機(jī)械外文翻譯--調(diào)直機(jī)的反饋方法

1、<p>　　A Closed Loop Feedback Method for a Manual Bar</p><p>　　Straightener</p><p>　　Robert J. Miklosovic, Zhiqiang Gao</p><p>　　Department of Electrical and Computer Engineerin

2、g</p><p>　　Cleveland State University</p><p>　　Cleveland, Ohio, USA</p><p>　　Abstract—Automation of a unique manually controlled industrial bar straightener is proposed. A continuou

3、s-time closed loop model is constructed in Simulink for an event-driven process through the use of asynchronous timers. The system is simulated with linear and nonlinear PD controllers. A nonlinear filter,called the trac

4、king differentiator, is introduced as an alternative to a linear approximate means of providing accurate derivative feedback in the presence of noise. In both cases, the nonlin</p><p>　　I. BACKGROUND</p&g

5、t;<p>　　Precision straightening of a cylindrical metal bar is largely based on the ability to precisely measure its geometry. A few fundamental measurements and how each influences the tolerance specification on s

6、traightness should first be understood. Methods for measuring roundness and straightness are covered to lay the groundwork for the problem formulation. The basic operation of the machine is outlined in Section II, and it

7、s fundamental limitations and need for automation are discussed. Section II</p><p>　　A. Measuring Roundness</p><p>　　Roundness is a quantity derived from comparing the shape of a cross-sectional

8、 area at one distinct point along a cylinder’s length against a circle. A round metal bar that is arbitrarily long with respect to its diameter has to be checked for roundness in many locations lengthwise and averaged to

9、 insure overall consistency. Roundness is approximated by rotating the work piece one revolution in a Vee block while measuring the surface with an indicator. Taking the difference between the minimum an</p><p

10、>　　B. Measuring Straightness</p><p>　　Straightness is a quantity derived from comparing the axial centerline of a specific section of a cylinder’s length against a straight line. A simple method for appro

11、ximating straightness is by rotating the bar one revolution between two Vee blocks that are a fixed distance (d) apart, while measuring in the center with an indicator. The distance that the axial centerline of the part

12、deviates from a theoretically straight centerline directly below the indicator equals the extent to which the part</p><p>　　1. From this, TIR is derived as:</p><p>　　TIR= IX –IN = (R + |Bow|)-(R

13、 – |Bow|) =2*|Bow| (1)</p><p>　　Deviations in roundness, outside diameter (OD) size, and finish can adversely affect the measurement.</p><p>　　Figure 1. Max. and min. indicator readings of a bow

14、ed part</p><p>　　C. Straightening</p><p>　　The straightening process, which can be broken into steps, simply involves correcting any error while checking for straightness. First, the part is mea

15、sured for straightness. Then, it is rotated so that the bow is oriented 180 degrees away from the Vee blocks with the maximum indicator reading facing upwards. Finally, a counter-bending force replaces the indicator and

16、straightens the work piece against the Vee blocks.</p><p>　　II. MACHINE OPERATION</p><p>　　The straightener to be automated uses a non-contact ultrasonic sensor in place of the indicator and rol

17、lers in place of the Vee blocks in an effort to minimize contact wear. The part slowly spirals through the machine. The indicator reading becomes a continuous sinusoid at the sensor’s output, having a peak-to-peak value

18、equal to the TIR each revolution. TIR is sampled from the sensor output and calculated each revolution, making the sample period of one revolution the minimum time between conse</p><p>　　Figure 2. The straig

19、htener to be automated</p><p>　　A. Process Limitations</p><p>　　There are aspects of the process that can limit the controller’s performance and slow it down by extending YSP. Each is observed a

20、nd taken into consideration when producing</p><p>　　an accurate simulation model:</p><p>　　1. The ultrasonic sensor introduces RFI noise into its. The use of a feedback filter is essential.</

21、p><p>　　2. A rough part surface finish adds distortion to the sensor output.</p><p>　　3. An out-of-round part superimposes harmonics on the sensor output sinusoid, placing a bound on the minimum st

22、eady state error that is achievable.</p><p>　　4. Inconsistent material density produces false measurements. The measured focal length of a</p><p>　　transducer is dependent on the density of the

23、material that is being measured [2]. The unit cannot measure accurately in the presence of a time-variant material density (i.e. hard spots). Although unavoidable, it can be detected, since TIR changes monotonically.<

24、/p><p>　　5. An inconsistent OD causes vertical shifts in the sensor output. A differential TIR measurement cancels these affects.</p><p>　　6. A twisted part condition is detected when the angular p

25、osition of the maximum indicator reading slowly moves with each revolution. This condition is created when the part is not straightened at the precise angular location and occurs because of the quantization affect of the

26、 digital readout used by the operator. The new controller will use a continuous signal and the part can be straightened 30 to 45 degrees ahead of the twist when encountered.</p><p>　　B. The Need for Automati

27、on</p><p>　　Replacement of the operator with electronic hardware is beneficial in several ways. The cost of the electronics is much less than the ongoing hourly wage and schedule of an operator. The limitati

28、ons associated with the digital readout are eliminated, which helps the machine to straighten faster with more precision. The process can be drastically sped up to produce more. Though the minimum sample time is one revo

29、lution, it does not need to be slow enough for human comprehension. A Programmable Lo</p><p>　　C. Research Methodology</p><p>　　The focus is split between modeling and control design, since this

30、 is a new control problem. The process is manually controlled rather than being strictly manual in operation, meaning the machine needs only a new controller. There is no need for a complete mechanical overhaul, so the b

31、est method of straightening is not researched. Typical of a small company, time and money are limited. Gao and Huang [3] presented a new error-based control design framework including such innovations as a nonlinea</p

32、><p>　　III. A CLOSED LOOP SOLUTION</p><p>　　The task of automation can begin once the process is well defined. A straightforward system block diagram is developed, and each block is modeled in Simu

33、link. Aspects of the hardware configuration are carefully considered.</p><p>　　A. From Open Loop to Closed Loop</p><p>　　The open loop multi-input-multi-output (MIMO) block diagram, in Fig. 3, r

34、epresents the manually controlled process. The operator calculates TIR and monitors the angular position (􀀀 P) of the bow with each revolution. When Y approaches a specified limit, the operator sets BT proportio

35、nal to Y and its</p><p>　　rate, and then pushes a button (BP) to straighten the part. The bend timer creates a pulse triggered by BP that counter-bends the part for BT minutes. This event forces the part to

36、have a specific rate for a period of time. After the event, the part takes on a new rate and the process perpetuates. Therefore, Y’ is a</p><p>　　piece-wise continuous function of time. The operator’s involv

37、ement in the process is represented as a block in Fig. 4.</p><p>　　Figure 3. Open loop block diagram</p><p>　　Figure 4. Operator block</p><p>　　Rearranging the blocks and breaking e

38、ach function down into smaller more-manageable blocks reduces the representation to a usable closed loop, single-input-singleoutput (SISO) form. Fig. 5 shows how a SISO plant is obtained by combining the process with the

39、 task of sampling the TIR, since it can be consistently computed.</p><p>　　Figure 5. SISO plant block</p><p>　　The plant has a variable-width pulse as the input (U) and the sampled TIR (Y) as th

40、e output to be controlled. The incoming changing position of the work piece is modeled as an unknown rate disturbance (D). The closed loop SISO block diagram is shown in Fig. 6.</p><p>　　Figure 6. Closed loo

41、p block diagram</p><p>　　B. Hardware Configuration</p><p>　　Depicted in Fig. 7, an encoder and a PLC are the only hardware needed for automation. The encoder feeds the angular position of the pa

42、rt back to the controller, and the PLC handles all of closed loop functions outside of the process.</p><p>　　Figure 7. Hardware configuration</p><p>　　Fig. 8 depicts the hardware layout for plan

43、t data acquisition during manual operation. The PLC is also used to calibrate a strip chart recorder, shown in Fig. 9, which simplifies calibration for the operator and removes room for error during data acquisition. Usi

44、ng the PLC for both data acquisition and</p><p>　　control keeps costs lower.</p><p>　　Figure 8. Open loop data acquisition configuration</p><p>　　Figure 9. Chart recorder settings f

45、or a .006” TIR signal</p><p>　　IV. MODELING</p><p>　　Simulation modeling involves the construction of Simulink blocks for each of the blocks in the SISO diagram. Three modular blocks are first d

46、esigned to accommodate the presence</p><p>　　of an event-driven plant in a continuous-time environment where the control variable is an asynchronous pulse width. They are all based on creating asynchronous t

47、imers in Simulink</p><p>　　by integrating a constant until it reaches a preset value, then shutting it off by feeding the input with a zero. Equation (2) will reach a value of one in the time interval equal

48、to the average of u(t) over it. The accuracy increases if the interval is very small or u(t) is a constant function. A suitable difference</p><p>　　equation is given in (3).</p><p>　　A. Triggere

49、d Sample-and-Hold (TSH) Block</p><p>　　The output and rate of the plant immediately after the part has been straightened are dependent on BT, and the previous Y and Y’, all of which occur over different time

50、 intervals. The function of the TSH block is to sample a value at one point in time so that it can be used at another time. Shown in Fig. 10, the block basically samples the input for a small period of time on the rising

51、 edge of an enabling pulse, and then holds that value constant at the output until the block is re-enabled. In S</p><p>　　Figure 10. TSH Simulink block</p><p>　　B. Pulse Block</p><p&g

52、t;　　The bend timer must be able to convert BT from a value into the timed pulse, U. The pulse block, illustrated in Fig. 11, was designed for this purpose. By incorporating (2), it creates a pulse that has a magnitude eq

53、ual to the sign of the input and a pulse width equal to the input’s absolute value.</p><p>　　Figure 11. Pulse Simulink block</p><p>　　The heart of the pulse block is the pulse subsystem block, s

54、hown in Fig. 12. The reciprocal block generates a divide-byzero error whenever its input is zero. Consequently, the reciprocal block needs to be isolated in a subsystem that is only enabled when supplied with a value th

55、at is larger than a user</p><p>　　defined constant.</p><p>　　V. CONTROL DESIGN</p><p>　　There are currently many different control structures available, the simplest of which is the

56、 PID controller design. For this reason, a PD controller is first applied to the closed loop model to verify its response and stability. It is used as a benchmark for other controllers. Next, a nonlinear PD (NPD) control

57、 scheme is introduced. It retains the tuning ease of the PD controller while improving performance. Last, a nonlinear filter, called the tracking differentiator (TD), is introduced as</p><p>　　an alternate m

58、eans of providing an accurate derivative feedback to the controller in the presence of noise, thus improving performance.</p><p>　　A. Linear PID Control</p><p>　　For many reasons, PID control is

59、 still used in 90% of industrial applications [3]. There are only three tuning parameters, each having direct physical significance to the error signal, not the model. This makes for easy tuning, without having to spend

60、considerable resources on the construction of a linear model. Linear models are often inaccurate and require re-tuning when the real-world plants</p><p>　　they represent are nonlinear and time varying. A con

61、trol structure that is error-based, and not model-based, is more resilient to model uncertainties [3]. The PID control law in (6) represents the direct physical meanings of the three parameters, where e is the error sign

62、al, KP is the proportional gain, KI is the integral gain, and KD is the derivative gain. </p><p>　　K e K edt K??P ??I _ ??D (6)</p><p>　　VI. SIMULATION</p><p>　　The PD controller an

63、d the NPD controller are simulated in Simulink on the closed loop model. The significance of practical initial conditions and disturbances are considered. Simulation results from the two cases are presented and discussed

64、. Next, the steady state error is compared under various noise and disturbance conditions. Finally, a second order approximation and a TD are implemented in the feedback loop and simulated with heavy feedback noise for b

65、oth controllers.</p><p>　　A. Transient Performance</p><p>　　Simulating the system and individually adjusting plant parameters to emulate various real world conditions verified the design. Fig. 1

66、9 graphs the output and rate response for the PD controller. Notice the control action occurs when the rate is at ±R2 (i.e. ±.001 in this case). The performance measures are defined as follows:</p><p

67、>　　1. The settling time (Ts) is the time it takes for Y to be within ±.001, since this is the general object of</p><p>　　straightening.</p><p>　　2. Overshoot (OS) is the maximum |Y| afte

68、r it has reached zero.</p><p>　　3. Steady state error (Ess) is the maximum |Y| in the steady state region.The linear and nonlinear PD controllers were tuned to achieve the same Ts. The results of the transie

69、nt responses of the two controllers for two widely different initial positions are tabulated in Table I, to compare OS and Ess. All units are in thousandths except for settling time.</p><p>　　4. The process

70、can be modeled in discrete time and/or with different software. Writing the entire plant and bend timer in C may simplify the simulation model. From this, a class of problems can be clearly studied.</p><p>　

71、　5. The process can be investigated and modeled as a finite state machine.</p><p>　　調(diào)直機(jī)的反饋方法羅伯特·J·Miklosovic，高志強(qiáng)電氣工程和計(jì)算機(jī)系克利夫蘭州立大學(xué)美國俄亥俄州克利夫蘭手動(dòng)控制獨(dú)特自動(dòng)化的建議。一個(gè)連續(xù)時(shí)間的閉環(huán)模型在Simulink中構(gòu)建事件驅(qū)動(dòng)的過程中，通過使用異步定時(shí)

72、器。該系統(tǒng)是模擬線性和非線性PD控制器。非線性濾波器，稱為跟蹤微分，介紹了作為一個(gè)替代的存在噪音，提供準(zhǔn)確的微分反饋線性近似手段。在這兩種情況下，非線性技術(shù)優(yōu)于線性，同時(shí)保留調(diào)整簡(jiǎn)單。一，背景主要是基于能夠精確地測(cè)量其幾何精度的圓柱形金屬條調(diào)直。應(yīng)先了解一些基本的測(cè)量和如何每個(gè)影響的直線度公差規(guī)范。圓度和直線度測(cè)量方法覆蓋問題制定奠定了基礎(chǔ)。本機(jī)的基本操作是在第二節(jié)，概述，并討論了其基本的限制和對(duì)自動(dòng)化的需要。第三節(jié)討論通過框圖的循

73、環(huán)過程中的作用和新的硬件關(guān)閉任務(wù)。第四節(jié)包含所有的塊，在Simulink建模描述。在第五節(jié)討論的線性和非線性控制器的設(shè)計(jì)，該系統(tǒng)是模擬在第六節(jié)，第七節(jié)總結(jié)。答：測(cè)量圓度圓度是從一個(gè)不同點(diǎn)，以及對(duì)一個(gè)圓圈一個(gè)圓柱體的長度比較的橫截面積形狀的數(shù)量。一個(gè)圓形的金屬條，其直徑是任意長有縱向許多地方要檢查圓，平均以確保整體一致性。圓度近似</p><p>　　單輸入單輸出植物塊擁有廠房作為輸入（U）和要控制的輸出采樣

74、的TIR（Y）的可變寬度的脈沖。傳入改變工件的位置是仿照干擾（四）作為一個(gè)未知的利率。閉環(huán)SISO框圖如圖。 6、閉環(huán)框圖7、編碼器和PLC是唯一的自動(dòng)化所需的硬件。編碼器反饋的角位置的一部分返回給控制器，PLC的處理進(jìn)程之外的所有閉環(huán)功能。圖7。硬件配置圖8描繪了在手工操作的工廠數(shù)據(jù)采集的硬件布局。 PLC也被用來校準(zhǔn)條狀圖表記錄，如圖。 9、簡(jiǎn)化操作的校準(zhǔn)和消除數(shù)據(jù)采集過程中的錯(cuò)誤空間。為數(shù)據(jù)采集使用PLC和控制使成本更低。

75、圖8開環(huán)數(shù)據(jù)采集配置圖9圖表記錄設(shè)置為.006“的TIR信號(hào)四。建模仿真建模涉及的Simulink模塊的建設(shè)，為每個(gè)塊在SISO圖。第一，三個(gè)模塊化塊設(shè)計(jì)，以適應(yīng)存在在連續(xù)時(shí)間的環(huán)境下控制變量是一個(gè)異步脈沖寬度的事件驅(qū)動(dòng)的植物。它們都是基于建立在Simulink的異步定時(shí)器通過整合不斷，直到它到達(dá)一個(gè)預(yù)設(shè)值，然后關(guān)閉它喂養(yǎng)零輸入。方程（2）將到達(dá)的時(shí)間間隔等于平均的U（T）的值。如果是非常小的時(shí)間間隔的準(zhǔn)確性增加或U（T）是一個(gè)