REVISION RECORD. Revision Description of change Prepared by: Eff. Date. 01 Initial Release Resonance Sub- 10/21/2003

Transcription

1 PAGE: 1 OF 44 REVISION RECORD Revision Description of change Prepared by: Eff. Date 01 Initial Release Resonance Sub- 10/21/2003 Committee 02 Modifications to Correlation Procedure 11/25/ Notch center def. integration 01/13/ Format changes to NC section, updated table of contents pages. 01/21/2004

2 PAGE: 2 OF 44 TABLE OF CONTENTS DESCRIPTION PAGE 1.0 Objective Scope Reference Documents Definitions General Definitions Shaker Excitation Windage Excitation Test Equipment Default Test Requirements 21 Appendix 23

3 PAGE: 3 OF OBJECTIVE: Establish standard definitions and requirements for the following HGA Resonance performance metrics. Frequency Response Functions (FRF) Windage Power Spectal Density (PSD) 2.0 DOCUMENT SCOPE: The follow document covers standardized definition information regarding HGA Level components for the purposes of evaluating resonance performance. 3.0 REFERENCE DOCUMENTS: IDEMA STANDARDS: T2-91, Rigid Disk Drive Head and Media Glossary

4 PAGE: 4 OF DEFINITIONS 4.1 General HGA: Head Gimbal Assembly Offset Height: The distance from HGA clamp surface to the slider mounting surface on suspension, when part is held at Z-height. (Does not include slider thickness.) Z-Height: The distance from the zero reference surface (bottom of the mount block / clamping surface) to the disk surface that the slider is flying on. (Typically set to midpoint of disc runout.) Parallelism: The angular orientation of the mount block clamp surface to the disk surface the slider is flying on. (Pitch & Roll) Skew Angle: See Diagram Pivot Point: See Diagram Figure Disk Rotation Skew Ansle Clamping Surface: The mounting surface area of the suspension assembly as referenced on the product print Clamping Reference: The datum surface edge geometry closest to the loadpoint of the suspension as referenced on the product print. Standard square base suspensions

5 PAGE: 5 OF 44 reference the distance from the center of the boss ID to the front edge of the clamping surface. (Other variations may exist, such as a radius clamping reference.) Clamping Force: The force applied to clamping surface to hold HGA to mount block for testing. Typically provided by specially design mounting screw and mesured in torque (inoz) Runout: The amount of axial deviation (peak to peak) from a true path of a rotating object during a single revolution Notch Curve: A relationship describing the interaction between a HGA s dynamics and Z height Background information on Z-height and resonance parameter (e.g. 1 st torsion) FRF gain. The U or W Notch Curve portrays the relationship between the resonance mode of interest to the Z-Height test condition. The shape of the U or W Notch Curve is dependent on the Z-Height sensitivity to the specific resonance mode. Modes without significant Z-Height sensitivity (i.e. sway) will typically produce trend graphs with no local minimum or maximum, but not produce significant notch graphs Definition of U or W Notch Curve: This definition will focus on the 1 st torsion mode (T1). The graph of 1st torsion FRF gain (db), as the Z-Height is varied, is referred to as the U or W Notch Curve (see Figure ). Depending on the HGA design and Z- Height test resolution, the shape of the curve can resemble either a U or W. For windage, the notch curve units will be displacement/sqrt(hz). In the case of the U Notch Curve, the notch center is the Z-Height at the minimum gain value. In the case of the W Notch Curve, the notch center is calculated and located using only the local center data points (per Appendix 7.2.2). Establishing the notch center can be accomplished several ways using analytical methods, but the chosen method must be very repeatable. One method is to curve fit the data using local extreme data, being careful that noise variations do not confound the data. Due to different excitation boundary conditions, the windage notch center could be at a slightly different Z-Height then the T1 gain notch center.

6 PAGE: 6 OF 44 Figure Notch Examples T1 gain (db) W-notch U-notch skewed U-notch Z-height (mm) Drive Performance related to U or W Notch Curve: Ideally, the notch center of the curve coincides with the Z-Height design target. In Figure , the arrow indicates the selected notch center for the U Notch Curve. The slope of the side legs, or the sensitivity, may be different for each leg (refer to Skewed U Notch Curve in Figure ). Low slope relates to low sensitivity and physically means tolerances in the drive have low impact on T1 gain. The purpose for selecting a HGA component notch center is to establish a reference point for correlation evaluation between different groups, process control, and relating component data to drive level data to optimize the Hard Disk Drive. The optimum Z-Height for drive level may be a compromise for several problem modes and quite different than the component T1 gain notch center. For example T1 and T2 notch centers may not be aligned with each other and therefore track misregistration tradeoffs shift the optimum Z-Height away from the T1 notch center.

7 PAGE: 7 OF Shaker Excitation (Base Excitation) Description: Part under test is attached to a rigid mounting block, which is coupled to and driven by a translational shaker Input or Excitation: Off-track motion (primarily in the radial direction) generated by the shaker and measured as displacement, velocity, or acceleration on or near the mounting block, and ideally representing the off-track motion at the base plate. Typically accelerometers or Laser Doppler Vibrometers are used to measure the input in the form of acceleration or velocity, respectively. In particular, note that the desired input is not force applied to the mounting block or to the part under test Output or Response: Off-track motion of a point on the edge of the slider near the trailing edge measured as displacement, velocity, or acceleration. Typically a LDV is used to measure slider response in the form of velocity, but other sensor types are theoretically possible Measured Frequency Response Function (FRF): Fourier transform of the output signal divided by the Fourier transform of the input signal. In symbols this can be expressed as Measured FRF(jω) = F { Output Signal } / F { Input Signal } where F { y(t) } is the Fourier transform of any signal y(t). These Fourier transforms will be calculated discretely most commonly using the Fast Fourier Transform (FFT) algorithm. As an example, consider a case where the input and output signals are in the form of acceleration and velocity, respectively. Under these circumstances a typical Measured FRF is represented by the following.

8 PAGE: 8 OF Gain of Measured FRF Amplitude (db) Frequency (Hz) x 10 4 Normalization of the Measured FRF is required to account for the signal types used and the sensitivities of the input and output sensors Normalized FRF: Normalized FRF represents the frequency response function between slider off-track response to prescribed base plate off-track motion with both measured in the same units. Normalized FRF(jω) = F { Slider off-track displacement } / F { Base plate off-track displacement } Depending on the specific form of the input and output signals (e.g. proportional to either displacement, velocity, or acceleration), it will be necessary to multiply the Measured FRF by the factor (jω) n, where j is the 1, ω is the frequency in rad/s, and the integer, n, is given in the table below. In terms of transfer functions based on Laplace transforms another interpretation is that poles or zeros at the origin are factored out. Input Output (Displacement) Output (Velocity) Output (Acceleration) Displacement n = 0 n = -1 n = -2 Velocity n = +1 n = 0 n = -1 Acceleration n = +2 n = +1 n = 0 Once this normalization is performed, the low frequency asymptote of the FRF will have a slope of zero db per decade, where the db value for any real positive amplitude ratio, A, can be expressed as db(a) = 20 log 10 (A)

9 PAGE: 9 OF 44 Conversion between amplitude or gain reported in db to linear units for a range of db values is illustrated in the following table. db(a) A After this normalization step is performed, the FRF appears as shown below. 120 Gain of Intermediate FRF 100 Amplitude (db) If the input displacement and output displacement are expressed in the same units, then in the limit as the frequency approaches zero the amplitude of the FRF needs to be unity when represented in linear units or equivalently 0 db when expressed logarithmically. Accordingly, a further normalization to the FRF is required to account for the calibration factors of the input and output sensors. This rescaling may be accomplished by dividing the FRF in linear units by its gain in the same units at very low frequency, or equivalently by subtracting from the FRF expressed in db units, the db level at very low frequency. Following this procedure,

10 PAGE: 10 OF 44 the measurement of the FRF is self-calibrating in that it is not necessary to calibrate individually either the input or output sensors. Depending on the relative polarities of the input and output signals, it may be necessary to mathematically reverse the polarity of one signal to achieve a phase angle of zero as the frequency approaches zero. This can be accomplished by multiplying one signal by negative one or equivalently by adding an appropriate integer multiple of 180 degrees to the phase curve to make the low frequency asymptote approach zero degrees. Physical interpretation is that at sufficiently low frequencies, the entire HGA behave as a rigid body with the base plate and slider moving in phase the same amount. As an example of interpreting the amplitude of the FRF, consider a resonant peak with a measured amplitude of 20 db. This represents a gain of 10 for any consistent units. In other words, drive the base plate 1 µin at this resonant frequency and the slider will move 10 µin. Similarly if the base plate is driven 1 nm at this frequency, then the slider will move 10 nm. Amplitude and phase plots for a typical Normalized FRF are depicted in the figures below: Gain of Normalized FRF Amplitude (db)

11 PAGE: 11 OF Phase of Normalized FRF 50 0 Phase (deg) Frequency (Hz) x 10 4 Although the formula for the Measured FRF is shown in a particularly simple form, there are certain practical matters to consider which are covered in many excellent texts on digital signal processing. One precaution which must be undertaken is to use appropriate low-pass analog filtering prior to any analog to digital conversion (ADC) to prevent a phenomena known as aliasing. Excitation level must be chosen and specified so as not to overdrive any system components, while at the same time providing acceptable signal to noise ratio. When using averaging techniques to reduce the effects of measurement noise, there are benefits to calculating the measured FRF using a ratio of the cross power spectral density between the output and input divided by the input power spectral density. In addition, windowing of various forms may be applied to the input and output time domain data to reduce an effect known as leakage. Lastly, band selectable Fourier analysis (zoom transform) may be used in conjunction with stitching techniques to achieve any desired frequency resolution. In any event normalization of the resulting FRF will proceed in the same manner as described above.

12 PAGE: 12 OF Choice of excitation signal F(t): For accurate system identification, there needs to be persistent excitation in the desired spectral bands with adequate energy. For linear, timeinvariant systems (like HGA s) the choice is usually between band-limited white noise and chirp signal. Strength of excitation should not influence the results as long as the system remains linear. However, measurement noise (such as windage) or channel range limitations might warrant tailoring the excitation signal in different spectral bands Controlling resolution bias error: Insufficient spectral resolution can cause bias error while estimating the true gains at modal peaks (see the figure below). Increasing the frequency resolution can be achieved by using a process called zooming. Zooming can be achieved through decreasing the sampling frequency (simultaneously frequency-shifting the zoom range) or increasing the time record (window) length. For a post-processing (non-real time) application like HGA resonance testing, the latter method might be preferred. Resolution Bias Error should be near the noise level of the system. Resolution Bias Error Gain Resolution Bias Error f Frequency Note on Coherence: Coherence determines how much of the measured output signal is linearly related to the measured input signal. It is used to check the quality of the measured FRF. It s a function of frequency and has a value between 0 and 1. It is defined as 2 2 [ Cross Power] γ ( f ) [ Input Power] [ Output Power] = G FF G FX ( f ) ( f ) G 2 XX ( f )

13 PAGE: 13 OF 44 A reason for low coherence in HGA resonance measurements is the presence of noise within the system. Noise W(f) F(f) (Base motion) H(f) (HGA) X(f) (Slider motion) X ( f ) By definition, Frequency Response Function H ( f ) where X(f) and F(f) are F( f ) the Fourier Transforms of the slider and base motions respectively. An estimate of H(f) is given by Hˆ ( f ) = Cross Spectrum Input Autospectrum G = G ( f ) = ( f ) when the output is contaminated by noise (such as windage in our case). F is the complex conjugate of F and N is the # of averages. FX FF 1 N 1 N ( F ( F X ) F) Gain & Frequency Values from Bode Plot: Gain and Frequency values of a given mode correspond to the maximum amplitude point on the gain axis. Three examples below, are representative of mode shapes found on standard bode plots. Mode Peak Example Mode Gain Gain Mode Frequency Frequency

14 PAGE: 14 OF 44 Mode Peak Example Mode Gain Gain Mode Frequency Frequency Mode Peak Example Mode Gain 20 Gain Mode Frequency Frequency

15 PAGE: 15 OF Windage Excitation Description: Part under test is attached to a rigid mounting block or specifed arm Driving Excitation: Part under test is driven by airflow created by the spinning disc or discs in the case of a dual disc test Response Variable: Velocity for example in the off-track direction measured with an LDV of a point on the edge of the slider near the trailing edge Windage Plot: Power Spectral Density of windage-induced slider displacement measured in displacement/ Hz plotted against frequency in a specified bandwidth. The displacement is usually in peak units. A typical windage plot is shown below PSDrms: Area under the Windage Plot in rms units is called PSDrms which is also equal to the Standard Deviation (1σ) of windage-induced displacement. It is expressed in displacement units. To obtain this, sum of the squares PSD values at every frequency is multiplied by the frequency resolution ( f) of the plot. Square root of this product gives PSDrms in the corresponding BW. Also note that if PSD values are given in peak units in the Windage Plot, they have to be divided by 2 (to get the rms values) before squaring and summing up.

16 PAGE: 16 OF Windage Signal Processing: Windage signal is largely non-periodic, stationary and Gaussian. Time averaging of the signal results in a zero mean value. A useful way to quantify windage, from a servo engineer s perspective will be finding the σ (Standard Deviation) from its discrete time samples (see figure below). +3σ Displacement -3σ time xi µ xi Now i = 1 i= 1 σ = = N N because µ = (1) The right hand side of (1) works out to be the same as the RMS- linear- average of the time series. In other words, (By definition, N 2 ( ) ( ) N 2 σ = x rms (2) x rms = x 2 = t0 1 2 T t0 T x ( t) dt gives the RMS linear average for a continuous signal. ) It is easy to calculate σ of any time series using equation (1). But it will not give a break up of contributions from various modes of the HGA. Fortunately, Parseval s theorem will help compute the RMS power from frequency domain. Parseval s theorem states that power computed in time domain is equal to the power computed in frequency domain. This can be expressed as follows: x rms ( = (3) x ( t) dt = x f ) df

17 PAGE: 17 OF 44 The right hand side of this equation may be computed just by summing up squares of the Fourier coefficients of x(t) and dividing it by 2, for a single-sided spectrum. I.e., σ = N 1 x 2 2 n (4) FFT and Windowing: FFT is an implementation of DFT (Discrete Fourier Transform) calculation. Key parameters involved are: N: # of samples; T: duration of data acquisition; t: sampling interval; f: frequency interval; Fs: sampling frequency; The relationships connecting these parameters are T = N t ; t = 1/Fs ; Fs = N f ; f T = 1. Which means that for better frequency resolution, one must either reduce Fs or increase N. Two major pitfalls of DFT and their solutions are listed below: Pitfall Cause Solution Aliasing sampling Use antialiasing filter Leakage Time limitation Use correct window; decrease f Any signal in the neighborhood of the sampling frequency F s (falling in a region defined by F s ± F a ) would appear as a ghost signal of frequency f a, which would alias with the real signals of that frequency. In other words, a chosen sampling rate f s is good only for signals lower than F s /2 and an anti-aliasing filter should be used to attenuate all the signals of frequency higher than F s /2. Spectral leakage is seen to happen, if there is a truncation in the time-domain for the recorded event. For example, if an event outlasts the recording window, FFT of the truncated event will have spurious spectral components at frequencies higher than the true composition of the signal. Similarly, inadequate frequency resolution can result in leakage (estimation errors) of amplitudes of fine spectral features. As mentioned earlier, windage phenomenon can be treated as a broadband, stationary, random process. For a specified fixed window length and sampling rate appropriate for the application at hand, FFT spectrum of windage does not depend upon the instant sampling is initiated. This means that there is no need to apply non-uniform windowing techniques on the signals, which is usually done to limit the spectral leakage of periodic, narrowband signals.

18 PAGE: 18 OF 44 However, windage is a mix of periodic and non-periodic signals. The periodic signals come from structural modes of the HGA. So it might be argued that an arbitrarily chosen, timewindow with uniform weighting, would introduce spectral leakage for the periodic part of signals. If the user prefers to apply a non-uniform window (to control this leakage), Hanning or Kaiser-Bessel functions are recommended. The latter method is known to resolve closely spaced modes better than Hanning. If applying a non-uniform window (on time domain data before performing FFT on them), care should be taken to apply a correction factor to the resulting power spectra to recover the actual heights of the peaks. Multiply the PSD values by a scalar given by, ( norm( w) ) 2 ( sum( w) ) 2 = ( w( n) ) 2 ( w( n) ) 2 (where w(n) is a vector representing the non-uniform window function), to get the true PSD values. It is also possible to apply overlaps between the windows to improve data weighting. For example, a 75% overlap between Hanning windows results in equal data weighing. Large overlaps, at the same time, will also reduce the effective BW of the system for real-time signal processing Spike removal: Windage PSD spectra usually contain spikes (plotted in red in the figure below, as an example) that occur at multiples of the spindle frequency. Magnitude of these spikes may not stabilize even after multiple spectral averaging. If one of these spikes rides over an HGA mode, it adds a random noise component to the modal peak. These spikes also contribute significantly to the PSDrms of the spectrum. PSD Frequency Removal of these peaks can be achieved while post-processing the PSD spectra. It involves determining the location of each peak and replacing it by a value representing the values and /or trends of the adjoining sections on the curve. The blue trace in the above

19 PAGE: 19 OF 44 figure illustrates a typical PSD curve with spikes removed. Median filtering is a simpler method capable of smoothening the spikes, but it also alters the modal gains. A median filter moves down a data sequence, replacing every element with the median of its immediate data neighborhood. Size of the data neighborhood for median processing is determined by rank of the median-filter. With an appropriately chosen rank, spikes in the windage spectra get rejected in any finite-length sequence formed by adjacent data points. This is because of the huge difference in magnitudes between the spikes and adjoining valleys (normal data) in windage spectra that force the local median into the range of nominal values Background noise reduction: Background noise dictates the resolution of LDV decoder. Noise limited resolution of a good velocity decoder is typically in the range of 0.2 µm/s/ Hz. It is the velocity equivalent of a voltage sine wave that has the same rms power as that of noise reflected from a stationary target of good reflectivity calculated at 1 Hz spectral bandwidth. It is advisable to check the noise spectra from time to time using a stationary slider target. There is no need to subtract it from the total windage power as long as the noise levels are acceptable.

20 PAGE: 20 OF Test Equipment: Block Diagram of test equipment Typical FRF System Typical Windage System

21 PAGE: 21 OF Default Test Requirements 5.1 The part should be clamped on the clamping surface, as indicated on the product print. 5.2 Use Clamping Force as indicated on product print. 5.3 Clamping cannot cause permanent distortion to part under test. 5.4 Required boundary condition (RPM, Radius, & Skew Angle) such that part being tested establishes a stable air bearing under the slider. 5.5 Off-track motion to be measured at trailing edge of the slider For LDV Devices laser Spot diameter to be less than 75% of the slider thickness. Spot to be centered a distance of approximately one spot diameter from trailing edge of slider. 5.6 Disc/Platter runout less than mm peak-to-peak. 5.7 Skew Angle absolute value less than For Shaker excitation, target Skew Angle is zero degrees at the centroid of the slider For Windage excitation, target Skew Angle is dependent on test condition desired. 5.8 Input Excitation signal: For Shaker excitation, use band-limited white-noise signal centered in the frequency range of interest. Setup signal to maximize signal-to-noise ratio and coherence function without overdriving system components For Windage excitation, not applicable. 5.9 Tooling Requirements Clamping Surface Clamping Force Parameter Specification Note Profile mm Corners defined as no break and to a depth of mm Flatness mm Surface Finish 203 nm Average Roughness Parallelism mm 35mm plane for Pitch & 15mm plane for Roll Parameter Specification Note Force 26.7 ± 2.2 Newton Swage Attach

22 PAGE: 22 OF HGA Position Reference Diagrams Parameter Specification Note Position mm HGA primary to mount block Angle 0 ± 2 Rotation about block Height Nominal ± mm Z Height

23 PAGE: 23 OF 44 APPENDIX TABLE OF CONTENTS DESCRIPTION PAGE A Gauge Repeatability & Reproducibility 24 B Gauge Correlation 27 C Windage Signal Processing with LDV 37 D Tooling Standards 40

24 PAGE: 24 OF 44 A: APPENDIX: Measurement Gauge Repeatability & Reproducibility DESCRIPTION TABLE OF CONTENTS 1.0 Objective 2.0 Scope 3.0 Reference Documents 4.0 Definitions 5.0 Procedure 6.0 Analysis 7.0 Appendix 1.0 OBJECTIVE: This instruction is established as a guide for performing GR&R studies for HGA Resonance equipment. 2.0 SCOPE: The scope of this procedure pertains to HGA Resonance measurement equipment. 3.0 REFERENCE DOCUMENT: N/A 4.0 DEFINITION / TERMINOLOGY Standard Deviation = x is the mean of the samples. n is the total number of samples. x is the individual sample. The summation is taken from 1 to n USL = Upper Specification Limit LSL = Lower Specification Limit

25 PAGE: 25 OF Tolerance Range = USL LSL for double sided specifications For single sided specifications: Tolerance Range = 2 * (USL Target) or 2 * (Target LSL) P/T Ratio = (5.15 * Gauge Standard Deviation) (Tolerance Range) % R&R = (Gauge Standard Deviation) 2 (Process Standard Deviation) PROCEDURE 5.1 GR&R Measurements and Data Recording Ensure measurement gauge has been calibrated with qualified calibration tooling prior to performing any measurements Select and label at least 10 samples that represent the normal process variation. [It is recommended to start with 15 samples to account for potential sample damage during measurements.] Identify two or three operators and refer to the operators as A, B, and C. (Using operators who will normally perform measurements with the gauge being evaluated.) Arrange the samples in random order and have operator A measure the samples. Record operator A s results Remove samples from measurement equipment & tooling between readings Have operator B measure the same samples without seeing operator A s results. Record operator B s results If using three operators, have operator C measure the same samples without seeing operator A or B s results. Record operator C s results Repeat steps through 5.1.6, using a different random order for measuring the samples. (Repeat cycle so each operator measures the samples three times.) See GR&R data record sheet, section 7.0 Note: If desired, the cycle listed in steps to can be performed independently for each operator. This allows the GR&R procedure to be performed when all operators are not present at the same time. 6.0 GR&R Analysis 6.1 Enter GR&R data into statistical software.

26 PAGE: 26 OF Analyze GR&R Compare Measurement Gauge to the following criteria. % R&R P/T Ratio Bad > 7.7% > 30% Acceptable 2 7.7% % Good 0-2% 0-10 % If the gauge under investigation does not pass above criteria, the responsible person is required to take appropriate corrective action. Note: The above rules are used for guidance. Some gauges may have approved deviation from above criteria. In such a cases, the acceptance criteria will be documented in the specific Gauge Management Plan. 7.0 Data Record Sheet GR&R Data Record Sheet Example Operator A Operator B Operator C Sample Meas 1 Meas 2 Meas 3 Meas 1 Meas 2 Meas 3 Meas 1 Meas 2 Meas

27 PAGE: 27 OF 44 B: APPENDIX: Gauge Correlation DESCRIPTION TABLE OF CONTENTS 1.0 Objective 2.0 Scope 3.0 Reference Documents 4.0 Definitions 5.0 Procedure 6.0 Analysis 7.0 Attachment & Examples 1.0 OBJECTIVE: Established a guide to perform measurement correlation between two or more gauges. 2.0 SCOPE: The scope of this procedure pertains to HGA Resonance measurement equipment. This document identifies the guidelines for sample preparation, correlation flow, and statistical analysis for measurement correlation. 3.0 REFERENCE DOCUMENT: GR&R Procedure

28 PAGE: 28 OF DEFINITION / TERMINOLOGY 4.1 Correlation: The linear relationship between two random variables. Usually measured by a correlation coefficient (such as Pearson s r ) where the value of the coefficient ranges between 1 to 1. The larger value of r in absolute value, the stronger the linear association between the two variables. 4.2 Correlation Samples: Designated parts for specific study. Can be actual product, tester standards, or artifacts. 4.3 R 2 : Squared value of the Pearson s coefficient, calculated from a regression fit between the gauges. 4.4 Offset: The difference between the observed values from two measurement gauges, calculated using a Paired t-test statistic. 4.5 Slope: Co-efficient of the regression equation between the gauges. The slope value is an indication of the linear sensitivity between the two gauges. The closer the value is to 1, the stronger indication that the two gauge s linear sensitivities are equivalent. (Very dependent on the spread of data and the sample size.) 5.0 PROCEDURE 5.1 Obtain correlation samples for study. 5.2 Ensure measurement gauge has been calibrated with qualified calibration tooling prior to performing any measurements. 5.3 Ensure gauge has been setup per product specific settings requirements. (See Appendix section 7.3) 5.4 GR&R should be completed and acceptable prior to performing the correlation exercise. In the event that an acceptable GR&R is not achieved, the following correlation procedure is still recommended for any resonance correlation exercise. 5.5 Select a quantity of parts to achieve at least 30 data points that spread across 80% to 120% of the specification range, for the correlation. 5.6 Measure all resonance parameters of interest on each part For parameters which are z-height sensitive, the study can include seven to ten HGAs measured at several different z-heights. (This method simulates the process of creating a larger distribution to assist with correlation analysis.) It is recommended to take steps of ~0.05 mm for the various z-heights targets. The z-height values should be centered about the targeted z-height listed on the print. 5.7 Record the measurement readings for each individual part at each tested z-height. (Example table for data recording in Appendix section 7.1.)

29 PAGE: 29 OF Upon completion of the above steps, send the correlation samples along with tester settings record and data to location of 2 nd gauge. 5.9 Repeat steps of 5.6 through 5.7 and record data for 2 nd gauge The entire procedure may be repeated with the original gauge to verify samples were not damaged during the test. 6.0 ANALYSIS 6.1 Analyze the correlation results using appropriate statistical software. General guideline for correlation acceptance criteria defined below. Statistical comparisons should be performed at a 95% confidence level. Parameter Criteria Offset Absolute value of Offset < 10 % Tol. Slope 1.0 +/-.10 R 2 > 85% Gauge Variance Test Variance Ratio of Gauges = 1.0 +/-.10 Notch Center Offset < mm 6.2 For the mechanical correlation of the equipment, use the Notch Center Offset criteria defined in table listed under section The intent of the Notch Center correlation is to validate the mechanical calibration and setup of the equipment used to perform the correlation measurements Determine the resonance parameter to use for Notch Center Offset analysis; typically this is 1 st Torsion gain. The parameter must have a significant Z-height sensitivity. (For the remainder of this procedure, we will assume the use of 1 st Torsion gain.) Calculate the average gain value for each Z-height, across all parts measured specific to the gauge Plot the Z-height versus Gain chart for both gauges and determine the Notch Center point for each gauge. (This would be the lowest gain point for a U-notch or the highest gain point between the two minimums for a W-notch. See section 7.2.1) Determine the corresponding Z-height for the Notch Center location obtained from both plots. (One method is to derive the equation of the plot, set the first derivative of the equation to zero, and solve for Z-height. See section 7.2.2) Calculate the difference in the two Z-heights and compare to the Notch Center Offset criteria defined in the table listed under Section 6.1. (See section 7.2.3)

30 PAGE: 30 OF For parameters with Z-height sensitivity; use Offset, Slope, R 2, and Gauge Variance test criteria defined in the table listed under section Align measured data by Z-height for each gauge. (As seen in the example data table.) Perform correlation analysis between gauges for parameter of interest and compare to criteria Offset obtained from a Paired t-test statistic comparison of the two gauges. (See section 7.4.2) Slope and R 2 obtained from a linear fit to the data between gauges. (See section 7.4.1) Gauge variance comparison performed using a Test for Equal Variances. (See section 7.4.3) 6.4 For parameters without Z-height sensitivity, use the Offset and Gauge Variance Test criteria defined in the table listed under section Perform correlation analysis between gauges for parameter of interest and compare to criteria Offset obtained from a Paired-T test statistic comparison of the two gauges. (See section 7.4.2) Gauge variance comparison performed using a Test for Equal Variances. (See section 7.4.3)

31 PAGE: 31 OF Attachments & Examples 7.1 Measured Data Record Sheet Correlation Data Record Sheet Example GAGE CORRELATION BETWEEN PARAMETER P/N AND SPEC TOTAL TOLERANCE GAGE: OPERATOR: DATE: GAUGE 1 GAUGE 2 GAGE: OPERATOR: DATE: Mode 1 Mode 1 Mode 2 Mode 2 Mode # Mode # Mode 1 Mode 1 Mode 2 Mode 2 Mode # Mode # Part # Z-Height Freq Gain Freq Gain Freq Gain Part # Z-Height Freq Gain Freq Gain Freq Gain

32 PAGE: 32 OF Notch Center Z-height & Z-height Offset Example: U-Notch & W-Notch Examples: Gain (db) U-Notch Graph for 1st Torsion Z-Height (inch) Gain (db) W-Notch Graph for 1st Torsion Z-Height (inch) Establish Notch Center Location from W or U Notch Curve: Relationship Equation: Gain = (Z-height) (Z-height^2) 1st Derivative (set = 0): 0 = (2)(Z-height) Notch Center (Z-height): Z-height = ( )/(2* ) =.026" Gain (db) W-Notch Graph for 1st Torsion y = x x R 2 = Z-Height (inch) Note: Notch Center calculation for a W notch are calculated using only the center data points Notch Center Correlation between two gauges (Gauge 1 & Gauge 2): Relationship Equation: EQ1: Gain = (Z-height) (Z-height^2) EQ2: Gain = (Z-height) (Z-height^2) 1st Derivative (set = 0): EQ1: 0 = (2)(Z-height) EQ2: 0 = (2)(Z-height) Notch Center (Z-height): Z-height = (3467.5)/(2*66683) =.026" Z-height = (3734.3)/(2*66683) =.028" Notch Center Offset: (Max - min) = =.002" Gain (db) Eq 1: W-Notch Graph for 1st Torsion y = x x R 2 = Eq 2: y = x x R 2 = Z-Height (inch)

33 PAGE: 33 OF Tester Settings Record Sheet: Product = Specified by HGA / suspension documentation Standard = IDEMA standard can be set Machine = recorded Hardware Settings Resonance FRF Windage Units Example Comments Spindle Speed Product Product RPM 10k Head Radius Product Product mm (inch) Head Skew Product Standard Degrees Windage standard OD Disk Diameter Product Product mm (inch) Disk Thickness Product mm (inch) Number of Disks Standard Product FRF stand 1 disk Disk Spacing Product mm (inch) HGA Mount Block Tilt Tolerance Standard Standard mrad Standards TBD HGA Mount Block Position Tolerance Standard Standard mm (inch) Standards TBD HGA Mount Block Skew Tolerance Standard Standard mrad Standards TBD Z-height Product Product mm (inch) Include range of interest Part Orientation Standard UP are tested as DOWN Part Population Standard Standard One HGA per arm Both FRF and Windage Standard of 1 HGA Tail ON/OFF Product Product Runout Standard mm (inch) Windage Standard to have Maximum Shroud Standard Windage Standard to be none Motion Sensor Settings Resonance FRF Windage Units Example Comments Input Sensor Location Standard HGA Mount Block (Off-track) Standard HGA Mount Block Input Sensor Type Machine LDV Input Low Pass Filter Machine khz 100 Standard Type Input Low Pass Filter Cutoff Frequency Machine 2 * high frequency Input Sensitivity Machine mm/s/v 25 Input A to D resolution Machine 16 bit Min Input A to D bandwidth Machine 4 * high frequency Min Input signal conditioning Machine Output Sensor Location Standard Standard Slider Trailing Edge (Off-track) Standard Trailing Edge Output Sensor Type Machine Machine LDV Output Low Pass Filter Machine Machine khz 100 Output Low Pass Filter Cutoff Frequency Machine Machine 2 * high frequency Output Sensitivity Machine Machine mm/s/v 25 Output A to D resolution Machine Machine 16 bit Min Output A to D bandwidth Machine Machine 4 * high frequency Min Output signal conditioning Machine Machine Signal Analyzer Settings Resonance FRF Windage Units Example Comments Test Type Freq. Response FFT Number of Frequency Buckets Product Product 1 Start Frequency Product Product khz 2.0 Frequency Span(s) Product Product khz 25.6 Resolution Product Product Lines 1600 Delta F Product Product Hz 16 Source Type Standard Periodic Chirp Periodic Chirp Average Type Standard RMS Number of Averages Product 200 If Sine Sweep, then settle time Windowing Type Standard Hanning Motion Input Transducer Gain Measure Measure EU/V 40 Motion Input Transducer Coupling Measure Measure AC Motion Output Transducer Gain Measure Measure EU/V 40 Motion Output Transducer Coupling Measure Measure AC DSP Calibration Wave Form Standard Standard V 25,000 Hertz 1 volt sine wave Software Settings Resonance FRF Windage Units Example Comments Resonance Magnitude Inspection Method Standard Peak to Mass Line Skaker Mode Cancellation Standard Signal Filtering Standard Median Baseline Noise Subtraction Standard X Axis Display Standard Standard khz Y Axis Display Standard Standard db, or nm (uinch) - RMS/sqrt Hz Power Spectral Density (PSD)

34 PAGE: 34 OF Correlation Example between Two Gauges: R 2 & Slope (Example): The Slope value is pulled from the chart below (Slope = 0.997). Based on the Slope criteria in section 6.1, the Slope result is deemed acceptable The Slope value is the X-coefficient of the regression equation. The slope value shows the relationship of how the two gauges see the same population of parts Resonance Correlation Example (1st Tosrion Gain) y = x R 2 = Gauge Series1 Linear (Series1) Gauge The R 2 value is pulled from the chart above (R 2 = 0.885). Based on the R 2 criteria in section 6.1, the R 2 result is deemed acceptable The R 2 value shows how the correlation data fits the regression equation, which provides insight to the error in the calculated slope value. (For example, an R 2 value of 90% equates to an error of 5% in the slope value.) If the R 2 value fails, then the error in the Slope value will be high. Engineering judgement should be applied prior to using the Slope value with a low R 2 value If the R 2 value passes, then the error in the Slope value will be low.

35 PAGE: 35 OF See Gauge Slope Comparison example chart below. When the Slope value departs from 1, the larger the discrepency between the gauges ability to see the same product specification range. In the chart below, it is seen that this situation will lead to higher misclassification risk If Gauge 1 was defining truth and the specification limit maximum was 16 db, then Gauge 2 could measure parts at 14 db and state acceptable while those same parts would fail on Gauge 1. Defining a Slope criteria will help control this scenario. (Note: This situation may also show up as Gauge Variance Comparison failure, since the variance on the samples will be smaller with Gauge 2.) Gauge 2 Gauge Slope Comparison Gauge 2 Upper = 12 db 16 Lower = 6 db Gauge 1 Upper = 16 db Lower = 4 db Slope Line Gauge Offset (Example): The 95% Confidence Interval for the mean difference includes zero and the P-Value is greater than.05. Therefore, the test statistic shows that the mean difference is not statistically significant. Paired T-Test and CI: Gauge 1, Gauge 2 N Mean StDev SE Mean Gauge Gauge Difference % CI for mean difference: (-1.001, 0.200) T-Test of mean difference = 0 (vs not = 0): T-Value = P-Value = The Offset value is pulled from table above (Offset = db). Based on the Offset criteria in section 6.1 (with a tolerance of 5 db), the Offset result is deemed acceptable.

36 PAGE: 36 OF If the test shows the Offset difference is statistically significant and the Offset value passes the criteria, then determine if the difference is practically significance and make an engineering judgement If the test shows the Offset difference is not statistically significant and the Offset value fails the criteria, then investigate to ensure the sample size is appropriate Gauge Variance Comparison (Example): The 95% Confidence Interval for the sigmas from each gauge overlap and the P- Value is greater than.05. Therefore, test statistic shows that any difference in the variance between the gauges is statistically insignificant. (See table below) Test for Equal Variances Confidence intervals for standard deviations: Lower Sigma Upper N Factor Levels Gauge Gauge 2 F-Test (normal distribution) Test Statistic: P-Value : The Variance Ratio of the gauges is.931. Based on the criteria listed in section 6.1, the Variance Ratio result is deemed acceptable. (Gauge 1 sigma) 2 / (Gauge 2 sigma) 2 = (3.912) 2 / (4.055) 2 = If the test shows the difference is statistically significant and the Variance Ratio passes the criteria, then determine if the difference is practically significance and make an engineering judgement If the test shows the difference is not statistically significant and the Variance Ratio fails the criteria, then investigate to ensure the sample size is appropriate.

37 PAGE: 37 OF 44 C: APPENDIX: Windage Signal Processing with LDV DESCRIPTION TABLE OF CONTENTS 1.0 Objective 2.0 Scope 3.0 Reference Documents 4.0 Equipments and settings 5.0 Attachment: Notes on channel setup 1.0 OBJECTIVE: Objective of this document is to provide general guidelines for data acquisition and processing for windage with an LDV system. 2.0 SCOPE: Scope of this procedure pertains to HGA Windage measurement equipments. 3.0 REFERENCE DOCUMENT: N/A

38 PAGE: 38 OF EQUIPMENTS & SETTINGS Major components of a windage/frf data acquisition system are shown in Fig.1. PC & software DAQ board & accessory LDV Controller LDV Sensor HGA under test Fig.1 Components of a windage data acquisition system They are: 4.1. LDV Sensor head: The sensor head contains a laser source and an interferometer to couple the reflected beam back to the source beam LDV Controller: The LDV controller conditions and decodes the signal to compute velocity and or displacement of the target. Settings of the controller may be briefed as follows: Measurement range: It is preferred to use the velocity decoder (rather than the displacement decoder) because of its higher signal-to-noise ratio for the derived displacements in large bandwidths. The velocity measurement range of the controller has to be selected so that the velocity, acceleration and frequency limits of the decoder channel are not exceeded and resolution is acceptable. The range has to be adjusted also to the data acquisition accessory input so that the DAQ channel does not saturate Tracking filter: It is preferred to turn off the filter to avoid any chance of the filter losing lock with the signals. The signal to noise ratio might instead be improved by adjusting the optical reflection conditions from the target Velocity filter: It can be used as an anti-aliasing filter. The cut-off frequency of this lowpass filter should be above the maximum frequency of interest. Parameter Spec. Remarks Range 5 mm/s/v Typical setting Tracking filter Off Velocity filter 50 khz minimum Nyquist= 100 khz 4.3. DAQ board and accessory: Data acquisition board and accessories convert the analog voltage signal into digital format, which is sent over to the PC for performing the spectral computations. The main criteria for choosing DAQ accessories are sampling rate, resolution and channel noise.

39 PAGE: 39 OF 44 Parameter Spec. Remarks Sampling rate 200 khz Typical setting Resolution 16 bit Typical Channel noise -90 db Typical value 4.4. Windage data processing system: The data is processed using a signal-processing algorithm. Important parameters pertaining to this process are time-length of data acquisition, sampling rate, windowing, averaging, number of FFT points (lines), nature of filtering and the algorithm used to calculate PSD (power spectral density). A brief description of windage signal processing is given in section 5.0 (Attachment: Notes on Channel Set-up). 5.0 ATTACHMENT: Notes on Channel set-up Analog signal Gain adjustment Antialiasing filter ADC Digital signal db Filter roll-off Filter dynamic range Aliasing Useful range F s /2 F s f A typical channel is shown above. Useful channel frequency range is determined by sampling frequency (Fs), filter dynamic range and roll-off rate of the filter. Phase distortion caused by the filter should also be considered. The input gain stage must be optimized to match the ADC range (2 N discrete values to cover the voltage range). It is desirable to have an estimate of noise from the channel components (including cabling) at least in the useful range.

40 PAGE: 40 OF 44 D: APPENDIX: Tooling Standards DESCRIPTION TABLE OF CONTENTS 1.0 Objective 2.0 Scope 3.0 Reference Documents 4.0 Tooling for FRF Measurement 5.0 Tooling for Windage Measurement 1.0 OBJECTIVE: Objective of this document is to provide general guidelines and specification for HGA tooling for the measurement of windage and FRF. 2.0 SCOPE: Scope of this procedure pertains to HGA Windage and Resonance measurement tooling. 3.0 REFERENCE DOCUMENT: N/A

41 PAGE: 41 OF TOOLING FOR FRF MEASUREMENT: Major components of a FRF measurement tooling system are shown in Fig.1. Figure 1, Numbered Items 1) Shaker or Excitation Source 2) Shaker Adapter 3) Mount Block 4) Disk Media Figure 1: FRF Measurement Tooling System As can be seen in Fig. 1, there are a number of different tools that are part of makeup the FRF measurement system. Each tool has a significant interaction with the design and use of the other tools. This document will illustrate a configuration of tooling typically used for FRF testing. Other configurations do exist and are used interchangeably to meet the same requirements.

42 PAGE: 42 OF Shaker Adapter: The shaker adapter is the mechanical connection between the excitation source and the product specific HGA mount block tooling. Figure 2: Shaker Adapter The shaker adapter allows the user to remove the mount block from the resonance tester. Removal of the mount block from the resonance measurement system allows for the following capabilities: Testing of multiple products each requiring different mount blocks. Alignment of the HGA to the primary HGA datum defined on the print. The design of a shaker adapter must take into consideration the following items: Mounting methods to the shaker and mount blocks such that mount blocks can be product specific and the shaker adapter is fixed. Modes of the shaker adapter and interactions with the shaker and the mount block. Datum for mount block. Tolerances for manufacturing. Maintenance. The shaker adapter can be attached to the excitation source by many methods. Commonly used methods include: threads, glue, and directly manufactured into the excitation source. Integration of the shaker adapter into the excitation source is generally a preferred method due to the reduction in the number of components.

43 PAGE: 43 OF Mount Block: The mount block contains the surface that represents the clamping surface of the HGA as defined on the HGA print. Figure 3. Typical Mount Block When combined with a shaker adapter the mount block is removed for the purpose of loading and unloading HGA s. During the loading of an HGA on to the mount block, a fixture may be used to position and align the HGA to the primary datum surface. Generally this fixture uses part features to align and position the HGA to the mount block. The mount block contains a feature for installation of a screw that holds the washer against the HGA. When the washer and screw are installed the HGA is clamped the primary mount surface. Other methods can be used to hold the HGA to the primary datum, such as: glue, wax, and/or springs. Mount blocks are generally constrained to the shaker adapter via a primary surface and pins for positioning. The mount block is held in place to the shaker adapter with a screw installed to a specified torque. 4.3 Washer and Screw: The washer and screw provide the clamping force to the HGA. Holding the HGA against the primary datum surface. The screw is installed using a torque wrench. The washer and screw features may also be combined into one piece of tooling.