Tape transport control based on sensor fusion

Transcription

1 Preprints o the 19th World Congress The International Federation o Atoatic Control Cape Town, Soth Arica. Agst 4-9, 14 Tape transport control based on sensor sion Angeliki Pantazi*, Gioanni Cherbini*, Eiji Ogra**, and ens elitto* *IBM Research Zrich, 883 Rüschlikon, Switzerland (e-ail: {agp, cbi, jje}@zrich.ib.co) **IBM STG, Tokyo, apan (e-ail: eogra@jp.ib.co) Abstract: Reliable and precise tape transport is o ndaental iportance to achiee larger oletric recording densities in tape storage systes. The perorance o the tape transport control syste, which is aected by ariations in the tape elocity and tension, ipacts the write and read qality o the data tracks and eentally the achieable areal recording density. Dring operation in crise elocity ode, distrbances in elocity and tension ay be indced or exaple by tape reel eccentricities. This proble is particlarly serios when the reel rotation reqencies are close to the resonance reqency deterined by the tape path. Typically, the tape elocity is estiated at the tape head ro a sero signal, which is obtained by reading pre-oratted sero inoration, and sed or tape transport control dring crise ode. In sch a schee, elocity distrbances obsered at the head cannot be attribted to an indiidal reel. Hence no targeted distrbance sppression can take place at the indiidal tape reels. Howeer, Hall sensors are typically inclded in tape dries to obtain additional tape elocity inoration ro the indiidal reels. This inoration is sed to achiee proper tape transport operation in the absence o alid elocity estiates ro the pre-oratted sero inoration, or exaple dring tape acceleration. In this paper, we present a noel control schee or tape transport that ses elocity easreents ro three sorces, the Hall sensors at the tape reels and the sero channel that yields elocity estiates at the head. Speciically, characterization o the ltiple-inpt ltiple-otpt (MIMO) tape transport syste proides an accrate syste odel and enables an optiized two-sensor control design. Frtherore, H iltering is eployed to peror sensor sion and the reslting state estiates are sed in the eedback control o the two reels. Experiental reslts are presented to illstrate the behaior and perorance o the proposed tape transport control syste. Keywords: Tape transport, elocity control, sensor sion, MIMO syste. 1. INTRODUCTION The continos growth o digital data the International Data Corporation (IDC, 1) projects the digital nierse to grow to 4 zettabytes by will reqire rther drastic iproeents o aordable and high capacity storage technologies to hold these ast aonts o data. Tape systes keep sering as a low cost storage tier by proiding the lowest total cost o ownership and are especially sited or inreqently accessed data sch as backp and archial storage (Reine and Kahn, 1). Tape prodct roadaps as well as the tape storage roadap set by the Inoration Storage Indstry Consorti (INSIC, 1) indicate that extreely high tape capacities o ore than 1 Tbyte appear easible within the next decade. The potential or rther signiicant tape areal density iproeents hae also been inestigated in Argedo et. al. (8) and Cherbini et. al. (11). Adances in seeral technology areas are necessary to achiee the aggressie tape capacity targets. Linear and track density increases as well as thinner tape edia are key ingredients to achieing higher storage capacities. In this paper we ocs on the design o an iproed reel-toreel syste to enable higher track density and to spport thinner tape edia throgh ltra-precise control o the tapetransport syste. In particlar, tight control o elocity, tension and potentially o tape diensional stability (TDS) ariations will gain in iportance when oing to thinner tape aterial in order to enable increased tape oletric density. In a conentional tape-transport syste seeral sensors are sed to deterine the tape-transport paraeters. Typically a priary tape elocity estiate is proided by reading a preoratted sero pattern ro the tape with one or ore sero readers and extracting the releant inoration sing a sero channel. The priary tape elocity proides a easre o the tape elocity at the tape head location. The liitation o sensing the priary elocity at the tape head is that elocity distrbances cannot be attribted to an indiidal tape reel. Howeer, additional Hall sensors are aailable and are sed to obtain secondary tape-elocity inoration ro the indiidal reels. This inoration is typically ore noisy than the priary elocity estiate, and is sed to control the tape-transport operation in the absence o priary elocity estiates, e.g. dring tape acceleration or deceleration or wheneer the sero readers are not aligned with the sero patterns on tape. The priary elocity response is typically assed to be the aerage o the elocity responses at the two reels (Mathr Copyright 14 IFAC 6849

2 Cape Town, Soth Arica. Agst 4-9, 14 and Messner, 1998). With this approach, no targeted distrbance sppression is possible at the indiidal tape reels. Howeer, a recent tape transport syste identiication stdy has shown that the priary tape elocity easred at the head is doinated by the elocity at the reel spplying the tape rather than by the aerage elocity o the spply and take-p reels (Cherbini et al., 13). In this paper, a noel tape transport control syste is proposed, which exploits the aailability o ltiple sensors or elocity estiation and ses a sensor sion approach that cobines priary and secondary elocity estiates to optially control the indiidal tape reels. First, the tape transport syste identiication procedre as deeloped in Cherbini et al. (13) is briely described. Measreents o the secondary tape elocities at the indiidal reels sing Hall sensors and the priary elocity at the tape head are sed or characterizing the relationships between the indiidal elocity contribtions. It will be shown that the elocity obsered at the head is essentially deterined by the reel spplying the tape. Based on these indings, the classical tape transport syste, which ses the priary elocity estiate to control both tape reels, will be extended. In a irst approach, the take-p reel will be controlled by the Hall elocity easred at that reel, and the spply reel will be controlled by the priary elocity estiate. In a second approach, sensor sion is introdced to optially control the spply reel sing both the priary elocity estiate and the Hall elocity estiate ro the spply reel. The concept o sensor sion has been sccesslly sed or nanopositioning in probe storage deices (Pantazi et al., 7). A detailed description o the concepts and controller design or nanopositioning with ltiple sensors can be ond in Sebastian and Pantazi (1). In a tape reel-to-reel control syste, sensor sion can be sed to cobine the low-noise priary elocity estiate at the tape head with the ore noisy easreent at the tape reels sing Hall sensors, which howeer contain only reel-speciic elocity coponents. The priary elocity estiates proide better noise perorance copared with the Hall sensor easreents especially at high reqencies. The sensor sion proble orlation is introdced, where the state ariable o interest is the spply reel elocity. An H iltering approach is sed or the estiator design. The optial state estiate is obtained ro the control signals and the easreents ro the two sensors, i.e., the Hall sensor and the sero channel. Experiental reslts show that the concept o sensor sion yields iproed perorance in ters o elocity ariation and tape tension ariation, especially itigating distrbances close to the tape path resonance reqency. The paper is organized as ollows. The tape-transport syste identiication procedre is described in Section. Tape transport control by elocity eedback is introdced in Section 3 and extended with the concept o sensor sion in Section 4. Experiental reslts are discssed in Section 5. The paper concldes with Section 6.. TAPE TRANSPORT SYSTEM IDENTIFICATION The block diagra o the tape transport and track-ollowing seroechaniss in a tape drie is shown in Fig. 1. For otion in the orward direction, the tape is transported ro the ile reel, acting as a spply reel, to the achine reel, acting as a take-p reel, throgh the tape path deterined by the rollers R1 to R4. In reerse direction, the roles o the ile reel and achine reels are reersed. Read/write operations are perored in contact with the tape by the read/write eleents or the sero and data channels that are present in the tape head. A digital sero channel proides estiates o the tape elocity, tape longitdinal position, and head lateral position with respect to the tape, which are deried ro sero signals that are read by sero readers in the head odle. Hall sensors are sed to obtain additional tape elocity inoration ro the indiidal reels, which typically is sed to achiee proper tape transport operation in the absence o alid paraeter estiates ro the sero channel, e.g., dring tape acceleration. A seroechanis is considered here with no eedback o tension. Sero channel Machine reel File reel otion Fig. 1. Block diagra o tape transport and track-ollowing seroechaniss. A brie oeriew is now gien o the tape transport odel and identiication ethod, described in detail in Cherbini et al. (13). The echanical behaior o the syste is goerned by second-order dierential eqations, which are obtained by eqating the change in anglar oent to the s o torqes or each reel, and are expressed in state-space or as x & = F x + G, (1) where the state ector is gien by x x = x tape position at achine reel = tape elocity at achine reel, () ( ) t tape position at ile reel tape elocity at ile reel the ector o control signals is 685

3 Cape Town, Soth Arica. Agst 4-9, 14 =, and 1 ( 1+ µ ) RKT ( 1+ µ ) RDT β ( 1+ µ ) RKT ( 1+ µ ) RDT = F 1 R KT R DT R KT R DT β RK G = R K (3), (4) where K T and D T are the tape spring constant and daper coeicient, K and K are the otor drier gains, β and β are the otor iscos riction coeicients, R and R are the radii, and are the oents o inertia o the achine reel and ile reel, respectiely, and µ is the Colob riction coeicient. Note that the radii and the oents o inertia are tie arying. The tape transport otpt signals are expressed or orward tape otion in ters o the state ector as y = H x τ =, where H = KT 1 D T K T 1, D T and where (t) and (t) are the tape elocities at the achine and ile reels, respectiely, and τ (t) represents the tape tension. With reerence to Fig. 1, the secondary elocities at the achine and ile reels, H, ( t) and H,, are easred by the Hall sensors, whereas the priary elocity (t) at the head is estiated by the sero channel. Syste identiication is perored in the reqency doain. The tape transport paraeter estiates and the reqency responses o the tape transport syste as described by Eqs. (1) to (5) are obtained at a gien longitdinal tape position by alternatiely speriposing chirp signals on the achine reel and ile reel otor inpt signals, while the tape transport seroechanis is operated in closed loop at a certain noinal tape elocity and tension. Closed-loop operation dring application o the chirp signals is considered to preent large deiations o the reglated ariables ro the noinal ales. The paraeter ales depend on longitdinal tape position, tape elocity, and tension, and are identiied by the iniization o a cost nction, which is gien by the L nor o the dierence between the reqency responses o the syste gien by Eqs. (1) to (5) and those obtained by experiental waeors. For a wide range o tape elocities and tensions, the reqency responses ro the inpt signals to the priary tape elocity closely atch the responses ro the inpt (5) signals to the secondary tape elocity at the spply reel, which is the ile reel or tape otion in the orward direction, and the take-p reel in the reerse direction. This reslt indicates that in steady state, contrary to the assption sally ade that the priary elocity response is the aerage o the elocity responses at the indiidal reels, see, e.g., Mathr and Messner (1998), the priary tape elocity atches the secondary tape elocity at the spply reel. Figre shows the reqency responses identiied and the reqency responses obtained ro the state-space odel by sbstitting the estiated ales o the paraeters K T and D T in the odel o Eqs. (1) and (5), or noinal ales o tape elocity eqal to 4.1 /s and or a longitdinal position corresponding to the beginning o tape with otion in orward direction. The coparison o the reqency responses ro the inpt signals to the priary elocity and to the Hall elocities indicates that the reqency responses ro the inpt crrents to the priary elocity closely atch those ro the inpt crrents to the Hall elocity at the ile reel, which or the considered case o orward tape otion plays the role o the spply reel, as entioned aboe, whereas there is no atch between the priary elocity and the Hall elocity at the achine reel. Aplitde (db) Phase (deg) Aplitde (db) Phase (deg) Machine reel otor crrent Velocity Machine reel secondary elocity Priary elocity Model: Fro achine crrent to ile reel elocity File reel secondary elocity Priary elocity Model:Fro ile crrent to ile reel elocity File reel otor crrent Velocity Fig.. Freqency responses ro achine reel otor crrent to the priary and achine reel tape elocities (top), and ro ile reel otor crrent to the priary and ile reel tape elocities (botto), ro Cherbini et al. (13). 3. CONTROL OF TAPE TRANSPORT BY VELOCITY FEEDBACK The block diagra o a classical tape transport control syste with eedback o priary elocity is shown in Fig. 3. The closed-loop syste or each reel consists o a eedback controller or tape-elocity control and a eedorward controller or tape-tension control. Fig. 4 illstrates the 6851

4 Cape Town, Soth Arica. Agst 4-9, 14 spectrogras o the secondary and priary elocity easreents with tape otion in orward direction. Large spectral coponents are shown as dark red cres, whereas sall spectral coponents are shown in light green. Note that the reqencies o the spectral coponents o the elocity at the achine reel, acting as take-p reel, decrease with tie, as the achine reel radis increases, whereas the reqencies o the spectral coponents o the elocity at the ile reel, acting as spply reel, increase with tie. More iportantly, the reqencies o the spectral coponents o the priary elocity tend to increase, indicating the strong copling o the elocity at the ile reel, or in general at the spply reel, with the priary elocity. Frtherore, this sggests that sing the eedback o the priary elocity to control the elocity at both reels ay lead to larger adjstent errors as copared to a control schee where the copling o priary and secondary elocity is taken into accont. r + Machine reel controller K FF K FB H, P TT τ H, File reel controller K FF Tape transport syste, R, R K FB + r Fig. 3. Block diagra o tape transport control syste with eedback o priary elocity. 4. CONTROL OF TAPE TRANSPORT BASED ON SENSOR FUSION The syste identiication and the elocity distrbance characterization presented in the preios sections hae shown that there is no atch between the priary elocity and the elocity easreent obtained by the Hall sensor at the take-p reel. Based on the aboe obseration, the classical tape transport control syste is odiied as shown in the block diagra o Fig. 5. For the take-p reel the elocity eedback is yielded by the elocity easreent obtained by the Hall sensor at that reel. For the spply reel, two approaches are considered. In the irst approach, the priary elocity easreent is sed directly or eedback control o the spply reel. Thereore, depending on the direction o tape otion a siple switching is perored in the elocity sed or eedback control between the priary and the secondary easreents. In the second approach, sensor sion is perored to obtain an optial estiate o the spply reel elocity. The estiate o this state ariable is sbseqently eployed or eedback control o the spply reel. r + Machine reel controller K FB P TT File reel controller K FF K FF Tape transport syste, R, R K FB + r Forward direction: H, Reerse direction: H, τ H, Forward direction: Reerse direction: H, Fig. 5. Block diagra o tape transport control syste with eedback o secondary and priary elocities. Fig. 4. Spectrogras o the secondary (top and iddle) and priary (botto) elocity easreents. Een thogh the identiication reslts hae shown that both the priary elocity estiated by the sero channel and the secondary elocity easred by the Hall sensor proide a good estiate o the elocity at the spply reel, a ore accrate estiate o the elocity at the spply reel can be obtained by cobining the inoration ro both sensors. The priary elocity estiation is deried by accrate sero inoration oratted in the tape that is read by a sero reader and sbseqently processed by the sero channel. Thereore priary elocity estiates hae better noise perorance than the Hall sensor easreents, especially at the high reqencies. Howeer, the priary elocity is obtained when the tape reaches the head location, hence lowreqency elocity ariations at the spply reel ay not be accrately captred. Frtherore, as entioned earlier, priary elocity is not aailable at ery low speeds and dring acceleration. Hence, it is essential to hae the Hall elocity easreents always aailable, in addition to the priary elocity. 685

5 Cape Town, Soth Arica. Agst 4-9, 14 d W d P n H, W H W p H, B G11 G1 G1 G + G 31 G3 G41 G4 Bn G1 Dd GDd d + G3Dd nh G D 4 d np Bp 4 ) d = n H n p w, K EST (a) P E n p ) z = H, x C = x 1 C p x p Dn d DH + nh Dp np ) (b) E H, y = Fig. 6. Sensor sion proble: block diagra representation (a), and LFT orlation (b). A block diagra representation o the sensor sion proble is illstrated in Fig. 6(a). An optial estiate o the reel elocity state ariable has to be obtained sing an estiator K EST. The releant spectral characteristics o easreent noise sorces and distrbances are captred by the weighting ilters W H, W p and W d haing speciic reqency responses. In particlar, W p is chosen to be a irst order low-pass ilter to captre inaccracies o the priary elocity estiate at ery low reqencies. For siplicity, W H and W d are chosen as scalar gains. The gain selection o the W H weight ais at achieing a reqency separation o the two elocity easreents in the elocity estiate. Filter selection with ore elaborate spectral shaping is possible, bt will reslt in a higher-order estiator. The agnitde responses o the weighting ilters or the noise sorces are shown in Fig. 7. The W d weight captres the characteristics o the distrbances that are assed to be white. The agented state-space eqations describing the dynaics o the tape transport syste inclding the weighting ilters are gien by A x& F11 F1 F13 F14 x & F1 F F3 F4 x& = F31 F3 F33 F34 x & F41 F4 F43 F44 x & p Apxp where F, G, denote the state-space atrices o the tape transport syste P deined in Eq. 4, and A p, B p, C p and D p are the state-space atrices o W p ; D H and D d are the constant gains o W H and W d, respectiely. For sensor sion, the state ariable o interest is the spply reel elocity, which corresponds to the ile reel elocity or the considered case o orward direction o tape otion. An H iltering approach is sed or the estiator design. The optial state estiate is obtained ro the control signals and the easreents ro the two sensors, i.e., the Hall sensor and the sero channel. A linear ractional transoration (LFT) orlation o the sensor sion proble is illstrated in Fig. 6(b). The generalized plant odel is gien by where C 1 = [ 1 ]. The objectie o the optiization proble is to design the estiator E sch that it bonds the ininity nor o the transer nction T zw = F l (P E,E). Magnitde (db) 4 4 W H W p Fig 7. Magnitde responses o the ilters W H and W p. 6853

6 Cape Town, Soth Arica. Agst 4-9, 14 The sensor sion achieed sing the H iltering approach is illstrated by the reqency responses presented in Fig. 8, where the indiidal transer nctions relate the estiated spply reel elocity with H,,, and, and the reqency separation achieed with the designed estiator is clearly isible. At ery low reqencies the elocity estiate is ostly deterined by the secondary elocity H,, whereas at high reqencies the estiate is essentially based on the priary elocity. The reliance o the elocity estiate on the inpt signals and is deterined by the selection o the weight chosen or W d. A siilar orlation can be introdced or the reerse tape otion direction, where the role o the spply reel is perored by the achine reel. The estiated tape elocity at the spply reel can be sed jointly with the elocity at the take-p reel easred by the second Hall sensor or the eedback control o the tape transport elocity, as shown in the next section. Magnitde (db) 5 1 Estiate based on H, Estiate based on Estiate based on Estiate based on Fig 8. Freqency responses relating the spply reel elocity estiate with the sensor easreents and the inpt crrents. 5. EXPERIMENTAL RESULTS The perorance iproeents oer a classical tape transport control syste, which are obtained by selectie eedback o tape elocity inoration or by estiated elocity inoration, are illstrated sing an experiental tape path. The tape path is deterined by the achine and ile reels and or roller eleents. The rollers are langeless and gide the tape along the path oer the head eleent. Tension is easred by two strain gages located on two o the or rollers in the tape path. A pictre o the experiental tape transport set-p is shown in Fig. 9. Fig 9. Experiental tape transport set-p. In Fig. 1 a coparison o the priary tape elocity power spectral densities is gien or the three considered control syste conigrations, naely a) eedback control o spply and take-p reels sing priary elocity, b) priary elocity sed or eedback control o spply reel and Hall easreent or eedback control o take-p reel, and c) eedback control o the spply reel sing the elocity estiated by the two sensors and o the take-p reel sing the Hall easreent. The elocity spectr shows that the ain distrbance appears at a reqency o ~15 Hz. Note that the two schees that tilize the Hall elocity inoration along with the priary elocity achiee redced elocity ariation, especially at the reqencies o the elocity distrbance. In ters o standard deiation o the tape elocity, the three schees yield a) 5.8 /s, b) 3.63 /s, and c) 3.57 /s. Power Spectral Density File: priary elocity, Machine: priary elocity File: priary elocity, Machine: secondary elocity File: estiated elocity, Machine: secondary elocity Fig 1. Power spectral density o priary elocity or dierent tape transport control conigrations. Figres 11 and 1 copare the tape elocity and tension ariations obtained by a classical syste based on the priary elocity eedback only, and the enhanced syste that ses the Hall elocity along with the estiated elocity or the take-p and spply reel, respectiely. Note in Fig. 11 the aster conerges o the tape elocity in the case where the 6854

7 Cape Town, Soth Arica. Agst 4-9, 14 Hall easreent is sed along with the estiated elocity. Note in Fig. 1 that the iproed elocity control proides also an iproed perorance in ters o tape tension, especially itigating the distrbances at reqencies close to the tape path resonance. In ters o tension standard deiation, the perorance iproes ro.1 N or the syste based on priary elocity eedback to.16 N or the schee with the siple switching to the secondary elocity easreent. The tension perorance rther iproes to.14 N or the syste sing joint elocity control. Tape elocity (/s) File: priary elocity, Machine: priary elocity File: estiated elocity, Machine: secondary elocity Tie (s) Fig 11. Coparison o tape elocity obtained by eedback o priary elocity only and by joint elocity control. Tape tension (N) File: priary elocity, Machine: priary elocity File: estiated elocity, Machine: secondary elocity Tie (s) Fig 1. Coparison o tape tension obtained by eedback o priary elocity only and by joint elocity control. 6. CONCLUSIONS A noel control schee or tape transport that tilizes elocity easreents ro three sorces - the Hall sensors yielding secondary elocity inoration at the tape reels and the sero channel proiding priary elocity estiates at the head location - has been introdced. An accrate syste odel has been deried sing an adanced characterization procedre, thereby enabling an optiized two-sensor control design. H iltering has been introdced to peror sensor sion or iproed eedback control o the elocity at the two tape reels. Experiental reslts show iproed perorance o the noel control concept in ters o redced elocity ariation and tape tension ariation when copared with a classical control schee that only relies on priary elocity eedback. An interesting extension o the concept will be obtained by inclding eedback o tension ariation estiates as well as tension sensor estiates into the tape transport control design. REFERENCES Argedo A., Beran D., Biskeborn R., Cherbini G., Cideciyan R., Eletherio E., Häberle W., Hellan D., Iaino W., elitto., dd K., bert P.-O., Lantz M.A., McClelland G.M., Mittelholzer T., Narayan S., and Ölçer S., (8). Scaling tape-recording areal densities to 1 Gb/in. IBM. Res. Deelop., 5, pp Cherbini, G. Cideciyan R.D., Dellann L., Eletherio E., Haeberle W., elitto., Kartik V., Lantz M., Oelcer S., Pantazi A., Rothizen H., Beran D., Iaino W., bert P.-O., McClelland G., Koeppe P., Tsrta K., Harasawa T., Mrata Y., Msha A., Nogchi H., Ohts H., Shiiz O., and Szki R., (11). 9.5 Gb/in Recording Areal Density on Bari Ferrite Tape. IEEE Trans. Magn., 47, No. 1, pp Cherbini, G., Pantazi, A., and elitto,. (13). Identiication o MIMO Transport Systes in Tape Dries. Proc. 13 IEEE/ASME International Conerence on Adanced Intelligent Mechatronics AIM 13, pp Wollongong, Astralia. IDC (1). THE DIGITAL UNIVERSE IN : Big Data, Bigger Digital Shadows, and Biggest Growth in the Far East, Aailable: INSIC (1). INSIC s 1- International Magnetic Tape Storage Roadap, Technology Oeriew, Aailable: oadap%-%techoeriew%v4.1.pd Mathr, P. D., and Messner, W. C. (1998). Controller deelopent or a prototype high-speed low-tension tape transport. IEEE Trans. Control Syst. Technol., ol. 6, pp Pantazi, A., Sebastian, A., Cherbini, G., Lantz, M., Pozidis, H., Rothizen, H., and Eletherio, E. (7). Control o MEMS-based Scanning-probe Data Storage Deice. IEEE Trans. Control Syst. Technol., ol. 15, no. 5, pp Reine, D. and Kahn, M. (1). In search o the long-ter archiing soltion - tape deliers signiicant TCO adantage oer disk, Clipper Notes. Aailable: Sebastian, A., and Pantazi, A. (1). Nanopositioning With Mltiple Sensors: A Case Stdy in Data Storage. IEEE Trans. on Control Syst. Technol., ol., no., pp ,