The axisymmetric model is used to generate T_{p{\mathrm{~and~}}}\ T_{{p}\infty} data for the parameters shown in Table 10.1.
Data, to five significant figures, are obtained as shown in Table 10.2.
Use these data and first-order Tikhonov regularization to compute the correction kernel vector H. Use a Tikhonov regularization coefficient \alpha_{T i k}= 10.
Table 10.1 | Simulation parameters for Example 10.1. | ||
Sensor depth, E, mm | 0.75 | Sensor radius, R, mm | 0.375 |
Wire length, L, mm | 20 | Domain radius, R_{D}, mm | 8 |
k_{w},\,\mathrm{W/m}-{K} | 25 | k, W/m-K | 0.625 |
C_{w},\,\mathrm{J/m^3}-{K} | 2.1E6 | C,\mathrm{J/m^{3}{\mathrm{K}}} | 4.2E6 |
q_{o},\,\mathrm{W/m^{2}} | 1E5 |
Table 10.2 | Sensor and undisturbed temperature data for heating with q=c o n s t=10^{5}\,\mathrm{W/m^{2}}. | ||||
t,s | T_{p},{C} | T_{p},{C} | t,s | T_{p},{C} | T_{p},{C} |
1 | 7.796 | 21.66 | 11 | 118.95 | 220.16 |
2 | 20.404 | 50.63 | 12 | 128.31 | 234.05 |
3 | 33.028 | 76.399 | 13 | 137.46 | 247.41 |
4 | 45.188 | 99.423 | 14 | 146.4 | 260.28 |
5 | 56.856 | 120.35 | 15 | 155.16 | 272.72 |
6 | 68.074 | 139.64 | 16 | 163.74 | 284.76 |
7 | 78.892 | 157.62 | 17 | 172.16 | 296.43 |
8 | 89.354 | 174.51 | 18 | 180.44 | 307.78 |
9 | 99.498 | 190.48 | 19 | 188.56 | 318.81 |
10 | 109.35 | 205.66 | 20 | 196.56 | 329.57 |
The sensitivity matrix, X, is constructed from the differences between the entries of the T_{p} vector, prepended with a “0” (just like the Δϕ values used to compute the sensitivity matrix for heat flux estimation). The first few values of this vector of differences are:
\left\lfloor7.7962-0\ \ 20.404-7.7962\ \ 33.028-20.404\ \ 45.188-33.028\ \cdots\right\rfloor ^T= \left\lfloor7.7962\\ 0\ \ 12.608\ \ 12.624\ \ 12.160\ \cdots\right\rfloor ^T (10.6)
These values are used to create the 20 × 20 matrix \mathbf{X}_{\phi_{p}}. The first few rows of this matrix are
{\bf X}_{\phi_{p}}= \begin{bmatrix} 7.7962 & 0 & 0 & 0 & 0 & 0 \cdots \\ 12.608 & 7.7962 & 0 & 0 & 0 & 0 \cdots \\12.624 & 12.608 & 7.7962 & 0 & 0 & 0 \cdots\\ 12.160 & 12.624 & 12.608 & 7.7962 & 0 & 0 \cdots\\ \vdots & \vdots &\ddots &\ddots &\ddots &\ddots \end{bmatrix}_{20\times 20} (10.7)
The \mathbf{X}_{\phi_{p}} matrix is used with first-order Tikhonov regularization to generate the filter matrix:
\mathbf{F}_{\phi_{p}}=\left(\mathbf{X}_{\phi_{p}}^{T}\mathbf{X}_{\phi_{p}}+\alpha_{\mathrm{Tik}}\mathbf{H}_{1}^{T}\mathbf{H}_{1}\right)^{-1}\mathbf{X}_{\phi_{p}}^{T} (10.8)
The first few entries of this matrix are:s
\mathbf{F}_{\phi_{p}}=\begin{bmatrix} 0.0584 & 0.0368 & − 0.0027 & − 0.0033 & 0.0002 & 0.0003\ \cdots \\ − 0.0576 & 0.0180 & 0.0369 & 0.0007 & − 0.0033 & − 0.0001 \cdots \\ − 0.0041 & − 0.0562 & 0.0207 & 0.0370 & 0.0005 & -0.0033 \cdots\\ 0.0056 & − 0.0003 & − 0.0563 & 0.0204 & 0.0370 & 0.0005 \cdots\\ \vdots & \vdots &\vdots &\vdots &\vdots &\vdots &\vdots & \end{bmatrix}_{20\times 20} (10.9)
The \mathbf{F}_{\phi_{p}} matrix multiplies into the “data” of \mathbf{T}_{p\infty}-\mathbf{T}_{p} in the inversion of Eq. (10.4)
\mathbf{T}_{p\infty\mathbf{-T}p}=\mathbf{X}_{\phi_{p}}\mathbf{H} (10.4)
to find the kernel vector H:
{\bf H}=\left({\bf X}_{\phi_{p}}^{T}{\bf X}_{\phi_{p}}+\alpha_{\mathrm{Tik}}{\bf H}_{1}^{T}{\bf H}_{1}\right)^{-1}{\bf X}_{\phi_{p}}^{T}\left\{{\bf T}_{p\infty}-{\bf T}_{p}\right\} =\mathbf{F}_{\phi_{p}}\{\mathbf{T}_{p\infty}-\mathbf{T}_{p}\} (10.10)
The result of these calculations is the following sequence:
Discussion:
Unlike many examples in this text, these calculations are carried out using dimensional data. Units on the entries of \mathbf{X}_{\phi_{p}} are Kelvin, and those on \mathbf{F}_{\phi_{p}} are K^{−1}. The correction kernel H is dimensionless.
Selection of the proper amount of regularization to use in the inversion is difficult. In this example, the value of \alpha_{T i k} = 10 was specified. This is a relatively large value in comparison to those used in solution of the IHCP in Chapter 7. In practice, the regularization should be large enough to produce a smooth H(t) curve. Experience suggests values from a rather broad range of the regularization parameter produce similar results.
Figure 10.4 displays results for H obtained using a range of regularization parameters. All of the values are much larger than used for IHCP solution. The results for the \alpha_{T i k}= 1 display irregularities in the result, indicating the regularization coefficient is too small. Results for \alpha_{T i k}= 100 are very smooth, but values are seen diminished compared to the nominal curve.
Another tool for selection of the regularization coefficient is consideration of a known heat flux input into the simulation model. See Example 10.2.
H_{1-7} | 1.6613 | 1.1935 | 0.9624 | 0.879 | 0.8156 | 0.7598 | 0.7151 |
H_{8-15} | 0.678 | 0.6468 | 0.6195 | 0.5945 | 0.5726 | 0.5532 | 0.5362 |
H_{16-20} | 0.5199 | 0.5053 | 0.4902 | 0.4775 | 0.4663 | 0.4574 |