Keywords

1 Background

The lock gate is a key point in river transport ecosystem for which an unanticipated damage can have a high financial impact. The failure of only one of it can lead to several million euros losses (Eick et al. 2017; Treece 2015). Old gates reach their limit lifetime and needs for predictive maintenance. But there is no standards or regulations to monitor these structures and follow damages on a lock gate.

In addition, lock gate access is highly constrained by navigation traffic continuity of service. Only few navigation stops dedicated to lock gate maintenance and inspection are planned over the year. The traditional way to manage these actions is to organize periodic onsite inspections of the structure. However, this method provides visual inspections of a limited set of areas of the structure, sometimes using instrumentation but often purely visual checks. Onsite inspection cannot access all parts of the structure; including the underwater areas of the structure where there are also potential areas of structural vulnerability. Even more, this implies a constrained time to detect defects and to provide an efficient action plan. To establish a predictive maintenance strategy, the use of monitoring solutions contributes to continuous and real time information, between navigation stops. The proposed approach aims at reducing the cost of the inspection and maintenance with a Digital Twin process to assist remote control of the structure when it is needed instead of a prescriptive uninformed calendar-based inspections.

CNR operates and maintains 18 development schemes on the Rhône River in France and takes a very close interest in the hydromechanics equipment that equip its main structures: hydroelectric power plants, mobile dams, and navigation lock-gates. The hydromechanics assets include more than 300 large-sized vantellerie structures: width between 10 m and 45 m, height between 6 m and 21 m and weight between 50 t and 200 t. The large number of these equipment, as well as their strategic importance in terms of safety and security of the installations, make them a priority asset. Among these vantellerie bodies, the lock-gates are, together with the downstream safety valves of the production units, the equipment that are both the most stressed and that which has undergone the greatest number of operating cycles. Half of operational lock-gates is over 50 years old, keeping in mind that this equipment are generally designed for a 70 years of service lifetime as detailed in DIN 19704-1 2014-11 (2014).

Infrastructure performance expressed either in terms of risk assessment or in terms of cost-effectiveness is established using specific indicators. The ultimate objective is to know either the real remaining lifetime, the damage on the structure, and identification of any related anomalies and the comparison between design, simulation, and real behavior for design optimization of the structure.

Digital twins are an emerging technology. Structural Digital Twins which simulate and monitor structures in near real time are now being considered by significant players in energy producing structures. Some companies provide Digital Twin products, a broad term covering a range of disparate technologies. But there is no commercial product today which can provide a complete structural integrity solution. An operator that wants a structural integrity monitoring tool today will need to build a costly set of customized hardware and software that will only partially answer the requirements and in addition will have the following limitations:

  • Current sensor technologies provide limited data by the sensors number, position, or static and dynamic precision, and are unsuitable for installation on large structure.

  • Current numerical modelling methods cannot process the structural complexity in real-time without gross simplifications or major economic trade-off.

In addition, there is no solution combining the two elements into a holistic model. Some solutions based on machine learning propose behavior tracking and alert generation, but they are based on patterns of known events and cannot anticipate unknown events or structurally based fatigue failure. The weakness of big data with Machine Learning to identify outliers is that there is no physical understanding of the cause of the problem when there is a deviation of indicators, nor 3D representation of the areas with corresponding stresses. It is thus difficult to plan an efficient maintenance response.

A Digital Twin for structural monitoring is a twin numerical model of an instrumented structure providing a physical understanding of phenomenon which allows better decision making. Based on merging data captured by the sensors with a numerical model of the instrumented structure, the proposed Digital Twin allows to get additional structure indicators (like stress) not directly accessible to the measurement. Then it became possible to process the real fatigue and so the remaining lifetime. In case of behavioral modification, the numerical model can immediately identify this change and send a notification to the duty holder/operator specifies. In addition, our methodology includes of a fully numerical model from the design stage so that the digital twin can support inspection planning and predictive maintenance support.

In the second part of this paper, we will present the global system overview. In the third part, the methodology applied to monitor the lock-gate using the NEURON system and water level measurements is introduced. A relevant indicator combining the maximum deformation of the lock-gate and the water level will be presented in detail. The fourth part will focus on the Live Digital Twin building, combining the data from the previous indicators, the 3D numerical model of the lock-gate and the continuous data flow from the monitoring system. Finally operational implementation and results of the Digital Twin for Avignon water lock gate located on the Rhône River. The estimation of the fatigue and the service lifetime of the lock-gate with the Digital Twin is emphasized and the use for the maintenance plan optimization of the owner is explained.

2 Solution Overview

Digital twins are of most use when object characteristics are changing over time or may be difficult to evaluate directly, thus making the initial model of the object invalid, and when measurement data that can be correlated with this change can be captured. To ensure real fatigue, damage and remaining lifetime estimation allowing decisions based on physical understanding of the observed phenomena, the proposed system is based onto 3 parts:

  • A monitoring system which allows to get all the needed static or dynamic indicators of the structure as well as the environmental or loads indicators

  • A 3D numerical model of the monitoring structure based on the knowledge of the structure which considers the environmental loads, the structural behavior, and the boundary conditions and fast enough to integrate sensor data, and operate in real-time physics-based simulations

  • A data hosting and processing platform architecture, between the measurement system and the physics-based numerical model of the structure, enabling the data-model convergence by dynamically updating the digital twin model and structural key indicators estimation to support operations and maintenance.

Based on the monitoring system, we proposed a Digital Twin updating in real time according to the endured loads by the structure and its integrity. The Digital Twin model is based on a 3D numerical model of the structure and is continuously feed with structural indicators (static, dynamic, modal, …) and environmental loads measurement (water height, wind states, sea state, temperature, …). The digital Twin delivers, from stress calculations, an estimate of fatigue, damage, and residual life at all points of the structure.

To provide an efficient Digital Twin, the developed solution simultaneously controls the measurement technologies (sensors, acquisition, transmission, processing), the 3D modelling of the structures (accurate, quick, and continuously updated, interfaces, boundary conditions), the structure load sources and their measurements (wind, swell, pedestrians, etc.), and a specific IT architecture to aggregate all the collection of data and simulation (Fig. 1).

Fig. 1.
figure 1

Digital Twin overview

Full size image

2.1 Industrial IoT Sensor Networks

Structural monitoring is based on two main types of indicators: structural and environmental. To allow qualitative correlation analyses of these measurements, the I-IoT system (Neuron) acquires and synchronizes data on the instrumented structure. Those data can then be correlated and merged into the calculations and digital models. The proposed monitoring I-IoT sensor network is comprised of a “plug-and-play” single cable hard-wired network of measurement nodes (synchronized high resolution tri-axis accelerometers, gyros, magnetometers, and a temp sensor) collecting static and dynamic deformation at a spatial density adapted to the structure and a software solution processing raw data into to specific indicators meaningful for the application (Fig. 2).

This network is suitable for large size structure (up to 30 nodes and 1km long), is available in different housings (IP65, 100m Subsea and ATEX Z1), and provide external sensors interoperability (strain gauges, analog and digital sensors) for load measurement. The mechanical indicators, static (tilts, roll, norm, torsion, deformation, convergence…) and dynamic (modal analysis, frequencies analysis and tracking, dynamic deformation), express the structural response to external loads and would be compared with Digital Twin simulated outputs. The environmental indicators measure external loads (stress, environment) applied on the structure: wind on wind turbines, hydrostatic load for lock gate, vehicles on a road bridge, trains on a railway bridge, the temperature for metal structures, humidity, corrosion. For each type of structure, only the environmental indicators which are significant for structural fatigue estimation are relevant for observation.

Data are pre-processed on the embedded gateway and send by 4G or Ethernet connection to an AWS cloud platform for data processing, long-term storage, and visualization.

2.2 Numerical Model

The Digital Twin provides an accurate description of objects that change over time: a precise, up-to-date copy of certain properties and states of the physical structure, such as its shape, position, state, and movements.

The model used in a Digital Twin is not a data-driven model, but it should produce results that are directly equivalent to measured quantities (so that the model updating process is data-driven), and it is likely that the model will take in other measured quantities as boundary conditions, loads, or material properties (Wright and Davidson 2020).

Digital Twins can use any type of model that is a sufficiently accurate representation of the physical object that is being twinned. In an ideal world, where computation would be instantaneous, and accuracy would be perfect, Digital Twins would use models derived directly from physics that took all phenomena likely to affect the quantities being measured and updated into account.

Digital models are mainly based on the finite element, boundary element or finite difference method, considering an almost infinite variety of types of structural geometry integrating all types of loads, material and boundary conditions. However, the drawback of a very complex model is that calculations may then take considerable time and means that real-time updating is not possible. Simplifying models to reduce these calculation times is therefore a real challenge. A model for a Digital Twin should be sufficiently physics-based in to manage meaningful updated parameters, sufficiently accurate to manage application useful updated parameters, sufficiently quick to run in comparison with the observed physical phenomenon. But the barrier of computational cost at high accuracy does not mean that physics-based approaches should be discarded altogether. Different solutions can be used to adapt or replace the computationally expensive model:

  • Some applications of Digital Twins do not require high-speed computation, because the time frame over which the twin is to be updated is hours rather than seconds.

  • Some applications of a Digital Twin can use local models of key parts of a structure or an object rather than considering the complete system.

  • Generate a surrogate model or metamodel based on a set of reliable results within the known operating parameter envelope of the physical object (pure data-driven model with Kriging).

  • Generate a reduced order model (ROM) by seeking to characterize the system being modelled in terms of a small number of functions or “modes”.

In the proposed solution, we choose to use FEM model based on the structural geometry and on ANSYS simulation tools with linear elastic assumption. To process unsymmetric events or dynamic approach, a full structure model is usually used. Load acting on the structure (like wind, ocean waves and marine current, water level, temperature, …) and physical parameters (cracks, erosion, …) will also be modelled as described in the Digital Twin section, to generate a specific reduced order model. This reduced order model will be used to simulate tilts for each node (for calibration and data-model convergence algorithm) and to provide stress estimation (for fatigue estimation).

3 Monitoring: From Tilts Measurement to Deformation Indicator

The maximum of deflection is a fundamental indicator of the static behavior of the structure, especially a water lock gate. This indicator is also correlated with other environmental phenomenon such as temperature, top and bottom level of water. Standard solutions, as detailed in Eick et al. (2017), propose indicators that focus on the structural behavior, especially the correlation between the constraints and the water level. These indicators can be generalized by using the maximum deformation instead of the local stress. But measuring stress on a lock gate is limited by the robustness of a stress gauges in water environment and measuring the deflection of the gate is tricky with optical external devices due to the presence of obstacles, water, and ‘dirty’ environment. The proposed solution is based on a high accurate accelerometer sensors networks distributed on the structure allowing to capture the deformation in real time and on the significant area of the lock gate as detailed in Carmona et al. (2019). For the chosen use case (Avignon located lock gate from CNR on Rhône River) the monitoring network is comprised of 9 nodes, fixed with magnets, and a software interoperability with CNR SCADA providing upstream water level.

Fig. 2.
figure 2

I-IoT Sensor network and water lock gate instrumentation

Full size image

Then, using tri-axis acceleration measurement, we estimate the tilt and then process, thanks to patented algorithms, the static deformation of the lock gate for each node.

The static information (tilts) leads to the estimation of spatial deformation, and dynamic data (accelerations) allow the modal behavior estimation. Using those data from all the nodes distributed along a vertical central line on the lock gate, our monitoring system can reconstruct the global deformation of all the lock gate, regardless sensor position, during different phases of the lock gate operation.

Using the global deformation measurement of the lock gate for each cycle of water charge/discharge we generate the maximum deformation indicator, in correlation with the water level. So based on this indicator, the structural behavior is monitoring in continuous, the measure data are easily accessible and depending on the measure phenomenon. In addition, loads and environmental indicators are available thanks to monitoring system interoperability. Data-driven model (Machine learning) can be derived from available measurement to detect prior to the Digital Twin deviant behavior and thus generate alarms. To compute the ROM (Reduced Order Model) model for the Digital Twin, we need tilts and water level during a long period, typically several days or weeks. All those data were then merged to process sensitivity studies leading to a reduction order of the FEM model (Fig. 3).

Fig. 3.
figure 3

Lock gate operation phases and maximum lock gate global deformation

Full size image

4 Digital Twin

4.1 Numerical Model

Digital Twin building starts with the CAD model and design study of the structure. The 3D numerical model will fulfill the requirements described in the previous model section. To define sensor position, a first simulation is done to determine maximum deformation areas. As we use acceleration to estimate tilts and global deformation, our sensors will be installed on these specific areas where the lock gate deformation is at a local maximum value.

As described in the next figure, the digital twin design process is divided into to 5 main steps (Fig. 4).

Fig. 4.
figure 4

Digital Twin design process

Full size image

4.1.1 Model Calibration

The calibration of the numerical model based on measurement is an essential step to provide estimation of the boundary condition and load correction. This step will allow the structural behavior to be validated with an acceptable error independently from the load applied on it. AS we are in static applications, calibration will be made on tilt angles indicators. After the installation of the monitoring system on the structure, comparison of simulated tilts with measurement is performed, to tune the simulation model to best fit of the measurement data. Needed inputs are sensor node position on the structure and measured structural indicators. The next figure shows model calibration based on measured and simulated tilt angle comparison for each node and for different boundaries conditions: number of contacts between the gate and the support wall, type of upper and lower blockage of the gate, and the stiffness coefficient of supports (Fig. 5).

Fig. 5.
figure 5

Model calibration based on tilt angles and estimated deformation

Full size image

4.1.2 Hotspots Identification

Hotspot identification aims at localizing the position of the most important stress on the structure where a fatigue assessment of the structure should be performed. The localization of the most important stress may be a welded or a non-welded area, regardless of the sensor position. The identification of this hotspot is often validated in collaboration with experts of the studied structure to provide the most possible feedback and then have a realistic hotspot. With a full picture of the stress distribution of the asset it is then possible to establish the actual level of the fatigue of the structure. In the case of Avignon located CNR lock Gate, we have identified 36 critical hotspots.

4.1.3 Loads Sensitivity Study and Reduced Order Model (ROM) Generation

A sensitivity study will also allow to define the uncertainties linked to the model/measurement correlation. This variational analysis is performed using: (i) external indicators as parametric inputs, (ii) tilts at sensor positions and stress at hotspots as parametric outputs. The main objective is to condense the relation between inputs and outputs to reduce the order of the simulated FEM model. As a result, simulations taking hours to days with FEM models decreased to seconds to minutes with a ROM model (Zhao et al. 1992; Chinesta et al. 2018).

The ROM is a collection of response surfaces with water levels (up and downstream) as inputs and with outputs such as stress on chosen hotspots and tilts angles for each node. Stresses on hotspot will be inputs for fatigue estimation and tilts angles will be inputs for data-model convergence monitoring.

The next figure introduces the ROM model for sensor node 1 between upstream and downstream water levels and tilt angles and for one specific hotspot, the weld number 28 near the edge of the lock gate between upstream and downstream water levels and stress amplitude (Fig. 6).

Fig. 6.
figure 6

Response surfaces examples: Sensor n°1 and Weld n°28

Full size image

4.1.4 Fatigue Estimation

After calibration of the load and structure, the fatigue assessment requires to extract the stress from the model which can be performed on the calibrated Reduced Order Model. The fatigue assessment, based on linear damage accumulation model, provides the remaining lifetime of the structure based on stress amplitude and cycling within the structure. The stress, provided by the ROM model is then counted using a standard Rainflow counting method (amplitude and cycling) and applied to S-N curves for damage accumulation estimation. The fatigue assessment implementation is based on standards and guidelines such as Eurocode 3, DNV or FKM (Forschungskuratorium Maschinenbau 2020). The delivered output of this module is the degree of structural performance and the remaining useful lifetime based on the historical usage (Fig. 7).

Fig. 7.
figure 7

From stress to fatigue estimation

Full size image

4.2 Digital Twin Processing

This phase corresponds to the operational deployment of the Digital Twin with an automatic and real time estimation of the fatigue depending on the real stress of the Lock gate. The structural estimation of the fatigue and the remaining lifetime are the main attending indicators. The ROM takes as inputs loads measurements (environment indicators) and deflections/tilts of the structure at sensor node positions and constraints/stress at hotspots as outputs. The model is updated when the difference between the simulated and measured tilts or deflections is higher than a selected threshold. In the same way, variations in physical parameters such as cracks size or erosion factor can lead to an update of the model (Fig. 8).

Fig. 8.
figure 8

Digital Twin process

Full size image

Based on this approach, the proposed digital Twin provides residual lifetime for a list of 36 critical hotspots on the Avignon water lock gate. The residual lifetime estimation based on loads history shows:

  • 20 uncritical welds

  • 6 welds are identified to be repaired (plastic deformation, zero residual lifetime)

  • 10 critical welds with low residual lifetime (<10 years).

Those results are in high correlation with the maintenance operations on the gate realized by CNR.

The next figure shows the remaining lifetime for these 10 welds and the model behavior. After a learning phase, the remaining lifetime evolves with the real and measured load cycles. A new value is calculated each day based on measures indicators (Fig. 9).

Fig. 9.
figure 9

Remaining lifetime estimation for critical hotspots

Full size image

5 Conclusion and Perspectives

To provide information on the remaining lifetime of every part of the structure, we have designed, tested, and validated a Digital Twin with a real-time, physics based numerical model of a full-scale asset to give greater insight into performance and operation. We deliver a Digital Twin of a whole complex structure which can track the loading, via sensors, and accurately predict potential failures and areas of fatigue.

Thanks to a collaboration with CNR on the Avignon Lock gate on Rhône River, our approach has been validated in real environment, and since this first demonstration, 3 others Digital Twins have been deployed on other hydraulic structures.

Residual lifetime estimation was a first application of our solutions, but with the same tools, we can also provide design validations on new structures or virtual sensors to monitor inaccessible areas of the structure or physical phenomena difficult to measure.

This approach is currently validating on other structures such as floating offshore wind turbine, large size bridge or nuclear plant.