# Structural Health Monitoring of Dams Based on Acoustic Monitoring, Deep Neural Networks, Fuzzy Logic and a CUSUM Control Algorithm

^{*}

## Abstract

**:**

## 1. Introduction

#### 1.1. Dams and Their Pathologies

- Hydraulic failures due to exceptional levels: includes, for example, overtopping and subsequent external erosion due to a spillway with insufficient discharge capacity, or even associated with gate damage or operating errors.
- Mass movements due to exceptional loads (except for the floods included in the previous item), inadequate material properties, or undetected geological singularities: includes, for example, slope instability (by limit equilibrium), deformations leading to overtopping, soil liquefaction, foundation instabilities, rapid subsidence associated with upstream slope sliding, and sliding of hillsides into the reservoir leading to overtopping.
- Internal erosion: includes, among others, development of piping in the dam core and erosion of foundation soils or joints (filling of discontinuities).

#### 1.1.1. Internal Erosion

#### 1.1.2. Sensors Available to Monitor Internal Erosion

#### 1.1.3. Monitoring Dams’ Acoustic Data

#### 1.2. Time-Lapse Monitoring and Control Algorithms for Scenario Classifications

- $\delta $ is the mean difference between the postchange and prechange regime normalized by the prechange standard deviation. This parameter can be considered as the minimum level of change on the average that one desires to detect. Note that if the change point one seeks only concerns the variance ${\mu}_{1}={\mu}_{0}$, then $\delta =0$, and ${C}_{1}=0$.
- q is the ratio of variances between the prechange and postchange regime. This parameter can be considered as the minimum level of variance change one desires to detect. If the change point one seeks only concerns the mean (no detection on the variance), ${\sigma}_{1}={\sigma}_{0}$, then $q=1$, so ${C}_{2}=0$.

## 2. Methodology

- Collect data from low-cost Raspberry Shake seismic stations during “normal” structural conditions (may be the current situation).
- Preprocess the data collected in order to reduce its dimensionality, i.e., calculate the Hjorth’s parameters for each data component.
- Split the “normal” dataset into train and test, to evaluate if the algorithm is able to indicate that the data is, in fact, “normal”.
- Process the Hjorth’s parameters for each training data component by calibrating an autoencoder. This DL algorithm should be able to reproduce the structure of the input data.
- Perform a scenario classification on the test data, checking the stability of the predictions, which should indicate a “normal” state of the dam. This scenario classification step involves passing the time series of errors outputted by the autoencoders though a CUSUM algorithm and a fuzzy logic classification code.

- Collect data from low-cost Raspberry Shake seismic stations during operation conditions.
- Preprocess the data collected in order to reduce its dimensionality, i.e., calculate the Hjorth’s parameters for each data component.
- Process the Hjorth’s parameters for each training data component by using the calibrated autoencoder to reproduce the input dataset.
- Perform a scenario classification on the data just collected, which should indicate the state of the dam. This scenario classification step involves passing the time series of errors outputted by the autoencoders though a CUSUM algorithm and a fuzzy logic classification code.

#### 2.1. Preprocessing

#### 2.2. Processing

#### 2.3. Scenario Classification

#### 2.4. Monitoring a Scale Model —SM

#### 2.4.1. Experimental Setup—Model Assembly

^{3}of concrete.

^{3}. The geometry of the dam can be seen in Figure 2.

#### 2.4.2. Seismic Monitoring System

## 3. Application of the Methodology to Monitor the Scale Model

- AM.R7D9F on the crest of the reduced model;
- AM.RFBA7 on the concrete block;
- AM.R016A on the GeoFluxo laboratory floor.

- The reservoir upstream of the dam was filling;
- The reservoir upstream of the dam was at a fixed top quota;
- The reservoir upstream of the dam was emptying.

- Homogeneous dam (homogeneous model and upstream reservoir at fixed top quota);
- Dam with heterogeneity and no flow through the pipe (model with sealed PVC tube and upstream reservoir at fixed top quota);
- Dam with heterogeneity and flow through the pipe (model after unsealing the PVC tube and upstream reservoir at decreasing top quota).

- If none of the indicators are of type “high” (i.e., greater than the threshold imposed by Wald’s inequality), the system state is green;
- If only one of the indicators is of the “high” type (i.e., greater than the limit imposed by Wald’s inequality), the state of the system is yellow;
- If only two of the indicators are of the “high” type (i.e., greater than the limit imposed by Wald’s inequality), the state of the system is orange;
- If all indicators are of the “high” type (i.e., greater than the Wald inequality threshold), the state of the system is red.

- Period used for benchmarking the algorithm (normal);
- Period used for benchmarking the algorithm (with heterogeneity and no flow);
- Period used for measuring the algorithm (with heterogeneity and with flow).

## 4. Conclusions

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Salin, J.; Balayssac, J.P.; Garnier, V. 1—Introduction. In Non-Destructive Testing and Evaluation of Civil Engineering Structures; Balayssac, J.P., Garnier, V., Eds.; Elsevier: Amsterdam, The Netherlands, 2018; pp. 1–20. [Google Scholar] [CrossRef]
- International Commission on Large Dams. Available online: https://www.icold-cigb.org/GB/world_register/general_synthesis.asp (accessed on 30 September 2019).
- Vianna, L.F.V. Metodologias de Análise de Risco Aplicadas em Planos de Ação de Emergência de Barragens: Auxílio ao Processo de Tomada de Decisão. Master’s Thesis, Escola de Engenharia, Universidade Federal de Minas Gerais, Belo Horizonte, Brazil, 2015. [Google Scholar]
- Melo, A.V. Análises de Risco Aplicadas a Barragens de Terra e Enrocamento: Estudo de caso de Barragens da Cemig. Master’s Thesis, Escola de Engenharia, Universidade Federal de Minas Gerais, Belo Horizonte, Brazil, 2014. [Google Scholar]
- Lafitte, R. Probabilistic risk analysis of large dams. Its value and limits. Int. Water Power Dam Constr.
**1993**, 45, 13–16. [Google Scholar] - Foster, M.; Fell, R.; Spannagle, M. The statistics of embankment dam failures and accidents. Can. Geotech. J.
**2000**, 37, 1000–1024. [Google Scholar] [CrossRef] - Ferdos, F. Internal Erosion Phenomena in Embankment Dams. Ph.D. Thesis, School of Architecture and Built Environment, KTH Royal Institute of Technology, Stockholm, Sweden, 2016. [Google Scholar]
- Ikard, S.J.; Revil, A.; Jardani, A.; Woodruff, W.F.; Parekh, M.; Mooney, M. Saline pulse test monitoring with the self-potential method to nonintrusively determine the velocity of the pore water in leaking areas of earth dams and embankments. Water Resour. Res.
**2012**, 48, W04201. [Google Scholar] [CrossRef] - Parekh, M.L. Advancing Internal Erosion Monitoring Using Seismic Methods in Field and Laboratory Studies. Ph.D. Thesis, College of Engineering and Computational Science, Colorado School of Mines, Golden, CO, USA, 2016. [Google Scholar]
- Pueyo Anchuela, O.; Frongia, P.; Di Gregorio, F.; Casas Sainz, A.M.; Pocoví Juan, A. Internal characterization of embankment dams using ground penetrating radar (GPR) and thermographic analysis: A case study of the Medau Zirimilis Dam (Sardinia, Italy). Eng. Geol.
**2018**, 237, 129–139. [Google Scholar] [CrossRef][Green Version] - Detecting the Movement of Soils Within Earthen Dams, Canals, Levees, and Their Foundations. Available online: https://www.usbr.gov/research/challenges/soilmovement.html (accessed on 9 March 2022).
- Planès, T.; Mooney, M.A.; Rittgers, J.B.R.; Parekh, M.L.; Behm, M.; Snieder, R. Time-lapse monitoring of internal erosion in earthen dams and levees using ambient seismic noise. Géotechnique
**2016**, 66, 301–312. [Google Scholar] [CrossRef][Green Version] - Belcher, W.; Camp, T.; Krzhizhanovskaya, V.V. Detecting Erosion Events in Earth Dam and Levee Passive Seismic Data with Clustering. In Proceedings of the 2015 IEEE 14th International Conference on Machine Learning and Applications (ICMLA), Miami, FL, USA, 9–11 December 2015; pp. 903–910. [Google Scholar] [CrossRef]
- Fisher, W.D.; Camp, T.K.; Krzhizhanovskaya, V.V. Anomaly detection in earth dam and levee passive seismic data using support vector machines and automatic feature selection. J. Comput. Sci.
**2017**, 20, 143–153. [Google Scholar] [CrossRef] - Fisher, W.D.; Camp, T.K.; Krzhizhanovskaya, V.V. Crack Detection in Earth Dam and Levee Passive Seismic Data Using Support Vector Machines. Procedia Comput. Sci.
**2016**, 80, 577–586. [Google Scholar] [CrossRef][Green Version] - Zhou, C.; Paffenroth, R.C. Anomaly Detection with Robust Deep Autoencoders. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Halifax, NS, Canada, 13–17 August 2017; Association for Computing Machinery: New York, NY, USA, 2017; pp. 665–674. [Google Scholar] [CrossRef]
- Chen, H.; Mao, Y.; Wang, L.; Qi, H. Spatial-Temporal Features Based Sensor Network Partition in Dam Safety Monitoring System. Sensors
**2020**, 20, 2517. [Google Scholar] [CrossRef] - Rastin, Z.; Amiri, G.G.; Darvishan, E. Unsupervised Structural Damage Detection Technique Based on a Deep Convolutional Autoencoder. Shock Vib.
**2021**, 2021, 6658575. [Google Scholar] [CrossRef] - Jolliffe, I.T.; Cadima, J. Principal component analysis: A review and recent developments. Philos. Trans. R. Soc. A Math. Phys. Eng. Sci.
**2016**, 374, 20150202. [Google Scholar] [CrossRef] - Pearson, K. LIII. On lines and planes of closest fit to systems of points in space. Lond. Edinb. Dublin Philos. Mag. J. Sci.
**1901**, 2, 559–572. [Google Scholar] [CrossRef][Green Version] - Bourlard, H.; Kamp, Y. Auto-association by multilayer perceptrons and singular value decomposition. Biol. Cybern.
**1988**, 59, 291–294. [Google Scholar] [CrossRef] [PubMed] - Baldi, P.; Hornik, K. Neural networks and principal component analysis: Learning from examples without local minima. Neural Netw.
**1989**, 2, 53–58. [Google Scholar] [CrossRef] - Hjorth, B. EEG analysis based on time domain properties. Electroencephalogr. Clin. Neurophysiol.
**1970**, 29, 306–310. [Google Scholar] [CrossRef] - Cocconcelli, M.; Strozzi, M.; Cavalaglio Camargo Molano, J.; Rubini, R. Detectivity: A combination of Hjorth’s parameters for condition monitoring of ball bearings. Mech. Syst. Signal Process.
**2022**, 164, 108247. [Google Scholar] [CrossRef] - Grover, C.; Turk, N. Rolling Element Bearing Fault Diagnosis using Empirical Mode Decomposition and Hjorth Parameters. Procedia Comput. Sci.
**2020**, 167, 1484–1494. [Google Scholar] [CrossRef] - Sahki, N.; Gégout-Petit, A.; Wantz-Mézières, S. Performance study of change-point detection thresholds for cumulative sum statistic in a sequential context. Qual. Reliab. Eng. Int.
**2020**, 36, 2699–2719. [Google Scholar] [CrossRef] - Lai, T.L. Sequential Changepoint Detection in Quality Control and Dynamical Systems. J. R. Stat. Soc. Ser. B (Methodol.)
**1995**, 57, 613–644. [Google Scholar] [CrossRef] - Egea-Roca, D.; Seco-Granados, G.; López-Salcedo, J.A. Comprehensive overview of quickest detection theory and its application to GNSS threat detection. Gyroscopy Navig.
**2017**, 8, 1–14. [Google Scholar] [CrossRef] - Tartakovsky, A.G.; Polunchenko, A.S.; Sokolov, G. Efficient Computer Network Anomaly Detection by Changepoint Detection Methods. IEEE J. Sel. Top. Signal Process.
**2013**, 7, 4–11. [Google Scholar] [CrossRef][Green Version] - Wald, A. Sequential Tests of Statistical Hypotheses. Ann. Math. Stat.
**1945**, 16, 117–186. [Google Scholar] [CrossRef] - Page, E.S. Continuous Inspection Schemes. Biometrika
**1954**, 41, 100–115. [Google Scholar] [CrossRef] - Lorden, G. Procedures for Reacting to a Change in Distribution. Ann. Math. Stat.
**1971**, 42, 1897–1908. [Google Scholar] [CrossRef] - Hooda, D.; Raich, V. Fuzzy Logic Models and Fuzzy Control: An Introduction; Alpha Science International Limited: Oxford, UK, 2017. [Google Scholar]
- Demicco, R.V. Review of Fuzzy Logic in the Geological Sciences: Where We Have Been and Where We Are Going. In Machine Intelligence: Quo Vadis? World Scientific: London, UK, 2004; pp. 235–250. [Google Scholar] [CrossRef]
- Manconi, A.; Coviello, V.; Galletti, M.; Seifert, R. Short Communication: Monitoring rockfalls with the Raspberry Shake. Earth Surf. Dyn.
**2018**, 6, 1219–1227. [Google Scholar] [CrossRef][Green Version] - Chu, C.R.; Chen, P.; Hsiao, Y.H.; Jang, J.P.; Lee, S.C. The Implementation of Debris Flow Seismic Detector With Raspberry Shake. AGU Fall Meet. Abstr.
**2019**, 2019, H11H-1565. [Google Scholar] - Anthony, R.E.; Ringler, A.T.; Wilson, D.C.; Wolin, E. Do Low-Cost Seismographs Perform Well Enough for Your Network? An Overview of Laboratory Tests and Field Observations of the OSOP Raspberry Shake 4D. Seismol. Res. Lett.
**2018**, 90, 219–228. [Google Scholar] [CrossRef][Green Version] - Barber, C.B.; Dobkin, D.P.; Huhdanpaa, H. The Quickhull Algorithm for Convex Hulls. ACM Trans. Math. Softw.
**1996**, 22, 469–483. [Google Scholar] [CrossRef][Green Version]

**Figure 5.**Collected data for sensor AM.R7D9F. From (

**a**) to (

**c**) the three monitoring stages are presented: homogeneous dam, heterogeneous dam without flow and heterogeneous dam with flow, respectively.

**Figure 6.**Collected data for sensor AM.R016A. From (

**a**) to (

**c**) the three monitoring stages are presented: homogeneous dam, heterogeneous dam without flow and heterogeneous dam with flow, respectively.

**Figure 7.**Collected data for sensor AM.RFBA7. From (

**a**–

**c**) the three monitoring stages are presented: homogeneous dam, heterogeneous dam without flow and heterogeneous dam with flow, respectively.

**Figure 8.**First and second principal components for the activity Hjorth parameter considering all the eight sensors.

**Figure 9.**First and second principal components for the mobility Hjorth parameter considering all the eight sensors.

**Figure 10.**First and second principal components for the complexity Hjorth parameter considering all the eight sensors.

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Ozelim, L.C.d.S.M.; Borges, L.P.d.F.; Cavalcante, A.L.B.; Albuquerque, E.A.C.; Diniz, M.d.S.; Góis, M.S.; Costa, K.R.C.B.d.; Sousa, P.F.d.; Dantas, A.P.d.N.; Jorge, R.M.;
et al. Structural Health Monitoring of Dams Based on Acoustic Monitoring, Deep Neural Networks, Fuzzy Logic and a CUSUM Control Algorithm. *Sensors* **2022**, *22*, 2482.
https://doi.org/10.3390/s22072482

**AMA Style**

Ozelim LCdSM, Borges LPdF, Cavalcante ALB, Albuquerque EAC, Diniz MdS, Góis MS, Costa KRCBd, Sousa PFd, Dantas APdN, Jorge RM,
et al. Structural Health Monitoring of Dams Based on Acoustic Monitoring, Deep Neural Networks, Fuzzy Logic and a CUSUM Control Algorithm. *Sensors*. 2022; 22(7):2482.
https://doi.org/10.3390/s22072482

**Chicago/Turabian Style**

Ozelim, Luan Carlos de Sena Monteiro, Lucas Parreira de Faria Borges, André Luís Brasil Cavalcante, Enzo Aldo Cunha Albuquerque, Mariana dos Santos Diniz, Manuelle Santos Góis, Katherin Rocio Cano Bezerra da Costa, Patrícia Figuereido de Sousa, Ana Paola do Nascimento Dantas, Rafael Mendes Jorge,
and et al. 2022. "Structural Health Monitoring of Dams Based on Acoustic Monitoring, Deep Neural Networks, Fuzzy Logic and a CUSUM Control Algorithm" *Sensors* 22, no. 7: 2482.
https://doi.org/10.3390/s22072482