摘 要
隨著全球經(jīng)濟的復(fù)蘇，制造加工業(yè)的需求日益增加，而作為現(xiàn)代機械制造業(yè)主要生產(chǎn)方式之一的高速切削，具有成本低、切削效率高以及加工件表面質(zhì)量好等優(yōu)點，正廣泛應(yīng)用于多個領(lǐng)域。但目前高速切削加工技術(shù)仍存在諸多問題，例如：切削研究多是針對特定的工藝條件，研究成果的通用性較差；高速切削是個復(fù)雜的物理化學(xué)過程，而現(xiàn)有研究針對其物理化學(xué)機理的探索方面已到達瓶頸狀態(tài)；控制加工質(zhì)量和加工成本始終是高速切削研究的永恒目標(biāo)，而以往的研究大多都浪費在了多次重復(fù)檢查錯誤中，工作效率低下。針對高速切削加工的以上難題，本文主要研究內(nèi)容如下：
（1）從實際需求出發(fā)設(shè)計可行的實驗方案，提出了均勻設(shè)計和正交設(shè)計逐步數(shù)據(jù)采集方法，在獲得實驗數(shù)據(jù)的同時，盡量減少試驗次數(shù)，有效的控制了實驗成本。
（2）采用MATLAB的數(shù)學(xué)模型處理功能，設(shè)計了一種數(shù)據(jù)處理方案，用多種常用的數(shù)學(xué)模型同時對采集到的數(shù)據(jù)進行擬合、預(yù)測，通過設(shè)定適當(dāng)?shù)膮?shù)指標(biāo)來比對所有模型的一致性，以獲得最優(yōu)模型，解決了以往研究中數(shù)學(xué)模型通用性差的問題。
（3）針對前階段優(yōu)化得到的最優(yōu)加工工藝，將統(tǒng)計過程控制理論（SPC）運用到高速切削加工工藝的監(jiān)測中，提出了一種高速切削加工在線監(jiān)控的思路。首先分析控制圖判斷過程是否可控，再進行過程能力評價，最后利用控制圖對切削過程進行監(jiān)控，若有異常則作出快速判斷，避免了實際生產(chǎn)中靠事后檢查去發(fā)現(xiàn)問題的被動局面，使整個切削加工過程處于事先預(yù)防的狀態(tài)。
（4）在上一階段SPC控制的基礎(chǔ)上，運用神經(jīng)網(wǎng)絡(luò)對SPC判斷的異常情形進行故障診斷，使得統(tǒng)計過程調(diào)整（SPA）得以實現(xiàn)。對高速切削加工故障類型進行條件簡化，選用合理的故障類型和系統(tǒng)輸入特征，建立概率神經(jīng)網(wǎng)絡(luò)，并對數(shù)據(jù)進行訓(xùn)練、分類，使得復(fù)雜的切削故障類型得到較好的診斷。
（5）采用Matlab軟件作為開發(fā)平臺，利用其矩陣處理功能對實驗數(shù)據(jù)進行處理，并完成SPC控制圖的繪制，運用神經(jīng)網(wǎng)絡(luò)功能模塊完成故障診斷的開發(fā)。
本課題通過理論分析、模型處理以及平臺開發(fā)，對于高速切削加工有以下改進效果：有效解決了高速切削數(shù)據(jù)分析通用性差的難題；同時采用統(tǒng)計學(xué)數(shù)據(jù)處理避開了物理化學(xué)機理研究的不精確性；運用統(tǒng)計過程控制及神經(jīng)網(wǎng)絡(luò)故障診斷來預(yù)防切削加工事后的被動檢查。

關(guān)鍵詞：高速切削加工； 多數(shù)學(xué)模型擬合； 統(tǒng)計過程控制；神經(jīng)網(wǎng)絡(luò)故障診斷



Abstract
With the global economic recovery, the demand of manufacturing industries gradually increasing. High-speed cutting as one of the main mode of production in modern manufacturing industries, which has become an important trend. However, there are still many problems in High-speed cutting field. Firstly, researches on cutting are most focus on special process conditions, which lead to the conclusion of the research cannot be widely used. Secondly, High-speed cutting is a complex physical-chemical process, at the same time the research on the mechanism of physical-chemical has reached a bottleneck. Finally, control processing quality and costs is always the target of High-speed cutting, however, previous studies are main rely on check and recheck to find error, which is inefficient and uneconomic. Besides, during recheck, a lot of resources are wasted on defective products. In order to solve these problems, main research contents of this thesis are as follows:
(1) Feasible experimental program is designed based on the characteristic of issue. Obtained experimental data at the maximum, while minimizing the number of tests. A data acquisition experimental program based on uniform design and orthogonal design is proposed, which effectively control the experiment cost.
(2) Combine with Mathematical model processing function of Matlab, a data processing program is designed. Several Mathematical models processing, fitting and forecasting the collected data at the same time. In order to obtain the optimal model, compare the consistency of all the models by setting certain parameters indicators, which solve the mathematical model cannot be used widely in previous studies.
(3) Based on the optimal model obtained above, SPC theory is applied to high-speed machining process monitoring. The idea of online monitoring for high-speed machining process is proposed. Firstly, analysis control charts are used to determine whether the process can be controlled or not. Then conducts a process capability eva luation. Finally, control charts are used to monitoring cutting process, judgments will be made if there are exceptions. This method solve the problem mentioned above, and put the cutting process in a precaution status.
(4) Based on SPC control mentioned above, Neural Networks is used to diagnose the exceptions judged by SPC, which realized SPA. By simplifying fault type of high-speed machining process and choosing reasonable fault type and system input characteristics to build up probabilistic neural network to train and classify data, which well diagnose the complex fault type.
(5) Matlab software is chosen as the develop platform and its matrix processing function is used to processing experimental data, then SPC control charts are completed, and finally neural network mode is used to develop fault diagnosis.
Based on the research mentioned above, the problem that mathematical model cannot be used widely is effectively solved. At the same time, use statistical data proces..