【正文】 1. IntroductionThe research of bining neural network with expert system has been paid much attention all over the word. At present, artificial intelligence expert system is widely used in CAPP, which is the main part of the integrated CAD/CAM. But there are serious problems in building CAPP system, which are the representation and inference of process knowledge. This paper uses feature technique, and bines artificial intelligence expert system with neural network to solve the above problems. Description of Workpiece Information The description and the input of workpiece information are a prerequisite torealize present, the geometry model, technique widely employed in CAD can’t be used in CAPP directly, for it provides only the geometry information of workpiece, and lacks the manufacturing information. Now the method based on feature technique is used to describe workpiece. The diagram workpieceMain direction levelPosition surface levelManufacturing feature levelGeometry informationManufacturing informationof workpieces based on feature hierarchy is shown in . The diagram of workpiece based on feature hierarchyThe main direction is defined by the normal line of workpiece surface. The position surfaces are surfaces which are at the different position, but have the same normal line direction. The manufacturing feature includes various surface, holes and grooves etc. Every feature is given by the corresponding parameters.3. The Representation of Process Knowledge Generally, the task of engineering design may be divided into a series of subtasks. The different subtask needs different knowledge. According to the characteristic of process design, the new representation method of multiknowledge is put forward:The representation of knowledge based on frame representation of knowledge based on frame structure can show not only the attribute of the thing, but only the subordination of different things. The frame structure expresses factual knowledge including blank fact, tool fact and machine tool fact etc. in our CAPP representation of knowledge based on production rule The characteristic of production rule is to use inplete elicitation knowledge to solve question. Its expression way is:IF W1(x) THEN W2(x)In this system, the production rule is used to represent the control knowledge whose content of knowledge is little and whose rule is representation of knowledge based on neural network Generally, selection knowledge has a large proportion in knowledge base of process design system, including the selection of blank, the machining method of feature unit, the machining tool and clamping apparatus etc. In existent CAPP system, the representation of knowledge usually es form the production rule of AI expert system. Although the great progress has been made in the research and application of expert system, there are also some problems:(1)“battle neck” and“narrow flight of steps”in acquiring knowledge.(2) The inference ability is weak. The problems of mach conflict and bination explosion usually arise.(3)The intelligence level is low. On the contrary, neural network technique is more effective in knowledge acquirement and inference. In this paper, BP network is regarded as rule memory. Knowledge acquirement is realized by the training of learning sample, and the inference is acplish by multiorder forward spread of network. In every neural network, knowledge is shown by network structure and value of joint weight [(,)] and threshold ((θ)). They are distributed in the whole neural network. The process of acquiring knowledge is shown in . In this system, every neural network is regarded as a module to represent and infer the partial knowledge. For example, system uses production rules to represent manufacturing knowledge of hole.xmynW(z1,y1)Y2Y1zkW(z1,y1)Z1W(zk,yn)xMzkX1W(xk,zk)( x1 ,…,xM,y1t…ytN){W()}qq—training order, ynt—the aim of yn==D or {W(,)}q={W}D—given precision {W}—constant weight array the process of acquiring knowledge It needs ten rules, and condition items are many, structure is plicated. If ten rule condition items as training sample. The rule condition items as input of network and condition items as output of network, a 10╳9╳4 BP neural network can be built. The knowledge of selecting hole process plan may be stored conveniently by network structure and value of joint weight and threshold. Where diameter is divided into four conditions。在此期間，我遇到了許多困難和疑惑，但在老師和同學的幫助下，得以在設計的期限內按要求順利完成設計。圖615 分析對話框　圖616 分析定義對話框　圖617 電動機設置對話框步驟4：結果回放，話框單擊主工具欄中“回放”按鈕，系統(tǒng)彈出回放對話框（如圖618所示），單擊“演示”，此時彈出動畫對話框（如圖619所示），單擊“播放”按鈕即可播放動畫，單擊“捕獲”按鈕，彈出“捕獲”對話框，保存畫片。步驟3：重復步驟2導入組件“”, 確定后系統(tǒng)彈出元件放置對話框，選擇連接類型為滑動桿，連接的名字命名為“xiamo1”，約束欄中“軸對齊”選擇元件和組件在導套和導柱各自的軸線，約束欄中“旋轉”選擇元件和組件中兩個平行的面作為它的旋轉約束，選項卡中顯示完成連接定義，單擊確定。子組件1：上模板部分步驟1：單擊菜單文件新建命令，打開新建對話框，選擇組件類型，輸入新建文件名稱“shangmo”,然后確定，進入組件工作環(huán)境，選擇子組件，進行裝配。如圖64所示。圖54 彈簧先復位結構圖第6章 模具三維造型及動畫設計 三維造型成型零件無論是動模各件還是定模各件，它們的畫法都基本相同，下面介紹應用Pro/ ENGINEER野火版定模固定板的畫法： 文件—新建命令，打開新建對話框，選擇零件類型，子類型選為實體，輸入新建文件名稱“dmgdb”,鉤選“使用缺省模塊”然后確定，在新文件選項窗口中選擇“mmnspartsolid”進入零件工作環(huán)境。當再次注射時，在模具閉合過程中，定模表面與反推桿接觸，并使反推桿推動推出機構一起返回原始位置。動模板的厚度h可用下面計算公式： H=K100 （52）F=pA式中 F為動模墊板受的總壓力 (N)。為使導柱比較順利地進入導套孔，在導套孔的前端應有倒角；導套孔的滑動部分按H7/f6間隙配合，；導套的材料硬度應低于導柱的硬度，這樣可以改善摩擦，以防止導柱或導套拉毛。，一副模具一般需要2～4個導柱。由于它主要是承受壓力，所以材料可選用HT200。模具外表面應光潔，加涂防銹漆。為克服這一現(xiàn)象的影響，用一個井穴將主流道延長以接收冷料，防止冷料進入澆注系統(tǒng)的流道和型腔，把這一用來容納注射間隔所產(chǎn)生的冷料的井穴稱為冷料穴。實際加工中，是先用圓形銑刀銑出直徑為Φ4和Φ6 的分流道，再將材料進行熱處理，后做一個銅公（電極）去放電，用電火花打出這個澆口來。我們將采用限制性澆口。實際加工時，用銑床銑出流道后，少為省一下模，省掉加工紋理就行了。取L=60 mm。主流道的一端常設計成帶凸臺的圓盤，高度為5~10mm,并與注射機固定模板的定位孔間隙配合，襯套的球形凹坑深度常取3~5mm,R2=R1+（1~2mm）。以塑件的基本高度尺寸20mm為例，得 H=(20+202％1/20) =同理，型腔的其它深度尺寸也應用式（3—5）計算。 為塑件外徑基本尺寸，(mm)。為澆注系統(tǒng)及飛邊等的塑料質量，(g)。，盡量使塑件開模時留在動模一邊。這樣要求注射機在開模結束后動、定模板之間的最小間