2.1 OTPA theory
Classical transfer path analysis (TPA) is a test-based method to seek the vibration or noise that the sound source transmits through the structure or air to a specified receiver, associated with the source contribution. The total response is as follows:
$$Y\left( \omega \right) = \mathop \sum \limits_{i} H_{i} F_{i} \left( \omega \right) + \mathop \sum \limits_{j} H_{j} Q_{j} \left( \omega \right),$$
(1)
where
\(F_{i} \left( {\upomega } \right)\) is the structural loads and
\(Q_{j} \left( {\upomega } \right)\) is the acoustic loads;
\(H_{i} \, {\text{and}} \, H_{j}\) are the transfer functions from the reference points to the receivers. The method of TPA focus on the measurement of the frequency response functions (FRF) and the load identification. However, some conditions limit the application of the TPA method. To avoid crosstalk between multiple transfer paths in the system, the frequency response functions of each transfer path between the excitation source point and the response point need to measure separately. Therefore, when there are multiple transmission paths between the source points and the response in the systems, the structure often needs to be disassembled. This also means a huge amount of testing work for a complex system, and since dismantling would change the boundary conditions of the system components. Moreover, the artificial excitation is not the actual excitation normally, and the accuracy of the frequency response function and the analyzed results could not be guaranteed.
The OTPA method makes up the deficiency of the traditional TPA. For this method could be used to measure the system's “source-transfer path-response” relationship and analyze the respective contribution in the actual operating condition of the system in the test, the process of load identification is eliminated. The test and analysis work are greatly simplified.
Based on the assumption of linear time-invariant system, the OTPA could expressed as
$$\varvec{Y} = \varvec{HX},$$
(2)
where
\({\varvec{Y}}\) is the output of the response point;
\({\varvec{X}}\) is the input of the excitation point;
\({\varvec{H}}\) is the matrix of the operational transmissibility. For vibration and noise transfer path analysis, the input and output of excitation source could be in the form of vibration, force or sound pressure signal, etc. By measuring the physical quantities of the excitation reference points and response points under the operational conditions of the system, the input and response sample matrices could be easily constructed as
$$\left( {\begin{array}{*{20}c} {\begin{array}{*{20}c} {a_{11} } & \cdots & {a_{1k} } \\ {a_{21} } & \cdots & {a_{2k} } \\ {a_{31} } & \cdots & {a_{3k} } \\ \end{array} } & {\begin{array}{*{20}c} {b_{11} } & \cdots & {b_{1m} } \\ {b_{21} } & \cdots & {b_{2m} } \\ {b_{31} } & \cdots & {b_{3m} } \\ \end{array} } \\ {\begin{array}{*{20}c} {a_{41} } & \cdots & {a_{4k} } \\ \vdots & \ddots & \vdots \\ {a_{n1} } & \cdots & {a_{nk} } \\ \end{array} } & {\begin{array}{*{20}c} {b_{41} } & \cdots & {b_{4m} } \\ \vdots & \ddots & \vdots \\ {b_{n1} } & \cdots & {b_{nm} } \\ \end{array} } \\ \end{array} } \right)\left( {\begin{array}{*{20}c} {H_{a1} } \\ \vdots \\ {\begin{array}{*{20}c} {H_{ak} } \\ {H_{b1} } \\ {\begin{array}{*{20}c} \vdots \\ {H_{bm} } \\ \end{array} } \\ \end{array} } \\ \end{array} } \right) = \left( {\begin{array}{*{20}c} {Y_{1} } \\ {Y_{2} } \\ {\begin{array}{*{20}c} {Y_{3} } \\ {Y_{4} } \\ {\begin{array}{*{20}c} \vdots \\ {Y_{n} } \\ \end{array} } \\ \end{array} } \\ \end{array} } \right),$$
(3)
where
n is the total number of working conditions,
k is the total number of vibration reference points, and
m is the total number of sound reference points;
\(Y_{n}\) is the response point under
nth working condition;
\(a_{nk}\) is the response of the
kth vibration reference point under the
nth working condition;
\(b_{nm}\) is the response of the
mth sound reference point under the
nth working condition;
\(H_{ak}\) is the transmissibility of the
kth vibration reference point to the target point; and
\(H_{bm}\) is the transmissibility of the
mth sound reference point to the target point.
Therefore, the transfer function matrix of the system under working condition is
$$\left( {\begin{array}{*{20}c} {H_{a1} } \\ \vdots \\ {\begin{array}{*{20}c} {H_{ak} } \\ {H_{b1} } \\ {\begin{array}{*{20}c} \vdots \\ {H_{bm} } \\ \end{array} } \\ \end{array} } \\ \end{array} } \right) = \left( {\begin{array}{*{20}c} {\begin{array}{*{20}c} {a_{11} } & \cdots & {a_{1k} } \\ {a_{21} } & \cdots & {a_{2k} } \\ {a_{31} } & \cdots & {a_{3k} } \\ \end{array} } & {\begin{array}{*{20}c} {b_{11} } & \cdots & {b_{1m} } \\ {b_{21} } & \cdots & {b_{2m} } \\ {b_{31} } & \cdots & {b_{3m} } \\ \end{array} } \\ {\begin{array}{*{20}c} {a_{41} } & \cdots & {a_{4k} } \\ \vdots & \ddots & \vdots \\ {a_{n1} } & \cdots & {a_{nk} } \\ \end{array} } & {\begin{array}{*{20}c} {b_{41} } & \cdots & {b_{4m} } \\ \vdots & \ddots & \vdots \\ {b_{n1} } & \cdots & {b_{nm} } \\ \end{array} } \\ \end{array} } \right)^{ - 1} \left( {\begin{array}{*{20}c} {Y_{1} } \\ {Y_{2} } \\ {\begin{array}{*{20}c} {Y_{3} } \\ {Y_{4} } \\ {\begin{array}{*{20}c} \vdots \\ {Y_{n} } \\ \end{array} } \\ \end{array} } \\ \end{array} } \right).$$
(4)
This matrix is non-square so the inverse is in a least square sense. The output contribution of each source and transfer path to the response point could be analyzed by matrix H quantitatively. Actually, there are often multiple energy transfer paths between the excitation point and response point in the application, which means the same input will generate responses on multiple paths/response points, namely crosstalk. Especially in the method of OTPA, the inter-coupling exists in the input and response matrixes, since the all the excitation and response are measured at the same time. Such crosstalk would lead to numerical problems in the process of matrix inversion, and the transfer relationship of each transfer channel could not be solved accurately. Therefore, the singular value decomposition (SVD) and principal component analysis (PCA) are applied in OTPA to achieve the CTC.
The response matrix of the reference point could be express as
$$\varvec{X} = {\varvec{{USV}}}^{\rm{T}},$$
(5)
where
X is an
r ×
m response matrix of the reference point,
U is an
r ×
m orthogonal unit matrix,
S is an
m ×
m singular value diagonal matrix, and
V is an
m ×
m orthogonal unit matrix. Thus, in the OTPA method, it is required that the number of test sample
r (number of working conditions) must be more than the number of input points
m.
After the SVD is performed on the input sample matrix
X to obtain the principal components (PCs) of the sample matrix, the rank of PCs is shown. In this process, the influence of background noise is reduced and the signal to noise ratio (SNR) is improved. Afterward, the PCs which have nothing to do with the response signal or have little influence can be considered as measurement bias or crosstalk, and it can be discarded and the corresponding singular value can be set to zero to improve the estimation accuracy of the transfer matrix. The PCA is used to reduce the influence of unwished frequency components, in order to reduce the crosstalk. The singular matrix
\({\varvec{S}}_{r}\) is expressed as:
$${ }{\varvec{S}}_{r} = \left[ {\begin{array}{*{20}c} {\begin{array}{*{20}c} {\sigma_{1} } & {} & {} \\ {} & \ddots & {} \\ {} & {} & {\sigma_{r} } \\ \end{array} } & 0 \\ 0 & {\begin{array}{*{20}c} 0 & {} & {} \\ {} & \ddots & {} \\ {} & {} & 0 \\ \end{array} } \\ \end{array} } \right].$$
(6)
The input pseudo-inverse matrix is
$${\varvec{X}}^{ + } = {\varvec{VS}}_{r}^{ - 1} {\varvec{U}}^{{\text{T}}} .$$
(7)
And, the achieved operational transfer relation matrix is
$${\varvec{H}}_{r} = {\varvec{VS}}_{r}^{ - 1} {\varvec{U}}^{{\text{T}}}\varvec{Y}.$$
(8)
Substituting Eq. (
8) into Eq. (2), the contribution of each reference point and total synthesized response signal can be derived as
$$\left( {\begin{array}{*{20}c} {\begin{array}{*{20}c} {a_{n1} } & \cdots & {a_{nk} } \\ \end{array} } & {b_{n1} } & {\begin{array}{*{20}c} \cdots & {b_{nm} } \\ \end{array} } \\ \end{array} } \right)\left( {\begin{array}{*{20}c} {H_{a1} } \\ {\begin{array}{*{20}c} \vdots \\ {H_{ak} } \\ {H_{b1} } \\ \end{array} } \\ {\begin{array}{*{20}c} \vdots \\ {H_{bm} } \\ \end{array} } \\ \end{array} } \right) = \mathop \sum \limits_{1}^{k} y_{na} + \mathop \sum \limits_{1}^{m} y_{nb} = Y_{n} ,$$
(9)
where
\(y_{na}\) and
\(y_{nb}\) represent the contribution of vibration and sound reference point under the
nth operational condition.
Before the OTPA analysis, the selection of calculation working conditions should be carried out first. When calculating working conditions, the number of working conditions should be guaranteed not less than the number of paths in order to ensure the invertibility of the transfer coefficient matrix, so as to improve the statically indeterminate degree of linear equations and improve the solution accuracy.