PROCESSING TIME SERIES DATA (VELOCITY, SHEAR STRESS) FROM THE PULSE WIDTH MODULATED-CONSTANT TEMPERATURE ANEMOMETER (PWM-CTA) J.F. Foss, Professor Michigan State University Mechanical Engineering Department East Lansing, MI 48824 T.J. Hicks, Chief Engineer Digital Flow Technologies, Inc. 4731 Nakoma Drive Okemos, MI 48864 ABSTRACT The digital output from the PWM-CTA represents the short duration (e.g., 20sec) true time average of the analog velocity (addressed herein) or shear stress. The developed techniques to extract the moments of the velocity (or shear stress) distribution and to create a regular time series are presented in this paper. INTRODUCTION The constant temperature anemometer (CTA) has been an important instrument for fluid mechanics research since its widespread availability in the 1960s. The use of the Wheatstone bridge and the feedback amplifier – whose differential input signals an increase/decrease in current to compensate for a momentary increase/decrease in the cooling velocity at the probe – provides a servo system that is capable of following rapid fluctuations of the velocity. The CTA has been the subject of numerous technical descriptions. Recommended ones include Perry1, Comte-Bellot2, Fingerson and Freymuth3. The Pulse Width Modulated-Constant Temperature Anemometer (PWM-CTA) provides a distinctly different electronic architecture (in comparison with the conventional (C)-CTA) with which to track high frequency velocity fluctuations. The PWM-CTA was introduced in the publication by Foss, et al.4 and protected by the subsequently issued patent: Hicks, et al.5. The PWM-CTA has been under continuous development since these early disclosures with the most recent description provided by Foss, et al.6 in which comparative measurements – with a Disa 55M system – in a slit-jet flow field were reported. Strong agreement, in terms of the first four moments of the velocity distribution as well as spectra, were shown in these comparative measurements. The present paper is to: i) further clarify the extraction of moments of the velocity (or shear stress) distribution from the PWM-CTA output, and ii) to report on the regularizing strategy that is employed to develop a periodic output for single and multiple sensor measurements. As shown below, this transformation of the data is required since the actual measurement period is a function of the effective cooling velocity that is experienced by a given sensor. The basic circuit configuration for the PWM-CTA is presented in the next section and the data processing observations and strategies are provided in the following two sections. BASIC FEATURES OF THE PWM-CTA Circuit A simplified block diagram of the PWM-CTA is presented in Fig. 1. When the electronic switch (S) is closed at the beginning of the next heating cycle, the adjustable DC voltage: EA, is imposed across the cable-impedance matching resistor (R1) and the hotwire (hot-film, shear stress, etc) sensor. The resulting current flow heats the sensor (resistance RS) with a corresponding increase in the temperature (and hence the resistance) of the sensor. The voltage: EB, across the probe itself, will exhibit a corresponding increase during this heating portion of the cycle. By adjusting the parameters of the circuit, the dual inputs to the comparator can be made to trigger its output-state change when RS(t)=RH=[1+OHR Rc] where OHR is the designated overheat ratio for the wire whose ambient temperature resistance is Rc. The high gain (104), differential amplifier [E B E Ref ] , ensures that a steeply rising signal will “sharply” trigger the comparator when the RS=RH condition is met. The triggered comparator output is used to: i) open the switch that connects EA to the sensor, ii) stop the accumulation of the counts from the master clock, and iii) present this digital value as the PWM-CTA output for the present cycle to the multiplexer (MUX) for storage in the host P.C. The accumulated counts represent the duration of the heating period: k (for the kth cycle) of a given sensor. As noted in Fig. 1, the PWM-CTA common unit (master clock, etc.) can serve multiple (up to 32) measurement channels. it is important to realize that this equilibration occurs in one cycle and that, in principle, there is no implied upper frequency limit. Specifically, the cycle time (T) can be arbitrarily reduced for arbitrary increases in the density-velocity product (V). Practical limitations in terms of the measurement resolution do, of course, exist. These are related to the frequency of the master clock. With the cessation of the heating current at time tK+K (for the Kth cycle), the following operations complete the cycle: a) the sensor is cooled in preparation for the next heating cycle, and b) the counter is reset. The multiple channel, common clock timing, digital output (no A/D required) operation is considered to make the PWM-CTA an excellent anemometer for multi-sensor (rakes, arrays) measurements. The period of one cycle (T) is a free parameter for the operation of the PWM-CTA. It was selected as 20sec (1/50 KHz) for the present data. The ~Velocity Magnitude (Q) Transfer Function The electrical attributes of Fig. 1 can be represented in terms of the sensor’s RS(t) behavior for several cycles: K1, K, K+1; see Fig. 2. The prescribed energy state for the sensor: RH, is reached at (tK1+K1) whereupon the wire cools in preparation for the next cycle. This same energy state is achieved in the next cycle at tK+K. The symbol: <T K> designates the time between these equal energy states as: <TK>=tK+K(tK-1+K-1)=T+(KK-1) The present unit, which operates at 50KHz, is considered to be appropriate for typical laboratory investigations. For example, at 50m/sec, the 20sec time period will deliver a “streamwise length averaged” measurement of 1mm. Since a typical sensor length and diameter are 1mm and 5m, this will provide the “spatially averaged velocity” within a 1mm 1mm 5m domain in space. It can also be noted that a 500KHz unit is in development. The first unit of this kind will be utilized by Prof. Wm. Saric (Arizona State Univ.) in his studies of boundary layer transition at M=2.4 The energy loss effects (conduction to the supports plus convection to the surrounding gas) can be equated to the electrical power input in the same time period. With reference to Fig. 2, the following expressions define this equilibration condition. (1) In a statistically stationary flow, the time average of the <TK> values will be equal to T. The (KK1) term in (1) is not a fluctuation (i.e., it is not like u=u(t) u ); rather, it is an indication of (u/t). A detailed feature of the PWM-CTA operating condition can be noted here: a nominal one percent change in RS: 0.99RS/RH1.00 is targeted for the anemometer’s operation. This is controlled by EA for a given T. Also, the targeted operating range for the values is 0.25/T0.5. This avoids starting transient effects (i.e., /T<0.25 for velocity = 0) and it permits adequate cooling of the sensor in the period tK+K<t<tK+T). Note that there are 4096 counts to define T. Hence, the value is resolved to nominally one count in 1024. Given that the key to the precise transfer function between K, <TK> and the temporally averaged <VK> velocity during the time period <T K>, is the return of the sensor to the preset energy state (RS=RH) at t=tK+K, t K τK t electrical energy liberated in <T K> by the sensor: (I S2 /R S ) dt K t K K t K t K K t 2 EA R S (t) dt R 1 R S (t) g(R S ; E A , R 1 ) dt (2) K where EA is an adjustable constant in the weakly time dependent g function. The circuit component: R1, is set at 50 for the hot-wire sensor of this study. Its value would be adjusted if, for example, RS1K for a shear stress sensor. Heat transfer from sensor to surroundings in the relevant time span: t K τK t K 1 τ K 1 [A BQ n ] dt [A B VK n ] TK the (3) where <VK> is the time average of the effective cooling velocity V(t) in the period <TK>. The preceding has clarified that (2) and (3) are equal in magnitude; hence t K K t g[R S ; E A , R 1 ] dt (4) K [A B VK n ] TK The coefficients (A,B) will be evaluated in a direct calibration. Since RS(tK)RS(t)RS(tK+K)=RH and since RS(tK)/RH0.99, the left hand side integrand is nominally constant and its time variable contribution can be absorbed into the coefficients of the following equation: τK A B VK n TK (5) A representative calibration data set for a 5m diameter, 1mm long tungsten sensor (with 30m diameter, 1mm long copper plated stubs) is shown in Fig. 3. The right-hand side of this expression, with the exception of its short time average characteristic (<V>), is identical to that for the conventional CTA. The latter would be written with E2 on the l.h.s. This observation leads to a primary difference between the PWM-CTA and the C-CTA; the former is more sensitive since it has a (nominal) second power: /<T> ~ <Vn>, instead of a fourth power: E2~Vn, response. That is, given that n0.5, an incremental change in /<T> can be expressed as (/<T> ~ 1 (V1/2)<V> whereas E ~ (V 1/ 2 ) V . E POST-ACQUISITION PROCESSING ALGORITHMS Stochastic Values – Single Sensor If single sensor data are acquired for a sufficiently long time period, the corresponding moments of the velocity magnitude distribution can be readily obtained as shown below. Each K value can be used to obtain <VK> for the corresponding <TK> time period. The first moment of the distribution (i.e., the time mean value) is V N K 1 VK TK / N K 1 TK (6) The extraction of the time mean value is a most efficient process using the PWM-CTA. It is instructive to compare this data processing to that of a digitized analog signal. Specifically, if the analog velocity signal were digitized at the same rate as the PWM-CTA samples were made, then the random deviations between the discrete sample and the short time temporal average (per the PWM-CTA) would have to be averaged to zero in order for the sampled analog values to represent the true time-mean value. The higher moments: =2,3,4, then follow as described in the next sub-section. The Higher Moments It can be expected that the true velocity, V(t), at the sensor will not be constant during the <T K> sample period albeit the magnitude of the velocity differences will approach zero as T0. Let [V(t)] be defined as [V(t)]K=V(t) - <VK>. (7) One can then show that, for the second moment of the fluctuations: ~ 1 V 2 lim T T T 0 (V(t) V) 2 dt , (8) the PWM-CTA will provide (for each time segment <TK>): [VK>+[V(t)]K V ]2=<VK>2+2[V(t)]K<VK> 2 +[V(t) ] K 2(VK+[V(t)]K) V + V 2. (9) This quantity can be integrated with respect to time for the period tK-1+K-1tK+K to obtain <[<VK>+[V(t)]K V ]2> = <VK>22<VK> V + V 2+<[V(t)] 2K > =[<VK> V ]2+<[V(t)]2> (10) where the first power: [V(t)]K, terms integrate to zero. Hence, the second moment of the velocity distribution would be: ~ V2 N [ VK V]2 TK / K 1 N K 1 N TK K 1 [ V(t)] 2 TK / N K 1 TK (11) It is evident that the PWM-CTA will provide too low a value of the second moment given that only the first term on the rhs can be evaluated using the data available from the PWM-CTA. However, if the sample rate is sufficiently large, the second RHS term can be made sufficiently small. The following strategy can be recommended to assess the influence of the [V(t) ]2K contributions. It is made feasible by post-processing the calibration and the flow field values as shown below. This strategy relies upon the observation that the error will be zero in the limit as T0. With reference to Fig. 2, the basic transfer function could also be established for the time period. tK-1+K-1tK+1+K+1. This will be referred to as the (2T) series. Similarly, the transfer function could be defined over the period: tK-1+K-1tK+2+K+2. This will be identified as the 3T series. It can be noted that these additional calibrations do not need to be acquired. Specifically, since the calibration velocity is constant, <T K>=T (for all K) and the /<T> values for the 2T and the 3T series are the same as /<T> for the 1T series. The original calibration data will then serve for the coefficients: A,B,n for each series. Next, the flow field data can be processed to yield ~ V 2 for series n=1 ~ V 2 for series n=2 ~ V 2 for series n=3 ~ V 2 (1) ~ V 2 (2) ~ V 2 (3) and the function: ~ V 2 (n)=a+bn+cn2 (12) ~ can be formed using the three known values: V 2 (1), ~ ~ V 2 (2), V 2 (3). ~ The desired V (n=0) value can be evaluated as ~ V ~ ~ V 2 (0)= V 2 (1) (1) n n 1 (13) where V b 2 c n 1 (14) ~ ~ The magnitude of the difference: [V 2 (0) V 2 (1)], will provide an evaluation of the “confidence level” in the measured second moment. Similar algebraic considerations will show the impact on the third and fourth moments. These details are not provided here. The Recovery of a Regular (Registered To T And Not <T>) Time Series Output The preceding sections have emphasized that an output value, from each sensor, is available with an associated time increment: [<T K>]j for each sensor: j. Clearly, if one wishes to process data from an x-array (j=1,2), a 3-sensor probe (j=1,2,3), etc., there must be a technique to temporally align these signals. This would also be true for single sensors that are to be processed for cross-correlations, etc. It will also be beneficial for spectral analyses to have access to a regularized time series. A regularized time series can be readily obtained in the post-processing phase of the data reduction for a single sensor as described in the next sub-section. Processing for a multi-sensor probe is discussed in the second sub-section. A Regularized Output – Single Sensor It is clear that a regularized output will be created at a frequency of (1/T). Centering the regularized output with respect to tK is, however, an arbitrary selection. Given that the values are to be limited to the range: 0.25/T0.5, it is suggested that /T=0.375 be selected as the starting point for the regularized output. With this condition, /T=0.875 will be the center point (or the designated time) for the output signal. Figure 4 schematically shows four scenarios for the possible values that will dictate the “regularized output” for the corresponding period: tK1+0.375TtK+0.375T. Condition “a”: K-10.375T, is mutually exclusive with respect to “b” for which K-1>0.375T. Conditions “a” and “b” can be paired with condition “c”: K0.375T or “d”: K>0.375T. The symbol: <VK>, represents, as before, the average velocity in the <TK> time period. This value, and its adjacent values, are shared for the regularized output in the time period tK1+3T/8<ttK+3T/8. The following conditions show the time-weighted velocity values that will contribute to the regularized velocity-time product. This net product will represent the data for tK1+3T/8tK+3T/8. if “a”: if “b”: if “c”: if “d”: <VK>5T/8 <VK-1>(K-1-3T/8) + <VK>(T-K-1) <VK>K + <VK+1>(3T/8-K) <VK>3T/8 The regularized value for the Kth time period is, then: =if “a” or if “b” plus if “c” or if “d”. involves the equivalent of plotting the 13 distinct Q values that correspond to the measured (or E2) values from the two sensors and identifying their intersection point. For the conventional anemometer, the sampled voltages are “instantaneous” and the Q and values are similarly defined for this very short time segment. The need to register the two PWM-CTA channels to each other presents a different issue. Namely, the 1 and 2 samples must be co-registered and then the Q, values can be evaluated. For this, the Q value corresponding to =0 will be used for the 1 and the 2 time series. This will then provide K-1, K}1 <QK(=0)>}1 (19) and K-1, K}2 <QK(=0)>}2 (15) (20) where these are regularized to a center point of The regularized velocity value will be designated as: VK=/T. An Algorithm for an X-Array Processing the values (1, 2) from an x-array (or other configurations for which the cannot be directly related to a velocity magnitude) requires a computing algorithm that includes, but goes beyond, the regularization noted above. One such algorithm (attributed to Scott C. Morris in terms of processing C-CTA signals) is provided here. Consider that each K/<TK> value of an x-array represents a combination of velocity magnitude (Q, in the plane of the x) and the pitch angle with respect to the probe shaft. The latter is designated as . Namely ( K )1 f1 (Q, ) TK 1 7 t K 8 T (16) The above described calibration data can be further processed to create a two-equation, two-unknown solution as: Q( ) Q K ( ) [ Q K (0) ]1 Q( 0) Q K (0) g 1 ( ) (21) Q( ) Q K ( ) [ Q K (0).] 2 Q( 0) 2 Q K (0) g 2 ( ) (22) and (17) for which there will be a unique and <QK>, <K> ( K ) 2 f 2 (Q, ) TK 2 (18) Consider a family of (Q) values for the discrete values (-36, -30, -24…0…24, 30, 36) of the full calibration data set. The Morris strategy (developed for the equivalent E 2Bridge (Q) for discrete values) combination. One can think of the calculation for (Q,) as identifying the intersection point of two curves in the (Q,) plane. If sensor 1 is oriented such that =+45o is perpendicular to the sensor and if =45o is perpendicular to sensor 2, then g1() is a monotonically decreasing function g2() is a monotonically increasing function and, plotting the curves (3.17) and (3.18) will lead to a unique intersection point. Note that the smoothed calibration data: /T=f[Q;], permit arbitrarily fine steps in Q(0) to define the respective “f” functions. For example h[;Q(0)]=a+b()+c(2)+… where a=a[Q(0)]. b=b[Q(0)], etc., and the a,b,c…coefficients are obtained from the (for example) 13 conditions: 0, 6o, 12o…36o. SUMMARY The PWM-CTA provides an alternative to the conventional (C)-CTA which has well served the experimental fluid dynamics community since the early 1960’s. The PWM-CTA offers a more sensitive transfer function (nominally a second vs. a fourth power response), a digital output, ease of simultaneous measurements and (not discussed above) relatively low cost. Its intrinsic attribute of providing a short (e.g., 20 microsec) time average leads to a rapid convergence to the true time average value (velocity or shear stress) and to higher moments that will (slightly) lower than the true values. The magnitude of the latter effects can be assessed using the strategy shown herein. The second intrinsic attribute of the PWM-CTA, to deliver an irregular time series, can be addressed using the schemes noted above. These will be addressed by representative examples in the August 2001 presentation and this additional information can be obtained from the author at <foss@egr.msu.edu> or by telephone at 517-355-3337. REFERENCES 1). Perry, A.E., 1982, “Hot-wire Anemometry,” Oxford University Press, New York. 2). Comte-Bellot, G., 1976, “Hot-Wire Anemometry”, Annu. Rev. Fluid Mech., vol. 8, pp. 209-231. 3). Fingerson, L.M. and Freymuth, P., 1996, “Thermal Anemometers”, Fluid Mechanics Measurements 2nd Edition, R.J. Goldstein, Ed., pp. 115-173. 4). Foss, J.F., Bohl, D.G. and Hicks, T.J. 1996, "The Pulse Width Modulated-Constant Temperature Anemometer," Meas. Sci. and Tech., vol. 7, pp. 13881395. 5). Hicks, T.J., Foss, J.F. and Bohl, D.G., 1997, "Pulse Width Modulated Constant Temperature Anemometer," U.S. Patent No. 5,654,507. 6). Foss, J.F., Hicks, T.J. and Morris, S.C., 2001, “Operating Experience With The Pulse Width Modulated-Constant Temperature Anemometer”, Sixth International Thermal Anemometry Symposium, Melbourne, Australia. MUX EA 16 bit out To other channels From other channels Terminator Board R1 R1 EB' + EB RS x11 (x6 optional) Comp FlipFlop R _ Q x10,000 S EREF Clock (163.84 MHz) Count = 0 4096 Figure 1: PWM-CTA Schematic Note: R1=50 for a typical (RC=3.5) hot-wire sensor RS(t) Figure 2: Timing Diagram for RS(t) 16bit Counter R Common Unit Figure 3: A representative /T=f(Q) calibration for a single sensor. RS(t) t a c b d tj t j -1 { * <Vj - 1> <Vj> <Vj -1> a <Vj> b <Vj> <Vj + 1> <Vj> c <Vj + 1> d * Shows the applicable velocity for the indicated values Figure 4: A representation of the algorithm which creates a regularized <V(t j)> time series.