【摘要】：讓機(jī)器變得智能是工控界的共同理想。在面對(duì)重復(fù)性加工任務(wù)時(shí),如果數(shù)控機(jī)床具有“自學(xué)習(xí)”功能,主動(dòng)根據(jù)之前零件的誤差信息指導(dǎo)后續(xù)加工,就會(huì)使加工誤差逐漸減小,成品率大幅提升。然而,國(guó)內(nèi)現(xiàn)有的數(shù)控機(jī)床都只能把大批量生產(chǎn)當(dāng)作標(biāo)準(zhǔn)單件生產(chǎn)的機(jī)械重復(fù),之前的加工信息得不到利用,經(jīng)過多道復(fù)雜的工序后,成品率低下。針對(duì)上述問題,本文提出為現(xiàn)有數(shù)控機(jī)床設(shè)計(jì)自學(xué)習(xí)控制器的想法。在眾多自學(xué)習(xí)算法中,選擇迭代學(xué)習(xí)控制算法。本著原理簡(jiǎn)單、便于應(yīng)用、魯棒性強(qiáng)的原則,最終選定控制器結(jié)構(gòu)為基于干擾觀測(cè)器的閉環(huán)PID(Proportional Integral Differential)型迭代學(xué)習(xí)控制器。并分為四個(gè)步驟展開工作,分別是數(shù)控系統(tǒng)的建模與辨識(shí)、自學(xué)習(xí)控制器結(jié)構(gòu)設(shè)計(jì)、學(xué)習(xí)增益參數(shù)優(yōu)化以及效果驗(yàn)證。在深入分析迭代算法后,發(fā)現(xiàn)對(duì)該算法的研究大多停留在仿真與半實(shí)物仿真階段,于是總結(jié)出制約算法應(yīng)用的兩大難題。其一:算法的理論支持大多是基于開環(huán)迭代進(jìn)行的,但開環(huán)系統(tǒng)穩(wěn)定性難以保證,實(shí)際系統(tǒng)大多是閉環(huán)系統(tǒng);其二,即使?jié)M足收斂性條件,在迭代的過程中仍存在誤差先減小后增大的過沖現(xiàn)象。為解決第一個(gè)難題,在綜合考慮迭代算法收斂精度、速度與可實(shí)現(xiàn)的復(fù)雜程度后,選定閉環(huán)迭代結(jié)構(gòu),并為其推導(dǎo)出一套學(xué)習(xí)增益參數(shù)優(yōu)化方案。同時(shí),將迭代算法只能抑制重復(fù)性擾動(dòng)的限制放寬,輔之以干擾觀測(cè)器抑制非重復(fù)性擾動(dòng);為解決第二個(gè)難題,提出條件啟停迭代機(jī)制,實(shí)現(xiàn)了從學(xué)習(xí)模式到生產(chǎn)模式的順利過渡。在三軸雕銑機(jī)床上進(jìn)行蝴蝶軌跡跟蹤效果驗(yàn)證時(shí),對(duì)比基于單純形法的最優(yōu)參數(shù)自整定方法,單軸跟蹤誤差為33.4μm;而本文設(shè)計(jì)的自學(xué)習(xí)控制器經(jīng)過十次迭代可將單軸跟蹤誤差降為5.705μm。在現(xiàn)有迭代學(xué)習(xí)的書籍文獻(xiàn)中,大多使用通篇的公式進(jìn)行原理闡述,對(duì)于初學(xué)者入門困難。本文將這些理論知識(shí)與數(shù)控機(jī)床相結(jié)合,以實(shí)際應(yīng)用為前提,以最簡(jiǎn)單方法實(shí)現(xiàn)為基礎(chǔ),用通俗易懂的語言描述了控制器設(shè)計(jì)的詳細(xì)過程,并將該方法寫入數(shù)控系統(tǒng)中,實(shí)現(xiàn)了算法的應(yīng)用價(jià)值。
[Abstract]:It is the common ideal of industrial control to make machines intelligent. In the face of repeated machining tasks, if NC machine tools have the function of "self-learning" and guide the follow-up machining according to the error information of the former parts, the machining errors will gradually decrease and the finished product rate will be greatly increased. However, the existing CNC machine tools in China can only regard mass production as a mechanical repetition of standard single-piece production, and the previous processing information is not utilized. After many complicated processes, the yield of finished products is low. In view of the above problems, this paper puts forward the idea of designing self-learning controller for existing NC machine tools. Among many self-learning algorithms, iterative learning control algorithm is chosen. Based on the principle of simple principle, easy application and strong robustness, the structure of the controller is selected as the closed-loop PID (Proportional Integral Differential) iterative learning controller based on disturbance observer. It is divided into four steps: modeling and identification of numerical control system, structure design of self-learning controller, optimization of learning gain parameters and effect verification. After deeply analyzing the iterative algorithm, it is found that the research of the algorithm is mostly in the stage of simulation and hardware-in-the-loop simulation, so two difficult problems restricting the application of the algorithm are summarized. First, the theoretical support of the algorithm is mostly based on the open-loop iteration, but the stability of the open-loop system is difficult to guarantee, and the actual system is mostly closed-loop system; second, even if the convergence condition is satisfied, In the iterative process, the error decreases first and then increases. In order to solve the first problem, after considering the convergence accuracy, speed and realizable complexity of the iterative algorithm, the closed-loop iterative structure is selected and a set of optimization scheme for learning gain parameters is derived. At the same time, the iterative algorithm can only restrain the limitation of repetitive disturbance, and the disturbance observer is used to suppress the non-repetitive disturbance. In order to solve the second problem, a conditional start / stop iterative mechanism is proposed. Realized the smooth transition from the learning mode to the production mode. When the butterfly track tracking effect is verified on a three-axis carving and milling machine tool, the optimal parameter self-tuning method based on simplex method is compared. The uniaxial tracking error is 33.4 渭 m, and the self-learning controller designed in this paper can reduce the uniaxial tracking error to 5.705 渭 m after ten iterations. In the existing iterative learning literature, most of them use the whole formula to explain the principle, so it is difficult for beginners to get started. In this paper, we combine these theories with NC machine tools, take the practical application as the premise, take the simplest method as the foundation, describe the detailed process of the controller design in a simple and understandable language, and write this method into the NC system. The application value of the algorithm is realized.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP273

【參考文獻(xiàn)】

相關(guān)期刊論文 前6條

1 張曙;;工業(yè)4.0和智能制造[J];機(jī)械設(shè)計(jì)與制造工程;2014年08期

2 江思敏;王先逵;;基于擾動(dòng)觀測(cè)的非圓零件CNC加工的迭代學(xué)習(xí)復(fù)合PID控制[J];機(jī)械工程學(xué)報(bào);2014年17期

3 王麗梅;郭宜興;;基于混合誤差迭代學(xué)習(xí)控制的XY平臺(tái)輪廓控制[J];制造業(yè)自動(dòng)化;2013年23期

4 姜曉明;王巖;王程;陳興林;;魯棒迭代學(xué)習(xí)控制及在高精密平臺(tái)中的應(yīng)用[J];系統(tǒng)工程與電子技術(shù);2013年03期

5 謝勝利,田森平,謝振東;基于向量圖分析的迭代學(xué)習(xí)控制非線性算法[J];控制理論與應(yīng)用;2004年06期

6 孫明軒,黃寶健,張學(xué)智;非線性系統(tǒng)的PD型迭代學(xué)習(xí)控制[J];自動(dòng)化學(xué)報(bào);1998年05期

相關(guān)碩士學(xué)位論文 前1條

1 毛琳琳;基于ILC的典型非圓截面零件數(shù)控切削系統(tǒng)應(yīng)用研究[D];華南理工大學(xué);2012年

，

本文編號(hào)：2225996