Triple decomposition and sparse representation for noisy pressure-sensitive paint data

Kubota, Koyo; Takagi, Makoto; Ikami, Tsubasa; Egami, Yasuhiro; Nagai, Hiroki; Kashikawa, Takahiro; Kimura, Koichi; Matsuda, Yu

doi:10.1063/5.0220866

Triple decomposition is a powerful analytical method for a deep understanding of the flow structure by extracting the mean value, organized coherent motion, and stochastic part from a fluctuating quantity. Here, we perform the triple decomposition of the spatial two-dimensional data, especially pressure-sensitive paint (PSP) data, since the PSP method is widely used to measure the pressure distribution on a surface in wind tunnel testing. However, the PSP data measuring near atmospheric pressure contain significant noise, and this makes it difficult to conduct the decomposition. To construct phase-averaged data representing an organized coherent motion, we propose a relatively simple method based on a multi-dimensional scaling plot of the cosine similarity between each PSP datum. Then, the stochastic part is extracted by selecting phase-averaged data with an appropriate phase angle based on the similarity between the measurement and phase-averaged data, and the PSP data are successfully decomposed. Moreover, we consider sparse optimal sensor positions, in which the data are effectively represented, based on the stochastic part as a data-driven approach. The optimal sensor positions are determined as a combinatorial optimization problem and estimated using Fujitsu computing as a service digital annealer. We reconstruct the pressure distribution from the pressure data at the optimal sensor positions using the mean value, organized coherent motion, and stochastic part obtained from the triple decomposition. The root mean square error between the pressure measured by a pressure transducer and the reconstructed pressure obtained by the proposed method is small, even when the number of modes and sensor points is small. The application of PSP measurement is expected to expand further, and the framework for calculating triple decomposition and sparse representation based on the decomposition will be useful for detailed flow analysis.

I. INTRODUCTION

The Reynolds decomposition is known as a basic method for extracting the stochastic part of flow. In the Reynolds decomposition, a fluctuating quantity

f (x, t)

at position vector

x

and time

t

is decomposed as

f (x, t) = \bar{f} (x) + f' (x, t),

(1)

where

\bar{f} (x)

is the mean value for time and

f' (x, t)

is the deviation from the mean value (the stochastic part). The triple decomposition was proposed for the analysis of a flow field consisting of organized coherent motion.¹ In the triple decomposition, a fluctuating quantity

f (x, t)

at position vector

x

and time

t

is decomposed as

f (x, t) = \bar{f} (x) + \tilde{f} (x, ϕ (t)) + f' (x, t),

(2)

where

\tilde{f} (x, ϕ (t))

is the contribution of the organized coherent motion and

ϕ (t)

is the phase angle at time

t

⁠. The decomposition is straightforwardly conducted when a reference signal is available. The phase angle is extracted from the reference signal, and

\tilde{f} (x, ϕ (t))

is obtained.^1,2 For example, a raw signal is divided into multiple phase angle ranges based on a reference signal, and the signals in each phase angle range are averaged to construct a phase-averaged signal.² However, reference signals are generally not always available. As a method that does not require a reference signal, an optimal frequency-domain filter deduced from power-spectral estimates is used to extract

\tilde{f} (x, ϕ (t))

⁠.³ These methods are useful for data measured by a point measurement technique such as a hot-wire anemometer and a laser Doppler velocimeter (LDV). The proper orthogonal decomposition (POD) is used to extract a coherent structure for the analysis of spatial two-dimensional (2D) data measured by techniques such as particle image velocimetry (PIV)⁴ and Schlieren method.⁵ The extracted coherent structure reconstructed by truncated POD modes was similar to that obtained by the phase-averaged fields.⁴ For example, the averaged pressure and acceleration structures are estimated from PIV data based on this method.⁶ It has also been applied to the analysis of the coherent velocity fluctuation of swirling turbulent jets.⁷ The triple decomposition based on the optimal mode decomposition (OMD)⁸ is proposed to decompose a flow field consisting of large and small coherent structures.⁹ The OMD method is a generalization of the dynamic mode decomposition (DMD) method¹⁰ because the OMD method optimizes a subspace basis, whereas the DMD method uses the POD modes as a subspace basis. This method is applied to the analysis of the effect of free-stream turbulence on the near-field growth of the near wake of a cylinder.¹¹

As a flow measurement method obtaining spatial 2D data, the pressure-sensitive paint (PSP) method^12–15 has been applied to various fields of aerodynamics such as aerospace engineering,^16–20 fluid machinery,^21–25 automotive engineering,^26–28 and micro-engineering.^29–31 The PSP method enables the measurement of the pressure on the surface to which the PSP coating is applied. The method utilizes oxygen quenching of photoluminescence, and the pressure is measured by the variation of the emission intensity of the PSP coating containing luminescent dye. Since the variation in the emission intensity is small for a small pressure variation, the detection of the small pressure variation is difficult due to the relatively large camera noise compared with the variation of the emission intensity. Therefore, the PSP data contain significant noise. For PSP data, by explicitly describing the measurement noise

n (x, t)

at the position vector

x

and time

t

⁠, we can decompose pressure

p (x, t)

as

p (x, t) = \bar{p} (x) + \tilde{p} (x, ϕ (t)) + p' (x, t) + n (x, t),

(3)

where

\bar{p} (x)

is the mean value for time,

\tilde{p} (x, ϕ (t))

is the contribution of the organized coherent motion,

ϕ (t)

is the phase angle at time

t

⁠, and

p' (x, t)

is the stochastic part. The difficulty with the decomposition for PSP data are to extract

\tilde{p} (x, ϕ (t))

and

p' (x, t)

from the large measurement noise of

n (x, t)

⁠. The POD and DMD analysis can extract a large flow structure, and several methods using POD and DMD are proposed to reduce the noise from time-series PSP data.^32–37 These methods enable the selection of the modes to reconstruct the pressure distribution with reduced noise. For example, the empirical selection of POD modes with small noise³² or mode selection by sparse modeling have been proposed.^35,37 Although these denoising methods effectively remove the noise, to the best of our knowledge, there are no reports on the application of POP and DMD to the triple decomposition for PSP measurement data. This is because even the first few POD/DMD modes with large energy contributions contain the noise;^32,35,36 thus, it is difficult to separate a large-scale structure from the noise. Moreover, data representing a large flow structure reconstructed from these modes do not always coincide with phase-averaged data. As a phase-average method for the PSP method, the conditional image sampling (CIS) method is proposed;³⁸ the CIS method requires the acquisition of a reference signal. In our previous study, we proposed a phase-average method leveraging a time-series clustering method without using a reference signal.³⁹

In this study, we conduct the triple decomposition for noisy PSP data. We estimate $\tilde{p} (x, ϕ (t))$ using a simple phase-average method based on our previously proposed method.³⁹ Then, the stochastic part $p' (x, t)$ is extracted, and the noise $n (x, t)$ is removed using the POD analysis and sparse modeling. After conducting the triple decomposition, we estimate optimal sensor positions to effectively represent the distribution using Fujitsu computing as a service digital annealer.^40,41 Sparse sensing is realized at optimal sensor positions, which are determined to be the solution to the optimal sensor placement problem.^27,42–45 As a demonstration, we apply the proposed method to the PSP data measuring the pressure distribution of the Kármán vortex in our previous studies.^46,47 The time-series pressure data are triple-decomposed and reconstructed as noise-suppressed data. The reconstructed pressure data are compared with those measured by a pressure transducer. This triple decomposition without using a reference signal is useful for analyzing flow field data with significant noise, such as the PSP data.

II. PROPOSED METHOD FOR TRIPLE DECOMPOSITION

The pressure distribution $p (x, t)$ measured by the PSP method can be decomposed into Eq. (3). First, the mean distribution for time is calculated. Second, the phase-averaged distribution is extracted as the coherent motion. Third, the stochastic part is obtained by calculating $p (x, t) - \bar{p} (x) - \tilde{p} (x, ϕ (t))$ ⁠. Finally, the noise of the stochastic part data is reduced. The details of the proposed method are provided in this section.

A. Calculation of mean distribution for time

Theoretically, the mean value of the pressure distribution

\bar{p} (x)

is defined as

\bar{p} (x) = lim_{T \to \infty} \frac{1}{T} \int_{0}^{T} p (x, t) d t .

(4)

However, pressure

p (x, t)

is measured over a time interval of

Δ t

by a camera in the PSP method, and the measurement time is always limited. The mean value of the measured pressure is calculated as follows:

\bar{p} (x) = \frac{1}{N} \sum_{n = 0}^{N} p (x, t_{n}),

(5)

where

t_{n}

is the measured time represented as

t_{n} = n Δ t

and

n = 0, 1, 2, \dots, N

⁠. The total number of measurements (images for the PSP method) is represented by

N

⁠.

B. Estimation of coherent motion

The pressure

\tilde{p} (x, ϕ (t))

organized coherent motion is estimated by a phase-average method. In our previous study, we proposed a phase-average method that does not use a reference signal.³⁹ The brief explanation of the method is provided in this paper. The cosine similarities between each data point are calculated, and the cosine similarity

cos θ_{i, j}

is defined as

cos θ_{i, j} = \frac{⟨p (x, t_{i}), p (x, t_{j})⟩}{{‖p (x, t_{i})‖}_{2} {‖p (x, t_{j})‖}_{2}},

(6)

where

⟨p (x, t_{i}), p (x, t_{j})⟩

is the inner product of

p (x, t_{i})

and

p (x, t_{j})

⁠, and

{‖p (x, t_{i})‖}_{2}

is the

ℓ 2

norm of

p (x, t_{i})

⁠. The integers

i

and

j

are within the range of

0 \leq i \leq N

and

0 \leq j \leq N

⁠, respectively. Since the pressure

p (x, t_{i})

are usually a 2D array,

p (x, t_{i})

is reshaped into a column vector in this similarity calculation. Then, the measured pressure data are presented on a 2D scatterplot using multi-dimensional scaling (MDS) based on the distance metric

|sin (θ_{i, j} / 2)|

between

p (x, t_{i})

and

p (x, t_{j})

⁠, where

|sin (θ_{i, j} / 2)|

is the absolute value of

sin (θ_{i, j} / 2)

⁠. The data on the circle with a radius of

1 / 2

in the MDS plot represent a periodic phenomenon. However, the data not on the circle are out of the periodic one. In our previous study, the data were classified as a combinatorial optimization problem using Fujitsu computing as a service digital annealer. As a simplified method, we manually classify the data for each phase angle in this study. Then, the phase-averaged data

{\tilde{p}}_{M} (x, ϕ_{m})

are obtained by averaging the data in each phase angle range. The subscript

M

denotes the number of divisions per period. When one period is divided into

M

subperiods, the range of each phase angle is

2 π / M,

and the

m

th phase angle

ϕ_{m}

⁠, which is defined as the center of each range, is written as

ϕ_{m} = (2 m - 1) \frac{π}{M},

(7)

where

m

is the integer within the range of

1 \leq m \leq M

⁠.

C. Estimation of the stochastic part

The sum of the stochastic part

p' (x, t_{n})

and noise

n (x, t)

at time

t_{n} = n Δ t

is calculated as

p^{'} (x, t_{n}) + n (x, t_{n}) = p (x, t_{n}) - \bar{p} (x) - \tilde{p} (x, ϕ (t_{n})) .

(8)

The problem is to determine the phase angle

ϕ (t_{n})

because a reference signal is not used in this study. Moreover, the phase-averaged data

{\tilde{p}}_{M} (x, ϕ_{m})

are obtained at the discrete phase angles of

ϕ_{m}

⁠. Following the POD/OMD approach proposed by Baj et al.,⁹ which focuses on flow velocity and not pressure,

\tilde{p} (x, ϕ (t_{n}))

is estimated by the minimization problem as follows:

\tilde{p} (x, ϕ (t_{n})) = \underset{\tilde{p} (x, ϕ (t_{n}))}{argmin} {‖p (x, t_{n}) - \bar{p} (x) - \tilde{p} (x, ϕ (t_{n}))‖}_{2},

(9)

\tilde{p} (x, ϕ (t_{n})) = \sum_{k} c_{i} (t_{n}) ψ_{k} (x),

(10)

where

ψ_{k} (x)

is the

k

th POD/OMD mode and

c_{i} (t_{n})

is the coefficient at time

t_{n}

corresponding to the mode

ψ_{k} (x)

⁠. However, this least squares method will result in overfitting for noisy data. That is, this method will estimate a part of

p^{'}

as a part of

\tilde{p}

⁠.

In this study, we propose an alternative method. We extend the discrete phase-averaged data using the following method: First,

{\tilde{p}}_{M} (x, ϕ_{m})

is reshaped to a column vector, which is written as

{\tilde{p}}_{M} (ϕ_{m})

⁠. Second, the matrix

{\tilde{P}}_{M}

with the column vectors arranged side-by-side is considered as follows:

{\tilde{P}}_{M} = [{\tilde{p}}_{M} (ϕ_{1}) {\tilde{p}}_{M} (ϕ_{2}) \dots {\tilde{p}}_{M} (ϕ_{M})] .

(11)

The singular value decomposition (SVD) of

\tilde{P}

is written as

{\tilde{P}}_{M} = U Σ V^{⊤},

(12)

where

U

and

V

are unitary matrices and the superscript

⊤

indicates the transpose. The matrix

Σ

is a diagonal matrix whose diagonal elements are singular values. The columns of

U

are left singular vectors that correspond to the POD modes. The columns of

V

are right singular vectors representing the phase history of each left singular vector. Third, fitting the amplitude variations

V^{⊤}

with a curve, the estimated data with an increased number of divisions

L (M < L)

per period can be obtained. In other words, we can estimate super-resolution data in the phase direction of the phase-averaged data as

{\tilde{P}}_{L} = U Γ,

(13)

where

Γ

is the matrix representing the amplitudes determined by the singular value

Σ

and the interpolated values by curve fitting. The number of the divisions

L

is determined by calculating the average of the residuals between

{\tilde{p}}_{L} (ϕ_{l})

and the next phase range

{\tilde{p}}_{L} (ϕ_{l + 1})

⁠, where

l

is the integer within the range of

1 \leq l \leq L

⁠. Then, the average of the residuals

R

is calculated as

R = \frac{1}{q L} {‖{\tilde{P}}_{L + 1} - {\tilde{P}}_{L}‖}_{F}^{2},

(14)

{\tilde{P}}_{L} = [{\tilde{p}}_{L} (ϕ_{1}) {\tilde{p}}_{L} (ϕ_{2}) \dots {\tilde{p}}_{L} (ϕ_{L})],

(15)

{\tilde{P}}_{L + 1} = [{\tilde{p}}_{L} (ϕ_{L}) {\tilde{p}}_{L} (ϕ_{1}) \dots {\tilde{p}}_{L} (ϕ_{L - 1})],

(16)

where

q

is the number of pixels of the image and the subscript

F

indicates the Frobenius norm. It is expected that

R

will converge for a sufficiently large

L

⁠. Finally,

\tilde{p} (x, ϕ (t_{n}))

is selected as

{\tilde{p}}_{L} (x, ϕ_{l})

⁠, which is equivalent to

{\tilde{p}}_{L} (ϕ_{l})

in the reshaped vector form, with the highest cosine similarity to

p (x, t_{n}) - \bar{p} (x)

⁠.

The stochastic part

p' (x, t_{n})

is separated from the noise

n (x, t_{n})

⁠. The sum of the stochastic part

p' (x, t_{n})

and noise

n (x, t)

is denoted as

\hat{p} (x, t_{n}) = p^{'} (x, t_{n}) + n (x, t_{n}) .

(17)

Here, we again use SVD. The sum

\hat{p} (x, t_{n})

is reshaped to a column vector

\hat{p} (t_{n})

⁠, and the matrix

\hat{P} = [\hat{p} (0) \hat{p} (Δ t) \hat{p} (2 Δ t) \dots \hat{p} (n Δ t) \dots \hat{p} (N Δ t)]

is considered. The SVD of

\hat{P}

is written as

\hat{P} = \hat{U} \hat{Σ} {\hat{V}}^{⊤},

(18)

where

\hat{U}

and

\hat{V}

are unitary matrices and

Σ

represents singular values. The stochastic part

p' (t_{n})

⁠, which is the reshaped form of

p' (x, t_{n})

⁠, is extracted by truncating the SVD; thus, the stochastic part

p' (t_{n})

is written as

p' (t_{n}) = \hat{U} α (t_{n}),

(19)

where

α (t_{n})

is the amplitude vector. The amplitude

α (t_{n})

is estimated using sparse modeling based on the least absolute shrinkage and selection operator (LASSO).^48–50 However, the estimation of

α (t_{n})

for all pixels of all measured images results in high computational cost. In our previous studies,^37,51 we proposed an estimation method for the pressure distribution

p (x, t_{n})

based on the selected optimal sensor points determined as the optimal sensor placement problem. Here, we propose the optimal points for detecting the stochastic part

p' (t_{n})

⁠. The same algorithm using Fujitsu computing as a service digital annealer⁵¹ can be applied, except that the pressure distribution is replaced by the stochastic part

p' (t_{n})

⁠. We briefly introduce the algorithm for determining optimal points. More details are provided in our previous study.⁵¹ The idea of the method is that the large variation of the POD mode at an optimal point well represents the characteristic of the distribution, and points with similar variation of the mode do not need to be selected. We consider the row vector

{\hat{u}}_{i}

of

\hat{U} \hat{Σ}

⁠, and

{\hat{u}}_{i}

represents the variation of the mode at point

i

⁠. The weight

w ({\hat{u}}_{i}, {\hat{u}}_{j})

between points

i

and

j

is defined as

w ({\hat{u}}_{i}, {\hat{u}}_{j}) = |{\hat{u}}_{i}| |{\hat{u}}_{j}| |sin θ|,

(20)

where

θ

is the angle between

{\hat{u}}_{i}

and

{\hat{u}}_{j}

⁠. The optimal sensor points are selected by solving the maximum clique problem from the undirected graph

G_{c}

⁠, whose vertices have the weights larger than

c

⁠. We solve the independent set problem of the complement graph

{\bar{G}}_{c}

of

G_{c}

⁠, because a maximum clique problem is equivalent to an independent set problem of the complement graph in general. This problem is known as an NP-hard problem and is solved by Fujitsu computing as a service digital annealer. The independent set is obtained by solving the following minimization problem:

min - \sum_{ζ \in {\bar{G}}_{c}} x_{ζ} + Λ \sum_{(ζ, η) \in Ω} x_{ζ} x_{η},

(21)

where

x_{ζ} = 0

or

1

for the

ζ

th vertex belonging to or not to

{\bar{G}}_{c}

⁠, respectively. The set

Ω

represents the edge set of

{\bar{G}}_{c}

⁠. Then, the amplitude

α (t_{n})

is determined by the following minimization problem of LASSO:³⁷

α (t_{n}) = \underset{α (t_{n})}{argmin} \frac{1}{2 q} {‖S \hat{p} (t_{n}) - S \hat{U} α (t_{n})‖}_{2}^{2} + λ {‖α (t_{n})‖}_{1},

(22)

where

{‖α (t_{n})‖}_{1}

is the

ℓ 1

norm of

α (t_{n})

and

S

is the matrix indicating optimal sensor positions. The element of

S

is

1

for the sensor position and

0

for other positions. Then, the stochastic part

p' (t_{n})

is estimated by Eq. using the calculated

α (t_{n})

⁠.

III. RESULTS AND DISCUSSION

A. Triple mode decomposition for PSP data

We applied the proposed method to the PSP data obtained in our previous study.^46,47 The data were the pressure distribution induced by the Kármán vortex behind a square cylinder. The Reynolds number was $1.1 \times 10^{5}$ ⁠. The time interval of $Δ t$ was $1.0 ms$ ⁠. Before applying the proposed method, the PSP distributions were processed by a $5 \times 5$ smoothing spatial filter. A typical example of the processed data are shown in Fig. 1. As shown in the figure, the noise was still large for further analysis.

FIG. 1.

Typical pressure distribution data measured by the PSP method, where pressure p is normalized by an atmospheric pressure pref. A 5×5 smoothing spatial filter is applied to original data. Reproduced with permission from Inoue et al., Phys. Fluids 33, 077105 (2021); licensed under a Creative Commons Attribution (CC BY) license.37

View large Download slide

Typical pressure distribution data measured by the PSP method, where pressure $p$ is normalized by an atmospheric pressure $p_{ref}$ ⁠. A $5 \times 5$ smoothing spatial filter is applied to original data. Reproduced with permission from Inoue et al., Phys. Fluids 33, 077105 (2021); licensed under a Creative Commons Attribution (CC BY) license.³⁷

Next, we calculated the cosine similarity for the 21 800 data points by Eq. (6). Since the data contained large amounts of noise, a truncated-SVD of rank five was used to reduce the noise. It is noted that these truncated-SVD data were only used for the calculation of the cosine similarity not for the following phase-averaging calculation. The MDS plot is shown in Fig. 2. The data on the circle with a radius of $1 / 2$ shown in the red dashed line represents a periodic phenomenon; thus, we obtained the phase-averaged data by dividing the data on the circle in 12 subsets (⁠ $M = 12$ ⁠) and averaging each subset. Here, the number of the divisions per period $M = 12$ was determined because the amount of data were not enough to reduce the noise by averaging for some phase range for larger $M$ ⁠. The phase-averaged data are shown in Fig. 3, where the phase angle $ϕ_{m}$ (⁠ $m = 1, 2, \dots, 12$ ⁠) is defined in Eq. (7). In this figure, the phase-averaged distributions of $\bar{p} (x) + {\tilde{p}}_{M = 12} (x, ϕ_{m})$ are presented for ease of visibility. The noise was significantly reduced by averaging.

FIG. 2.

View large Download slide

MSD plot based on cosine similarity between measurement data. The data on the circle with a radius of 1/2 shown in red dashed line represent a periodic phenomenon.

FIG. 3.

Phase-averaged pressure data p¯x+p̃M=12x,ϕm, where the pressure is normalized by the atmospheric pressure pref and ϕn indicates the n th phase.

View large Download slide

Phase-averaged pressure data $\bar{p} (x) + {\tilde{p}}_{M = 12} (x, ϕ_{m})$ ⁠, where the pressure is normalized by the atmospheric pressure $p_{ref}$ and $ϕ_{n}$ indicates the $n$ th phase.

Based on these discreate phase-averaged data, the super-resolution data in the phase direction of the phase-averaged data were estimated. The SVD of the phase-averaged data was calculated as Eq. . As an example, the first six POD modes (column vectors of

U

⁠) are shown in Fig. 4. We employed the first four modes for estimating the super-resolution data based on the threshold determination method proposed by Donoho and Gavish.⁵² The amplitudes for the first and third POD modes are shown in Fig. 5. The discrete amplitudes for phase angle were interpolated by following function:

v_{h} (ϕ) = \sum_{k = 1}^{2} a_{k} sin (b_{k} ϕ + c_{k}),

(23)

where

v_{h} (ϕ)

is the amplitude for

h

th mode at phase

ϕ

⁠. For the first mode,

a_{1} = 0.41

⁠,

b_{1} = 1.00

⁠,

c_{1} = 0.61

⁠,

a_{2} = 0.01

⁠,

b_{2} = 3.00

⁠, and

c_{2} = - 1.00

⁠. The first mode contains the fundamental and third harmonics. However, the effect of the third harmonic is small. For the third mode,

a_{1} = - 0.01

⁠,

b_{1} = 0.97

⁠,

c_{1} = - 1.11

⁠,

a_{2} = - 0.41

⁠,

b_{2} = 2.00

⁠, and

c_{2} = - 2.10

⁠. The third mode contains the fundamental and second harmonics. Then, the number of divisions per period was determined to be

L = 96

based on Eq. , where

R

converged almost to

2.0 \times 10^{- 8}

⁠. Then, the super-resolution data were estimated by Eq. . Then,

\tilde{p} (x, ϕ (t_{n}))

was selected as

{\tilde{p}}_{L} (x, ϕ_{l})

with the maximum cosine similarity to

p (x, t_{n}) - \bar{p} (x)

⁠. Here, the amplitude of

{\tilde{p}}_{L} (x, ϕ_{l})

was also determined by maximizing the similarity because the amplitude of the vortex varied. Figure 6 shows the selected phase for

p (x, t_{n}) - \bar{p} (x)

over time up to

128 ms

⁠. As shown in the figure, the selected phase-averaged image changed periodically over time. Figures 7 and 8 show typical images of

\hat{p} (x, t_{n})

shown in Eq. and the first six POD modes of

\hat{p} (x, t_{n})

⁠, respectively. While the significant noise was observed on

\hat{p} (x, t_{n})

⁠, the noise on the POD modes is small due to the SVD calculation based on a large number of images (21 800 images). Based on the POD modes, we calculated optimal sensor positions by Eq. and show an example of sensor positions in Fig. 9. The optimal sensors were placed where the vortices passed. This trend was similar to the results calculated based on the POD mode of the Kármán vortex in previous study.⁵¹ Then, the stochastic part

p' (x, t_{n})

was extracted from

\hat{p} (x, t_{n})

by Eq. as shown in Fig. 10. Finally, the noise reduced pressure distribution was obtained by calculating the sum

\bar{p} (x) + \tilde{p} (x, ϕ (t_{n})) + p' (x, t_{n})

as shown in Fig. 11, and the pressure distributions were successfully reconstructed by the method. The twin vortices aligned just behind both ends of the square cylinder as shown at

t_{6}

in Fig. 11 were sometimes observed. In such a case, the similarity was small for any

{\tilde{p}}_{L} (x, ϕ_{l})

and the amplitude of

{\tilde{p}}_{L} (x, ϕ_{l})

was also small. This correlation between the similarity and amplitude was clearly observed as shown in Fig. 12. This indicates that the contribution of

{\tilde{p}}_{L} (x, ϕ_{l})

is small for the phenomena that deviate from the periodic coherent motion.

FIG. 4.

View large Download slide

First six POD modes of phase-averaged pressure data.

FIG. 5.

View large Download slide

Typical example of amplitude of first and third POD modes. The horizontal and vertical axes are phase and amplitude, respectively. The fitting curves are used for interpolating the data between discrete phases.

FIG. 6.

Typical example of selected phase for px,tn−p¯x over time up to 128 ms. The horizontal and vertical axes are time and phase, respectively.

View large Download slide

Typical example of selected phase for $p (x, t_{n}) - \bar{p} (x)$ over time up to $128 ms$ ⁠. The horizontal and vertical axes are time and phase, respectively.

FIG. 7.

Typical example of the extracted stochastic part containing noise p̂x,tn, where pref is the atmospheric pressure.

View large Download slide

Typical example of the extracted stochastic part containing noise $\hat{p} (x, t_{n})$ ⁠, where $p_{ref}$ is the atmospheric pressure.

FIG. 8.

View large Download slide

First six POD modes of the stochastic part.

FIG. 9.

View large Download slide

Example of optimal sensor positions calculated based on the POD mode of the stochastic part. The number of sensors is 25.

FIG. 10.

View large Download slide

Typical example of the denoised stochastic part $p' (x, t_{n})$ ⁠, where $p_{ref}$ is the atmospheric pressure.

FIG. 11.

Typical example of the denoised pressure distribution px,tn, where pref is the atmospheric pressure. The pressure px,tn is obtained by p¯x+p̃Lx,ϕl+p′x,tn.

View large Download slide

Typical example of the denoised pressure distribution $p (x, t_{n})$ ⁠, where $p_{ref}$ is the atmospheric pressure. The pressure $p (x, t_{n})$ is obtained by $\bar{p} (x) + {\tilde{p}}_{L} (x, ϕ_{l}) + p^{'} (x, t_{n})$ ⁠.

FIG. 12.

View large Download slide

Relation between the maximum value of similarity and amplitude. The amplitude is small for small similarity resulting in a small contribution of coherent motion ${\tilde{p}}_{L} (x, ϕ_{l})$ to pressure distribution $p (x, t_{n})$ ⁠.

B. Sparse sensing at optimal sensor points

As a data-driven approach, the pressure distribution can be reconstructed from the pressure data at the optimal sensor points estimated by the POD modes of $\hat{p} (x, t_{n})$ based on triple decomposition. The mean value $\bar{p} (x)$ ⁠, organized coherent motion ${\tilde{p}}_{L} (x, ϕ_{l})$ ⁠, and optimal sensor positions $S$ were given/pre-calculated as a prior knowledge. The reconstruction of the pressure distribution was conducted by the following method. First, we calculated the cosine similarities between the pressure data at optimal sensor positions and ${\tilde{p}}_{L} (Sx, ϕ_{l})$ (⁠ $l = 1, 2, \dots, L$ ⁠) and selected ${\tilde{p}}_{L} (x, ϕ_{l_{max}})$ at the phase $ϕ_{l_{max}}$ that gives the maximum similarity. Second, $\hat{p} (x, t_{n})$ was calculated using $\bar{p} (x)$ and ${\tilde{p}}_{L} (x, ϕ_{l_{max}})$ ⁠. Third, the SVD of $\hat{p} (x, t_{n})$ was calculated and the amplitudes of the POD modes were determined following Eq. (22) to extract $p^{'} (x, t_{n})$ using Eq. (19) from the noisy data of $\hat{p} (x, t_{n})$ ⁠. Finally, the pressure distributions were obtained by $\bar{p} (x) + {\tilde{p}}_{L} (x, ϕ_{l_{max}}) + p^{'} (x, t_{n})$ ⁠.

The root mean square error (RMSE) between the pressure data measured by a pressure transducer through a pressure tap and the reconstructed data are shown in Fig. 13. The position of the pressure tap is shown in Fig. 13(a). For comparison, the RMSE results obtained by the method based on the previous study^37,51 were also shown. The pressure distributions were reconstructed from the POD modes calculated from the observed data $p (x, t)$ ⁠, and the amplitudes of these POD modes were determined by LASSO at the optimal sensor positions that were determined by these POD modes in the previous method. That is, the difference between these methods is whether the sensor position is determined based on the POD modes of $\hat{p} (x, t_{n})$ ⁠. The horizontal axis in Fig. 13(b) represents the number of POD modes used to reconstruct the pressure distribution, and this was varied by varying $λ$ in LASSO. The RMSE was calculated for $128$ time steps, and the standard deviation of the number of modes used for each time step is shown in the figure. The results for the optimal sensor points of 25 and 60 are shown. The RMSE values were small for a small number of optimal sensor points and POD modes in the proposed method. This result indicates that triple decomposition enables a sparser representation.

FIG. 13.

View large Download slide

Comparison between the proposed and previous methods for pressure reconstruction performance. (a) The position of the pressure tap. (b) The RMSE between reconstructed pressure and pressure measured at pressure tap. The horizontal axis is the number of POD modes used to reconstruct the pressure distribution.

IV. CONCLUSIONS

We conducted the triple decomposition of noisy pressure-sensitive paint (PSP) data that are decomposed to the mean value, organized coherent motion, and stochastic part. We proposed a relatively simple method based on the multi-dimensional scaling (MDS) plot of the cosine similarity to construct phase-averaged data representing an organized coherent motion for periodic flows. Since the constructed phase-averaged data are typically discrete with respect to the phase angle direction, we produced the data with a sufficient phase angle step size by interpolating the data between the phase angles. Then, the stochastic part can be extracted by selecting phase-averaged data with an appropriate phase angle based on the similarity between the measurement data and phase-averaged data. The noise of the stochastic part was suppressed by superposing the proper orthogonal decomposition (POD) modes, whose amplitudes were determined by the least absolute shrinkage and selection operator (LASSO) at optimal sensor positions. We used Fujitsu computing as a service digital annealer for the estimation of optimal sensor positions based on the POD modes of the stochastic part. In this manner, the triple decomposition of the noisy PSP data was performed. The application of PSP measurements is expected to expand further, and triple decomposition will be very useful.

The pressure distribution was successfully reconstructed from the pressures at the optimal sensor positions using the phase-averaged data and the POD modes of the stochastic part. The reconstructed pressures agreed well with those measured by a pressure transducer, even when the number of sensors was small compared with that in the previous study. As a result, a sparser representation was achieved using the triple decomposition.

ACKNOWLEDGMENTS

A part of this work was supported by the Collaborative Research Project (J23I041 and J24I064) and the Low Turbulence Wind Tunnel Facility of the Advance Flow Experimental Research Center at the Institute of Fluid Science, Tohoku University. The authors express their gratitude to Dr. Yasuhumi Konishi, Mr. Hiroyuki Okuizumi, and Mr. Yuya Yamazaki for their assistance during the wind tunnel testing. We also gratefully appreciate Tayca Corporation for providing the titanium dioxide.

AUTHOR DECLARATIONS

Conflict of Interest

The authors declare the following competing interest: Takahiro Kashikawa and Koichi Kimura are employees of Fujitsu Ltd. All other authors declare no competing interest.

Author Contributions

Koyo Kubota: Conceptualization (supporting); Data curation (equal); Formal analysis (equal); Investigation (lead); Methodology (supporting); Software (lead); Validation (equal); Visualization (equal). Makoto Takagi: Investigation (supporting); Methodology (supporting); Software (supporting). Tsubasa Ikami: Data curation (supporting); Investigation (supporting); Resources (equal); Validation (supporting); Writing – review & editing (supporting). Yasuhiro Egami: Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Resources (equal); Writing – review & editing (supporting). Hiroki Nagai: Investigation (supporting); Resources (equal); Writing – review & editing (supporting). Takahiro Kashikawa: Methodology (supporting); Software (supporting); Validation (supporting). Koichi Kimura: Methodology (supporting); Software (supporting); Validation (supporting). Yu Matsuda: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Funding acquisition (lead); Investigation (supporting); Methodology (lead); Project administration (lead); Resources (equal); Software (supporting); Supervision (lead); Validation (equal); Visualization (equal); Writing – original draft (lead); Writing – review & editing (lead).

DATA AVAILABILITY

The part of the flow measurement dataset is available in Zenodo at https://doi.org/10.5281/zenodo.10215642 (Ref. 47).

REFERENCES

1.

A. K. M. F.

Hussain

and

W. C.

Reynolds

, “

The mechanics of an organized wave in turbulent shear flow

,”

J. Fluid Mech.

41

,

241

(

1970

).

https://doi.org/10.1017/S0022112070000605

Google Scholar

Crossref

2.

B.

Cantwell

and

D.

Coles

, “

An experimental study of entrainment and transport in the turbulent near wake of a circular cylinder

,”

J. Fluid Mech.

136

,

321

(

1983

).

https://doi.org/10.1017/S0022112083002189

Google Scholar

Crossref

3.

G. J.

Brereton

and

A.

Kodal

, “

A frequency-domain filtering technique for triple decomposition of unsteady turbulent flow

,”

J. Fluids Eng.

114

,

45

(

1992

).

https://doi.org/10.1115/1.2909998

Google Scholar

Crossref

4.

R.

Perrin

,

M.

Braza

,

E.

Cid

,

S.

Cazin

,

F.

Moradei

,

A.

Barthet

,

A.

Sevrain

, and

Y.

Hoarau

, “

Near-wake turbulence properties in the high Reynolds number incompressible flow around a circular cylinder measured by two- and three-component PIV

,”

Flow Turbul. Combust.

77

,

185

(

2006

).

https://doi.org/10.1007/s10494-006-9043-5

Google Scholar

Crossref

5.

M.

Yang

,

G.

Wang

,

M.

Guo

,

Y.

Tian

,

Z.

Zhong

,

M.

Xu

,

L.

Li

,

J.

Le

, and

H.

Zhang

, “

Intelligent flow field reconstruction based on proper orthogonal decomposition dimensionality reduction and improved multi-branch convolution fusion

,”

Phys. Fluids

35

,

115106

(

2023

).

https://doi.org/10.1063/5.0174660

Google Scholar

Crossref

6.

R.

Shanmughan

,

P.-Y.

Passaggia

,

N.

Mazellier

, and

A.

Kourta

, “

Coherent pressure and acceleration estimation from triply decomposed turbulent bluff-body wakes

,”

Exp. Fluids

65

,

53

(

2024

).

https://doi.org/10.1007/s00348-024-03774-4

Google Scholar

Crossref

7.

A. S.

Lobasov

,

S. V.

Alekseenko

,

D. M.

Markovich

, and

V. M.

Dulin

, “

Mass and momentum transport in the near field of swirling turbulent jets. Effect of swirl rate

,”

Int. J. Heat Fluid Flow

83

,

108539

(

2020

).

https://doi.org/10.1016/j.ijheatfluidflow.2020.108539

Google Scholar

Crossref

8.

A.

Wynn

,

D. S.

Pearson

,

B.

Ganapathisubramani

, and

P. J.

Goulart

, “

Optimal mode decomposition for unsteady flows

,”

J. Fluid Mech.

733

,

473

(

2013

).

https://doi.org/10.1017/jfm.2013.426

Google Scholar

Crossref

9.

P.

Baj

,

P. J. K.

Bruce

, and

O. R. H.

Buxton

, “

The triple decomposition of a fluctuating velocity field in a multiscale flow

,”

Phys. Fluids

27

,

075104

(

2015

).

https://doi.org/10.1063/1.4923744

Google Scholar

Crossref

10.

P. J.

Schmid

, “

Dynamic mode decomposition of numerical and experimental data

,”

J. Fluid Mech.

656

,

5

–

28

(

2010

).

https://doi.org/10.1017/S0022112010001217

Google Scholar

Crossref

11.

K. S.

Kankanwadi

and

O. R. H.

Buxton

, “

Influence of freestream turbulence on the near-field growth of a turbulent cylinder wake: Turbulent entrainment and wake meandering

,”

Phys. Rev. Fluids

8

,

034603

(

2023

).

https://doi.org/10.1103/PhysRevFluids.8.034603

Google Scholar

Crossref

12.

T.

Liu

,

J. P.

Sullivan

,

K.

Asai

,

C.

Klein

, and

Y.

Egami

,

Pressure and Temperature Sensitive Paints

(

Springer

,

Cham

,

2021

).

Google Scholar

Crossref

13.

J. H.

Bell

,

E. T.

Schairer

,

L. A.

Hand

, and

R. D.

Mehta

, “

Surface pressure measurements using luminescent coatings

,”

Annu. Rev. Fluid Mech.

33

,

155

(

2001

).

https://doi.org/10.1146/annurev.fluid.33.1.155

Google Scholar

Crossref

14.

J. W.

Gregory

,

K.

Asai

,

M.

Kameda

,

T.

Liu

, and

J. P.

Sullivan

, “

A review of pressure-sensitive paint for high-speed and unsteady aerodynamics

,”

Proc. Inst. Mech. Eng., Part G

222

,

249

(

2008

).

https://doi.org/10.1243/09544100JAERO243

Google Scholar

Crossref

15.

D.

Peng

and

Y.

Liu

, “

Fast pressure-sensitive paint for understanding complex flows: From regular to harsh environments

,”

Exp. Fluids

61

,

8

(

2020

).

https://doi.org/10.1007/s00348-019-2839-6

Google Scholar

Crossref

16.

K.

Nakakita

,

M.

Kurita

,

K.

Mitsuo

, and

S.

Watanabe

, “

Practical pressure-sensitive paint measurement system for industrial wind tunnels at JAXA

,”

Meas. Sci. Technol.

17

,

359

(

2006

).

https://doi.org/10.1088/0957-0233/17/2/017

Google Scholar

Crossref

17.

Y.

Egami

,

C.

Klein

,

U.

Henne

,

K.

de Groot

,

J. B.

Meyer

,

C.-P.

Krückeberg

, and

F.

Boden

,

In-Flight Application of Pressure Sensitive Paint

(

Springer

,

Berlin

,

2013

).

Google Scholar

Crossref

18.

Y.

Sugioka

,

S.

Koike

,

K.

Nakakita

,

D.

Numata

,

T.

Nonomura

, and

K.

Asai

, “

Experimental analysis of transonic buffet on a 3D swept wing using fast-response pressure-sensitive paint

,”

Exp. Fluids

59

,

108

(

2018

).

https://doi.org/10.1007/s00348-018-2565-5

Google Scholar

Crossref

19.

D.

Yorita

,

U.

Henne

, and

C.

Klein

, “

Time-resolved pressure-sensitive paint measurements for cryogenic wind tunnel tests

,”

CEAS Aeronaut. J.

15

,

37

(

2023

).

https://doi.org/10.1007/s13272-022-00637-8

Google Scholar

Crossref

20.

T.

Lutz

,

J.

Kleinert

,

A.

Waldmann

,

L.

Koop

,

D.

Yorita

,

G.

Dietz

, and

M.

Schulz

, “

Research initiative for numerical and experimental studies on high-speed stall of civil aircraft

,”

J. Aircr.

60

,

623

(

2023

).

https://doi.org/10.2514/1.C036829

Google Scholar

Crossref

21.

J. W.

Gregory

, “

Resolving vortex-induced pressure fluctuations on a cylinder in rotor wake using fast-responding pressure-sensitive paint

,”

Phys. Fluids

31

,

055106

(

2019

).

https://doi.org/10.1063/1.5092944

Google Scholar

Crossref

22.

K. J.

Disotell

,

D.

Peng

,

T. J.

Juliano

,

J. W.

Gregory

,

J. W.

Crafton

, and

N. M.

Komerath

, “

Single-shot temperature- and pressure-sensitive paint measurements on an unsteady helicopter blade

,”

Exp. Fluids

55

,

1671

(

2014

).

https://doi.org/10.1007/s00348-014-1671-2

Google Scholar

Crossref

23.

N.

Ge

,

R.

Li

,

L.

Gao

,

L.

Wang

, and

B.

Ouyang

, “

An accumulated imaging method with phase-locking for rotor pressure-sensitive paint measurements

,”

Exp. Fluids

64

,

113

(

2023

).

https://doi.org/10.1007/s00348-023-03653-4

Google Scholar

Crossref

24.

D.

Peng

,

L.

Jiao

,

Y.

Yu

,

Y.

Liu

,

T.

Oshio

,

T.

Kawakubo

, and

A.

Yakushiji

, “

Single-shot lifetime-based PSP and TSP measurements on turbocharger compressor blades

,”

Exp. Fluids

58

,

127

(

2017

).

https://doi.org/10.1007/s00348-017-2416-9

Google Scholar

Crossref

25.

Z.

Dong

,

F.

Gu

,

R.

Chen

,

J.

Yang

,

W.

Zhou

,

Y.

Liu

, and

D.

Peng

, “

Three-dimensional pressure field measurement on turbine guide vanes using a dynamic distortion correction endoscopic pressure-sensitive paint technique

,”

Exp. Fluids

64

,

97

(

2023

).

https://doi.org/10.1007/s00348-023-03643-6

Google Scholar

Crossref

26.

T.

Yamashita

,

H.

Sugiura

,

H.

Nagai

,

K.

Asai

, and

K.

Ishida

, “

Pressure-sensitive paint measurement of the flow around a simplified car model

,”

J. Vis.

10

,

289

(

2007

).

https://doi.org/10.1007/BF03181696

Google Scholar

Crossref

27.

R.

Inoba

,

K.

Uchida

,

Y.

Iwasaki

,

T.

Nagata

,

Y.

Ozawa

,

Y.

Saito

,

T.

Nonomura

, and

K.

Asai

, “

Optimization of sparse sensor placement for estimation of wind direction and surface pressure distribution using time-averaged pressure-sensitive paint data on automobile model

,”

J. Wind Eng. Ind. Aerodyn.

227

,

105043

(

2022

).

https://doi.org/10.1016/j.jweia.2022.105043

Google Scholar

Crossref

28.

R.

Inoba

,

K.

Uchida

,

Y.

Iwasaki

,

K.

Yamada

,

A.

Jebli

,

T.

Nagata

,

Y.

Ozawa

, and

T.

Nonomura

, “

Improved estimation of yaw angle and surface pressure distribution of Ahmed model with optimized sparse sensors by Bayesian framework based on pressure-sensitive paint data

,”

Exp. Therm Fluid Sci.

156

,

111210

(

2024

).

https://doi.org/10.1016/j.expthermflusci.2024.111210

Google Scholar

Crossref

29.

C.

Huang

,

J. W.

Gregory

, and

J. P.

Sullivan

, “

Microchannel pressure measurements using molecular sensors

,”

J. Microelectromech. Syst.

16

,

777

(

2007

).

https://doi.org/10.1109/JMEMS.2007.892914

Google Scholar

Crossref

30.

Y.

Matsuda

,

T.

Uchida

,

S.

Suzuki

,

R.

Misaki

,

H.

Yamaguchi

, and

T.

Niimi

, “

Pressure-sensitive molecular film for investigation of micro gas flows

,”

Microfluid. Nanofluid.

10

,

165

(

2011

).

https://doi.org/10.1007/s10404-010-0664-6

Google Scholar

Crossref

31.

C.-Y.

Huang

,

Z.-C.

Lin

, and

M.-K.

Wu

, “

Pressure-sensitive paint measurements for microscale supersonic flow with wedge models

,”

Aerosp. Sci. Technol.

127

,

107713

(

2022

).

https://doi.org/10.1016/j.ast.2022.107713

Google Scholar

Crossref

32.

M.

Pastuhoff

,

D.

Yorita

,

K.

Asai

, and

P. H.

Alfredsson

, “

Enhancing the signal-to-noise ratio of pressure sensitive paint data by singular value decomposition

,”

Meas. Sci. Technol.

24

,

075301

(

2013

).

https://doi.org/10.1088/0957-0233/24/7/075301

Google Scholar

Crossref

33.

D.

Peng

,

S.

Wang

, and

Y.

Liu

, “

Fast PSP measurements of wall-pressure fluctuation in low-speed flows: Improvements using proper orthogonal decomposition

,”

Exp. Fluids

57

,

45

(

2016

).

https://doi.org/10.1007/s00348-016-2130-z

Google Scholar

Crossref

34.

M. Y.

Ali

,

A.

Pandey

, and

J. W.

Gregory

, “

Dynamic mode decomposition of fast pressure sensitive paint data

,”

Sensors

16

,

862

(

2016

).

https://doi.org/10.3390/s16060862

Google Scholar

Crossref

PubMed

35.

X.

Wen

,

Y.

Liu

,

Z.

Li

,

Y.

Chen

, and

D.

Peng

, “

Data mining of a clean signal from highly noisy data based on compressed data fusion: A fast-responding pressure-sensitive paint application

,”

Phys. Fluids

30

,

097103

(

2018

).

https://doi.org/10.1063/1.5046681

Google Scholar

Crossref

36.

T.

Nonomura

,

H.

Shibata

, and

R.

Takaki

, “

Extended-Kalman-filter-based dynamic mode decomposition for simultaneous system identification and denoising

,”

PLoS One

14

,

e0209836

(

2019

).

https://doi.org/10.1371/journal.pone.0209836

Google Scholar

Crossref

PubMed

37.

T.

Inoue

,

Y.

Matsuda

,

T.

Ikami

,

T.

Nonomura

,

Y.

Egami

, and

H.

Nagai

, “

Data-driven approach for noise reduction in pressure-sensitive paint data based on modal expansion and time-series data at optimally placed points

,”

Phys. Fluids

33

,

077105

(

2021

).

https://doi.org/10.1063/5.0049071

Google Scholar

Crossref

38.

D.

Yorita

,

H.

Nagai

,

K.

Asai

, and

T.

Narumi

, “

Unsteady PSP technique for measuring naturally-disturbed periodic phenomena

,” AIAA Paper No. 2010-307,

2010

.

Google Scholar

39.

T.

Inoue

,

K.

Kubota

,

T.

Ikami

,

Y.

Egami

,

H.

Nagai

,

T.

Kashikawa

,

K.

Kimura

, and

Y.

Matsuda

, “

Clustering method for time-series images using quantum-inspired digital annealer technology

,”

Commun. Eng.

3

,

10

(

2024

).

https://doi.org/10.1038/s44172-023-00158-0

Google Scholar

Crossref

40.

M.

Aramon

,

G.

Rosenberg

,

E.

Valiante

,

T.

Miyazawa

,

H.

Tamura

, and

H. G.

Katzgraber

, “

Physics-inspired optimization for quadratic unconstrained problems using a digital annealer

,”

Front. Phys.

7

,

48

(

2019

).

https://doi.org/10.3389/fphy.2019.00048

Google Scholar

Crossref

41.

S.

Matsubara

,

M.

Takatsu

,

T.

Miyazawa

,

T.

Shibasaki

,

Y.

Watanabe

,

K.

Takemoto

, and

H.

Tamura

, “

Digital annealer for high-speed solving of combinatorial optimization problems and its applications

,” in

25th Asia and South Pacific Design Automation Conference (ASP-DAC)

(

2020

).

Google Scholar

Crossref

42.

S.

Joshi

and

S.

Boyd

, “

Sensor selection via convex optimization

,”

IEEE Trans. Signal Process.

57

,

451

(

2009

).

https://doi.org/10.1109/TSP.2008.2007095

Google Scholar

Crossref

43.

K.

Manohar

,

B. W.

Brunton

,

J. N.

Kutz

, and

S. L.

Brunton

, “

Data-driven sparse sensor placement for reconstruction: demonstrating the benefits of exploiting known patterns

,”

IEEE Control Syst. Mag.

38

,

63

(

2018

).

https://doi.org/10.1109/MCS.2018.2810460

Google Scholar

Crossref

44.

R.

Herzog

,

I.

Riedel

, and

D.

Uciński

, “

Optimal sensor placement for joint parameter and state estimation problems in large-scale dynamical systems with applications to thermo-mechanics

,”

Optim. Eng.

19

,

591

(

2018

).

https://doi.org/10.1007/s11081-018-9391-8

Google Scholar

Crossref

45.

Y.

Saito

,

T.

Nonomura

,

K.

Yamada

,

K.

Asai

,

Y.

Sasaki

, and

D.

Tsubakino

, “

Determinant-based fast greedy sensor selection algorithm

,”

IEEE Access

9

,

68535

(

2021

).

https://doi.org/10.1109/ACCESS.2021.3076186

Google Scholar

Crossref

46.

Y.

Egami

,

A.

Hasegawa

,

Y.

Matsuda

,

T.

Ikami

, and

H.

Nagai

, “

Ruthenium-based fast-responding pressure-sensitive paint for measuring small pressure fluctuation in low-speed flow field

,”

Meas. Sci. Technol.

32

,

024003

(

2021

).

https://doi.org/10.1088/1361-6501/abb916

Google Scholar

Crossref

47.

K.

Kubota

,

T.

Inoue

,

T.

Ikami

,

Y.

Egami

,

H.

Nagai

, and

Y.

Matsuda

(

2023

). “Flow measurement dataset: Kármán vortex street measured by pressure-sensitive paint technique,”

Zenodo.

https://doi.org/10.5281/zenodo.10215642

48.

R.

Tibshirani

, “

Regression shrinkage and selection via the Lasso

,”

J. R. Stat. Soc. Ser. B

58

,

267

(

1996

).

https://doi.org/10.1111/j.2517-6161.1996.tb02080.x

Google Scholar

Crossref

49.

T.

Hastie

,

R.

Tibshirani

, and

J. H.

Friedman

,

The Elements of Statistical Learning: Data Mining, Inference, and Prediction

(

Springer

,

2009

).

Google Scholar

Crossref

50.

T.

Hastie

,

R.

Tibshirani

, and

M.

Wainwright

,

Statistical Learning with Sparsity: The Lasso and Generalizations

(

CRC Press

,

2015

).

Google Scholar

Crossref

51.

T.

Inoue

,

T.

Ikami

,

Y.

Egami

,

H.

Nagai

,

Y.

Naganuma

,

K.

Kimura

, and

Y.

Matsuda

, “

Data-driven optimal sensor placement for high-dimensional system using annealing machine

,”

Mech. Syst. Signal Process.

188

,

109957

(

2023

).

https://doi.org/10.1016/j.ymssp.2022.109957

Google Scholar

Crossref

52.

M.

Gavish

and

D. L.

Donoho

, “

The optimal hard threshold for singular values is

4 / \sqrt{3}

,”

IEEE Trans. Inf. Theory

60

,

5040

(

2014

).

https://doi.org/10.1109/TIT.2014.2323359

Google Scholar

Crossref

Triple decomposition and sparse representation for noisy pressure-sensitive paint data

I. INTRODUCTION

II. PROPOSED METHOD FOR TRIPLE DECOMPOSITION

A. Calculation of mean distribution for time

B. Estimation of coherent motion

C. Estimation of the stochastic part

III. RESULTS AND DISCUSSION

A. Triple mode decomposition for PSP data

B. Sparse sensing at optimal sensor points

IV. CONCLUSIONS

ACKNOWLEDGMENTS

AUTHOR DECLARATIONS

Conflict of Interest

Author Contributions

DATA AVAILABILITY

REFERENCES

Citing articles via

Submit your article

Sign up for alerts

Resources

Explore

pubs.aip.org

Connect with AIP Publishing

Triple decomposition and sparse representation for noisy pressure-sensitive paint data

I. INTRODUCTION

II. PROPOSED METHOD FOR TRIPLE DECOMPOSITION

A. Calculation of mean distribution for time

B. Estimation of coherent motion

C. Estimation of the stochastic part

III. RESULTS AND DISCUSSION

A. Triple mode decomposition for PSP data

B. Sparse sensing at optimal sensor points

IV. CONCLUSIONS

ACKNOWLEDGMENTS

AUTHOR DECLARATIONS

Conflict of Interest

Author Contributions

DATA AVAILABILITY

REFERENCES

Citing articles via

Submit your article

Sign up for alerts

Resources

Explore

pubs.aip.org

Connect with AIP Publishing

This Feature Is Available To Subscribers Only