applications. This approach would be given usefulness elsewhere.

In future, this technique will be implemented for designing experimental neural network predictor on thrust bearing system. Also, this kind of neural predictor will be suitable for complex bearing systems.

A new type of neural network is used to investigate film thickness of thrust bearing system. Quick propagation neural network has given superior performance for designing of model of thrust bearing system. As described and shown in figures and tables, this kind of neural predictor could be employed for analysing such systems in practical analyses.

This paper aims to improve the performance and speed of artificial neural network (ANN)‐ball‐bearing fault detection expert systems by eliminating unimportant inputs and changing the ANN structure.

An algorithm is used to select the best subset of features to boost the success of detecting healthy and faulty ball. Some of the important parameters of the ANN are also optimized to make the classifier achieve the maximum performance.

It was found that better accuracy can be obtained for ANN with fewer inputs.

The method can be used for other machinery condition‐monitoring systems which are based on ANN.

The results are useful for bearing fault detection systems designers and quality check centers in bearing manufacturing companies.

The algorithm used in this research is faster than in previous studies. Changing ANN parameters improved the results. The system was examined using experimental data of ball‐bearings.

To develop a non‐linear modelling technique for modern air vehicles with an application to a twin rotor multi‐input‐multi‐output system (TRMS) which resembles the dynamics of a helicopter to a certain extent and presents formidable control challenges.Design/methodology/approach – A Non‐linear AutoRegressive process with eXternal input (NARX) approach with a feedforward neural work and a resilient propagation (RPROP) algorithm is used to model the system. The RPROP algorithm possesses direct weight update capability without considering the size of the partial derivative. The obtained model is shown to be adequate by carrying out convincing tests such as correlations, cross‐validations and prediction based on predicted output and, therefore, is deemed to be reliable.Findings – It is shown that the combination of the feedforward neural networks and RPROP algorithms is very useful and effective in modelling systems with high non‐linearity and other complex characteristics. It is always important to attain a model with minimum number of neurons in different layers of the network by overcoming the possibility of getting stuck in the shallow local minimum of error function by using RPROP algorithm.Research limitations/implications – The system is modelled off‐line. On‐line modelling will be required for real‐time control purpose.Practical implications – The non‐linear modelling approach presented in this study is shown to be appropriately applicable to model new generations' air vehicles and other complex mechatronic systems such as TRMS. So, the approach will be appealing to industrial applications.Originality/value – This paper addresses the problems of modelling modern sophisticated non‐linear systems with complex characteristics and uncertain dynamics.

The purpose of this paper is to enable construction project team members to understand the factors that they must closely monitor to complete projects with a desired quality and also to predict quality performance during the course of a project. With quality being one of the prime concerns of clients in their construction projects, there is a definite need to monitor its performance.

The study discussed here is an extension of past research in which 55 project performance attributes were identified based on expert's opinion and literature survey which after analysis resulted in 20 factors (11 success and nine failure). The results of the second stage questionnaire survey conducted have been used to develop the quality performance prediction model based on artificial neural networks (ANN).

The analyses of the responses led to the conclusion that factors such as project manager's competence, monitoring and feedback by project participants, commitment of all project participants, good coordination between project participants and availability of trained resources significantly affect the quality performance criterion. The best prediction model was found to be a 5‐5‐1 feed forward neural network based on back propagation algorithm with a mean absolute percentage deviation (MAPD) of 8.044 percent.

Project professionals can concentrate on certain factors instead of handling all the factors at the same time to achieve the desired quality performance. Also the study may be helpful for the project manager and his/her team to predict the quality performance of the project during its course.

The present study resulted in a model to predict the quality performance based on the factors identified as critical using ANN. With the control of the identified critical factors and usage of the prediction model, the desired quality performance can be achieved in construction projects.

The purpose of this paper is to identify a workable methodology to prioritise those crime scenes which have the greatest opportunity of a forensic recovery to enable effective Crime Scene Investigator (CSI) resource deployment.

The motivation behind this work stemmed from an abundance of volume crime scenes that required examination and a lack of resources that could be deployed. Within a data mining application environment, two supervised learning algorithms were used to model Northamptonshire Police's forensic data to provide a computer‐based model that could predict the outcome of finding a forensic sample at the currently unattended scene of a crime.

Based on past data, a computer model could be produced to predict the probability of finding useful fingerprints, DNA and/or footwear marks at the scene of a volume crime. In this paper, volume crime means burglary dwelling, burglary in commercial buildings, theft of and theft from motor vehicles. The model was 68 percent accurate. CSIs were 41 percent accurate in their predictions. This has been tested within five different police forces each having differing computer systems, demonstrating that the methodology is portable.

The model, when connected to either a crime recording system or an incident recording system, can produce a prioritised crime scene attendance list within minutes and assess crimes/incidents as they are reported. This list can be seamlessly used in conjunction with other attendance criteria if required, e.g. vulnerable victim, etc.

This paper provides a scientific solution to CSI resource attendance management being proved in five different UK police forces.

Many strategies have been put forward for training deep network models, however, stacking of several layers of non-linearities typically results in poor propagation of gradients and activations. The purpose of this paper is to explore the use of two steps strategy where initial deep learning model is obtained first by unsupervised learning and then optimizing the initial deep learning model by fine tuning. A number of fine tuning algorithms are explored in this work for optimizing deep learning models. This includes proposing a new algorithm where Backpropagation with adaptive gain algorithm is integrated with Dropout technique and the authors evaluate its performance in the fine tuning of the pretrained deep network.

The parameters of deep neural networks are first learnt using greedy layer-wise unsupervised pretraining. The proposed technique is then used to perform supervised fine tuning of the deep neural network model. Extensive experimental study is performed to evaluate the performance of the proposed fine tuning technique on three benchmark data sets: USPS, Gisette and MNIST. The authors have tested the approach on varying size data sets which include randomly chosen training samples of size 20, 50, 70 and 100 percent from the original data set.

Through extensive experimental study, it is concluded that the two steps strategy and the proposed fine tuning technique significantly yield promising results in optimization of deep network models.

This paper proposes employing several algorithms for fine tuning of deep network model. A new approach that integrates adaptive gain Backpropagation (BP) algorithm with Dropout technique is proposed for fine tuning of deep networks. Evaluation and comparison of various algorithms proposed for fine tuning on three benchmark data sets is presented in the paper.