NEW DEVELOPMENTS IN SERVO ACCELEROMETERS March 1, 1968 by Dr. Earl D. Jacobs For Presentation to the INSTITUTE OF ENVIRONMENTAL SCIENCES V4th Annual Technical Meeting April 29, 1968 St. Louis, Missouri <= INERTIAL INSTRUMENT ENDEVCO i CTEEPHONE DIVISION 714-540-2390 9018 SOUTH KiEON OFIVE SANTA ANA, CALEORMA 82707 gig Wy ‘Owisien of Becton, ickneon and Comeany S¥al3WOWTIDDV OAYIS NI SLNAWdOT2A30 M3N (s12jewose}e009 sees 91 |e 941 Jo uordis959q y)NEW DEVELOPMENTS IN SERVO ACCELEROMETERS By: Dr. Eat D. Jacobs, Endesco Corporation Dr, Jacobs holds a PRD in Physics from the California Institute of Technology where he also received his B.A. and M.S. degrees, He was on the faculty of the California State College at Los Angeles for 10 years serving as Professor of Physics and (Chairman of the Department. He has worked with several large ‘corporations in instrumentation problems, including. Autonetics Division of ‘North American Aviation where he managed the Inertia Instrument Reseach Activities for thre years. He is known for his work on advanced instrument concepts and ie cnventor ofthe Laser accelerometer Dr, Jacobs is now a vice president of Endeveo Corporation and Manager of the Inertial Instruments Division ABSTRACT |A.new high accuracy accelerometer which combines wide dynamic range and excellent frequency characteristics is described, ‘This instrument, the QFLEX accelerometer is an adaptation of| the Eyestone-Wilton sensor, utilizing a combination of high nickel- iron alloys and fused natural quart to provide log term stability, low thermal sensitivity and excellent shock and. vibration characteristics. ‘The design parameters of the QFLEX accelerometer ar dis ‘cused and a version of the instrument specifically intended for low frequency low g environmental and vibration measurements {s described. Data showing performance of the instrument is pre- sented, INTRODUCTION During the past few years there has been a dramatic increase in the requests for the measurement of low frequency, low ampl- tude acceleration. The growing requirement for these measure- ‘ments, both in terms of precision and quantity, has resulted from several technological advances. Improvement inthe state of the at ofthe design of structures has resulted in the optimization ofthe Gesign of these structures, Buildings, bridges, launch towers, ar~ frames, all demand better weightatrength factors and less ‘dead materia” As components become more complex and sensitive, better isolation of the mechanical interaction between indepen dent systems is required. We live in 2 more complex society and fare more concermed with bridge damage elevator collapse and all types df catastrophic failure of our envionment. Each ofthese has imposed an increased requirement for precise knowledge of low ‘requeney vibration and acceleration environment Continued improvements of existing transducers, particulaly Lunbonded strain gages using piezoresstive elements have been the Iajor thrist to provide means for making these measurements, However, these instruments are finally limited by their inherent acceleration bandwidth product, Accurate measurements of low frequency vibration down to steady state and at low acceleration levels require instruments with dynamic ranges far beyond those avilable from even the most venturesome concepts of strain gape transducers The natural solution i some type of fore rebalance lor servo type device Recognizing the inherent limitations of the existing instr: iments and the necesity of an advanced state of the art accelero- meter, Endevco scientists began a search for an ideal mechaniza- tion fora serv accelerometer It became clear very early that one could isolate the fundamental limitations on performance and then determine from trade-off studies, optimum configurations Since cost, aswell as performance, must finally become one of the limiting factors in design, this must also be taken into considera tion. The various trade-offs will be discussed later after brief ‘description ofthe final device. However, one point is especially interesting, that i the consideration of the element of cost in design, When cost war not considered, utilizing state of the art technology, many approaches to achieve the desired accuracy seemed feaibe. When cost was included it was necesary to com promise some essential ingredient of performance: ether the flexure, the magnet design, the servo electronics, ete, and in every cas, the final limitations were related to the fundamental roperties of materials as much asthe specific configuration ofthe particular element. Hence, st became clear that an advancement in the state of the at was required, particularly inthe materials used, in order to achive the desired sensor. Fortunately, the technology requited for sucha sensor was avalable. AS a result of work on an advanced inert grade sensor Started several years eater, an accelerometer capable of making ‘measurements to an accuracy of better than one part per milion had been developed. Combining the success of the inertial grade instrament with the demanding requirements of an instrument ‘useful for the environmental studies, the QFLEX Sensor was Incorporated into Endeveo's QA-116 Accelerometer ‘THE QFLEX SENSOR ‘The QFlex design of the Eyestone Wilson sensor is torque balance mechanism using specially processed fused natural quartz Tor the sensitive element. ‘The Eyestone Wilson sensor shown in figure 1, is « dynamic type, linear, single axis electro-mechanical device for measuring acceleration. The operation is based on measuring une component Of the force required to constrain 2 proof mass to partake in the ‘motion of the accelerated base, The sensor consists ofthe follow- Ing key elements: 1. A proof mass, pendulously supported and ideally constrained So's to allow only one degreeof freedom about a well efined axis fixed within the sensor 2. Apickoff which can sense extremely small dieplacements of ‘the proof mass about is fixed axis.ett Fig 1. The Eyestone-Wilson Sensor. 3. A torquer, consisting of a coil disposed within a permanent ‘magnetic field, attached to the proof mass which allows an ‘exactly’ linear force to be applied to the proof mass in response toa current passed through the coi 4. restorer circuit oF servo which causes an electrical curent to flow through the torquer col in response to the pickot? signal such that the resulting electromotive force exactly balances the inertial reactive forces. In this manner, the current passing through the torquer becomes an exact mes ture of the acceleration. ‘CONSTRUCTION In the QFlex design, the basic problem of constraining the proof mass in a way which is highly compliant in only one rection, and which at the same time is free of the fection effects associated with jewel pivots and bearings, but without the hyster sis, instability and fatigue associated with metal fexuresis solved by forming the flexure and proof mass from a single piece of specially procesed quart. ‘The proof mass, exure and flexure support shown in figure 2, are formed from a single fused quartz blank. A slots cut in the bank in auch a manner a8 to form an annular section and central disc connected by narrow bridge. The central dic serves as the Thin film pickoft and Torquer Leads Proof Mass. talized Pickolf Plate. Fig. 2. The Proof Mass, Flexure and Flexure Support Sub- ‘Assembly. The base structure is formed chemically from a single Piece of fused quart, proof mass, the bridge as a Mexure, and the annular section a the flexure support. The bridge section i chemically milled in order to increase its compliance in a direction normal to the central die and to improve its physical characteristics. A portion of the cen- tral disc is made conductive by vapor depositing metalic films to provide electrical surface as requied forthe capacitance pickof®. Conducting leads forthe pickof€ signals andthe torquer coil dive current are formed across the flexure by similarly vacuum depouit- ing metalic films. Finally attachment ofthe torquer cols onto the central dis completes the proof mas, flexure, and flexure support subassembly. This subassembly, is then clamped between the ‘magnet structures as shown in figure 3, ‘The balanced capacitance bridge pickoff is formed by the small gaps between the metalized portion of the quartz proof ‘mass and the fixed reference plates in the magnet structure. ‘The small preciion gap between the magnet housings and Fig, 3. An Exploded View of the Q-Flex Sensor Assembly.From this analysis we can examine the effect of the error sources which result from pendulum defection in a pendulous ‘accelerometer, the cross axis eror given by (2) and the ibropen- ulous error given by (3). “The magnitude of these errors is related to the servo stiff ness since the angle © is determined by the ability of the servo to hold its mull. Inthe Q-Flex accelerometer, the proof mass i held to its null to better than wavelength of sodium ight throughout its entre rated frequency and acceleration range. Thus O is main tained to less than 5 x 10 radians and k is considerably less than 5x 108 radians/g. This means that errors from these effects are Jess than 10 parts per millon inthe Q-Fex senor, Even in pendu lous accelerometers which have much mare poorly designed servos, these terms are generally small compared fo other error sources, |A more important error source which also gives rise to cross axis and rectification terme are effects related to shifts of the input axis and distortion ofthe support inthe presence of aceler ation. Considering the normal mico-ceep of materials and the normal values of YoungyModulus it can be seen the real design ‘problem is that of providing a well-defined and stable suspension. [tis these practical considerations which wil determine the super ofity of a particular pendulous or non-pendulous support. Consideration of these factors leads to the conclusion that the most preferable suspension technique is that which puts the least demand on the materials forming the suspension. For example, a non pendulous acelerometer in which the suspension consists of 2 voice coil support requires that thee flexures be stably located with respect to each other and that, under the influence of aecleration, al three flexures exhibit identically the same mechanical effects. On the other hand, in the Eyestone: Wilson sensor all the forces are nulled about a single exure and the use of fused quarts assures a stable isoeastic support. The experimental results on these two suspensions bear out this analy- sis: The value reported by the manufacturer of one of the best voice coil nor-pendulous accelerometers forthe DC output from vibration (measured with Ig along the input axis and a vibration of 10g peak at 250 Ra) Rectification Error (Voice Coil) = 0.001 tole ‘The result ofa similar measurement made with the Q-Flex sensor (Gee figure 6): Rectification Error (Q-Flex 10001 gale Frequency) Fig, 6, Change in DC output of QA-116 accelerometer and 20g vibration at stated frequency, Ordinate is ploted in units of change in output per of vibration amplitude LINEARITY ‘Accelerometer nonlinearity is another important effect which can lead to rectification errors. The output of an acceler- ‘meter, nesecting crosvaxis terms can be expresied as a power Series in terms ofthe acceleration along the input axis, as follows om ko+ky 9) +k242 + OF where Ag isthe output sgnal expressed in units fg, isthe input eleration, ko is the zero offset, ky is the scale factor, k2is the Second order non-linearity, et We will sume that terms higher than second order are negligible. ‘The effect ofthe second order term in expression (4) i clear. If ais a sinusoidal vibration, the second order term will result in a rectification of the AC signal wth a resulting change inthe steady- tate output In the electromagnetically restored accelerometer, the inertial reaction force is balanced by the force existing between a curent- ‘amying conductor and a magnetic field (sometimes called “pon- ‘eromotive force”). The rebalancing force given by Ampere's aw Feiss o ‘where i isthe current in the conductor, s the length of conductor land B the average magnetic field. In equation (5), F isa linear function of the curent. This is only true Ifthe magnetic field B inindependent of curent. However, asa result ofthe curent flow- ing through the coil, there wil be an armature reaction effect ‘which will introduce a current dependent term in the magnetic field, This can be expressed a follows: B=(Bo+ib) o where Bg isthe field generated by the permanent magnet and ib [nthe amature reaction Feld. Substituting (6 into (5): Ba abit soit o “Thus, infact, the force i not a linea function of the current, but has 2 second order term resulting from the armature reaction effets, This is corected in the Q-Flex accelerometer by appropri ttely arranging the relationship between a pir of torquercoils and magnets as shown in figure 7. The effect of the current i is to Increase the magnetic field in ne gap and decrease it in the other. Fig. 7. Correction of the armaturereaction effect in the Q-Flex Senor,Ac Ani =o Fig, 4. The QA-116 Accelerometer Restorer Electronics Block Diagram, RESTORER ELECTRONICS ‘A block diagram of the instrument servo is shown in figure 4, ‘The microelectronic package is 2 hybrid mix of discrete com- ponents and integrated circuits mounted on gas+-epoxy boards. ‘The servo electronics includes a power distribution circuit (not shown), 4 100KHz reference oscillator and the contol amplifier. ‘The input state of the control amplifier is an ac amplifier having high open loop gain and large common mode rejection at the reference frequency. It amplifies the suppressed cartier dix placement signal ftom the differential capacitor pickofT and provides isolation between the sensor and remaining electronics. ‘The demodulator compares the phase of the pickoff signal with the reference oscillator signal, generating an extremely low noise output proportional to the proof mass displacement. This Aisplacement signal is amplified by the de power amplifier and ‘nally impressed across the instrument torquer in series withthe readout resistor. ‘The demodulator and the integrated circuit de amplifier ‘exhibit very low offsets minimizing amplifier contribution to rll instability. The high input impedance of the integrated ‘operational amplifiers permits the use of small size capacitors and high value resistors. This allows the cater signal present inthe demodulated signal to be attenuated to an acceptable level with simple RC networks PENDULOUS VERSUS NON-PENDULOUS SUSPENSION Essential to any analysis of servo accelerometers isa discus sion of the suspension mechanisms. Several systems have been used and examined by various manufacturers, These can be roughly clasified into two categories, (1) the pendulous beam in which an element is constrained to move about a fixed axis and (@) the translational suspension in which the proof mast con strained to move along a lnear path. Although many theoretical arguments have been posed a8 10 which is superior, ultimately, ‘this determination must be made by examining the practicability of any given slution For example, it has been sugested that an advantage of the nompendulous suspension is that it eliminates croseaxis and vibra: tion errors due tothe angular deflection of the proof mass in the ‘pendulous accelerometer. The thrust ofthis argument i iustrated in figure 5. Consider two inertial reaction fore, Aj along the {input axis and Ac normal to the input axis. In the pendulous accel tometer, the force Aj will cause x torque about the suspension xis with a resulting deflection ofthe pendulum. The deflection angle @ is proportional to the input aceleration and is gven by kA; wo ‘As a result ofthis deflection a component ofthe cross axis force, ‘Ac will appear along the true input axis which is normal tothe proof mass. This will aul in an erroneous output signal Ac sin ©, of for small values of ©, the error wil be: Ee Cross AxitEnor a Ac (2) ICA\ and Ac are the components of ap inertial reaction fore, A, at say 45° to the input axis, Ac= Ais 4A or solving fr the errr (normally called the vibropendulous err) By = Vibropendulous eror= kA? (3) ‘As can be seen from (3) if Ais sinusoidal such as would result from vibration, change In the steady state output ofthe acceler ometer will result. Fig. 5, Deflection of Proof Mas inthe Pendulous Accelerometer.‘The magnetic fed in each ofthe gaps are asin equation (6) By = By ib @ B2= By-ib ° vielding for th average magnetic fed, p=BLEB2 = By (10) Hence, by providing a symmetric arrangement of the magnets and coils, the nomdinear effects due to armature reaction can be ancl, ‘The highly linear behavior of the O-Flex sensor makes the ‘measurement ofits non inearties vey difficult, Extensive measurements of the linearity in the zero t0 Ig range have been made by appropristely orienting the instrument in te earth's gravitational fed. The results ofthese measurements have yielded: p= 5x 10Spig? £5 x 10% pig? ‘The rectification effects have been determined by mounting the accelerometer with the input axis in the vertical direction and superimposing a 20g sinusoidal bration along the input axis ‘The variation inthe steady-state value of the accelerometer output {s measured with and without the 20g vibration present. The out- put signal remained ‘constant throughout the experiments to within the 0.01% resolution of the voltmeter. THRESHOLD AND DYNAMIC RANGE ‘The usefulness of an acceleration sensor for environmental studies, quite apart from the acuracy required for any given mea surement, depends on the “smallness” of the accelerations it can recognizably detect and the “largenest” of the acceleration envi- ronment it can operate in, The “smallnes or threshold of an accelerometer is deter: mined by many complicated factors, note, energy dispation, intemal stress and strain, fiction effects, et. In the final analysis however, since most accelerometers infact are force messuring devices, that is they are dynamic as opposed to kinematic sensors, the threshold of an aeelerometer must be related to the relative magnitude ofthe intemal frces present as compared to the ace! eration induced forces. Iti fr jst this reason that major efforts have been expended in the search for physlealy durable, stable ‘highly compliant flexures. Concomitant to the flexure problem, the pickoff must provide a means of preciely measuring proof ‘mass motion without affecting that motion, Given any accelerometer concept, once the threshold and ‘dynamic range for that concept have been evaluated, the upper acceleration range of an accelerometer utilizing that concept can be adjusted within rather wide Limits using standard engineering practices. For example, by reducing the mast of the proof mass by 4 factor of two, or doubling the magnetic field strength the upper acceleration limit of an electromagnetic accelerometer can be doubled. The final limit or input range, of couse, i related tothe ‘mechanical integrity ofthe structure itself ‘As of this writing, there is no experimental data avilable on the upper acceleration limit of the Q-Flex sensor. A standa Sensor using a specially built high power servo hasbeen used to make measurements of a centrifuge operating at 200g. Extensive shock testing of the sensor used in the QA-II6 series accelero- ‘eters has been undertaken, but these have been limited to 100 shocks. None of these studies have uncovered any bac limitation (on acceleration range and its anicpated that vey high acceer- ‘ometers could be built using the Q-Fex design, The restorer elec ‘wonics used in the QA-116 accelerometers are designed for a ‘maximum operating input of 208, ‘The experimental determination ofthe threshold ofan accel trometer which is known to have a very low threshold is very dificult, Since the world is constantly in motion, ita challeng- ing problem to induce and detect a motion in the acelerometet Which might be small compared to other real but uncontrolled input motions For the purpose of making these lowevel measurements & 30,000 pound concrete pad was built, extending into the ground and isolated from the building concrete floor. A QFlex sensor With a low noise servo was mounted on the pad with the input axis vertical. The resulting one output was measured with a differential volumeter and recorded on 8 wale of 10 microg's A Portion of the record chart from this testis shown in figure 8. The ambient noise was approximately 0.1 microg me. The disturbance which occurred at 11:05 pim., PST, December 27, 1967, was compared with data from the California Institute of ‘Technology Seismology Laboratory and traced to.an earthquake in Chile, $600 miles away. The points marked a,b, c,d, are for iden- tification and comparison purposes only. From ths data itis con- cluded that the upper limit of the threshold, characteristic of the (QFlex design, is of the order of 0.1 micro-g. This means that for all practical purposes the threshold ofthe instrument willbe lim ited by factors other than the senor itself ito} pm.PsT_ oegeween 2, iz : Fig 8, Seismic Disturbance detected with the Q-Flex Sensor.Relative Reepone: ve 30 100 ‘500 1000 2e Frequency (Ha) Fig. 9. Frequency Response ofthe QA-116 THERMAL EFFECTS ‘The combination of the low coefficients of thermal expan sion of te fused quartz and invar used forthe structural elements and the thin film conducting leads used to bring power into the proof mass result in unusually small thermal effects in the QFlex Sensor. Even the normal temperature coefficients associ permanent magnets have been compensated by balancing net strength temperatute dependence against the vars reluctance, It possible, in principle, to build Q-Flex sensors in Which al of the thermal sensitivities are ofthe order of 1 ppm/°F and, in fact, the accelerometers intended for guidance applications ‘using basically the same sensor consistently demonstrate temper ature coefficients of zerounbalance and scale factor of less than 10 ppmjPF. However, tis impractical to attempt to achieve this low value in more economical acelerometers, largely because of the additional cost of testing to verify the low coefficients andthe larg increas in the cos of the related electronic equipment. ‘The thermal zero shift and thermal sensitivity sift of the accelerometers intended for environmental studies are typically less than 0.01% /F, FREQUENCY RESPONSE. ‘The smal gap between the proof mass and the reference plate hich forms the capacitance pickoff also provides gas damping Overs wide region of temperstures and presures forthe dimen: sions involved, the gas behaves ike an ideal Newtonian fui, and hence is relatively independent of temperature, The normal gas damping is between 0.6 and O38 of ei Because of the very nature of sero accelerometers, the fre: quency response is extremely flat at the lower frequency range ‘The response at higher frequencies depends on both the character. istics of the sensor and the servo. Addition of lead or lag in the hase response of the servo is equivalent to adding a negative or Positive electrical damping term. The natural frequency of the Accelerometers determined by the combination of the mechanical charitenstis ofthe sensor and servo gain, ‘The frequency response of the QA-II6 accelerometer is shown i Figure 9, Fig, 10, The Endeveo QA-116-15 Low G Accelerometer Fig. 11. Outline Drawing ofthe QA-116-15 Accelerometer.‘THE QA-116-15 ACCELEROMETER Endeveo is curently producing several models of the QFlex sensor intended to be used for environmental studies. These have been designated the QA-116 series accelerometer. One of these, the QA-II6-5 is shown in figures 10 and 11 A varity of compli: mentary electronic and mounting hardware is avaiable for this, instrument. It is also compatible with many existing measurement systems including some design for piezoresistive accelerometer. ‘The performance specifications of the QA-I16-15, which are ‘typical ofthe QA-116 series are: Range (Full Seale) £15¢ Sensitivity (Nominal) 250 mv/g Excitation Voltage 28 Votts De ‘Output Resistance 250 ohms Resolution (0c) ©.00001g Threshold o.000018, Linearity (Full Scale) 0.059%, Hysteresis (Full Scale) 0.05% Frequency Response (5%) DC to 500 Hz Natural Frequency (Nominal) 1000 Hz Zera Unbalance (Less Than) 20.02 ‘Thermal Zero Shift (Max.) o.00028/"F Thermal Sensitivity Shift (Max.) 0.01% /"F Transverse Sensitivity 0.0028/8, ‘ACKNOWLEDGEMENT ‘The author wishes to express special appreciation to Mr. Kenneth R. Jackman of General Dynamics for his support and Patience in the preparation of ths paper. The work described, of| course, is the result of many individual scientists and engineers, too numerous t ist.