Optimal sensor placement in hydropower house based on improved triaxial effective independence method

Ji-jian LIAN, Long-jun HE, Hai-jun WANG*

State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University, Tianjin 300072, P. R. China

Optimal sensor placement in hydropower house based on improved triaxial effective independence method

Ji-jian LIAN, Long-jun HE, Hai-jun WANG*

State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University, Tianjin 300072, P. R. China

The capacity and size of hydro-generator units are increasing with the rapid development of hydroelectric enterprises, and the vibration of the powerhouse structure has increasingly become a major problem. Field testing is an important method for research on dynamic identification and vibration mechanisms. Research on optimal sensor placement has become a very important topic due to the need to obtain effective testing information from limited test resources. To overcome inadequacies of the present methods, this paper puts forward the triaxial effective independence driving-point residue (EfI3-DPR3) method for optimal sensor placement. The EfI3-DPR3 method can incorporate both the maximum triaxial modal kinetic energy and linear independence of the triaxial target modes at the selected nodes. It was applied to the optimal placement of triaxial sensors for vibration testing in a hydropower house, and satisfactory results were obtained. This method can provide some guidance for optimal placement of triaxial sensors of underground powerhouses.

triaxial sensor; optimal placement; EfI3-DPR3 method; hydropower house; vibration testing

1 Introduction

The hydropower house is the center of an electricity generation hub. It is an integrated infrastructural component, made up of the hydraulic structure, hydraulic machinery, and electrical equipment. It is also a complex coupled system of water flow, units, and the powerhouse (Lian et al. 2007). The vibration of hydropower units and houses is controlled increasingly strictly as the plant scale increases, the operation of the power station gains stability, and the working conditions improve. Field testing is an important method for research on dynamic identification and vibration mechanisms. Research on optimal sensor placement has become a very important topic due to the need to obtain effective testing results from limited test resources. The optimal sensor placement is usually applied to simple structures such as trusses, frames, plates, beams, and pillars to carry out health monitoring anddamage diagnosis. Research has been carried out in the field of civil engineering, bridges, and machinery. However, there have been fewer studies on hydraulic structures, especially on the hydropower house (Ma et al. 2006). Thus, it is very significant to introduce the theory of optimal sensor placement to the hydropower house and other hydraulic structures, and it is also needed for research on health monitoring and dynamic identification of hydropower house structures.

The most important stage is the selection of appropriate optimization algorithms for optimal sensor placement. At present, sensor placement optimization algorithms can be generally divided into traditional algorithms and non-traditional algorithms. Traditional algorithms mainly include the effective independence (EfI) method (Kammer 1991), kinetic energy method (Chung and Moore 1993; Worden and Burrows 2001; Heo et al. 1997), QR decomposition method (Schedlinski and Link 1996), and modal assurance criterion (MAC) method (Carne and Dohrmann 1995). All of these algorithms have their limitations (Swann and Chattopadhyay 2006). Non-traditional algorithms mainly consist of the genetic algorithm, particle swarm algorithm, and simulated annealing algorithm. But these algorithms are only the transformations of traditional algorithms without essential changes, and their search capability and reliability are always unsatisfactory.

In the field vibration test of a hydropower house, triaxial displacement or acceleration sensors are always needed to be placed at many selected nodes. On the one hand, placing the triaxial sensors at these selected nodes is more convenient for field operation. On the other hand, the triaxial sensors can monitor the spatial vibration of concerned parts more comprehensively. But, using the optimization algorithms mentioned above, we can only select the sensors in each direction independently, leading to the result that the selected sensors can only meet the optimal rules in one or two directions in most cases, so the requirement of optimal triaxial sensor placement cannot be satisfied. Because of this, an optimal placement method for triaxial sensors is needed. Kammer and Tinker (2004) introduced the triaxial effective independence (EfI3) method based on the EfI method, which simultaneously maximized the linear independence of the target modes in three directions of triaxial sensors. However, the nodes selected with this method were not always distributed at the locations with more modal kinetic energy, and this led to the loss of effective information in field tests with strong environmental noise.

Based on the EfI3 method, this paper proposes the triaxial effective independence driving-point residue (EfI3-DPR3) method in order to improve low modal kinetic energy at the nodes selected with the EfI3 method. The EfI3-DPR3 method used the driving-point residue coefficient reflecting the weight of the triaxial kinetic energy to improve the EfI3 method, and can simultaneously maximize the linear independence of triaxial target modes and triaxial modal kinetic energy at selected locations for the triaxial sensors. Finally, the method was applied to the optimal placement of triaxial sensors for vibration test in an underground powerhouse.

2 Methods of optimal triaxial sensor placement

2.1 EfI3 method

The core concept of EfI method is reserving the nodes which have greater contribution to the linear independence of the target modes so that more modal information can be obtained in the condition of limited sensors. In general, candidate nodes have six associated degrees of freedom: three translations and three rotations. However, the usual practice is to place sensors at selected nodes to measure translational motion. The EfI3 method is a means of seeking the linear independence of the triaxial modes for triaxial sensors based on the EfI method. By means of a finite element model (FEM), the target modal matrix of each node can be partitioned into three rows according to three degrees of translational freedom. Therefore, the problem of triaxial sensor placement is ascribed to the estimation of the corresponding Fisher information matrix (FIM) given by

whereQ3iis the triaxial FIM of theith node,?3iis the triaxial target modal matrix of theith node, andWis a weighting matrix.Wis set to be an identity matrix for simplicity. Thus, Eq. (1) can be expressed in the fo llowing form:

whereNis the number of candidate nodes, andQis the total triaxial FIM. Modal identification is equivalent to estimating the modal response based on the measured sensor data. Maximization ofQresults in the minimization of the corresponding error covariance matrix, which results in the best estimate of the modal response. Based on this decomposition,Q3′i, the total triaxial FIM with theith node being deleted, has a determinant that can be expressed in terms of the determinant ofQin the following form:

whereI3is a 3×3 identity matrix, andE3ican be expressed as

where?3i1,?3i2, and?3i3are the mode shapes of theith node in three orthogonal directions.E3iis a 3×3 fully dense matrix whose diagonal elements containing the modal contribution values of theith node in three orthogonal directions. Note that the determinant ofQ3′idecreases to zero and the target modes are no longer independent whenE3ihas an eigenvalue ofλ=1.0. This indicates that theith node is vital to the independence of the target modes. Therefore, at least one of the three degrees of freedom associated with theith node must have a modal contribution value of 1.0. Furthermore, the EfI3 method can be defined by thefollowing equation:

whereFirepresents the fractional change in the determinant of the total triaxial FIM if theith node is deleted from the candidate nodes. If theith node is vital to target mode independence, the correspondingFivalue is 1.0, and if theith node contributes nothing to the independence, itsFivalue is 0. Using this measure, the candidate nodes can be ranked and sorted, and the lowest ranked node is deleted. In an iterative way, the candidate set of nodes can be paired down to the desired number of triaxial sensors for modal test.

2.2 EfI-DPR3 method

The EfI3 method solves the optimal placement problem of triaxial sensors. But the nodes selected with this method are not always distributed at locations with more modal kinetic energy, and this method is likely to lead to the loss of effective information in the field test with strong environmental noise. Therefore, based on research on the driving-point residue coefficient of uniaxial sensors (Imamovic 1998; Wu et al. 2006), this paper presents the EfI3-DPR3 method, which adopts the triaxial driving-point residue coefficient reflecting the weight of the triaxial kinetic energy to improve the EfI3 method.

In order to avoid the loss of effective information, the sum of all contributing modes of theith node, which reflected the modal kinetic energy of theith node, was defined in the following expression:

whereeiis the weighted sum of all contributing modes of theith node,mis the number of concerned modes,ωjis the frequency ofjth target mode, andψijis thejth mode of theith node in the uniaxial state. While uniaxial sensors are to be placed, the uniaxial driving-point residue coefficient is defined as

whereCiis the uniaxial driving-point residue coefficient of theith node.

Therefore, the triaxial driving-point residue coefficient for triaxial sensor placement can be expressed as

whereC3iis the triaxial driving-point residue coefficient of theith node;Cxi,Cyi, andCziare the driving-point residue coefficients of theith node in thex,y, andzdirections, respectively; andψxij,ψyij, andψzijare thejth mode o f theith node in thex,y, andzdirections, respectively.

Through Eq. (5) and Eq. (8), this paper proposes the EfI3-DPR3 method expressed by the following equation:

whereis the improved triaxial effective independent distribution value of theith node.FiandC3irepresent the linear independence of the target modes and the triaxial modal kinetic energy of theith node, respectively. Thus,can incorporate both linear independence of the triaxial target modes and the improvement of signal-to-noise ratio at the selected nodes.

The steps are as follows for utilizing the EfI3-DPR3 method to implement the optimal placement of triaxial sensors: (1) the FEM of the study object is built and the modal analysis is conducted; (2) concerned modes and initial nodes are chosen, and?3iis obtained; (3) based on the EfI3 method,Fiis calculated with Eq. (5); (4)C3iis obtained with Eq. (8); (5)is obtained with Eq. (9); (6) all of the nodes are sorted with the values of(i=1, 2,…,N), and the node with the minimum value ofis deleted; (7) a new modal matrix is formed according to the residual candidate nodes, and steps (3) to (6) are repeated until the desired number of triaxial sensors for the modal test is obtained.

3 Optimal placement of triaxial sensors for hydropower house

Prototype observation is an effective method for study on the vibration behavior of the hydropower house and units. Sufficient test preparation, especially the scheme of sensor placement, is a prerequisite of successful vibration test. The locations and number of the sensors are the major factors affecting the test results. In previous vibration tests of the hydropower house, the sensors were always placed according to previous experience, so the selection of measurement points was blind and random. This study used the EfI3-DPR3 method for optimal placement of triaxial sensors in the hydropower house, using a large underground powerhouse in China as an example.

3.1 Research purpose and mode selection

During the operation of the underground powerhouse, some problems have occurred, including a bracket and beam crack, a turbine runner and flow guide vane crack, and a runner cone falling off. In order to determine the reason for the strong vibration on the bracket, beam, and floor, and to confirm vibration sources and dynamic response characteristics, the authors carried out several vibration tests. At that time, sensor placement mainly relied on site experience.

The FEM used one unit of the underground powerhouse as the research object. Due to the high stiffness of substructures such as the turbine foundation, only the upper structure above the turbine layer was simulated, and the bottom of the model adopted the consolidation constraints in order to focus on the vibration characteristics of the bracket, beam, and floor.The range of model simulation extended downward to the bottom of turbine layer (with an elevation of 1 007.5 m) and upward to the top of the generator layer (with an elevation of 1 019.0 m). The FEM adopted the faying surfaces of the outer concrete and rock of the spiral cases as the upstream and downstream boundaries, and the joints of units as the left and right boundaries, as illustrated in Fig. 1.

Fig. 1 FEM of upper structure of underground powerhouse

The model coordinates were in the Cartesian coordinate system. The origin of the coordinates was in the center of the units with an elevation of 1 012.5 m. Thex-axis was perpendicular to the flow direction. They-axis was parallel to the flow direction, and the upstream direction was positive. Thez-axis was consistent with the centerline of units, and the upward direction was positive. The whole system followed the right-hand screw rule. The FEM has a total of 8 097 elements and 10 090 nodes.

We obtained the first 50 modes according to the finite element modal analysis. The previous studies showed that when the natural vibration frequency is close to the excitation frequency induced by mechanical, hydraulic, or electromagnetic factors, and the vibration energy is sufficient, the powerhouse may generate resonance. Through the empirical formula, we know that the excitation frequency induced by the non-uniform flow field in the spiral case is 30.96 Hz, which is close to the frequencies of the 7th to 10th modes (29.08 Hz, 30.39 Hz, 31.03 Hz, and 34.75 Hz). This may be the main reason for strong vibration in the bracket, beam, and floor. Therefore, this study took brackets in the generator layer of the underground powerhouse as research objects, selected the 7th to 10th modes as concerned modes, and then made optimal sensor placement for the brackets in order to achieve the desired test effect and determine the reason for intensive vibration.

3.2 Selection of initial nodes

According to the finite element modal analysis, the brackets of the upper structure have a strong vibration response in the second and third quadrants. Thus, three mid-span brackets in this area were chosen for study and numbered I, II, and III, as shown in Fig. 2. In accordancewith the FEM grid partition, six candidate nodes were chosen in each bracket, and the candidate nodes for the three brackets were numbered from 1 to 18, as illustrated in Fig. 3. In the finished field vibration test, a triaxial displacement sensor was placed on the No. I bracket, and its position was close to node 1, shown in Fig. 3 with a black solid circle.

Fig. 2 FEM of brackets

Fig. 3 Candidate nodes of three brackets

3.3 Implementation and verification of EfI3-DPR3 method

On the basis of the principles described above, the EfI3 and EfI3-DPR3 methods were implemented in MATLAB (Zhang 2003). Through running the program with the mass-normalized triaxial modal matrix as input data, we obtainedFiandC3i. We then calculated, and deleted the candidate node with the minimum value ofin each iteration. With six candidate nodes remaining, the computation process is shown in Table 1. Node 3 and node 16 have the same value ofFi, but because node 3 has the higher value ofC3i, node 16 has the minimum value of. Therefore, node 16 is deleted in this iteration. Then, we can obtain a five-node placement scheme in which nodes 1, 3, 4, 13, and 15 are included.

Table 1 Computation process with EfI3-DPR3 method (six nodes remaining)

According to the different numbers of selected nodes, triaxial sensor placement schemes of the EfI3 and EfI3-DPR3 methods were obtained. Table 2 shows the triaxial sensor placement schemes for the vibration test of brackets when the numbers of remaining nodes are 1 to 18. It shows that the placement schemes of the two methods are the same when the numbers of remaining nodes are 2, 3, 4, 14, 15, 16, 17, and 18, and there are some differences with other numbers.

Table 2 Triaxial sensor placement schemes of EfI3 and EfI3-DPR3 methods

Corresponding evaluation criteria need to be built to judge the quality of the sensor placement algorithms. In view of the limited locations and strong environmental noise for the powerhouse field test, this study selected the 2-norm of the total triaxial FIM and the triaxial modal strain energy (TMSE) to judge the two sensor placement algorithms. The 2-norm of the total triaxial FIM reflects the contribution degree of selected nodes to the linear independence of target modes. While the value of the contribution degree is higher, the scheme is better under this evaluation criterion. The scheme with higher modal strain energy can adapt to the field test with strong environmental noise better, and improve the signal-to-noise ratio when the measurement of dynamic response is carried out. Based on the schemes in Table 2, the ratios of the 2-norm of total triaxial FIM of remaining nodes to the 2-norm of total triaxial FIM of initial nodes (the ratios of the 2-norm of FIM for short) were calculated, as illustrated in Fig. 4(a), and the ratios of the TMSE of remaining nodes to the TMSE of initial nodes (the ratios of the TMSE for short) were also calculated, as shown in Fig. 4(b).

Fig. 4(a) shows that the ratios of the 2-norm of FIM increase with the number of remaining nodes, and the ratios of the EfI3-DPR3 method are all larger than or equal to those of the EfI3 method, no matter what the number of remaining nodes is. Obviously, the EfI3-DPR3 method is better than the EfI3 method under this criterion. Fig. 4(b) reveals that the law of the ratio of the TMSE is the same as the law of the ratio of the 2-norm of FIM. Thus, the EfI3-DPR3 method is also better than the EfI3 method under the modal strain energy criterion.

Fig. 4 Comparison of ratios of 2-norm of FIM and ratios of TMSE between EfI3-DPR3 and EfI3 methods

To sum up, the EfI3-DPR3 method proposed in this study makes an improvement over the EfI3 method, and the solution determined by the EfI3-DPR3 method can provide reliable technical support for optimal placement of triaxial sensors for the powerhouse vibration test.

In addition, when the number of selected nodes is one, node 1 is the selected node determined by the EfI3-DPR3 method, and it is very close to the location of the triaxial senor placed in the finished field test. This indicates that the selection of nodes in the finished test is reasonable, the main vibration sources can be identified, and the dynamic vibration response can be evaluated with the test signals. Fig. 5 and Fig. 6 respectively show the vertical displacement signal and power spectrum obtained by the vertical sensor on the bracket of the selected unit under the conditions of a head of 145 m and a load of 200 MW.

Fig. 5 Vertical displacement signal of bracket under head of 145 m and load of 200 MW

Fig. 6 Power spectrum of vertical vibration of bracket under head of 145 m and load of 200 MW

Through analysis of Fig. 5, we can draw a conclusion that the standard deviation of vertical vibration displacement is 1.03 μm at the selected node of the bracket under theoperating condition. As shown in Fig. 6, the dominant vibration frequency of the bracket is about 31 Hz. This can also be obtained by the signals of thexandydirections. On the one hand, the phenomenon above reveals that the excitation frequency induced by the non-uniform flow field in the spiral case is the main vibration source, which induces the vibration of the brackets in many conditions. On the other hand, it also demonstrates that the EfI3-DPR3 method proposed in this paper is effective for optimal placement of triaxial sensors.

4 Conclusions

Based on the EfI3 method, this paper proposes the EfI3-DPR3 method in order to improve low modal strain energy at the nodes selected with the EfI3 method. The EfI3-DPR3 method used the driving-point residue coefficient reflecting the weight of the triaxial kinetic energy to improve the EfI3 method, and was applied to the optimal placement of triaxial sensors in a field test in an underground powerhouse to determine the reason for strong vibration on the brackets. The main conclusions are as follows:

(1) The triaxial sensor placement schemes were obtained based on the EfI3 and EfI3-DPR3 methods when the numbers of selected nodes were 1 to 18. The placement schemes of the two methods are the same when the numbers of selected nodes are 2, 3, 4, 14, 15, 16, 17, and 18, and there are some differences with other numbers.

(2) The study made a comparison between the quality of the EfI3-DPR3 and EfI3 methods judged by two common evaluation criteria of optimal sensor placement algorithms. The result shows that the EfI3-DPR3 method has certain advantages in ensuring the contribution to the linear independence of the triaxial target mode and anti-noise performance.

(3) The selected node in the finished field test is in accordance with the node selected with the EfI3-DPR3 method when the number of selected nodes was one. Based on the test signal analysis, it is concluded that the excitation frequency induced by the non-uniform flow field in the spiral case is the main vibration source, inducing the vibration of the brackets under many conditions.

(4) It is proved that the EfI3-DPR3 method proposed in this paper is effective for optimal placement of triaxial sensors. This method should be further studied for optimal placement of triaxial sensors in other types of hydraulic structures.

Carne, T. G., and Dohrmann, C. R. 1995. A modal test design strategy for modal correlation.Proceedings of the 13th International Modal Analysis Conference, 927-933. Bethel: SPIE-The International Society for Optical Engineering.

Chung, Y. T., and Moore, J. D. 1993. On-orbit sensor placement and system identification of space station with limited instrumentations.Proceedings of the 11th International Modal Analysis Conference, 41-46. Bethel: Society for Experimental Mechanics.

Heo, G., Wang, M. L., and Satpathi, D. 1997. Optimal transducer placement for health monitoring of long span bridge.Soil Dynamics and Earthquake Engineering, 16(7-8), 495-502. [doi:10.1016/S0267-7261 (97)00010-9]

Imamovic, N. 1998.Validation of Large Structural Dynamics Models Using Modal Test Data. Ph. D. Dissertation. London: Imperial College.

Kammer, D. C. 1991. Sensor placement for on-orbit modal identification and correlation of large space structures.Journal of Guidance, Control, and Dynamics, 14(2), 251-259.

Kammer, D. C., and Tinker, M. L. 2004. Optimal placement of triaxial accelerometers for modal vibration tests.Journal of Mechanical Systems and Signal Processing, 18(1), 29-41. [doi:10.1016/S0888-3270(03)00017-7]

Lian, J. J., Wang, H. J., and Qin, L. 2007.Hydropower House Structure Study. Beijing: China Water Power Press. (in Chinese)

Ma, Z. Y., Zhang, K. H., and Chen, J. 2006. Application of optimal sensor placement algorithms for vibration characteristics analysis of power house in hydropower station.Journal of Dalian University of Technology, 46(2), 262-265. (in Chinese)

Schedlinski, C., and Link, M. 1996. An approach to optimal pick-up and exciter placement.Proceedings of 14th International Modal Analysis Conference, 376-382. Bethel: Society for Experimental Mechanics.

Swann, C., and Chattopadhyay, A. 2006. Optimization of piezoelectric sensor location for delamination detection in composite laminates.Engineering Optimization, 38(5), 511-528. [doi:10.1080/ 03052150600557841]

Worden, K., and Burrows, A. P. 2001. Optimal sensor placement for fault detection.Engineering Structures, 23(8), 885-901. [doi:10.1016/S0141-0296(00)00118-8]

Wu, Z. Y., Dai, F. J., and Yang, H. F. 2006. Research on optimal sensor placement methods for structure health monitoring.Journal of Zhengzhou University (Engineering Science), 27(4), 92-96. (in Chinese)

Zhang, Z. Y. 2003.Master MATLAB 6.5. Beijing: Beihang University Press. (in Chinese)

(Edited by Yan LEI)

This work was supported by the National Natural Science Foundation of China (Grant No. 50909072), the New Teachers’ Fund for Doctor Station, the Ministry of Education of China (Grant No. 20090032120082), and the Communication Research Item for the West Area, the Ministry of Communications of China (Grant No. 2009328000084).

*Corresponding author (e-mail:bookwhj@tju.edu.cn)

Received Aug. 26, 2011; accepted Dec. 6, 2011