An accurate model for prediction of autoignition temperature of pure compounds

Accurate prediction of pure compounds autoignition temperature (AIT) is of great importance. In this study, the Artificial Neural Network-Group Contribution (ANN-GC) method is applied to evaluate the AIT of pure compounds. 1025 pure compounds from various chemical families are investigated to propose a comprehensive and predictive model. The obtained results show the squared correlation coefficient of 0.984, root mean square error of 15.44K, and average percent error of 1.6% for the experimental values.     

Validation and data splitting in predictive regression modeling of honing surface roughness data

Model validation is critical in predicting the performance of manufacturing processes. In predictive regression, proper selection of variables helps minimize the model mismatch error, proper selection of models helps reduce the model estimation error, and proper validation of models helps minimize the model prediction error. In this paper, the literature is briefly reviewed and a rigorous procedure is proposed for evaluating the validation and data splitting methods in predictive regression modeling. Experimental data from a honing surface roughness study will be used to illustrate the methodology. In particular, the individual versus average data splitting methods as well as the fivefold versus threefold cross-validation methods are compared. This paper shows that statistical tests and prediction errors evaluation are important in subset selection and cross-validation of predictive regression models. No statistical differences were found between the fivefold and the threefold cross-validation methods, and between use of the individual and average data splitting methods in predictive regression modeling.     

Autoignition Temperature Prediction Using an Artificial Neural Network with Particle Swarm Optimization

The autoignition temperatures of organic compounds were estimated using a hybrid method that includes a simple group contribution method (GCM) implemented in an artificial neural network (ANN) replacing a standard back-propagation algorithm with particle swarm optimization (PSO). A data set of 250 compounds was used for training the network. The optimal condition of the network was obtained by adjusting various parameters by trial-and-error. The capabilities of the designed network were tested in the prediction of the autoignition temperature of 93 compounds not considered during the training step. The proposed model is shown to be more accurate than those of other published works. The results show that the proposed GCM + ANN + PSO method represent an excellent alternative for the estimation of this property with acceptable accuracy (AARD = 1.7%; AAE = 10K).     

Prediction of flow curves and forming limit curves of Mg alloy thin sheets using ANN-based models

Multivariable empirical models based on artificial neural networks were developed in order to predict the flow curves and forming limit curves of AZ31 magnesium alloy thin sheets, in warm forming conditions, vs. process parameters and fibre orientation. Experimental tensile and hemispherical punch tests were carried out in order to obtain the experimental data set, in terms of flow curves and forming limit curves, to be used to train the artificial neural networks. A preliminary study, based on the leave one-out-cross validation methodology, has proven the very good predictive capability of the ANN-based models in modelling both flow curves (flow stress level, curve shape and strain at the onset of necking) and forming limit curves (curve shape, major strain values and minor strain limit) under different process conditions and fibre orientations. Then, the generalisation capability of the neural models in capturing the effect of process parameters and fibre orientation on flow curves and formability has been proven by the excellent agreement, in terms of the high correlation coefficients, low relative errors and average absolute relative errors, between predicted and experimental results not investigated in the training set.     

Densities and Viscosities of the Quinary System: Cyclohexane (1) + \varvec{m}-Xylene (2) + Cyclooctane (3) + Chlorobenzene (4) + Decane (5) and Its Quaternary Subsystems at 308.15 K and 313.15 K

The volumetric and viscometric properties of the quinary system: cyclohexane + $m$ m -xylene + cyclooctane + chlorobenzene + decane, were measured over the entire composition range at 308.15 K and 313.15 K. The experimental data obtained in the course of the present study were employed to analyze the predictive capability of six semi-theoretical and empirical well-known viscosity models reported in the literature, namely, the generalized McAllister three-body interaction model, the pseudo- binary McAllister model, the group contribution model, the generalized corresponding states principle model, the Allan and Teja correlation, and the Grunberg and Nissan law of viscosity. The predictive capabilities of the models were compared using the percentage average absolute deviation (%AAD). The final results showed that the generalized McAllister model gives the lowest AADs of 3.3 % and 3.7 % at 308.15 K and 313.15 K, respectively.     

Prediction of mechanical properties of a warm compacted molybdenum prealloy using artificial neural network and adaptive neuro-fuzzy models

Predictive models using artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) were successfully developed to predict yield strength and ultimate tensile strength of warm compacted 0.85 wt.% molybdenum prealloy samples. To construct these models, 48 different experimental data were gathered from the literature. A portion of the data set was randomly chosen to train both ANN with back propagation (BP) learning algorithm and ANFIS model with Gaussian membership function and the rest was implemented to verify the performance of the trained network against the unseen data. The generalization capability of the networks was also evaluated by applying new input data within the domain covered by the training pattern. To compare the obtained results, coefficient of determination (R²), root mean squared error (RMSE) and average absolute error (AAE) indexes were chosen and calculated for both of the models. The results showed that artificial neural network and adaptive neuro-fuzzy system were both potentially strong for prediction of the mechanical properties of warm compacted 0.85 wt.% molybdenum prealloy; however, the proposed ANFIS showed better performance than the ANN model. Also, the ANFIS model was subjected to a sensitivity analysis to find the significant inputs affecting mechanical properties of the samples.     

Extension of a prediction model to estimate vapor pressures of perfluorinated compounds (PFCs)

Perfluorinated compounds (PFCs) are persistent and have been found globally as environmental contaminants. Release into the environment can occur from manufacturing, industrial and consumer uses. The vapor pressure is an important physical property influencing both the release and the environmental partitioning, but few reliable experimental determinations are available. Here we update a previous PLS regression model to cover also this compound class, using only a few calibration compounds. The recalibration is accomplished by applying a leverage-based weighting scheme that is generally applicable in updating structure-property relationships. The predictive performance is validated with an external validation set and is considerably better than for other standard estimation software, both with regard to accuracy and precision. The model can be given a chemical interpretation and the prediction error for the liquid vapor pressure is within 0.2 log units of Pa. Finally, the model is applied and vapor pressure estimates are reported for more than 200 PFCs where no reliable experimental data are available.     

A neural network for predicting normal boiling point of pure refrigerants using molecular groups and a topological index

An artificial neuron network based on genetic algorithm is presented to predict the normal boiling point (T_b) of refrigerants from 16 molecular groups and a topological index. The 16 molecular groups used in this paper can cover most refrigerants or working fluids in refrigeration, heat pump and organic Rankine cycle; the chosen topological index is able to distinguish all the refrigerant isomers. A total of 334 data points from previous experiments are used to create this network. The calculated results, which are based on a developed numerical method, show a good agreement with experimental data; the average absolute deviations for training, validation and test sets are 1.83%, 1.77%, 2.13%, respectively. A performance comparison between the developed numerical model and the other two existing models, namely QSPR approach and UNIFAC group contribution method, shows that the proposed model can predict T_b of refrigerants in a better accord with experimental data.     

Further exploring rm metrics for validation of QSPR models

Quantitative structure-property relationship (QSPR) models are widely used for prediction of properties, activities and/or toxicities of new chemicals. Validation strategies check the reliability of predictions of QSPR models. The classical metrics like Q² and R²_pred (Q²_ext) are commonly used, besides other techniques, for internal validation (mostly leave-one-out) and external validation (test set validation) respectively. Recently, we have proposed a set of novel r_m² metrics which has been extensively used by us and other research groups for validation of QSPR models. In the present attempt, some additional variants of r_m² metrics have been proposed and their applications in judging the quality of predictions of QSPR models have been shown by analyzing results of the QSPR models obtained from three different data sets (n = 119, 90, and 384). In each case, 50 combinations of training and test sets have been generated, and models have been developed based on the training set compounds and subsequently applied for prediction of responses of the test set compounds. Finally, models for a particular data set have been ranked according to the quality of predictions. The role of different validation metrics (including classical metrics and different variants of r_m² metrics) in differentiating the “good” (predictive) models from the “bad” (low predictive) models has been studied. Finally, a set of guidelines has been proposed for checking the predictive quality of QSPR models.