WO2014196892A1

WO2014196892A1 - System for leakage and collapse detection of levees and method using the system

Info

Publication number: WO2014196892A1
Application number: PCT/RU2013/000649
Authority: WO
Inventors: Alexey Petrovich KOZIONOV; Ilya Igorevich MOKHOV; Alexander Leonidovich PYAYT
Original assignee: Siemens Aktiengesellschaft
Priority date: 2013-06-04
Filing date: 2013-07-30
Publication date: 2014-12-11
Also published as: EP2989434A1

Abstract

The present invention relates to a system for leakage and collapse detection of levees with at least one sensor (3) and to a method for anomaly detection using the system. The at least one sensor (3) is a temperature sensor and/or pressure sensor. The method for anomaly detection using the system combines at least one time frequency measurement and at least one one-side classification method.

Description

SYSTEM FOR LEAKAGE AND COLLAPSE DETECTION OF LEVEES

AND METHOD USING THE SYSTEM

Description

The present invention relates to a system for leakage and collapse detection of levees with at least one sensor and to a method for anomaly detection using the system.

Dike safety monitoring is a challenging task, to prevent flooding the environment and damaging buildings, infrastructure and nature. It requires the application of structural safety monitoring methods for prediction of dike failure.

There are different scenarios of dike destruction. One is the overtopping of a primary dam structure. A water level, normally below the high of the dam structure, with water on one side of the dam can rise up to the top of the dam and flow over the dam. The dam becomes submerged and fails keeping the water on one side of the dam.

Another failure of a dam, levee respectively dike occurs due to foundation defects at different settlements, due to slides, slope instability, uplift pressure and due to foundation seepage. Failure of a dam can further occur due to piping and seepage.

A detection of levee failure at an early stage is important to prevent a total collapse of the levee and flooding of the protected side of the levee. Actions like mounting of sandbags on one side can be used to support the levee structure and prevent a total failure. An early detection allows also a warning and evacuation of people in danger due to a possible failure.

There are different approaches of levee collapse detection. One is the visual inspection of a dike. A disadvantage of visual detection is the high possibility of late detection of failure developments. Not all developments are visible from outside a dam and visual inspection is usually a very rare event.

A second approach is physical modeling of a dike. Physical models can be classified as following:

- simple physical equations for example applied for levee behavior assessment and

- virtual dike models constructed e.g. on basis of a finite element model.

In both cases the physical model calculates a probability of levee collapse using information like dike geometry and soil structure. It requires a good knowledge of the dike, research and produces additional costs. Virtual dike modeling provides an accurate piping prediction and detection, but requires significant computational efforts. This costs time, computational capacities and produces costs.

The object of the present invention is to present a system and method for leakage and/or collapse detection of levees solving the presented problems, being cost-effective and reliable, giving information about possible levees failure in an early stage, making it possible to take early actions like evacuation and levee reinforcement and/or stabilization.

The above objects are achieved by the system for leakage and/or collapse detection of levees according to claim 1 , and the method for anomaly detection according to claim 5, particularly using the system. Advantageous embodiments of the present invention are given in dependent claims. Features of the main claims can be combined with each other and with features of dependent claims, and features of dependent claims can be combined together.

The system for leakage and/or collapse detection of levees according to the present invention comprises at least one sensor, particularly a temperature sensor and/or a pressure sensor.

A cost-effective and reliable detection of leakage and/or collapse of levees is possible with the system according to the present invention. It gives information about possible levee failures in an early stage, making it possible to take early actions like evacuation of people, levee reinforcement and/or stabilization. Detection within levees with the at least one sensor can give information about leakage before a collapse of the levee occurs. Temperature and pressure sensors are cheap and easy to use, stable and reliable also over a long time.

The sensor can be and/or comprise a fiber optic cable. The cable can be heat able and/or comprises at least one device to heat the cable environment. Fiber optic cable installed at or in a dike/levee allows to measure temperature with high resolution. A process of piping, that means water destabilizing the dike by flowing through it, can be detected as abnormal temperature in a place of seepage. In winter time it can be a relatively high temperature compared to the normal temperature or in summer time it can be a relatively low temperature compared to the normal temperature of the surrounding environment. In autumn and spring it can be difficult to detect abnormal temperatures. There are times where the temperature of soil does not differ significantly from water temperature. An active principle, like known from GTC Kappelmeyer, can be used to monitor and detect abnormal temperatures even in autumn and spring. The fiber optic or its surrounding is heated from time to time, particularly in defined, periodic intervals of time. This allows robust/reliable detection of leakage even in autumn and spring, but also in summer and winter times.

A network of sensors can be comprised by, particularly spatial distributed in and/or at a levee and/or a levee system. This allows not only locally restricted monitoring and detection, but stability monitoring for a whole dike/levee and/or dike/levee system. Data from measurements collected and transmitted from a sensor network can be analyzed and processed e.g. by algorithms in online mode. Different types of sensors can be used for e.g. piping detection, like fiber optic cable temperature sensors and/or pore pressure sensors in a network.

A electronic data evaluation system can be comprised by the system for leakage and/or collapse detection of levees according to the present invention, particularly a computer and data transformation equipment to online collect, transfer and/or process data from the at least one sensor, particularly spatial distributed sensor network. This gives, i.e. compared to visual inspection of dikes the possibility to measure in short time intervals and automatically, from a distance. That safes personal costs and increases the probability to detect a failure of the levee in time.

The Method for anomaly detection according to the present invention, particularly with a system for leakage and/or collapse detection of levees as described before, combines at least one time frequency measurement and at least one one-side classification method.

By classifying the measurement it is possible to distinguish between normal events and anomalies. Leakage and/or collapse of levees can be detected and identified as anomaly. Special actions can be taken to stabilize the dam or to evacuate the area in risk of being flooded.

A wavelet transform on time series of the at least one time frequency measurement can be comprised. This enables for example in graphical representations of measurements and/or in computer calculations, i.e. automatic analysis an easy identification of anomalies like leakage or collapse. The location in space in the levee and/or levee system can be determined, where the anomaly occurs.

The Method can include at least a training and at least a test mode. The training mode can at least comprise the following steps

- Maximum Overlap Wavelet Transform (MODWT), and/or

- Normalization of data, and/or

- Use of normalized data for training of one-side classifier, particularly with at least one Neural Cloud used as one side classifier. The test mode can comprise a step of the use of trained one-side classifier for anomaly detection.

This enables the system to be run automatically and to give a signal when an anomaly is detected. The system "learns" to distinguish between normal and abnormal values. With different iterations within the training mode the range of values of normal behavior can be trained or identified/determined. In the test mode a value out of the range is identified as anomaly.

Values substantially 1 of the at least one sensor in spatial time series, after a predefined interval of heat up time, after Maximum Overlap Wavelet Transform (MODWT) and normalization, can be classified as normal behavior, particularly as no leakage and/or collapse of the at least one levee. Values substantially 0 can be classified as anomaly, particularly as a leakage and/or collapse of the at least one levee to be detected.

Values of the at least one sensor and calculated results from it can be used to construct a Neural Cloud.

A combination of representation of the data in Neural Cloud, particularly in 2 dimensional representation, and measured temperature with fiber optic cable length is used to distinguish between abnormal and normal behavior and/or anomaly classification and/or detection. An identification in space, in the levee along the cable or cable network is possible.

At least one predefined interval of heat up time of at least one sensor can be in the range of 10 minutes. This time range can also be used for cooling down. The time range of 10 minutes is long enough to heat up and/or cool down the surrounding of a fiber optical cable, particularly in sand. It can be adjusted to the season, and be dependent on soil structure and other features. The heat up time can also differ from the cool down time, for example be longer in winter or shorter in summer than the cool down time.

The method can be used for monitoring at least one levee, particularly for leakage and/or collapse detection in the test mode. The method can be used to spatially localize abnormal behavior/anomalies, particularly leakage and/or collapse of at least one levee within the levee and/or levee system. A reliable, cost effective detection, particularly in the long run of leakage and/or collapse of a single levee or a levee system is possible with the method and localization in space is feasible. This enables in time actions for stabilization of the levee and/or for the safety of the environment protected by the levee. People and animals can be informed and/or rescued if an anomaly is detected.

The measurement and/or data processing can be automatically, online, remote and/or continuous with time and/or in fixed time intervals, particularly for test mode. Compared to manual investigation of levees it saves money with time, is more reliable due to short intervals of measurement and enables detection within levees and its structure. The remote control can be performed in a center/from a central point from where for example actions are planned.

The advantages in connection with the described method for anomaly detection according to the present invention are similar to the previously, in connection with the system for leakage and/or collapse detection of levees described advantages and vice versa.

The present invention is further described hereinafter with reference to illustrated embodiments shown in the accompanying drawings, in which:

Fig. 1 illustrates a cross-section of a levee 1 with fiber optic cable 2 and sensors 3, and

Fig. 2 a cross-section alongside of the levee 1 of Fig. 1, and

Fig. 3 a workflow of the method according to the present invention with at least a training I and at least a test mode II, and

Fig. 4 a workflow of data preparation in training mode I, and

Fig. 5 a workflow of data preparation in test mode II,

and

Fig. 6 a workflow of training data preparation in more detail, and Fig. 7 experimental data of temperature T change along a GTC Kappelmeyer fibre optic cable with length 1 in m at different times t during heat-up process, and

Fig. 8 normalized data after MODWT application along the cable length from A 1^st iteration of "cold" fiber to F 6^th iteration of heating-up, and

Fig. 9 retaining spatial dam leakage detection after 6^th iteration with calculated confidence values C, and

Fig. 10 3 dim view of constructed Neural Clouds from calculated confidence values C analogous Fig. 9 corresponding to coefficients of 2^nd and 3^rd level of decomposition, and

Fig. 1 1 experimental data of temperature T change along a Rhine levee GTC

Kappelmeyer fibre optic cable with length 1 in m at different times t during heat-up process, and

Fig. 12 normalized data after MODWT application along the cable length 1 from A 1^st iteration of "cold" fiber to F 6^th iteration of heating-up, and

Fig. 13 experimental data for air bubble detection of temperature T change along a fiber optic cable with length 1 in m after a heat-up time t of 10 min, and

Fig. 14 calculated confidence values C from experimental data of Fig. 13, and Fig. 15 2 dim view of constructed Neural Clouds with 2^nd and 3^rd level of decomposition after pre-processing (X- and Y-axis respectively) corresponding to data of Fig. 13.

In Fig. 1 a cross-section of a levee 1 with a fiber optic cable 2 and sensors 3 is shown. The sensors 3 and fiber optic cable 2 are comprised by a monitoring system for a levee and/or dike system, to predict and/or detect dike failure. A network of sensors 3 distributed for example homogenous in the dike system in connection with fiber optic cable 2 and data processing equipment are monitoring the condition of the dike system. The monitoring can for example be performed continuously with time or in certain predefined time intervals. Overtopping of the primary dam structure, piping and seepage of levees can be detected by the system and defects of the dam structure due to differential settlement, slides, slope instability, uplift pressure and foundation seepage can be predicted. A detection and prediction of possible failures in an early stage enables to take special actions like reinforcement of weak regions of the levee 1 with for example sand or sandbags. This can prevent a complete failure of the levee and/or dike system. An eroding and water breakthrough of levee 1 in connection with harm to people and assets can be prevented.

Fig. 2 shows a cross-section alongside of the levee 1 of Fig. 1 with a regular distance of sensors 3 from each other. In other cases like different composition of the ground the sensors 3 can also be arranged in and/or at the levee in irregular distances, for example cumulated in critical regions. All possible distributions of sensors 3 and cables 2 connecting sensors 3 in and at the dam system can be used to set up a monitoring system/network. The cable 2 can also be used as, comprise or be the sensor 3.

The evaluation unit and/or data processing equipment in data communication with the network is for simplicity not shown in Figures. It can comprise a computer and data output devices like printer and/or monitors, and can be connected to the internet or an intranet. The data processing can be located near to the sensor and dike system, or distant like at a special place, for example a control room. From this special place actions can be planned and coordinated after prediction or detection of dike failure by the system. Examples for actions are the information and/or evacuation of people, the evacuation of animals, the reinforcement of critical regions of the levee with particularly sand and/or sandbags and/or the reduction of water pressure at critical points by usage of polder areas.

In Fig. 3 a workflow of the method according to the present invention is shown, with at least a training I and at least a test mode II. Both modes respectively phases I and II require data-processing. In Fig. 4 a workflow of data preparation in the training mode I is shown. In Fig. 5 a workflow of data preparation in the test mode II is shown.

To monitor the stability of a levee 1 and/or to predict the failure an analysis of measurements collected from a sensor network installed in and/or at the levee 1 is performed. The information from the sensor network is processed by algorithms in online mode. Different types of sensors 3 can be used for e.g. piping detection. Two possible sensor types are fibre optical cable temperature sensors and/or pore pressure sensors. Fiber optic cable installed in a levee 1 allows to measure temperature T with high resolution. A process of piping can be detected as abnormal temperature T in a place of seepage, for example by a relatively low measured temperature T in summer time or a relatively high temperature T in winter time. During autumn and spring time the detection of seepage with this method can be difficult. During those times the temperature T of soil differs not significantly from temperature T of water. A clear distinction of normal and abnormal behavior, that means detection of for example seepage is not possible.

From GTC Kappelmeyer the use of an active principle of monitoring is known. A fiber optic respectively it's surrounding is heated from time to time and temperature T is measured during and/or after a heat up cycle. With this method a reliable detection of leakage even in summer and autumn is possible.

According to the present invention two main phases I and II are necessary to reliable detect for example leakage of a levee 1 during all periods of a year with a sensor network. Hereinafter as an example a fibre optic cable as sensor network is distributed in a levee 1 and/or levee system. It is also possible to use other sensor networks which are not described for simplicity in the Figures.

Several spatial series measured in time using the fiber optic cable 2 as shown in Fig. 1 and 2 are performed. The two main phases I and II include a training mode I and a test mode II, as shown in Fig. 3. Both modes require data pre-processing. The measurements in phase I take place with a "cold" cable before start of heat-up and several first iterations after start of heat-up. The respective data processing procedure is shown in Fig. 4. Leakage might be visually identified in figures derived from measurements after data processing only with a significant number of iterations of heat up. Real abnormal behavior has not to be included into the training set. With the training set derived out of several spatial series measured in time a one-side classifier is trained in phase I.

Leakage is usually presented as a rapid change in time series. There are different approaches to detect a rapid change in time series, e.g. Student's T-test. Wavelets can be used for abrupt fault detection. For example at "Wivenhoe" Dam it is known that wavelets were applied for analysis of water temperature measurements. Each signal was decomposed using wavelets into daily, sub-annual and annual (DSA) components. Each of the components was used for further analysis.

According to the present invention the Maximum Overlap Discrete Wavelet

Transform (MODWT) can be selected as a pre-processing procedure, see Fig. 4 and 5. After that each level of decomposition can be normalized. This data can be used as input of one-side classifier training, see Fig. 4. In Fig. 5 the procedure for test mode II is shown. For the examples shown in Figures Neural Clouds are used as one-side classifier.

Fig. 6 shows the visualization of training data preparation. Several spatial series measured at times t, t+1 , t+2 in the period At are used for data analysis. MODWT is applied to each spatial series. From 1-D data the MODWT coefficients are derived. The results are respectively normalized, i.e. the MODWT coefficients after normalization are derived. This data are used for training the one-side classifier.

GTC Kappelmeyer described for testing of a developed anomaly detection method a real example of abnormal behavior, registered in measurements collected at an earth filled dam with bitumen sealing. The dam had a total length 1 greater than 2 km. It contained a leakage of bitumen sealing made of asphalt-coated gravel and bitumen binder of the dam. This anomaly can be presented in spatial time series as a rapid drop in the interval around 150 meter, see the example of temperature T measurement in grad Celsius along a fibre optic cable length 1 in m at different heat-up times t in Fig. 7. The leakage 4 in the region between 74 and 163 m of the cable is seen as minimum in temperature (see region enclosed by dotted line). The measurements 5, 6 and 7 represent the temperature along the length 1 of the cable after 1 ^st iteration of heat-up (25 min), after 2^nd iteration of heat-up (30 min) and after 4 iterations (20 min) of cold cable respectively.

In Fig. 8 the normalized data after MODWT application along the cable length from A 1^st iteration of "cold" fiber to F 6^th iteration of heating-up are shown. Calculated confidence intervals, see Fig. 9 show that the part of the cable between 150 m and 160 m is classified as abnormal. Calculated confidence values C close to 1 are related to normal behavior and close to 0 are related to anomalies. With calculated confidence values C close to 0 after 6^th iteration in Fig. 9 a dam leakage is detected.

In Fig. 10 a 3 dim view of constructed Neural Clouds from calculated confidence values C analog Fig. 9 is shown. The values correspond to coefficients of the 2^nd and 3 ^rd level of decomposition. D₂ is the coefficient corresponding to the 2^nd level of decomposition and D₃ is the coefficient corresponding to the 3^rd level of decomposition.

In Fig. 1 1 to 15 the data analysis according to the present invention is performed for data from a Rhine dam (GTC Kappelmeyer). Fig. 1 1 shows experimental data of temperature T change along a Rhine levee GTC Kappelmeyer fibre optic cable, with length 1 in m at different time's t during a heat-up process. No piping is detected but a so-called "air bubble" 8, caused due to problems with the fibre optic cable installation. A cable is surrounded with air, which has a lower thermal conductivity in comparison to the ground that covers the rest of the cable. The temperature across the whole cable is more or less the same during the heat-up process, substantially within a range of 28 to 30 degree Celsius. Between 209 m and 221 m cable length 1 the temperature T rises significantly. This effect is interpreted as an anomaly. The air bubble 8 is positioned between 209 m and 221 m of cable length 1.

In Fig. 1 1 the regions 9 of the cable are close to air. The temperature T curve 10 along the cable 1 represents 5 min of "cold" cable. Curve 1 1 represents the cable after the 1^st minute of heat-up, curve 12 the cable after the 2^nd minute of heat-up and curve 13 the cable after the 3^rd minute of heat-up. The region with air bubble 8 is marked by enclosing with a dotted line.

In Fig. 12 normalized data after MODWT application along the cable length, shown in Fig. 1 1 from A 1^st iteration of "cold" fiber to F 6^th iteration of heating-up are given. Data represented are from the least asymmetric wavelet with 8 levels of decomposition. Coefficients corresponding to third and fourth levels of decomposition after pre-processing, see Fig. 8 were used as input for the one-side classifier. Measurements related to the cold fiber and several first minutes of heat-up were used for the NC training. The series A to F of the temperature T curves in grad Celsius along the cable length 1 in m gives the normalized 1^st level of decomposition to the 6^th level of decomposition respectively after 10 minutes of heat up and the "cold" fiber. The air bubble is marked by enclosing the region 8 in a dotted line.

In Fig. 13 the spatial time series after a heat-up time t of 10 min is shown. The experimental data of temperature T change along a fiber optic cable with length 1 in m with marked special values are shown for air bubble detection. This special values are given as points in Fig. 14, where the calculated confidence values C derived from the experimental data of Fig. 13 are presented along the fibre optic cable length 1 in m. Values close to 1 correspond to normal behavior and values close to 0 are interpreted as anomalies. Fig. 15 gives a 2 dim view of constructed Neural Clouds with 2^nd and 3^rd level of decomposition after pre-processing, X- and Y-axis respectively, corresponding to data of Fig. 13. The classification of points corresponding to special values in Fig. 13 is clarified and/or confirmed by the position of points in Fig. 15. A cluster with normal data, related to the training set I, the test set II and detected outliers are presented in Fig. 8. MODWT does not give perfect localization properties according to the Gibbs phenomenon. Due to this fact some points are not correctly classified as abnormal behavior in Fig. 14. Lower points 14 in Fig. 15 correspond to normal temperatures in Fig. 13. Higher points 15 in Fig. 15 correspond only to abnormal temperatures T in Fig. 13. In the representation of Fig. 15 normal and abnormal behavior can be distinguished. By combination of the described representations of values a principal function of the method is proved.

By a combination of time-frequency methods for pre-processing and one-side classification a dam leakage detection using measurements collected from distributed temperature sensors 3, particularly fiber optic cables is achievable. GTC Kappelmeyer fiber optic is able to detect leakages even in spring and autumn time. It is difficult to detect leakages reliable with sensors without heat-up, i.e. passive measuring systems. With post-processing as described before active measuring systems are able to detect leakages reliable during all times of the year.

The method and system according to the present invention is applicable in real world, particularly at dam structures and/or systems. A localization of the region comprising the leakage is possible. The system can be used for monitoring of a wide range of distributed objects like dam, levees, tunnel and so on.

The same approach can be used for other physical value analysis, particularly another type of fibre optics e.g. for strain measurements. If the pattern of possible failures is similar to leakage pattern, particularly with a rapid change in spatial time series, the failure can be detected. The approach can also be used for tests of the correctness of overall system installations, for example for air bubble detection. For training I data related to normal behavior of the object can be sufficient.

Features described before can be used single or in combination, and in combination with embodiments known from the state of the art. Additional models might be used for testing of the described approach and generation of additional training sets with patterns of abnormal behavior.

Claims

1. System for leakage and/or collapse detection of levees with at least one sensor (3), particularly a temperature sensor and/or pressure sensor.

2. System according to claim 1 , characterized in that the sensor (3) is and/or comprises a fiber optic cable (2), particularly heatable and/or with at least one device to heat the cable environment.

3. System according to any one of the claims 1 or 2, characterized in that a network of sensors (3) is comprised, particularly spatial distributed in and/or at a levee (1) and/or levee system.

4. System according to any one of the claims 1 to 3, characterized in that a electronic data evaluation system is comprised, particularly a computer and data transformation equipment to online collect, transfer and/or process data from the at least one sensor (3), particularly spatial distributed sensor network.

5. Method for anomaly detection, particularly using a system according to any one of the claims 1 to 4, combining at least one time frequency measurement and at least one one-side classification method.

6. Method according to claim 5, comprising a wavelet transform on time series of the at least one time frequency measurement.

7. Method according to any one of the claims 5 or 6, comprising at least a training (I) and at least a test mode (II), particularly comprising steps in the training mode (I)

- Maximum Overlap Wavelet Transform (MODWT), and/or

- Normalization of data, and/or

- Use of normalized data for training of one-side classifier, particularly with at least one Neural Cloud used as one side classifier, and/or

particularly comprising a step of the test mode (II)

- Use of trained one-side classifier for anomaly detection.

8. Method according to claim 7, characterized in that values of the at least one sensor (3) in spatial time series after a predefined interval of heat up time after Maximum Overlap Wavelet Transform (MODWT) and normalization substantially 1 are classified as normal behavior, particularly no leakage and/or collapse of at least one levee (1), and values substantially 0 are classified as anomaly, particularly a leakage and/or collapse of at least one levee (1) to be detected.

9. Method according to claim 8, characterized in that values of the at least one sensor (3) are used to construct a Neural Cloud.

10. Method according to claim 9, characterized in that a combination of representation of data in Neural Cloud, particularly in 2 dimensional representation, and measured temperature (T) with fiber optic cable length (1) is used to distinguish between abnormal and normal behavior and/or anomaly classification and/or detection.

1 1. Method according to any one of the claims 8 to 10, characterized in that at least one predefined interval of heat up time of at least one sensor (3) is in the range of 10 minutes.

12. Method according to any one of the claims 5 to 1 1 , characterized in that the method is used for monitoring at least one levee (1), particularly for leakage and/or collapse detection in the test mode (II).

13. Method according to any one of the claims 5 to 12, characterized in that the method is used to spatially localize abnormal behavior/anomalies, particularly leakage and/or collapse of at least one levee (1) within the levee (1) and/or levee system.

14. Method according to any one of the claims 5 to 13, characterized in that the measurement and/or data processing is automatically, online, remote and/or continuous with time and/or in fixed time intervals, particularly for test mode (II).