27 KiB
Spindle Analysis
- Notes
- Data Processing
- Time Domain Data
- Model of the spindle
- ZIP file containing the data and matlab files
- Schematic of the model
- Parameters
- Compute Mass and Stiffness Matrices
- Compute resonance frequencies
- From model_damping compute the Damping Matrix
- Define inputs, outputs and state names
- Define A, B and C matrices
- Generate the State Space Model
- Bode Plot
- Save the model
- Frequency Domain Data
- ZIP file containing the data and matlab files
- Load the processed data and the model
- Compute the PSD
- Plot the computed PSD
- Compute the response of the model
- Plot the PSD of the Force using the model
- Estimated Shape of the PSD of the force
- PSD in [N]
- PSD in [m]
- Compute the resulting RMS value [m]
- Compute the resulting RMS value [m]
- Functions
The report made by the PEL is accessible here.
Notes
Date | 2017-04-25 |
Location | PEL Lab |
The goal is to estimate all the error motions induced by the Spindle
Data Processing
<<sec:spindle_data_processing>>
ZIP file containing the data and matlab files ignore
All the files (data and Matlab scripts) are accessible here.
Load Measurement Data
spindle_1rpm_table = readtable('./mat/10turns_1rpm_icepap.txt');
spindle_60rpm_table = readtable('./mat/10turns_60rpm_IcepapFIR.txt');
spindle_1rpm_table(1, :)
spindle_1rpm = table2array(spindle_1rpm_table);
spindle_60rpm = table2array(spindle_60rpm_table);
Convert Signals from [deg] to [sec]
speed_1rpm = 360/60; % [deg/sec]
spindle_1rpm(:, 1) = spindle_1rpm(:, 2)/speed_1rpm; % From position [deg] to time [s]
speed_60rpm = 360/1; % [deg/sec]
spindle_60rpm(:, 1) = spindle_60rpm(:, 2)/speed_60rpm; % From position [deg] to time [s]
Convert Signals
% scaling = 1/80000; % 80 mV/um
scaling = 1e-6; % [um] to [m]
spindle_1rpm(:, 3:end) = scaling*spindle_1rpm(:, 3:end); % [V] to [m]
spindle_1rpm(:, 3:end) = spindle_1rpm(:, 3:end)-mean(spindle_1rpm(:, 3:end)); % Remove mean
spindle_60rpm(:, 3:end) = scaling*spindle_60rpm(:, 3:end); % [V] to [m]
spindle_60rpm(:, 3:end) = spindle_60rpm(:, 3:end)-mean(spindle_60rpm(:, 3:end)); % Remove mean
Ts and Fs for both measurements
Ts_1rpm = spindle_1rpm(end, 1)/(length(spindle_1rpm(:, 1))-1);
Fs_1rpm = 1/Ts_1rpm;
Ts_60rpm = spindle_60rpm(end, 1)/(length(spindle_60rpm(:, 1))-1);
Fs_60rpm = 1/Ts_60rpm;
Find Noise of the ADC [$\frac{m}{\sqrt{Hz}}$]
data = spindle_1rpm(:, 5);
dV_1rpm = min(abs(data(1) - data(data ~= data(1))));
noise_1rpm = dV_1rpm/sqrt(12*Fs_1rpm/2);
data = spindle_60rpm(:, 5);
dV_60rpm = min(abs(data(50) - data(data ~= data(50))));
noise_60rpm = dV_60rpm/sqrt(12*Fs_60rpm/2);
Save all the data under spindle struct
spindle.rpm1.time = spindle_1rpm(:, 1);
spindle.rpm1.deg = spindle_1rpm(:, 2);
spindle.rpm1.Ts = Ts_1rpm;
spindle.rpm1.Fs = 1/Ts_1rpm;
spindle.rpm1.x = spindle_1rpm(:, 3);
spindle.rpm1.y = spindle_1rpm(:, 4);
spindle.rpm1.z = spindle_1rpm(:, 5);
spindle.rpm1.adcn = noise_1rpm;
spindle.rpm60.time = spindle_60rpm(:, 1);
spindle.rpm60.deg = spindle_60rpm(:, 2);
spindle.rpm60.Ts = Ts_60rpm;
spindle.rpm60.Fs = 1/Ts_60rpm;
spindle.rpm60.x = spindle_60rpm(:, 3);
spindle.rpm60.y = spindle_60rpm(:, 4);
spindle.rpm60.z = spindle_60rpm(:, 5);
spindle.rpm60.adcn = noise_60rpm;
Compute Asynchronous data
for direction = {'x', 'y', 'z'}
spindle.rpm1.([direction{1}, 'async']) = getAsynchronousError(spindle.rpm1.(direction{1}), 10);
spindle.rpm60.([direction{1}, 'async']) = getAsynchronousError(spindle.rpm60.(direction{1}), 10);
end
Save data
save('./mat/spindle_data.mat', 'spindle');
Time Domain Data
<<sec:spindle_time_domain>>
ZIP file containing the data and matlab files ignore
All the files (data and Matlab scripts) are accessible here.
Load the processed data
load('./mat/spindle_data.mat', 'spindle');
Plot X-Y-Z position with respect to Time - 1rpm
figure;
hold on;
plot(spindle.rpm1.time, spindle.rpm1.x);
plot(spindle.rpm1.time, spindle.rpm1.y);
plot(spindle.rpm1.time, spindle.rpm1.z);
hold off;
xlabel('Time [s]'); ylabel('Amplitude [m]');
legend({'tx - 1rpm', 'ty - 1rpm', 'tz - 1rpm'});
<<plt-matlab>>
Plot X-Y-Z position with respect to Time - 60rpm
The measurements for the spindle turning at 60rpm are shown figure fig:spindle_xyz_60rpm.
figure;
hold on;
plot(spindle.rpm60.time, spindle.rpm60.x);
plot(spindle.rpm60.time, spindle.rpm60.y);
plot(spindle.rpm60.time, spindle.rpm60.z);
hold off;
xlabel('Time [s]'); ylabel('Amplitude [m]');
legend({'tx - 60rpm', 'ty - 60rpm', 'tz - 60rpm'});
<<plt-matlab>>
Plot Synchronous and Asynchronous - 1rpm
figure;
hold on;
plot(spindle.rpm1.time, spindle.rpm1.x);
plot(spindle.rpm1.time, spindle.rpm1.xasync);
hold off;
xlabel('Time [s]'); ylabel('Amplitude [m]');
legend({'tx - 1rpm - Sync', 'tx - 1rpm - Async'});
<<plt-matlab>>
Plot Synchronous and Asynchronous - 60rpm
The data is split into its Synchronous and Asynchronous part (figure fig:spindle_60rpm_sync_async). We then use the Asynchronous part for the analysis in the following sections as we suppose that we can deal with the synchronous part with feedforward control.
figure;
hold on;
plot(spindle.rpm60.time, spindle.rpm60.x);
plot(spindle.rpm60.time, spindle.rpm60.xasync);
hold off;
xlabel('Time [s]'); ylabel('Amplitude [m]');
legend({'tx - 60rpm - Sync', 'tx - 60rpm - Async'});
<<plt-matlab>>
Plot X against Y
figure;
hold on;
plot(spindle.rpm1.x, spindle.rpm1.y);
plot(spindle.rpm60.x, spindle.rpm60.y);
hold off;
xlabel('X Amplitude [m]'); ylabel('Y Amplitude [m]');
legend({'1rpm', '60rpm'});
<<plt-matlab>>
Plot X against Y - Asynchronous
figure;
hold on;
plot(spindle.rpm1.xasync, spindle.rpm1.yasync);
plot(spindle.rpm60.xasync, spindle.rpm60.yasync);
hold off;
xlabel('X Amplitude [m]'); ylabel('Y Amplitude [m]');
legend({'1rpm', '60rpm'});
<<plt-matlab>>
Model of the spindle
<<sec:spindle_model>>
ZIP file containing the data and matlab files ignore
All the files (data and Matlab scripts) are accessible here.
Schematic of the model
The model of the spindle used is shown figure fig:model_spindle.
$f$ is the perturbation force of the spindle and $d$ is the measured displacement.
Parameters
mg = 3000; % Mass of granite [kg]
ms = 50; % Mass of Spindle [kg]
kg = 1e8; % Stiffness of granite [N/m]
ks = 5e7; % Stiffness of spindle [N/m]
Compute Mass and Stiffness Matrices
Mm = diag([ms, mg]);
Km = diag([ks, ks+kg]) - diag(ks, -1) - diag(ks, 1);
Compute resonance frequencies
A = [zeros(size(Mm)) eye(size(Mm)) ; -Mm\Km zeros(size(Mm))];
eigA = imag(eigs(A))/2/pi;
eigA = eigA(eigA>0);
eigA = eigA(1:2);
From model_damping compute the Damping Matrix
modal_damping = 1e-5;
ab = [0.5*eigA(1) 0.5/eigA(1) ; 0.5*eigA(2) 0.5/eigA(2)]\[modal_damping ; modal_damping];
Cm = ab(1)*Mm +ab(2)*Km;
Define inputs, outputs and state names
StateName = {...
'xs', ... % Displacement of Spindle [m]
'xg', ... % Displacement of Granite [m]
'vs', ... % Velocity of Spindle [m]
'vg', ... % Velocity of Granite [m]
};
StateUnit = {'m', 'm', 'm/s', 'm/s'};
InputName = {...
'f' ... % Spindle Force [N]
};
InputUnit = {'N'};
OutputName = {...
'd' ... % Displacement [m]
};
OutputUnit = {'m'};
Define A, B and C matrices
% A Matrix
A = [zeros(size(Mm)) eye(size(Mm)) ; ...
-Mm\Km -Mm\Cm];
% B Matrix
B_low = zeros(length(StateName), length(InputName));
B_low(strcmp(StateName,'vs'), strcmp(InputName,'f')) = 1;
B_low(strcmp(StateName,'vg'), strcmp(InputName,'f')) = -1;
B = blkdiag(zeros(length(StateName)/2), pinv(Mm))*B_low;
% C Matrix
C = zeros(length(OutputName), length(StateName));
C(strcmp(OutputName,'d'), strcmp(StateName,'xs')) = 1;
C(strcmp(OutputName,'d'), strcmp(StateName,'xg')) = -1;
% D Matrix
D = zeros(length(OutputName), length(InputName));
Generate the State Space Model
sys = ss(A, B, C, D);
sys.StateName = StateName;
sys.StateUnit = StateUnit;
sys.InputName = InputName;
sys.InputUnit = InputUnit;
sys.OutputName = OutputName;
sys.OutputUnit = OutputUnit;
Bode Plot
The transfer function from a disturbance force $f$ to the measured displacement $d$ is shown figure fig:spindle_f_to_d.
freqs = logspace(-1, 3, 1000);
figure;
plot(freqs, abs(squeeze(freqresp(sys('d', 'f'), freqs, 'Hz'))));
set(gca, 'XScale', 'log'); set(gca, 'YScale', 'log');
xlabel('Frequency [Hz]'); ylabel('Amplitude [m/N]');
<<plt-matlab>>
Save the model
save('./mat/spindle_model.mat', 'sys');
Frequency Domain Data
<<sec:spindle_psd>>
ZIP file containing the data and matlab files ignore
All the files (data and Matlab scripts) are accessible here.
Load the processed data and the model
load('./mat/spindle_data.mat', 'spindle');
load('./mat/spindle_model.mat', 'sys');
Compute the PSD
n_av = 4; % Number of average
[pxx_1rpm, f_1rpm] = pwelch(spindle.rpm1.xasync, hanning(ceil(length(spindle.rpm1.xasync)/n_av)), [], [], spindle.rpm1.Fs);
[pyy_1rpm, ~] = pwelch(spindle.rpm1.yasync, hanning(ceil(length(spindle.rpm1.yasync)/n_av)), [], [], spindle.rpm1.Fs);
[pzz_1rpm, ~] = pwelch(spindle.rpm1.zasync, hanning(ceil(length(spindle.rpm1.zasync)/n_av)), [], [], spindle.rpm1.Fs);
[pxx_60rpm, f_60rpm] = pwelch(spindle.rpm60.xasync, hanning(ceil(length(spindle.rpm60.xasync)/n_av)), [], [], spindle.rpm60.Fs);
[pyy_60rpm, ~] = pwelch(spindle.rpm60.yasync, hanning(ceil(length(spindle.rpm60.yasync)/n_av)), [], [], spindle.rpm60.Fs);
[pzz_60rpm, ~] = pwelch(spindle.rpm60.zasync, hanning(ceil(length(spindle.rpm60.zasync)/n_av)), [], [], spindle.rpm60.Fs);
Plot the computed PSD
The Amplitude Spectral Densities of the displacement of the spindle for the $x$, $y$ and $z$ directions are shown figure fig:spindle_psd_xyz_60rpm. They correspond to the Asynchronous part shown figure fig:spindle_60rpm_sync_async.
figure;
hold on;
plot(f_1rpm, (pxx_1rpm).^.5, 'DisplayName', '$P_{xx}$ - 1rpm');
plot(f_1rpm, (pyy_1rpm).^.5, 'DisplayName', '$P_{yy}$ - 1rpm');
plot(f_1rpm, (pzz_1rpm).^.5, 'DisplayName', '$P_{zz}$ - 1rpm');
% plot(f_1rpm, spindle.rpm1.adcn*ones(size(f_1rpm)), '--k', 'DisplayName', 'ADC - 1rpm');
hold off;
set(gca, 'XScale', 'log'); set(gca, 'YScale', 'log');
xlabel('Frequency [Hz]'); ylabel('ASD [$m/\sqrt{Hz}$]');
legend('Location', 'northeast');
xlim([f_1rpm(2), f_1rpm(end)]);
<<plt-matlab>>
figure;
hold on;
plot(f_60rpm, (pxx_60rpm).^.5, 'DisplayName', '$P_{xx}$ - 60rpm');
plot(f_60rpm, (pyy_60rpm).^.5, 'DisplayName', '$P_{yy}$ - 60rpm');
plot(f_60rpm, (pzz_60rpm).^.5, 'DisplayName', '$P_{zz}$ - 60rpm');
% plot(f_60rpm, spindle.rpm60.adcn*ones(size(f_60rpm)), '--k', 'DisplayName', 'ADC - 60rpm');
hold off;
set(gca, 'XScale', 'log'); set(gca, 'YScale', 'log');
xlabel('Frequency [Hz]'); ylabel('ASD [$m/\sqrt{Hz}$]');
legend('Location', 'northeast');
xlim([f_60rpm(2), f_60rpm(end)]);
<<plt-matlab>>
Compute the response of the model
Tfd = abs(squeeze(freqresp(sys('d', 'f'), f_60rpm, 'Hz')));
Plot the PSD of the Force using the model
figure;
plot(f_60rpm, (pxx_60rpm.^.5)./Tfd, 'DisplayName', '$P_{xx}$');
set(gca, 'XScale', 'log'); set(gca, 'YScale', 'log');
xlabel('Frequency [Hz]'); ylabel('ASD [$N/\sqrt{Hz}$]');
xlim([f_60rpm(2), f_60rpm(end)]);
<<plt-matlab>>
Estimated Shape of the PSD of the force
s = tf('s');
Wd_simple = 1e-8/(1+s/2/pi/0.5)/(1+s/2/pi/100);
Wf_simple = Wd_simple/tf(sys('d', 'f'));
TWf_simple = abs(squeeze(freqresp(Wf_simple, f_60rpm, 'Hz')));
% Wf = 0.48902*(s+327.9)*(s^2 + 109.6*s + 1.687e04)/((s^2 + 30.59*s + 8541)*(s^2 + 29.11*s + 3.268e04));
% Wf = 0.15788*(s+418.6)*(s+1697)^2*(s^2 + 124.3*s + 2.529e04)*(s^2 + 681.3*s + 9.018e05)/((s^2 + 23.03*s + 8916)*(s^2 + 33.85*s + 6.559e04)*(s^2 + 71.43*s + 4.283e05)*(s^2 + 40.64*s + 1.789e06));
Wf = (s+1697)^2*(s^2 + 114.5*s + 2.278e04)*(s^2 + 205.1*s + 1.627e05)*(s^2 + 285.8*s + 8.624e05)*(s+100)/((s+0.5)*3012*(s^2 + 23.03*s + 8916)*(s^2 + 17.07*s + 4.798e04)*(s^2 + 41.17*s + 4.347e05)*(s^2 + 78.99*s + 1.789e06));
TWf = abs(squeeze(freqresp(Wf, f_60rpm, 'Hz')));
PSD in [N]
Above $200Hz$, the Amplitude Spectral Density seems dominated by noise coming from the electronics (charge amplifier, ADC, …). So we don't know what is the frequency content of the force above that frequency. However, we assume that $P_{xx}$ is decreasing with $1/f$ as it seems so be the case below $100Hz$ (figure fig:spindle_psd_xyz_60rpm).
We then fit the PSD of the displacement with a transfer function (figure fig:spindle_psd_d_comp_60rpm).
figure;
hold on;
plot(f_60rpm, (pxx_60rpm.^.5)./Tfd, 'DisplayName', '$\sqrt{P_{xx}}/|T_{d/f}|$');
plot(f_60rpm, TWf, 'DisplayName', 'Wf');
plot(f_60rpm, TWf_simple, '-k', 'DisplayName', 'Wfs');
set(gca, 'XScale', 'log'); set(gca, 'YScale', 'log');
xlabel('Frequency [Hz]'); ylabel('ASD [$N/\sqrt{Hz}$]');
xlim([f_60rpm(2), f_60rpm(end)]);
legend('Location', 'northeast');
<<plt-matlab>>
PSD in [m]
To obtain the PSD of the force $f$ that induce such displacement, we use the following formula: \[ \sqrt{PSD(d)} = |T_{d/f}| \sqrt{PSD(f)} \]
And so we have: \[ \sqrt{PSD(f)} = |T_{d/f}|^{-1} \sqrt{PSD(d)} \]
The obtain Power Spectral Density of the force is displayed figure fig:spindle_psd_f_comp_60rpm.
figure;
hold on;
plot(f_60rpm, pxx_60rpm.^.5, 'DisplayName', '$\sqrt{P_{xx}}$');
plot(f_60rpm, TWf.*Tfd, 'DisplayName', '$|W_f|*|T_{d/f}|$');
set(gca, 'XScale', 'log'); set(gca, 'YScale', 'log');
xlabel('Frequency [Hz]'); ylabel('ASD [$m/\sqrt{Hz}$]');
xlim([f_60rpm(2), f_60rpm(end)]);
legend('Location', 'northeast');
<<plt-matlab>>
Compute the resulting RMS value [m]
figure;
hold on;
plot(f_60rpm, 1e9*cumtrapz(f_60rpm, (pxx_60rpm)).^.5, '--', 'DisplayName', 'Exp. Data');
plot(f_60rpm, 1e9*cumtrapz(f_60rpm, ((TWf.*Tfd).^2)).^.5, '--', 'DisplayName', 'Estimated');
hold off;
set(gca, 'XScale', 'log');
xlabel('Frequency [Hz]'); ylabel('CPS [$nm$ rms]');
xlim([f_60rpm(2), f_60rpm(end)]);
legend('Location', 'southeast');
<<plt-matlab>>
Compute the resulting RMS value [m]
figure;
hold on;
plot(f_1rpm, 1e9*cumtrapz(f_1rpm, (pxx_1rpm)), '--', 'DisplayName', 'Exp. Data');
plot(f_1rpm, 1e9*(f_1rpm(end)-f_1rpm(1))/(length(f_1rpm)-1)*cumsum(pxx_1rpm), '--', 'DisplayName', 'Exp. Data');
hold off;
set(gca, 'XScale', 'log');
xlabel('Frequency [Hz]'); ylabel('CPS [$nm$ rms]');
xlim([f_1rpm(2), f_1rpm(end)]);
legend('Location', 'southeast');
<<plt-matlab>>
Functions
getAsynchronousError
<<sec:getAsynchronousError>>
This Matlab function is accessible here.
function [Wxdec] = getAsynchronousError(data, NbTurn)
%%
L = length(data);
res_per_rev = L/NbTurn;
P = 0:(res_per_rev*NbTurn-1);
Pos = P' * 360/res_per_rev;
% Temperature correction
x1 = myfit2(Pos, data);
% Convert data to frequency domain and scale accordingly
X2 = 2/(res_per_rev*NbTurn)*fft(x1);
f2 = (0:L-1)./NbTurn; %upr -> once per revolution
%%
X2dec = zeros(size(X2));
% Get only the non integer data
X2dec(mod(f2(:), 1) ~= 0) = X2(mod(f2(:), 1) ~= 0);
Wxdec = real((res_per_rev*NbTurn)/2 * ifft(X2dec));
%%
function Y = myfit2(x,y)
A = [x ones(size(x))]\y;
a = A(1); b = A(2);
Y = y - (a*x + b);
end
end