Trained ANN Model Research Articles

Artificial neural network modelling to predict hot deformation behaviour of as HIPed FGH4169 superalloy

The hot deformation behaviour of as HIPed FGH4169 superalloy was studied by single stroke compression test on MMS-200 test machine at the temperatures of 950–1050°C and the strain rates of 0·004–10 s−1. Based on the experimental results, a back-propagation artificial neural network model and constitutive equation method were established to predict the flow stress of FGH4169 superalloy. The predictability of two different models was compared. The correlation coefficients of experimental and predicted flow stress with the trained BP ANN model and constitutive equation were 0·9995 and 0·9808 respectively. The average root mean square error (RMSE) values of the trained ANN model and constitutive equation are 0·39 and 2·21 MPa respectively. And the average absolute relative error (AARE) values of the trained ANN model and constitutive equation are 1·79 and 7·47% respectively. The results showed that the ANN model is an effective tool to predict the flow stress in comparison with constitutive equation.

Development of experimental design approach and ANN-based models for determination of Cr(VI) ions uptake rate from aqueous solution onto the solid biodiesel waste residue

In the present work, the evaluation capacities of two optimization methodologies such as RSM and ANN were employed and compared for predication of Cr(VI) uptake rate using defatted pongamia oil cake (DPOC) in both batch and column mode. The influence of operating parameters was investigated through a central composite design (CCD) of RSM using Design Expert 8.0.7.1 software. The same data was fed as input in ANN to obtain a trained the multilayer feed-forward networks back-propagation algorithm using MATLAB. The performance of the developed ANN models were compared with RSM mathematical models for Cr(VI) uptake rate in terms of the coefficient of determination (R(2)), root mean square error (RMSE) and absolute average deviation (AAD). The estimated values confirm that ANN predominates RSM representing the superiority of a trained ANN models over RSM models in order to capture the non-linear behavior of the given system.

An intelligent system for low-pressure die-cast process parameters optimization

Low-pressure die-cast (LPDC) is widely used in manufacturing thin-walled aluminum alloy products. Since the quality of LPDC parts are mostly influenced by process conditions, how to determine the optimum process conditions becomes the key to improve the part quality. In this paper, a combining artificial neural network and genetic algorithm (ANN/GA) method is proposed to optimize the LPDC process. In this method, considering the more complicated preparation process of thin-walled casting, an ANN model combining learning vector quantization and back-propagation (BP) algorithm is proposed to map the complex relationship between process conditions and quality indexes of LPDC. Meanwhile, the orthogonal array design and numerical simulation is applied to obtain the training samples instead of carrying out a real experiment for the sake of cost saving. The genetic algorithm is employed to optimize the process parameters with the fitness function based on the trained ANN model. Then, by applying the optimized parameters, a thin-walled component of 300 mm in length, 100 mm in width, and 1.5 mm in thickness is successfully prepared. The results indicate that the proposed intelligent system is an effective tool for the process optimization of LPDC.

Design Rule Extraction from a Trained Ann Model using Ga for Product form Design of Mobile Phones

Abstract An artificial neural network (ANN) model and rule extraction from a trained ANN using genetic algorithm (GA) are applied to predict and advise on the rules for optimal product form design for a particular customer feeling. To map design elements and the affected impressions, principal component analysis (PCA) is employed to determine the essential dimensions for data analysis. By using ANN to examine the relationships between perceptual value and form elements, black-box ANN knowledge can be extracted by applying GA to generate design rules. A case study on the product form of a mobile phone design was conducted to implement the proposed approach. The resultant rules can be used to help product designers to better understand key design elements and to verify optimal solutions suggested by using ANN models.

Optimization of LPDC Process Parameters Using the Combination of Artificial Neural Network and Genetic Algorithm Method

In this article, the low-pressure die-cast (LPDC) process parameters of aluminum alloy thin-walled component with permanent mold are optimized using a combining artificial neural network and genetic algorithm (ANN/GA) method. In this method, an ANN model combining learning vector quantization (LVQ) and back-propagation (BP) algorithm is proposed to map the complex relationship between process conditions and quality indexes of LPDC. The genetic algorithm is employed to optimize the process parameters with the fitness function based on the trained ANN model. Then, by applying the optimized parameters, a thin-walled component with 300 mm in length, 100 mm in width, and 1.5 mm in thickness is successfully prepared and no obvious defects such as shrinkage, gas porosity, distortion, and crack were found in the component. The results indicate that the combining ANN/GA method is an effective tool for the process optimization of LPDC, and they also provide valuable reference on choosing the right process parameters for LPDC thin-walled aluminum alloy casting.

Discussion of “Daily Pan Evaporation Modeling in a Hot and Dry Climate” by J. Piri, S. Amin, A. Moghaddamnia, A. Keshavarz, D. Han, and R. Remesan

The discusser wishes to thank the authors for examining the potential of artificial neural networks ANNs in estimation of evaporation in a hot and dry region. The discusser would like to present the following important points of view, which the authors and potential researchers need to consider: 1. In the study, the gamma test GT was proposed as an alternative model for the determination of optimum input combination of an ANN model. However, the reliability of the GT was not taken into consideration. The GT results were given for the five different input combinations in Table 3. The performances of the ANN for each input combination could be investigated to examine the reliability of the GT see Kisi 2008a . Or, the GT technique could be compared to another method, e.g., principle component analysis Kisi 2009a . According to the GT, the relative importance of inputs was found as W Ed RH T in the study. This finding contradicts with the related literature Sudheer et al. 2002; Keskin and Terzi 2006; Kisi 2009b . Keskin and Terzi 2006 modeled the pan evaporations using an ANN technique and found that the T is more important than the W. Kisi 2009b investigated the accuracy of the ANN in pan evaporation estimation and found that the RH is more significant than the W. In addition, on page 804 the authors say that “Using temperature data alone, Sudheer et al. 2002 found that a properly trained ANN model could reasonably estimate the evaporation values at their study area in a temperate region.” This implies that the T is more important than the other variables i.e., RH, W on evaporation. On page 809 the authors also state that “Tan et al. 2007 found the important variables were: solar radiation, relative humidity, air temperature, and surface wind speed.” However, Tan et al. 2007 also found that the RH was more important in estimation pan evaporation than the W. This also contradicts with the relative importance W Ed RH T that was found in this study. 2. The ANN and NNARX models whose inputs are W, T, RH, and Ed were compared with simple empirical models, Linacre comprising T, Tdew and Marciano comprising U, es ,ea inputs, in Table 4. This is not a valid comparison. It does not make sense to compare simple empirical methods with more complicated methods ANN/NNARX containing four parameters. In addition to the comparison shown in Table 4, the ANN/NNARX and empirical methods could also be compared using the same input parameters. In addition, the comparison with empirical pan evaporation models has not been

Cyclic behavior of seismogenic sources in India and use of ANN for its prediction

Endeavors to realistically model physical processes responsible for earthquake occurrence and sustained large uncertainties in the results have lead to the application of techniques like artificial neural network for estimation of rate/probability of earthquake occurrence in future. The earthquake occurrence in India has been re-visited and artificial neural networks have been applied to learn the cyclic behavior of seismicity in the independent seismogenic sources to predict their future trends. As a prerequisite, the whole country has been divided into 24 seismogenic sources for which the seismicity cycles were studied. Their cyclic behavior has been captured in form of four stages of earthquake occurrence and the future trends have been predicted using ANN. To validate the trained ANN model, testing has been carried out in two ways: first, by giving the samples that are not used in training (NT) and second, by giving the total samples (T). As a method of testing, standard errors and correlation coefficients between the network output patterns and observed patterns of the testing sample given were considered. The outcome of the ANN is used to interpret the future seismicity of each of the 24 seismogenic zones in terms of various stages of the future seismicity cycles.

Reliability analysis of structures using artificial neural network based genetic algorithms

A new class of artificial neural network based genetic algorithms (ANN-GA) has been developed for reliability analysis of structures. The methods involve the selection of training datasets for establishing an ANN model by the uniform design method, approximation of the limit state function by the trained ANN model and estimation of the failure probability using the genetic algorithms. By effectively integrating the uniform design method with the artificial neural network based genetic algorithms (ANN-GA), the inherent inaccuracy of the selection of the training datasets for developing an ANN model in conventional ANN-GA has been eliminated while keeping the good features of the ANN-GA. Due to a small number of training datasets required for developing an ANN model, the proposed methods are very effective, particularly when a structural response evaluation entails costly finite element analysis or when a problem has a extremely small value of failure probability. Three numerical examples involving both structural and non-structural problems illustrate the application and effectiveness of the methods developed, which indicate that the proposed methods can provide accurate and computationally efficient estimates of probability of failure.

A new artificial neural network-based response surface method for structural reliability analysis

This paper presents a new artificial neural network-(ANN)based response surface method in conjunction with the uniform design method for predicting failure probability of structures. The method involves the selection of training datasets for establishing an ANN model by the uniform design method, approximation of the limit state function by the trained ANN model and estimation of the failure probability using first-order reliability method (FORM). In the proposed method, the use of the uniform design method can improve the quality of the selected training datasets, leading to a better performance of the ANN model. As a result, the ANN dramatically reduces the number of required trained datasets, and shows a good ability to approximate the limit state function and then provides a less rigorous formulation in the context of FORM. Results of three numerical examples involving both structural and non-structural problems indicate that the proposed method provides accurate and computationally efficient estimates of the probability of failure. Compared with the conventional ANN-based response surface method, the proposed method is much more economical to achieve reasonable accuracy when dealing with problems where closed-form failure functions are not available or the estimated failure probability is extremely small. Finally, several important parameters in the proposed method are discussed.

An artificial neural network application to produce debris source areas of Barla, Besparmak, and Kapi Mountains (NW Taurids, Turkey)

Abstract. Various statistical, mathematical and artificial intelligence techniques have been used in the areas of engineering geology, rock engineering and geomorphology for many years. However, among the techniques, artificial neural networks are relatively new approach used in engineering geology in particular. The attractiveness of ANN for the engineering geological problems comes from the information processing characteristics of the system, such as non-linearity, high parallelism, robustness, fault and failure tolerance, learning, ability to handle imprecise and fuzzy information, and their capability to generalize. For this reason, the purposes of the present study are to perform an application of ANN to a engineering geology problem having a very large database and to introduce a new approach to accelerate convergence. For these purposes, an ANN architecture having 5 neurons in one hidden layer was constructed. During the training stages, total 40 000 training cycles were performed and the minimum RMSE values were obtained at approximately 10 000th cycle. At this cycle, the obtained minimum RMSE value is 0.22 for the second training set, while that of value is calculated as 0.064 again for the second test set. Using the trained ANN model at 10 000th cycle for the second random sampling, the debris source area susceptibility map was produced and adjusted. Finally, a potential debris source susceptibility map for the study area was produced. When considering the field observations and existing inventory map, the produced map has a high prediction capacity and it can be used when assessing debris flow hazard mitigation efforts.

Non-destructive sensing and its inverse model for canopy parameters using texture analysis and artificial neural network

Utilisation of micro-precision agriculture is essential as an optimization method in the production system to increase quality of product in plant factory. Plant canopy is one of essential indicators of its quality degradation. In this study, samples of cultured Sunagoke moss Rhacomitrium canescens were used. It has been utilized as an active greening material in a forestation technology to mitigate the urban heat island effect. Canopy parameters for moss as low stature plants remain difficult to be quantified accurately. The direct measurement of canopy parameters was considered relatively inefficient and destructive to the plants. Therefore, artificial intelligence for diagnosing quality degradation is proposed using texture analysis and artificial neural network. It consists of two models covering non-destructive sensing as canopy parameters prediction model and its inverse model as a simple vision model. This paper discussed building both models to monitor canopy parameters using experimental result, as an example of how micro-precision agriculture can be implemented in plant factory. A black box relationship between canopy image and canopy parameters is proposed. Non-destructive sensing is defined that canopy parameters can be predicted from its image while the inverse model refers to the opposite way. Canopy patterns were identified using texture analysis for extracting image based on grey level co-occurrence matrix in to energy, local homogeneity and contrast features. The canopy parameters from experiment were determined as textural class, moisture content, leaf area index and leaf water potential. Finally, artificial neural network was used to model this both way and complex relationship. Artificial neural network model performance was tested successfully to describe the relationship between textural features and canopy parameters using back-propagation supervised learning method and inspection data. Both trained ANN models produced satisfied correlation between measured and predicted value and minimum inspection error. The research result is applicable not only for determining canopy parameters, but also as a simple vision model supporting application of speaking plant approach in plant factory.