Research On Knowledge Discovery Of Sheet Metal Forming Based On Data Mining

During stamping process design, it already became an important factor for a company to analyze the obtained information, extract knowledge and use the knowledge to assist and support designers. Because the technology of CAX is widely used, it is easier that people get the data. It is more urgent and important to extact the value knowledge and information hided in data and feedback to guide design. To research and develop data mining technology in stamping process design is significant for improing competition of stamping tool design and pushing the development of tooling industry.According to the feature of information flow of stamping process design and state of design, data mining technology is creatively brought in intelligent design system of stamping process design.. Based on analysis of design theory, the function of design system, input of information flow and the objective related the function are closely linked up. The ideal and flow of intelligent design system for stamping process design integrated data mining are firstly proposed. It emphasizes the importance of data value. On the foundation of new design ideal and flow, the architecture of the system is established. According to ideology of system approach, the hiberarchy model of the system is designed by abstractly stratifying and classifying system, which implements synchronous and harmonious information communication in the system. The information flow becomes bidirectional in the new system. The flow of design is not from design to validation but design to validation to design. The result of experiment data and numerical simulation can be fed back to guide stamping process design.The synthetical information model of system is put forward and established, which includes geometrical information of product, process information, knowledge of process planning and information of die and equipments. The structure and representation way of these information sub models are respectively studied. Especially, on basis of non-linear process planning (NLPP) theory, the representation way of stamping process design based AOS tree is set forth. The information of geometry and process is intergrated in this model. In this paper, process and knowledge information in the synthetical information model, hiberarchy of geometry, process and knowledge information and mechanism of information delivery are proposed to realize integration of feature, process and knowledge information by applying OOM and hierarchical structure of classes.According to experiment and numical simulation data feature of stamping process design, the paper provides discretization algorithm for continuous attribute based on fuzzy similar-ratio theory.By profoundly research on theories of rough set, principal component analysis and Artificial Neural Network, the paper creatively designs several data mining algorithms. The following data mining algorithms are proposed based on study on existing algorithm, which mainly focuses on research on rough set. These algorithms include heuristic reduction algorithm based on discerniblity ability of attribute, attribute reduction algorithm based on roughness of attribute, value of condition attribute reduction algorithm, rough set induction learn algorithm, principal component analysis algorithm and artificial neural network algorithm. The traits of these algorithms are compared with existing algorithms. In stamping process design, the objection of extracting process knowledge can be achieved by using these algorithms.Based on double bases cooperating mecha-nism, the model of data mining is initially presented by integrating blan process planning task with control of data mining, which is composed of analyzing data, main body module of data mining and control module. Moreover, the mechanism of controlling and harmonizing is provided. In the paper, in order to implement communication of data mining, the structural data mining language-FEDML for engineering field is designed by adopting analogous syntax with SQL, which is based on Backus-Naur Form. Moreover, the way of implement is given. By analyzing characteristic of FEA result data and Object Oriented hiberarchy modeling of simulation data, neutral files model of numerical simulation result data and its EXPRESS description are established based on STEP standard. It establishes the relation between entity objective by using elements, nodes or serial number of material parameters and process parameters table as id of data table.In this paper, knowledge management model for data mining is proposed. The knowledge extracted is applied into stamping process design and embodied into value of product. The evaluation algorithm of knowledge mined is designed. The paper brings forward algorithm and strategy of mined knowledge refinement which is based on Artificial Neural Network structure. Meanwhile, based on ideology of decomposing complex problem to get solution, the task of stamping process design is decomposed into some subtask. The mechanism of communication about design task and related knowledge objective is provided. Structure model of blanking process planning knowledge is set forth. The task of blanking process planning is decomposed into multi-subtasks. The reasoning model of blanking process planning based on black board model is established.At last, research works are validated by four cases analysis. To character of experiment data, the paper uses two cases to validate application of data mining. The one is multi step stamping process design of mini type and high precision part. The other one is calculation of developed length for bending part. In two cases, the paper discusses issues of current sheet metal forming in detail. The method of data mining is used to analyze experiment data. The validity of data mining in application of stamping process design to experiment data is proven by practice. Meanwhile, during study on discovering knowledge from numerical simulation data by data mining, the paper also uses two cases to explain application of data mining. The one is spring back analysis during U shape bending. The other is numerical simulation cases analysis of inner door panel of automobile. The algorithms of data mining designed in the paper are applied to thewith two cases. Knowledge discovery from numerical simulation data is realized.