Papers by Keyword: Predictive Model

Abstract: Flotation recovery is an important index of flotation process, in order to change the existing detection methods of low accuracy , a soft measurement model of flotation recoveries is proposed based on improved weighted LS - SVM. According to the flotation foam characteristics and the corresponding relation of flotation recovery, the fuzzy C-means clustering method is used for flotation characteristics of data processing , the image characteristic values as prediction model input and using genetic algorithm to optimize the parameters of the model. The result show that the modified algorithm can overcome a prediction standard model LS - SVM algorithm parameter optimization shortage, and have better forecasting effect which provide effective protection for flotation process operation and flotation operation stable operation optimization .

Abstract: An accurate and reliable control method of irregularity in yarn fineness is presented. It's based on the theory of the influence of cotton fiber length distribution on the evenness of yarn, and it improves the predictive model of irregularity in yarn fineness. Under the producing condition of controlling, we set a value or value range of predetermined term of or/and predictive model. And according to the given material and yarn parameter and predicted yarn fineness, it can be calculated that the parameter range of material fiber which is to be controlled. Finally, the yarn is span according to the practical production requirement, come up to the desired yarn irregularity, and the control of irregularity in yarn fineness is completed. The built predictive model is consistent with practical production. The control method on the basis of the model is precise and convenient. It's also avoiding such problems as either works and material wasting or poor quality of yarn which happened in controlling the irregularity of yarns by experience. The control of irregularity in yarn fineness becomes convenience and practical by using this method which is with much higher manipulability.

Abstract: Empirical analysis on typical product categories, product series and Price Index of every level of single species is made by using classical ARMA models as well as ARCH models, which based on the actual data sampling and network. This study sets up AR models with ARCH effect of timing of product operations Index that judged by LM test used as model identification, and then establishes corresponding mathematical quantitative model for prediction. All of these are carried out by the Metrical Economics and the Eviews software. With time series, the fitting and prediction for running change-trend of silk are also in the theoretic confidence interval, which can also verify the degree of accuracy and precision of the established model.

Abstract: Paper printability is an essential property affecting printing technology and quality. In this research, a predictive model for paper comprehensive printability is presented based on a simple analytic expression. Five properties of two-sided offset paper are tested with smoothness(k1), whiteness(k2), glossiness(k3), ink absorption(k4) and surface strength(k5). Weights for the properties are calculated by principal components regression (PCR) with software SPSS. Then the expression M(k1 , k2 , k3 , k4 , k5) between paper properties and printability synthetic value is obtained. In the same way, weights for the printing quality performances including density(l1), hue error(l2), gray scale(l3) and saturation(l4) are calculated by PCR and the expression N(l1 , l2 , l3 , l4) between printing quality performances and synthetic value is got. Then the fitted curve Y(x) (x is paper printability synthetic value; Y is printing quality synthetic value) is derived and x is replaced by expression M(k1 , k2 , k3 , k4 , k5). As a result the predictive model for two-sided offset paper comprehensive printability can be developed by the paper properties tested. It is shown that the predictive model can provide a practical and quantitative method for evaluating the comprehensive printability of two-sided offset paper fast and can meet the industry analysis requirements with good precision.

Abstract: The effects of seven key technological properties(ink viscosity, anilox screen, type of doctor blade, plate shore hardness, plate relief depth, type of mounting tape and printing speed) in flexographic printing on printing quality properties(printing density, print contrast and run lengths) were investigated. And the best interactions were analyzed by orthogonal design of seven factors and two levels experiments. Predictive models were built by linear multiple regression analysis(MRA) and BP(Back Propagation) neural network respectively. A comparison of accuracy for models were discussed by a brief statistical analysis of their predictive errors. Results indicated that the most important factor influencing print density, print contrast and run length was type of doctor blade, plate relief depth and print speed respectively. The model of BP neural network provided higher prediction than that use of linear MRA. However, the linear MRA model was convenient to be used in production.

Abstract: Micro abrasive air jet machining technology is being increasingly used in the fields of micro cutting. Since the aspect ratio is a major interest characteristics of kerf in micro cutting, an experimental investigation is carried out to study the effect of cutting process parameters on the aspect ratio in this study. It is found that the aspect ratio increases with an increase in air pressure, abrasive flow rate and jet incidence angle, while decreases with an increase in nozzle traverse speed. Furthermore a predictive model for aspect ratio is developed using the dimensional analysis technique. It is shown that the model predictions are in good agreement with the experimental results. The research results may be meaningful to efficiently control the aspect ratio.

Abstract: The insufficient information about the mechanism of EDM process affects decision-making related to operation, parameter adjustment, and automatic control. Without a thorough understanding of the process dynamical features, it is difficult to determine a suitable model to describe such processes. This research employs the chaotic analytical techniques to determine the level of complexity of an EDM process; the correlation dimension method. With the understanding of the process complexity and its composition by correlation dimension analysis, a well-defined model is proposed. Finally, by using this model structure and size, an online time-varied predictive model was developed and verified by another experiment.

Abstract: Springback in drawing of an axial symmetry sheet metal part was studied. First of all, according to a springback formula of simple bending, the moment was modified by concerning of axial symmetry shape of part to construct a formula for predicting the springback. Then, springback process was analyzed with FE software Dynaform. Finally, drawing experiment was carried out. Use the reverse engineering and software Geomagic to obtain the shape and springback of the part. Excluding errors of measurement and simulation, the results of experiment and finite element simulation and predictive model were compared, and it shows that the predictive model works well.

Abstract: In this paper, a methodology for determining the optimum machining parameters for a face mill operation is presented. Taguchi design of experiments was used in company with orthogonal arrays in order to study the effects and interactions that selected parameters had on the cutting behaviour. The parameters that were included in the study were cutting speed, feed rate, depth of cut and the effects of cutting fluid. Using the orthogonal array experimental runs were set up and carried out on specimens of metal plate on a CNC milling centre. The studied material was 13% Austenitic Manganese Steel. The response factors for the experimental runs were surface finish, spindle load, material removal rate, cutting forces and surface hardness. From the results and using the orthogonal arrays a set of empirical models were derived for each response factor which could be used to predict the optimal controllable factor settings for a given production criteria.

Abstract: In the paper, genetic algorithm is introduced in the study of network authority values of BP neural network, and a GA-NN algorithm is established. Based on this genetic algorithm-neural network method, a predictive model for fatigue performances of the pre-corroded aluminum alloys under a varied corrosion environmental spectrum was developed by means of training from the testing dada, and the fatigue performances of pre-corroded aluminum alloys can be predicted. The results indicate that genetic algorithm-neural network algorithm can be employed to predict the underlying fatigue performances of the pre-corroded aluminum alloy precisely, compared with traditional neural network.