Coal is the most abundant fossil fuel known in China,and also a vital global energy source.Today,the main consumption of coal is pulverized coal combustion for power generation which leads to environmental pollution and wastes.Direct coal liquefaction is an advantageous approach for the clean and effective utilization of coal,and also an effective way to solve the energy shortage problem in our country.For the successful process design and its engineering of coal liquefaction,it is inevitable to implement the liquefaction modeling study,i.e.,exploring the reaction schemes involved,deriving the kinetics rate expressions and identifying the major influencial process parameters,so as to quantitatively describe the complicated reaction process taking place.This research was performed with a partial support of the National Basic Research Program(973 Program)of China,aiming at the modeling and simulations of the kinetics of Shenhua coal direct liquefaction carried out in batch reactors.The objectives of the present work are 1)to develop the liquefaction kinetics model for Shenhua coal by incorporating thermodynamic calculations,and 2)to build reactor models of direct coal liquefaction by using artificial neural networks(ANN)and support vector machine(SVM)for simulating the performances of batch reactors, respectively.First,a batch reactor model for Shenhua coal liquefaction was developed,taking the co-existing vapor and liquid phases into account. Based on the kinetics data reported in literature,we derived quantitatively the reaction kinetics of Shenhua coal liquefaction.It is demonstrated that it is necessary in the interpretation of the kinetics data to calculate the concentrations of each component present in the liquid phase by thermodynamic calculations.The coal was divided into three parts,i.e., easy reactive part,hard reactive part and unreactive part.The easy reactive part generates directly the oil plus gas(O+G);the hard reactive part generates the preasphaltene(P).The counterreaction from P to(O+G)is inoperative,however the counterreaction from asphaltene(A)to P will response at high tempretures.Under lower tempretures,the conversion of P to A is the rate limited step for coal liquefaction,while at a high tempreture, the conversion of A to(O+G)is dominant.The simulation results of artificial neural networks(ANN)reactor model and support vector machine(SVM)for batch reactors show that, while the elective range of coal and solvent is narrow,the most influencial factor in coal liquefaction is tempreture;at the same time,the type and particle of the coal is assignable cause for oil yield.When the system researched is extended into eleven influencing factors and wide factor range,Coal variables(coal type and particle size)are the most important factors in the coal liquefaction system.In addition,particle size, temperature and gas pressure have a profound influence on the result of coal conversion,to the contrary,solvent type is a more important factor toward oil yield than coal conversion,those four factors(degree of fill, mixing,reaction time and solvent/coal ratio)have the great effect on coal conversion and oil yield.Compared with other factors,heat-up time and drying plays a less important role in liquefaction process.
Post about "artificial neural networks"
Applications of Chemometrics in Photometrical Simultaneous Determination of Multi-components Mixtures
In this paper, chemometric methods were used to assist the spetrophotometry in the simultaneous test of multi-component reactive dye mixtures with the stimulant industrial conditions.The samples of fixed multi-components were reactive red195, reactive orange122, reactive yellow145 which maximal absorbency were 540nm, 490nm, 417nm, respectively, and the spectrograms overlapped badly. Orthogonal table designed the experiments. The data obtained from spetrophotometry were be used after calculated the correlations between different wavelength points. If the correlation approached 1.0, we considered they were the reiterate informations. If the correlation approached 0.0, they were considered had no useful information. Then the data selected were processed by different methods. The average differences between the 3-component prediction results of the 2 sets of unknown samples obtained from the methods of K-matrix, principal component regression (PCR) and artificial neural networks (ANN) were 6.243%, 2.546%, 25.42%; 6.243%, 2.546%, 25.42%; 3.375%, 2.388%, 2.670%, respectively.The influence of pH values, water hardness, surfactants concentration, etc., on the test was investigated. pH values were tested if different values will interfere the determination in the range between 8.5 and 12.5, the results of test indicated that 3-component mixtures had a best stability in pH=10 and had no influence in multi-component simultaneous determination. There also have no influences in there levels designed by water hardness standard. OP and sodium dodecylbenzenesufonate, two surfactants as the textile auxiliaries were selected, results of experiments indicated the determination errors under 1%, within the error of spetrophotometry.Another part of this paper, the method established was be used in calculating the discolorations of multi-component dying wastewater after adsorbed by improved fly ashes (FA). Ascertained the best ways to improve the FA, the best adsorption time, dosage of FA and adsorption temperature, etc. through experiments. The most befitting temperature was 500℃, the ratio of Ca(OH)2/FA was 1:9. It can easily get the concentration of each component before and after adsorption in the stimulant wastewater by the FA with the method of ANN assist spetrophotometry. Discolorations can be obtained from the formula designed for calculating it. From the results of this application, this method has the significative meanings in the fields such as sorbent selection and preparation, techniques’improvement, etc.
At present,the ceramics study of the material developed is the traditional “trial-error”. researchers need do a lot of repeated experimental work. In response to this problem, computational intelligent was applicated to optimize the design of ceramic materials in the paper.Computational intelligence,known as a”soft approach”,is based on a numerical calculation of intelligent methods. Its flexibility, versatility and rigour are significantly better than knowledge-based artificial intelligence technology. Its biggest character is that it does not require setting up a precise model itself, and does not rely on the knowledge description, but directly processes the input data and giving the results.Therefore,the development of computational intelligence expands the traditional computing model and intelligent theory as an emerging discipline,and lays the foundation for the material intelligent design.This paper introduced the BP neural network, and analyzed its application on the optimizal design of ceramic materials.It briefly introduced the neural network toolbox of MATLAB software,and sum up the concept,structure, parameter selection of neural network technology, and the principle of using the toolbox to design network.Using this method,it established the mechanical properties of composite ceramic material with the group distribution ratio between the nonlinear mapping and the neural network model,and develops a ceramic material simulation system with the neural network :MATLAB software.And analyzed the case of Al2O3/SiC/Ti（C,N） composite ceramic materials. The results showed that the mean square error of data forecasting could be controlled within five percent, and the relative error could be controlled within 3 percent.Supportted from the MATLAB software platform,using of successive steps return method induction,it had be summed up material assigns the mathematics expression between the ratio and each functions index.The optimum of every function parameter and corresponding material ratio set had been gotten . The best hardness value was 19.5632GPa, SiC was 13.66%,Ti （C, N） was 32.48% respectively; The best tearing tenacity for was 5.3162 MPa.m1/2 ,SiC and Ti （C , N） contents was 14.563%,15.225% respectively; the best bending strength value was 723.82 MPa,SiC was 1.48%,Ti （C, N） was 28.15% respectively.Finally,combining the genetic algorithms and neural networks,a group of good percentage of the optimal ratio of components of Al2O3/SiC/Ti （C,N） composite ceramic material with better integrated performance was obtained, Al2O3 was 74%,SiC was 10%,Ti （C,N） was 16%.According to the ratio,composite ceramic die materials was preparated,and the performance test is carried out.
Casting industry has a decisive influence on the national economic development. As the casting industry developing to the scale and sophistication of the production process, the request about sand binder is getting higher and higher. As CO2 hardening alkaline phenolic resin is synthesized by a cumbersome process, the neural network model and the combination of traditional experiments and computer application are used to optimize CO2 hardening alkaline phenolic resin synthesize process. This method can improve the performance of sand moulds and save test materials use.Artificial neural network has started the rapid development as a non-linear science from the end of the 1980s. Artificial neural network from the experimental data through self-learning automatic access to a unique mathematical model of the superiority of its people without pre-set formula to the form, but in the experimental data based on a limited, iterative calculations, we can get a experimental data reflect the inherent law of the mathematical model .Neural network excels at dealing with the problems with unobvious regulation and too many variable components.The synthesis principle of CO2 hardening alkaline phenolic resin was analyzed first in this paper, a certain amount of experiments were carried on after comprehending the synthesize process, so enough data for neural network model study and verification were prepared. In accordance with the function and characteristic of artificial neural networks technology, used the antipropagation algorithm （the BP algorithm） to establish the reflection phenolics synthesis craft and the performance neural network model. In the training process, to speed up the network study speed and avoid the network shaking fiercely, an additional momentum method was used to improve the BP algorithm. And the BP neural network model was established finally. Simultaneously, the CO2 hardening alkalinity phenolics cementing agent’s flying into a rage quantity, defeated and dispersed shop characteristics and so on nature were also taken to carry on deeper tests, which proved the CO2 hardening alkalinity phenolics cementing agent’s superiority further.. Finally the efficient analysis about the CO2 hardening alkalinity phenolics cementing agent was made, indicating that it has prominent production use value and economic efficiency as a kind of casting cementing agent.
Prediction of coal spontaneous combustion is very important for preventing coal self-ignition. Testing of self-ignition trend and time is the basis of prediction technique of coal spontaneous combustion. The way to test self-ignition time accurately is spontaneous combustion simulating experiment, which takes more than one month and over 1 ton of coal sample. In this text, an improved oil-bath programmed temperature oxidation experimental device is adopted, which can test oxygen consumption and gas generation rate of coal during self-heating process precisely. It is thinked that there is corresponding relation between self-ignition duration and oxygen consumption rate, carbon monoxide as well as carbon dioxide generation rate of the coal at different temperature of self heating process. According to the corresponding relation,three prediction models are built in this text which are fuzzy clustering and fuzzy model identification method、minimal two multiplication method and artificial neural networks method.These three models can predict the coal spontaneous combustion.Fuzzy clustering and fuzzy model identification method :in the first ,using fuzzy clustering classfies the known coal self-ignition duration data,then according to the programmed temperature oxidation experimental data, using fuzzy model identification predicts the unknown coal self-ignition duration.Because using this method predicts according to the classfication result, so the calssfication result is the fuzzy result and we can not receive the precise self-ignition time.On the basis of theoretical analysis ,the theoretical model of corresponding relation between self-ignition duration and oxygen consumption rate, carbon monoxide as well as carbon dioxide generation rate of the coal at different temperature is built.Using minimal two multiplication predicting method can obtain coefficient of the model. So we can predict the coal self-ignition time and analyse the coal self-ignition trend. Because the model build on the theory basis,using this method we can receive preciser result ,but the calculation process is intricacy.Artificial neural network method is built according to the corresponding relation between oxygen consumption rate, carbon monoxide as well as carbon dioxide generation rate of the coal at different temperature and self-ignition duration .Train the artificial neural network using known coal self-ignition duration experiment data, so we can receive the joint strenght of nerve cell. Substitute the programmed temperature oxidation experimental data and coal quality analysis data in the artificial neural network,so we can calculate the coal experiment self-ignition time.This method is convenience and we can estimate precision of prediction.But abundant sample is needed to train the net.In practical work,we can apply different mathematicl predicting methods to predict the coal spontaneous combustion according to the known data and requirement.
The plastic materials have more and more widely applications nowadays, with higher demands on molding quality. It is increasingly urgent to study the injection molding process optimization technology for optimal process conditions.In order to avoid the blindness of injection molding design, an optimization method adopting twice orthogonal experiments is proposed, together with signal to noise ratio analysis, range analysis, variance analysis and CAE simulation, so that the sufficient and exact experimental information is obtained with the significantly reduced experiment times. The optimal injection molding process parameters is then acquired by analyzing and comparing the experimental information. Excellent products are accordingly manufactured in actual process.The main work and results are:1. Filling time, mold temperature, melt temperature, packing time and pressure are selected as process optimization objects based on the flow nature analysis of polymer melt.2. Two different types of orthogonal experiments are consequently introduced. The injection molding process parameters are refined and then the optimal parameters are achieved conveniently by considering the interaction and importance of process parameters.3. As for a mass of experimental data, the error of single set of data is controlled through signal to noise ratio analysis, the error of entire experiment data sample is controlled through variance analysis, and the quality trend of the molded parts along with the process parameters is obtained through range analysis.4. To improve further the prediction speed of injection molding process parameters, the neural network method is introduced. After training the obtained good experimental sample, prediction error of neural network for corresponding process parameters is smaller. Therefore the neural network can be used for trend prediction, but not suitable for quantitative prediction.5. The corresponding molds should be completed before practical molding tests, so the improper molds with structural defect must be repaired. CAE technology and experimental optimization techniques are applied to optimize the injection molding process, so that the process optimization precedes and directs the mold manufacturing. Hence optimal injection molding process parameters are obtained keeping away from the molds with serious defects.6. An injection molding process query system is tentatively studied aiming to store the optimized parameters and provide references for later optimization work, in view of many kinds of plastic parts, various molding process and an initial value in optimizing the process parameters.7. The proposed optimization method is exemplified to be of certain guiding significance in practical injection molding technology. In the future research, computer technology applied to injection molding process optimization should be lucubrated based on the demand of manufacturing information engineering and the injection molding process research should be integrated with mold manufacture for better research results.
Study on the Applications of the Water Environment Model in Baotou Section of the Yellow River Based on Artificial Neural Network
Yellow River is one of most multi-sediment-laden river in the world. The soil erosion is serious because of its fragile ecological environment of the upper and middle reaches of the Yellow River coupled with the disorder in the development of water resources. There was a severe effect on the water resource environment of the middle and lower reaches region of the Yellow River whose water quality had further deterioration because of the territory of another by municipal wastewater, industrial wastewater non-ferrous metals, and paper industrial wastewater pollution when it entered Baotou City. The water quality management of results and future trends of Baotou Section of the Yellow River has become very concerned about the issue, therefore, it is particularly important to accurate evaluate and forecast the conditions of the water environment.In this paper, the Baotou Section of the prediction of the Yellow River water quality monitoring sections related indicators and statistical manual neural network model was established with the use of the DO, COD, ammonia nitrogen, heavy metals, such as the concentration of water quality objectives for 10 values of output parameters that were easy to collect the water quality indicator concentration time series as a training set. And it was used to evaluate the water quality with satisfactory results and Initial examinations of the practical value of the model, the establishment of a water quality model is simple and practical to meet the supervision and management of water resources protection needs, but also it could carry out similar work experience for the future of the Yellow River.Zhaojun tomb monitor area of Baotou Section of the Yellow River as a test section water quality data set has been established on the BP network tested, BP network comprehensive evaluation of water quality in line with reality, than any other evaluation method has higher accuracy, but the water quality category will be more quantify. The water samples obtained at or near the water quality of the type of fall. This type of water that is more objective, comprehensive evaluation of water quality testing BP network to the water quality of category examples of evaluation results to make more specific and accurate.
Sepharose is a carrier for preparation of the protein A immunoadsorption column,and its industrial-scale process of activating reaction has been the key issues of batch production for the blood purification material.In the activating reaction process of the sepharose,the optimization and scale-up technology of the stirred tank reactor is a crucial factor to industrial processes.Although there are a number of experimental and theoretical studies for stirred tank reactor,the relative theory and design method is still not perfect.In recent years,researches on flow and mixing characteristics of stirred tank reactor by using Computational Fluid Dynamics method has shown remarkable advantages,and leads the development of mixing techniques.In this paper,the single-phase flow field is simulated with standardκ-εturbulence model and multi-reference frame method of steady-state flow using FLUENT software.To verify the feasibility of the CFD,the results are compared with the experimental data of literature.Then the single-phase flow field of sepharose is simulated.Through the researches of the flow pattern, velocity field and shear force field under the different type impellers conditions,it is found that 45°pitched blade down flow turbines(PBTD45) impeller has higher circulation flowrate and lower shearing effect,its performance is superior to rushton disk turbine(DT).So the PBTD45 impeller is recommended for the activating reaction of sepharose.The suspended characteristics of solid-liquid in stirred tank reactor,as solid catalyst and sepharose,are simulated in this paper.Through the researches of solid phase suspended performance,it is found that PBTD45 is superior to DT and PBTU45.Just-suspended impeller speed of the three impellers is respectively simulated as 948rpm,570rpm and 1152rpm,and the results are compared with the Zwietering correlation.The maximum relative deviation is 13.4%. In conclusion,the optimal operating conditions for activating catalyzed reaction of sepharose in 20L stirred tank reactor are:with PBTD45 impeller,N=570rpm and D=0.10m.Additionally,artificial neural networks(ANN) technology is introduced,with MATLAB as a software platform,to establish a prediction model for sepharose activated reaction,which is a 3-layer BP network structure.The predictions for two experimental conditions,shaked flask and stirred flask,are respectively made using ANN models.Compared to the experimental data,the average absolute relative deviation is less than 6%. Finally,scale-up technology of bioreactor for preparation of blood purification materials is studied.Then the large scale biorcactor is simulated and optimized,which can provide some theoretical basis and design guidance for bioreactor optimization design and scale-up.
Fabric drape is a deformation phenomenon that the fabric sags due to its weight and it is an important factor which determines visual beauty of the fabric. With the clothing of good drape performance, it is able to form a smooth surface modeling, and it can fit properly with the human body, and it gives visual comfort. Therefore, the fabric drape performance is one of the main factors, which is related to fashion design and marketing, one of the main factors. Today, as people constant run after beautiful appearance of clothing, as well as the emergence of a variety of new fabrics and ultra-fine fibers, researching the fabric drape performance has become increasingly important.The paper combines the expertise of clothing textiles and design ideas of software engineering, realizes prediction of fabric drape performance. First, researching on the fabric weave construction parameters and the fabric drape performance parameters, and then introducing neural network and its application of the fabric drape performance prediction. Then a detailed analysis of the BP neural network model structure and BP Algorithm. The intelligent neuron is introduced, which improves the BP neural network. Next, according to data features of the fabric weave construction parameters and the fabric drape performance parameters, as well as the comparison of different algorithms, finalized a more reasonable prediction algorithm–Levenberg-Marquardt algorithm, established prediction model which could more accurately predict the fabric drape performance.The general framework of the final from the system analysis and design and back-office databases, further details to clarify the design process and the implementation processes of system. According to requirement of the system, based on the national standard which is related to the textile industry and referencing the ID rules of the library call number and combines the management and inquiry of database to research the sample code of fabric. And design the standard codes to meet the system requirements. Finally, Established the fabric drape performance prediction system. The system achieves inquiry function of fiber; yarn, fabric-related properties and fabric drape performance indicators, as well as prediction function from the known parameters of the fabric weave construction to drape performance indicators.
Simulation and Optimization of the Stamping-forging Hybrid Forming for Miniature Aluminum Alloy Bearing Block
The stamping-forging hybrid forming is a new process for non-uniform thickness sheet, which is combined of the sheet forming and the bulk forming. It can substitute for many traditional processes such as casting, welding, cutting and so on, for both having the advantages of sheet forming–big distortion area–and bulk forming–non-uniform thickness, and not only raise the production efficiency, but also improve mechanical property of work piece. In view of certain type of miniature bearing block, the stamping-forging compound forming process was used, the forming rules of essential location were studied, and the combination plan of optimal process parameters was gained through optimization in this article.Firstly, the extrusion process was chosen to form the small columns on flange through the process analysis of the miniature bearing block, and the feasibility of the method was confirmed by analyzing the flowing rules of material during forming. Secondly, in order to avoid the influence between two connectional stations the combined blanking and extrusion process was taken into consideration to form the cylinder-shaped location. The numerical simulation analysis of the forming process was carried on to obtain some forming rule with the help of the professional finite element software DEFORM. The result indicates that the deformation zone of metal materials is located under the punch during the combined blanking and extrusion process, and below the binder board only the elastic deformation takes place which does not have the influence on the formed shape of the previous step.After analyzing the rules of material deformation during the combined blanking and extrusion forming process in general, the method combining numerical simulation and orthogonal design was selected to analyze the effects of the three key parameters such as friction coefficient, the force of counter-punch and BHF towards the damage. The sensitivity of index arm on processing parameters was estimated by range analysis. The result shows that friction coefficient is the main influential factor to the damage, the force of counter-punch is medium one, and BHF is the smallest one. Based on the consequence of the orthogonal design, the damage was chosen as the optimized goal, and the genetic algorithm combined with artificial neural networks was taken to optimize the processing parameters. The optimized result indicates that after the optimization the processing parameters can obtain the lowest damage value, so enhance the forming property of material enormously, and the quality of work piece obtained correspondingly is optimal as well.Finally, the fine blanking process was used to form the final shape of the work piece and the simulation of the process was carried on to obtain the reasonable processing parameters for producing a satisfied work piece. Meanwhile, the germination and the expansion of crack during fine blanking process in different fracture criterion were discussed through fine blanking simulations to obtain the influence law of stress triaxiality on fine blanking crack and to provide the essential basis in choosing an appropriate fracture criterion for simulating the fracture process effectively. The result indicates that the smaller the stress triaxiality is, the later the crack germinates; while the tooth ring with tooth profile can enhance the hydrostatic stress and cut down the stress triaxiality. The simulation of germination and the development of the crack in thin steel sheet fine blanking with the N-CL model are more coincided with the practicality.The results of this research have the instruction function to the practical production.