Study On Support Vector Machine-Based Modeling Methods And Their Applications In Material Processing

It is well known that heat processes are nonlinear, multivariable, timevarying and strong coupled. Thus, the heat processes are typical complex processes. It is too hard to obtain accurate mathematical models of hot work processes using classical modeling methods. Obtaining knowledge models of hot working processes is helpful for understanding the processes, obtaining and even abstracting the experiential knowledge of human and realizing the intelligent control of these processes. Thus, it is of great significance to obtain knowledge models of hot processing of metal.In recent years, it is becoming the focus to get knowledge models of hot work processes with the fuzzy set method, the neural network method, the rough set methods and their combination. Many interesting results have been obtained, however, they could not completely satisfy the practical needs. Further modeling techniques are necessary.Considering the complexities of hot working process, this dissertation proposes Support Vector Machine-based Fuzzy Rules Discovery System (SVM-FRDS) and C-weighted On-line support vector regression (COSVR) kowledge modeling method of complex processes based on the SVM and fuzzy system theory. Key algorithms of the method are studied in detail. Verifying experiments using chaotic Mackey–Glass are carried out. The experimental results show that the SVM-FRDS model possesses good generalization capability as well as high comprehensibility, the COSVR method can change the model more effectlively. The proposed approaches are applied in the heat processes. The proposed SVM-FRDS and COSVR methods are applied in modeling and control for the aluminum alloy pulse GTAW welding dynamic process. For controlling of carbon potential using an oxygen sensor we analyse the influencing factors and study the amendment model.The experimental results prove that the SVM-FRDS and COSVR method can effectively obtain the knowledge model in the field.The main work is as follows:1. In general, comprehensibility is one of the required characteristics of reliable systems. Especially in the complex hot work field, if the model is comprehensive, or even revisable, the model may be more reliable and adaptive. This paper discusses a support vector machine-based Fuzzy Rules Discovery System (SVM-FRDS). The character of SVM in extracting support vector provides a mechanism to extract fuzzy IF–THEN rules from the training data set. We construct the fuzzy inference system using fuzzy basis function. The gradient technique is used to tune the fuzzy rules and the inference system. Main algorithm of SVM-FRDS is given. We theoretically analyze the proposed SVM-FRDS on the rule extraction and the inference method comparing with other fuzzy systems. Comparative tests are performed using chaotic Mackey–Glass benchmark data. Comparative analysis and tests about SVM-FRDS with respect to other fuzzy systems show that the new approach possesses satisfactory generalization capability as well as high comprehensibility.2. There are many influencing factors, even uncontrollable factors in the hot work processes, so the time-varying model is more acceptable. Based on the on-line support vector reression, this paper investigates C-weighted On-line Support Vector Reression approach (COSVR). In COSVR parameter C varies for different samples. This approach efficiently updates the trained function whenever a sample is added to the training set. Comparative tests are performed using chaotic Mackey–Glass benchmark. The experimental results show that the method can change the model more effectlively.3. The proposed SVM-FRDS and COSVR methods are applied in modeling and control for the aluminum alloy pulse GTAW welding dynamic process. 1) Verifying experiments on obtaining knowledge models for GTAW welding process using the proposed SVM-FRDS and COSVR modeling method are carried out. For SVM-FRDS, the model possesses good generalization capability as well as high comprehensibility, and the SVM-FRDS adaptive inverse control method is feasible in welding process control. For COSVR, the method can change the model more effectively. 2) A new adaptive inverse control method based on SVM-FRDS is proposed and applied in GTAW process control. The proposed adaptive inverse control method can automatically extract control rule from the weld process data. The welding experiments results show that the SVM-FRDS adaptive inverse control method can achieve uniform weld formation during the pulsed GTAW welding.4.The oxygen sensor is widely used in measuring the carbon potential (CP). There exists a deviation between the real CP value and the measured CP value using an oxygen sensor. It is very important to study the amendment model for controlling of CP using an oxygen sensor. We select the relevant variables according to the practical experience. We measure and record the relevant data according to the practical experience. We analyse the influencing factors and build sinle-variable and multi-variable models using the SVM approach. Under the guidance of the knowledge model, based on the carbon potential relevant theory, we build the mechanics model. We build the knowledge model using the proposed COSVR, and the model has higher precision. The COSVR modeling approach builds an important foundation for the practical application of amendment model.5. C-weighted On-line Support Vector Regression Based Knowledge Discovery System for Hot Working (COSVRKDSHW) is designed and developed. The software system includes all functions needed by on-line support vector regression and c-weighted on-line support vector regression modeling method and integrates some auxiliary functions.