SLIDE 1 Polynomial nonlinear state-space modeling
- f the F-16 aircraft benchmark
Koen Tiels
2017 Workshop on Nonlinear System Identification Benchmarks
SLIDE 2
Introduction: F16 ground vibration test
2/20
SLIDE 3 Goal: Capture system dynamics
System Model + Inputs Outputs Modeled
Error signals
+ –
3/20
SLIDE 4 Goal: Capture system dynamics
System Model + Inputs Outputs Modeled
Error signal
+ –
4/20
SLIDE 5 Goal: Capture system dynamics
System Model + Inputs Outputs Modeled
Error signals
+ –
5/20
SLIDE 6
Set-up Nonparametric analysis Parametric modeling
6/20
SLIDE 7
Two inputs and three outputs are provided in the benchmark
Signal generator Shaker F16 Voltage Force 145 accelerations
Two inputs:
reference input: Voltage actual input: Force
Three outputs:
Acceleration at excitation location Acceleration on right wing Acceleration on payload
7/20
SLIDE 8 Multisine excitation with random frequency grid
8/20
41 −0.4 0.4 Time (s) Voltage (V) Voltage signal 1 60 −100 −10.5 Frequency (Hz) Amplitude (dB) Voltage spectrum
u(t) =
F
Akcos(2πkf0t + φk) 3 amplitude levels (12.2, 49.0, and 97.1 N RMS) 3 periods per amplitude level (two in steady state) 10 input realizations per level (9 for estimation, 1 for testing) 16384 points per period
SLIDE 9
Set-up Nonparametric analysis Parametric modeling
9/20
SLIDE 10
The FRFs and the distortion levels are estimated
10/20
SLIDE 11 The FRFs and the distortion levels are estimated
11/20
20 40 60 −100 −50 Frequency (Hz) FRF (dB) Output 1 20 40 60 −100 −50 Frequency (Hz) FRF (dB) Output 2 20 40 60 −100 −50 Frequency (Hz) FRF (dB) Output 3
SLIDE 12 Focus on frequency band [4.7, 11] Hz
12/20
4.75.2 6.6 7.3 9.1 11 −71 −45 −15 Frequency (Hz) FRF (dB) Output 1 4.75.2 6.6 7.3 9.1 11 −65 −39 −10 Frequency (Hz) FRF (dB) Output 2 4.75.2 6.6 7.3 9.1 11 −65 −40 −10 Frequency (Hz) FRF (dB) Output 3
SLIDE 13 Amplitude level 2 has the largest nonlinear distortion to signal ratio
13/20
4.7 5.2 6.6 7.3 9.1 11 −65 −40 −10 Frequency (Hz) FRF (dB) Output 3 Amplitude 2: FRF Amplitude 3: FRF Amplitude 1: FRF Amplitude 2: Total distorion Amplitude 3: Total distorion Amplitude 1: Total distorion Amplitude 1: Noise distorion Amplitude 2: Noise distorion Amplitude 3: Noise distorion
SLIDE 14 The nonlinear distortions at 7.3 Hz are about 25 dB larger than the noise
14/20
4.7 5.2 6.6 7.3 9.1 11 −65 −40 −10 Frequency (Hz) FRF (dB) Output 3 Amplitude 2: FRF Amplitude 3: FRF Amplitude 1: FRF Amplitude 2: Total distorion Amplitude 3: Total distorion Amplitude 1: Total distorion Amplitude 1: Noise distorion Amplitude 2: Noise distorion Amplitude 3: Noise distorion
SLIDE 15
Set-up Nonparametric analysis Parametric modeling
15/20
SLIDE 16
A linear state-space model captures dynamic behavior
16/20 x(t + 1) = A x(t) + B u(t) + E ζ(x(t), u(t)) y(t) = C x(t) + D u(t) + F η(x(t), u(t)) linear state-space model polynomials in x and u
SLIDE 17
A polynomial nonlinear state-space model captures nonlinear dynamic behavior
17/20 x(t + 1) = A x(t) + B u(t) + E ζ(x(t), u(t)) y(t) = C x(t) + D u(t) + F η(x(t), u(t)) linear state-space model polynomials in x and u
with e.g. ζ(x, u) =
x2
1
x1x2 x1u . . . x2
2u
u3 . . .
SLIDE 18 Identification of a polynomial nonlinear state-space model
18/20 x(t + 1) = A x(t) + B u(t) + E ζ(x(t), u(t)) y(t) = C x(t) + D u(t) + F η(x(t), u(t)) linear state-space model polynomials in x and u
θ =
vec(A) vec(B) vec(C) vec(D) vec(E) vec(F) vec(x(0)) vec(u(0))
ǫ(k, θ) = Y (k, θ) − Ymeas(k) KWLS(θ) =
NF
ǫH(k, θ)W (k)ǫ(k, θ) ˆ θ = arg min
θ
KWLS
SLIDE 19 Nonlinear interaction only within first resonance and coupled interaction in other three resonances
19/20
4.75.2 6.6 7.3 9.1 11 45 50 53 71 Output 2
Amplitude (dB) 4.75.2 6.6 7.3 9.1 11 44 49 52 71 Output 3
Output Linear error PNLSS error Linear error PNLSS error Output
SLIDE 20
Conclusions and future work
Nonparametric analysis:
High-quality measurements Some room for improvement with nonlinear modeling
Parametric modeling:
Frequency weighting possible Challenging benchmark
Future work:
Non-polynomial basis functions
20/20
SLIDE 21
SLIDE 22
Two inputs and three outputs are provided in the benchmark
Signal generator Shaker F16 Voltage Force 145 accelerations
Two inputs:
reference input: Voltage actual input: Force
Three outputs:
Acceleration at excitation location Acceleration on right wing Acceleration on payload
Sampling frequency:
400 Hz (upsampled from 200 Hz)
21/20
SLIDE 23 Amplitude level 2 has the largest nonlinear distortion to signal ratio
22/20
4.7 5.2 6.6 7.3 9.1 11 −65 −40 −10 Frequency (Hz) FRF (dB) Output 3 Amplitude 2: FRF Amplitude 3: FRF Amplitude 1: FRF Amplitude 2: Total distorion Amplitude 3: Total distorion Amplitude 1: Total distorion Amplitude 1: Noise distorion Amplitude 2: Noise distorion Amplitude 3: Noise distorion
SLIDE 24 The nonlinear distortions at 7.3 Hz are about 25 dB larger than the noise
23/20
6.6 7.3 −65 −40 −10 Frequency (Hz) FRF (dB) Output 3 Amplitude 1: Noise distorion Amplitude 2: Noise distorion Amplitude 3: Noise distorion Amplitude 1: FRF Amplitude 2: FRF Amplitude 1: Total distorion Amplitude 2: Total distorion Amplitude 3: Total distorion Amplitude 3: FRF
SLIDE 25 Focus on main resonance with second-order linear and nonlinear model (lowest amplitude)
24/20
7.3 8.6 21 47 59
Amplitude (dB) Output 1 7.3 8.6 27 53 65
Output 2 7.3 8.6 27 53 64
Output 3 Output Linear error Output Linear error Output Linear error PNLSS error PNLSS error PNLSS error
95 parameters
SLIDE 26 Focus on main resonance with second-order linear and nonlinear model (amplitude level 2)
25/20
7.3 8.6 38 51 66
Amplitude (dB) Output 1 7.3 8.6 44 57 71
Output 2 7.3 8.6 44 57 71
Output 3 Output Output Output Linear error Linear error Linear error PNLSS error PNLSS error PNLSS error
95 parameters
SLIDE 27
Limit the number of parameters by allowing nonlinear interaction only within a resonance
x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12
+
=
A1 A2 A3 A4 A5 A6
x(t) + Bu(t) +
f1(x1, x2) f2(x1, x2) f3(x3, x4) f4(x3, x4) f5(x5, x6) f6(x5, x6) f7(x7, x8) f8(x7, x8) f9(x9, x10) f10(x9, x10) f11(x11, x12) f12(x11, x12)
y(t) = Cx(t) + Du(t) + g1(x1, x2) + g2(x3, x4) + g3(x5, x6) + g4(x7, x8) + g5(x9, x10) + g6(x11, x12)
26/20
SLIDE 28
Limit the number of parameters by allowing nonlinear interaction only within a resonance
x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12
+
=
A1 A2 A3 A4 A5 A6
x(t) + Bu(t) +
f1(x1, x2) f2(x1, x2) f3(x3, x4) f4(x3, x4) f5(x5, x6) f6(x5, x6) f7(x7, x8) f8(x7, x8) f9(x9, x10) f10(x9, x10) f11(x11, x12) f12(x11, x12)
y(t) = Cx(t) + Du(t) + g1(x1, x2) + g2(x3, x4) + g3(x5, x6) + g4(x7, x8) + g5(x9, x10) + g6(x11, x12)
27/20
SLIDE 29 Nonlinear interaction only within a resonance (12th order models)
28/20
4.75.2 6.6 7.3 9.1 11 37 47 50 53 71 Output 2
Amplitude (dB) 4.75.2 6.6 7.3 9.1 11 37 46 49 52 71 Output 3
Output Output Linear error PNLSS error Linear error PNLSS error 350 parameters (full model would have 7826 parameters)
SLIDE 30 Nonlinear interaction only within first resonance and coupled interaction in other three resonances
29/20
4.75.2 6.6 7.3 9.1 11 45 50 53 71 Output 2
Amplitude (dB) 4.75.2 6.6 7.3 9.1 11 44 49 52 71 Output 3
Output Linear error PNLSS error Linear error PNLSS error Output 882 parameters (full model would have 7826 parameters)