Machining and Monitoring Strategies Part 3 doc

10 287 0
Machining and Monitoring Strategies Part 3 doc

Đang tải... (xem toàn văn)

Thông tin tài liệu

9.3 HSM – with Non- Orthogonal Machine Tools and Robots Variax/Hexapod – Design Concept Non-orthogonal machine tools such as the one simu- lated, designed and developed for HSM applications is typically illustrated in various ways in Figs. 223 to 225: utilising ‘virtual’ six axes kinematics (i.e. namely: X-, Y-, Z-, A-, B- and C-axes), therefore these axes operate without having any ‘true’ slideways. is par- ticular kinematic concept has actuators that cross each other forming X’s instead of meeting at apexes to form triangles, as they occur in aircra ight simulators, which uses conceptually similar mechanisms – known as Stewart platforms, these congurations being a form of ‘parallel kinematic link mechanism’ (Fig. 223). To develop this new concept for a machine tool, the manufacturer utilised computer-aided technology which played a pivotal role in creating the structural design (Fig. 223). In particular, the application of a totally three-dimensional design environment was employed, utilising both nite element analysis (FEA) in conjunction with kinematic analyses. However, the ‘Variax’ design uses a range of uncomplicated, or stan- dard mechanical components in its design. While new forms of motion actuators were discarded in favour of conventional and well-proven ballscrew technology, with its accompanying motor and drive machinery. Even the gimbals that secure the legs at the base and the spindle carrier, are relatively simple devices. With the design of such a high thrust machine, a signicant problem to overcome was the connection of the spindle to the six legs (Fig. 223a). e answer to the connection problem was a simple space frame design, allowing all the forces to be either in tension, or compression along the structural elements – similar to a bridge design. If one compared this ‘Variax design’ with that of a ‘plate-type design’ to secure the spindle to the legs, then the former space frame concept im- proves the mass-to-stiness ratio by 275%. While an- other key development problem to be overcome was that of how a spindle supported and driven by six axes kinematically moves in space, moreover, was it even mathematically possible to control the motional members? For example and by way of illustration of this complex mathematical/control problem: a simple ‘X-axis’ linear kinematic translation requires all six legs to simultaneously move, but each leg will move at dierent speeds, either accelerating, or decelerating at dierent rates through the whole ‘linear movement’ – requiring very complex multi-axes mathematical solu- tions to achieve this action. By employing a system of novel mathematical transformation runs in real-time by the CNC’s multi-processor this mathematical trans- lation action was achieved, but from a programmer’s viewpoint, conventional ‘word-address format’ 17 of programming knowledge was all that was needed to successful operate the machine tool. Non-Orthogonal Versus Orthogonal Machine Tool Designs e ‘Variax’s’ machine capabilities and benets dif- fer signicantly from those found on conventional slideway-based orthogonal machine tools. So, on an orthogonal machine the slideways must be perfectly straight, parallel and at 90° to each other. On these machines, an axis must have accuracy and precision control along the slideway having linear and rotary degrees of freedom carefully managed by the ground way being scraped to minimise any impending errors/ uncertainties. As mentioned earlier, these orthogonal axes have kinematically 21 degrees of freedom, with: linear motion; rotational – i.e. yaw, pitch and roll; plus 17 ‘Word-address format’ of CNC programming, can be consid- ered as: A system of coding instructions whereby each word in a block is addressed by using one, or more alphabetic characters identifying the meaning of the word. [Source: Valentino and Goldenberg, et al. 2000] For example, some typical ‘G- and M-codes’ are: G00 – rapid movement/ traverse of the tool (modal); G01 – linear interpolation (ie. tool moved at a prescribed fee- drate) (modal); G02 – circular interpolation clockwise – CW (modal); G03 – circular interpolation counter clockwise – CCW (modal); G04 – programmed dwell (*non-modal); G40 – Cancel cutter diameter compensation (modal); G41 – tool diameter cutter compensation (i.e. radial-oset) on le-hand side of workpiece (modal); G42 – tool diameter cutter compensation (i.e. radial-oset) on right-hand side of workpiece (modal); M00 – Program stop; M02 – End of program; M03 – spindle on (CW); M04 – spindle on (CCW). NB Many more codes/auxiliary functions exist, utilised in ‘word-address format’ programs.*Non–modal commands are only active in that actual block. [Source, Smith et al., 1993] Machining and Monitoring Strategies  Figure 223. Basic kinematics of the non-orthogonal machine tool, with a simulated rendition of the es- sential elements of the machine’s structure. [Courtesy of Giddings & Lewis] . squareness errors to be considered, if the machine tool is to perform within its intended specication. In ef- fect, with an orthogonal machine tool three axes are controlled (i.e. the X-, Y- and Z-axes), and the rota- tional and squareness parameters must be ‘perfect’. us, these errors/uncertainties can be established by laser-calibration, then compensated for up to a certain extent to create and ‘stable machine’. If for example, one considers the worst case scenario for a machine tool conguration, namely that of a horizontal ma- chining centre. Here, if we just deal with one axis, then the spindle travels up and down the column’s  Chapter  face – as it moves to various points on the workpiece’s geometry. e tooling is held here and in eect, is a cantilevered beam (i.e. the tool is normally only sup- ported at one end only). In this situation the tooling assembly can be equated to that of boring tooling, then its inherent rigidity will decrease by the cube of the stand-o distance from the ‘gauge line’ 18 . When a ma- chine tool builder calibrates this orthogonal-designed machine tool, they will measure along the base of the column – either optically (i.e. by for example, autocol- limation 19 instrumentation), or with laser interferom- etry, thus feeding-back any potential error sources for CNC’s dynamic compensation, during execution of the part program. Nonetheless, the column can either minutely: twist; bend; or even move; if its temperature varies by only just a few °C. In fact, it has been widely reported that up to 70% of the total errors present on a machine tool are thermally-related. On conventional machine tools – with slideways, essentially one does not really know what is occur- ring in all of the degrees of freedom at any particu- lar time. Although even if we did know the dynamic status, unless some form of ‘adaptive compensation system’ 20 was tted to the machine, having some so- phisticated multiple sensors positioned on and within critical positions on the machine’s structure, then oth- erwise, it would be almost impossible to compensate 18 ‘Gauge line’ of the tool, is the distance from where its eective length is taken, specically when the tool length compensa- tion distance is needed for the CNC programming details. Normally, this ‘gauge line’ is datumed from a known point on the male spindle taper, located in, and standing slightly proud of its female counterpart. is measurement is either prefer- ably set in a tool presetting machine, or it can be obtained on the machine tool by some form of table-mounted touch-trig- ger cube. Alternatively, the tool’s length can also be accurately measured by some form of laser-activated instrumentation, strategically-positioned at a suitable part of the machine tool’s structure – where it will not interfere with any subsequent machining operations. 19 ‘Autocollimation’ was the ‘traditional’ technique that was em- ployed for any form of machine tool calibration. Autocollima- tion instruments and equipment, utilise highly-evolved ac- curate and precision optical apparatus to measure alignments and squarenesses of the machine tool’s axes, which can then be utilised to compensate for any error sources detected in the machine tool’s kinematics. [Sources: Taylor Hobson, 1984/ Spectrum Metrology, Leicester (UK)] 20 ‘Adaptive error compensation’ , basically utilises sophisticated geometric algorithms that allow for compensation of the geo- metrical elements, via a range of strategically-positioned sen- sors on orthogonal machine tools. (Source: Ford, 1993) for such error sources – as they occur. Clearly then, it is no coincidence that both the ‘Variax’ operates with six legs and, it also has six degrees of freedom. ese leg orientations have three pairs of two legs that cross each other and which are secured, but free to swivel at either end (i.e. see the simulated renditions in Fig. 223 and partial assembly of an actual ‘Variax’ machine tool in Fig. 224). With these six legs, all of the degrees of freedom are controlled, which in turn, eradicates the usual sources of errors exhibited by conventional multiple and ‘stacked’ slideway axes. Some of the ma- jor benets obtained from the ‘Variax’s’ design congu- ration are: • Extremely rigid machine tool – coupled to small mass making it 500% more rigid than a conven- tional machine, due to the fact that all forces whether they are in compression, or tension are fed through the six legs and its associated space frame, • Exceedingly fast in operation – as all the ballscrews must move together, they only have a light mass to contend with, making it up to 5 times faster than an orthogonal machine tool, • Very high continuous thrust – due to concurrent and synchronised use of the six ballscrews, • High accuracy and precision – this is due to the machine’s inherent rigidity, coupled to the fact that all six degrees of freedom are controlled, with laser feedback through the centre of the ballscrews 21 , • No supporting structure – the ‘Variax’ is self-con- tained, so it does not need foundations. e ma- chine has a three-point location and the weight of the spindle head is neutralised, as far as the ballscews are concerned – by the three large gas- spring supports 22 (i.e. see Figs. 224 and 225a). 21 ‘Laser-controlled feedback’ , through the centre of the ballscrews, negates the Abbé oeset error, with any errors in the legs (i.e. axes) being averaged, rather than ‘stacked-up’ as is the situation in conventional machine tools. Hence, volu- metric accuracies are in line with that of a Co-ordinate Meas- uring Machine CMM – see details in Fig. 223a. 22 ‘Gas-spring supports’ , are tted outside the actuator platform (i.e. see Figs. 224 and 225a), they are strategically positioned so that they carry actuator platform’s gravity-induced load. us, these gas-spring support placements and their operation, means that the actuators have to only overcome inertia and momentum to move the machining head. is has the kinematic benet of enabling the machine spindle to move through space faster than any single actuator changes its length, this combination allows the ‘Variax’ to move and accelerate up to ve times faster than a conventional machine tool. Machining and Monitoring Strategies  NB e machine tool also has large rubberised compliance mountings, so when it is in operation it seems to ‘visually-bounce around’ , but this is something of an optical illusion, as the overall rigid integrity of the ‘Variax’ is maintained. Variax/Hexapod – Specification and Machining Performance e ‘Variax’ is equipped with a 630 mm square pallet and can both move and cut at 66 m min –1 (i.e. see Fig. 225b), accelerating at >1g while providing >3 tonnes of continuous thrust in the Z-axis. e machine tool is compliant with Standard: ASME B5.54, obtaining ac- curacies of <10 µm, while its rigidity is >175 Nm µm –1 . Pallet change time is 10 seconds, with a tool change time of 6 seconds (chip-to-chip), having a tool stor- age capacity of 50 tools (i.e. expandable). e standard tool spindle cartridge is rated at 22 kW , with spindle speed ranging from 100 to 16,000 rev min –1 , having an angle of tool spindle inclination of up to 25°. In demonstrations and in-house trials at the man- ufacturer’s premises, the initial prototype ‘Variax’ ma- chine was operated and machining components for three years. On one large aerospace machining appli- cation of a critical component, the original cycle-time was 19.3 hours on a conventional machining centre, but when this same part was cut on the ‘Variax’ , it took just 3 hours to complete, with the additional benet of being both more accurate and precise. Equally, when a smaller aerospace component – a landing bracket – was originally machined it took 1.65 hours on the con- ventional machine tool, but when the same part was placed on the ‘Variax’ , it took only 0.55 hours cycle- time to complete. Figure 224. The Variax/ Hexapod Machining Cen- tre here seen during nal assembly at Nottingham University, equipped with an up-rated high-fre- quency spindle: 40,000 rev min –1 @ 40 kW power-rat- ing, with a ski taper. NB Laser transducers – giving an accurate and precise control, can clearly be seen located on each of the actuator legs. [Courtesy of Giddings & Lewis/The University of Nottingham (UK)] .  Chapter  Figure 225. Variax/Hexapod oers a unique and rigid design solution, with many production benets. [Cour- tesy of Giddings & Lewis] . Machining and Monitoring Strategies  e capital outlay for a ‘Variax/Hexapod’ machine tool, costs about the same as a similar specication – component size capacity, to that of a conventional ve- axis machining centre, but with the above additional performance benets. Robotic Machining Robotic machining applications have been utilised for some years, currently up to a thousand such installa- tions are to be found world-wide. Probably the biggest user of robotic machining is the aerospace industries, although the automotive industry is catching up fast. Most of the current research work into robot machin- ing is undertaken with an anthropometric type of robot conguration, usually having either ve, or six axes (i.e. see Fig. 226). If a six axis robot is employed (Fig. 226), it gives several benets over say, a ve-axis machining centre, with probably the greatest production advantage being access to the workpiece’s surface features (Fig. 226b). is extra degree of freedom, allows the alternative wrist positions to achieve identical tool positional orientation – regardless of the workpiece contour, enabling the robot axis the ability to rotate the wrist about the tool’s axis. Moreover, where speed and ac- celeration are important, robots normally out-perform the more traditional machine tool structures, mainly due to both their low weight and minimal inertial ef- fects. Within the robot’s programming language, pa- rameters exist that allow a balance to be made between robotic arm speed and accuracy. So, when roughing- out higher speeds can be employed – at the expense of accuracy, then the parameters can be changed to a slower speed, but with greater nished machined part accuracies. Limitations also occur when using robots for ma- chining, with perhaps the most obvious one being their intrinsic lack of rigidity, when compared to that of virtually any machine tool. If a robot is employed at conventional spindle speeds the tool’s cutting forces are simply too great, creating both vibrations and de- ections in the robotic arm, which badly impacts on the component’s machined surface texture and di- mensional accuracy. In order to mitigate against any cutting force and accuracy deciencies, it is essential to utilise HSM spindles, in combination with taking small D OC ’s, to minimise these eects. e robot’s kinematic structural complexity also causes positional problems, due to innate manufactur- ing tolerance eects to the location of the joints in the robotic arm. is dimensional and geometric toler- ancing build-up, means that the mathematical model used to control individual joint positions in space, will minutely dier from the actual reality to that of the joint placements. is positional dierence being most apparent and exaggerated around the perimeter of the robot’s working envelope. Oen, it is impossible to calibrate the mathematical model for these com- plex inter-related errors, as the overall complexity of robot’s control algorithms then become such that they cannot execute the kinematic motions fast enough to dynamically control the robot at the required axis tra- jectories. In calibration trials on a typical six-axis anthropo- metric robot undertaken by Young (1999) while work- ing in cartesian co-ordinates for the robot’s positional accuracy – when assessed with laser interferometer instrumentation, the results produced linear errors of ± 0.8 mm in each cartesian co-ordinate (i.e. X-, Y- and Z-axes) – across the complete working range. e er- ror curves produced were in fact, symptomatic of the kinematic structure of these robots. Characteristics included a decrease in backlash to almost zero – to- ward the perimeter of the working envelope (i.e. due to gravitational eects), while a combination of linear and sinusoidal eects combined to produce the total error. Even though these robot errors are large with respect to those found on a machine tool, they should not deter robotic usage for suitable HSM applications. e repeatability shown by most robots is usually far superior to that of its accuracy. In practice, any multiple axis robotic arm utilised for sculptured machining applications, needs to have their axes biased and oset in order to eliminate the ‘paradox’ that might overwhelm them when all axes are attempting to keep the end-eector (i.e. tool) nor- mal to the contoured surface (Fig. 226b). For example, on a three-roll wrist (Fig. 226), the combination of the kinematic linear and rotational build-up, may result in instead of one of these axes angularly moving just 1°, it causes it to actually move 359° instead – thereby scrapping the workpiece in the process! To alleviate this problem, if the workstation/stand, is oset to one side and angled, this oset and compound angle will minimise the axis predicament that can aict the ro- bot’s subsequent programmed motions. e worksta- tion position in its ‘known space’ with respect to the robot axes datums must be known and accurately cali- brated, thus ensuring that the angled grid-plate for re- sulting workpiece xturing is both ‘xed and qualied’. By slowing down the robotic axis trajectories in the -  Chapter  Figure 226. Robotic high-speed machining with a multi-axis anthropometric robot, equipped with HSM milling head. [Courtesy of Southampton Solent University/Westwind Air Bearing Ltd./Smith, G.T.] . Machining and Monitoring Strategies  nal machining passes over a sculptured workpiece sur- face, a certain degree of accuracy can be achieved, but even here, it does not approach that of any moderately accurate and precise machine tool. Tool centre-point programming is the preferred option, as when utilising cartesian co-ordinates, the programmed points repre- sent the cutter’s end – once the required oset has been established. is tool centre-point program allows the programmer to eectively disregard the tool’s length in any subsequent machining applications. However, care must be taken with any exible coupling connec- tions, such as pneumatics, water, or electrical wiring to the HSM spindle – allowing sucient slack in the piping (i.e see Fig. 226), thus ensuring that the robot will not inadvertently de-couple these services, as it at- tempts to manufacture the complex part. 9.4 HSM – Toolholders/ Chucks Introduction Rotational speeds for tooling assemblies subjected to HSM applications must of necessity be very high, this can create several problems for any tooling utilised in such machining activities. Notably, due to the centrif- ugal force the toolholder could swell and slacken its gripping force on the tool’s shank. In an extreme situa- tion due to the application of cutting forces this might cause the tool to speedily exit from its holder, in so do- ing either scrapping the workpiece, or become a severe safety hazard to any operator in attendance at the ma- chine tool. erefore, both a good t and connection is an essential requirement of any HSM machining ap- plication. us, the mechanical interface between the toolholder and the machine’s spindle, together with that of the tool’s shank and its respective toolholder are the prerequisites for a successful HSM application. .. Toolshank Design and Gripping Pressures Tool and Sleeve – Mechanical Interface e latter point briey mentioned in the previous section concerning the tool shank’s tment in the holder, is an important criterion in any HSM applica- tion. Nonetheless, of greater interest and note, is the actual gripping pressure exerted at the toolholder’s mechanical interface with its mating tool shank. Some interesting toolholder designs have been attempted to increase the gripping pressure here, with the level ex- erted at the free-end of the holder not being too great a problem, due to elastic compliance of the sleeve in this region. e notable diculty arises as the shank is being gripped toward the ange end of this sleeve, where with most conventional toolholder sleeve de- signs, the gripping pressure drops-o signicantly at this locality. At high rotational speed and under the application of the cutting forces, the cutter will tend to become unstable and present a distinct ‘wobbling motion’ due to the lack of gripping pressure and sup- port here, being exacerbated by the higher imparted centrifugal forces. In order to try to alleviate this lack of grip problem, signicant design eort has been expended over the years to attempt to increase elas- tic behaviour and hence amplify gripping pressure at this sleeve’s region. In the graph shown in Fig. 227b, the elastic behaviour and its associated contraction are plotted from the tool sleeve’s free-end in linear steps along the sleeve and toward the ange. In most prior designs the sleeve contraction on the tool’s shank near the ange was approximately 50 µm (i.e. shown by plot ‘B’). By introducing a radially-plunged undercut to the sleeve’s outside diameter at the juncture of the ange and sleeve (i.e. see Fig. 227a), this creates signicantly increased elastic behaviour and hence improved grip- ping pressure at this region of the sleeve, as illustrated by the plotted relationship of the test results – produc- ing a sleeve contraction of 145 µm at the ange, as in- dicated by ‘A’ , in Fig. 227b. is vastly improved grip - ping pressure with the ‘undercut ange’ is of the order ≈150%, when compared to other design techniques, which restricts any attempt at the tool’s ‘wobbling be- haviour’ as a result of the improved elastic contraction on the shank. In order to obtain a good overall elastic gripping pressure along the length of the sleeve, multiple hard- ened needle rollers are positioned – at a slight angular inclination, both around and along the male shallow- angled tapered sleeve’s periphery (i.e. items 2 and 4 in Fig. 227a). As the female tapered sleeve (item 3) is ro- tated, the caged inclined needle rollers rotate and be- gin to steadily move up the male tapered sleeve toward the ange (item 1) and, in so doing, elastically com- press the sleeve (item 2), which in turn will tighten on a tool’s shank. As the tapered female sleeve (item 3) is fully tightened with its ‘C-spanner’ , it will com -  Chapter  Figure 227. High-speed milling toolholders (chucks), holding cutter’s shank along its whole gripping length. [Courtesy of Di- ashowa Tooling] . Machining and Monitoring Strategies  press its face against the rubberised contact seal and the ange (item 1), thereby acting as both a vibration damper and sealing against particle/debris ingress into the mechanism. e front face already being sealed against such potential particle ingress (i.e. see Fig. 228 sectional detail diagram). In order to increase the elas- tic behaviour of the male tapered sleeve (item 2 – Fig. 227) and improve its gripping pressure here, the end of the sleeve is partially slit along its length and around it at eight equally-spaced positions (i.e. see Fig. 228). is mechanical design solution for the tightening of the male tapered sleeve, gives possibly the optimum gripping pressure for such a mechanical interface on the tool’s parallel shank. Tool and Sleeve – Hydraulic Designs For an alternative tool-gripping pressure design con- cept, the hydraulic toolholder (i.e. see Fig. 228 – top right – for a section through the tool sleeve), oers an ideal alternative to the mechanical tool interface previ- ously mentioned above. Here, the hydraulic toolholder manufacturer guarantees a 0.005 mm tolerance-in- roundness (TIR) at 100 mm from the front face, with <0.0013 mm repeatability for such hydraulic toolhold - ers. e normal contraction for a φ25 mm toolholder is <0.13mm, although the contraction is proportionally higher for larger diameters, conversely, it will be lower for smaller diameters. e benets of utilising hydrau- lic toolholders are that they provide both rigidity and balance, while holding the tool’s shank axially-straight along its own centreline. It is also claimed that hydrau- lic toolholders produce less vibrations resulting in im- proved machined surface textures, with a possibility of less chatter. is latter benet it is claimed to be the result of the hydraulic uid in the toolholder acting as a natural vibration damper and impact cushion. Hydraulic toolholders are balanced to ISO/ANSI Standards, to G2.5 at 15,000 rev min –1 in practice, but are claimed to perform successfully in spindles rotat- ing at 50,000 rev min –1 . Moreover, other benets are made that tool presetting is much quicker to achieve than mechanical designed toolholder assemblies, as it is stated that due to the hydraulics having an automatic centring action, only the tool’s length needs to be pre- set. Further, as there are fewer moving parts, these toolholders need little maintenance and the sleeve and piston should have a fatigue life of >100,000 cycles, prior to servicing, this ‘in-service life’ representing many years of practical usage. e disadvantages to hydraulic toolholders are few, but may prove a signicant obstacle to their introduc- tion, including the fact that their purchase costs are higher than their mechanical counterparts, with the other limitation being a new hydraulic sleeve is re- quired for dierent tool shank diameters. However, this latter point can be mitigated against, by conduct- ing a programme of: rationalisation; consolidation; and optimisation on the various production require- ments for toolholder varieties and shank diameters (i.e. see Chapter 1, Sections 1.1.1 to 1.1.3 for details of such tool survey). Tool and Sleeve – Thermal and Cryogenic Designs Shrink-t systems require specially made tool hold- ers, being designed for a specic tool shank diameter, although they can accommodate any style of shank. Once in-situ the tool in its thermal sleeve behaves in a very similar way to that of a one-piece tool. e high gripping pressure coupled to excellent concentric- ity (i.e. <5 µm), allows these toolholders to increase speeds and feeds by >20%, when compared to most of their mechanically-designed counterparts. Due to greater rotating concentricity of shrink-t toolholders, there is a better wear pattern developed on the milling cutter’s teeth, or drill’s lips, etc., which it is claimed, increases tool life by >30% over conventional tool- holders. In a similar fashion to that of hydraulic tool- holders, thermal contraction occurs both around the periphery and along the whole mechanical interface, which will automatically centre the tool’s shank within its mating bore. is complete toolholder-to-tool t- ment, minimises centrifugal force when operating in an HSM mode. us, the contraction of the toolholder rigidly locks the tool in-situ, this gripping pressure is at least 500% greater than for conventional toolhold- ers. In fact the pressure exerted here, is even greater than that of the pull-stud (i.e. retention knob), mean- ing that in the presence of high forces, the whole as- sembly would be pulled out of the spindle before the tool would be released by its mating holder. e signicant component of a shrink-t toolhold- ing system is the induction heating unit, as schemati- cally depicted in Fig. 229a. Typically, the solid-state and self-contained unit is relatively compact and in operation to change tools, the user positions the tool holder in a receptacle built into a shelf at the front of the unit. An induction coil is located beneath the shelf and encircles the toolholder’s sleeve (i.e. collar), which  Chapter  . ‘word-address format’ programs.*Non–modal commands are only active in that actual block. [Source, Smith et al., 19 93] Machining and Monitoring Strategies  Figure 2 23. Basic kinematics of the non-orthogonal. Tooling] . Machining and Monitoring Strategies  press its face against the rubberised contact seal and the ange (item 1), thereby acting as both a vibration damper and sealing against particle/debris. Specification and Machining Performance e ‘Variax’ is equipped with a 630 mm square pallet and can both move and cut at 66 m min –1 (i.e. see Fig. 225b), accelerating at >1g while providing >3

Ngày đăng: 05/08/2014, 21:20

Tài liệu cùng người dùng

  • Đang cập nhật ...

Tài liệu liên quan