A method for managing structural operations for modification of structures

Abstract

Disclosed is a computer-implemented method, executed by processing circuitry of a processing system, for managing structural maintenance and / or modification a structure. The method comprises acquiring measurement data representing operational conditions of the structure from a plurality of sensors; maintaining, in a memory of processing system, a digital twin representing the structure as a physics-based model derived from structural design data; integrating acquired measurement data into the digital twin to update state variables of the digital twin representing mechanical behaviour of the structure; processing, by processing circuitry, measurement data and updated digital twin using damage detection logic to detect structural damage.

Description

[0001] A METHOD FOR. MANAGING STRUCTURAL OPERATIONS FOR MODIFICATION OF STRUCTURES

[0002] TECHNICAL FIELD

[0003] The present disclosure relates to computer-implemented methods executed by processing circuitry of processing systems for managing structural maintenance and / or modification of structures. Moreover, the present disclosure relates to processing systems for managing structural maintenance and / or modification of structures. Furthermore, the present disclosure relates to computer program for executing the methods for managing structural maintenance and / or modification of structures.

[0004] BACKGROUND

[0005] Typically, structural maintenance and retrofitting of buildings, bridges and other infrastructure rely on periodic inspections, sensor monitoring and engineering assessments to evaluate condition and plan interventions.

[0006] Existing practices often depend on manual or reactive procedures that incur increased costs, extended downtime and inefficient allocation of resources. Current techniques can fail to localise and quantify damage reliably, and there is limited capability to assimilate measurement data into a physics-based computational representation of a structure to update its mechanical state. Moreover, available systems commonly lack robust tools to predict structural response under predefined loading while explicitly accounting for detected damage characteristics, and they do not provide automated mechanisms to translate analysis and prediction into timely maintenance or modification actions.Therefore, in light of the foregoing discussion, there exists a need to overcome the aforementioned drawbacks.

[0007] SUMMARY

[0008] The aim of the present disclosure is to provide a computer-implemented method executed by processing circuitry of a processing system for managing structural maintenance and / or modification of a structure; a processing system for managing structural maintenance and / or modification of a structure; and a computer program that, when executed by processing circuitry of a processing system, causes the processing system to perform the computer-implemented method. The aim of the present disclosure is achieved as defined in the appended independent claims to which reference is made to. Advantageous features are set out in the appended dependent claims.

[0009] Throughout the description and claims of this specification, the words "comprise" and "contain" and variations of the words, for example "comprising" and "comprises", mean "including but not limited to", and do not exclude other components, integers or steps. Moreover, the singular encompasses the plural unless the context otherwise requires: in particular, where the indefinite article is used, the specification is to be understood as contemplating plurality as well as singularity, unless the context requires otherwise.

[0010] BRIEF DESCRIPTION OF DRAWINGS

[0011] FIG. 1 illustrates high level method steps, in accordance with an embodiment of the present disclosure.FIG. 2 illustrates expert system advantages, in accordance with an embodiment of the present disclosure;

[0012] FIG. 3 represents a machine learning model, in accordance with an embodiment of the present disclosure;

[0013] FIG. 4A depicts low severity damage, in accordance with an embodiment of the present disclosure;

[0014] FIG. 4B depicts medium severity damage, in accordance with an embodiment of the present disclosure; and

[0015] FIG. 4C depicts high severity damage, in accordance with an embodiment of the present disclosure.

[0016] DETAILED DESCRIPTION OF EMBODIMENTS

[0017] According to a first aspect, the present disclosure provides a computer-implemented method, executed by processing circuitry of a processing system, for managing structural maintenance and / or modification of a structure, the method comprising:

[0018] (i) acquiring measurement data representing operational conditions of the structure from a plurality of sensors;

[0019] (ii) maintaining, in a memory of the processing system, a digital twin representing the structure as a physics-based computational model derived from structural design data;

[0020] (iii) integrating the acquired measurement data into the digital twin to update state variables of the digital twin representing mechanical behaviour of the structure to obtain an updated digital twin;

[0021] (iv) processing, by the processing circuitry, the measurement data and the updated digital twin using damage detection logic to detect structural damage;(v) determining, by the processing circuitry, damage characteristics of the detected structural damage within the updated digital twin;

[0022] (vi) simulating, using the digital twin and a simulation module, at least one structural response of the structure under predefined load conditions taking into account the determined damage characteristics; and

[0023] (vii) automatically initiating, by a control module, at least one maintenance and / or modification action based on the simulated structural response.

[0024] Acquisition of measurement data from a plurality of sensors and integration of those data into a digital twin improves the fidelity of the computational representation so that state variables (for example stresses, strains, stiffness distributions, modal properties and explicit defect parameters such as crack width and location) reflect the current physical condition rather than only the as-designed condition. This reduces model error and uncertainty in subsequent analyses. Use of damage detection logic operating on both measurements and the updated digital twin allows detection of damage with finer spatial and temporal sensitivity than either measurement-only or model-only approaches. Determination of damage characteristics using the digital twin (for example a localized reduction of stiffness, a crack geometry and its temporal growth rate) provides parameters which, when taken into account by a simulation module running a finite element or other physicsbased analyses under predefined load conditions (such as service loads, wind or traffic loading sequences), produce accurate predictions of structural response. Finally, automatic initiation of maintenance and / or modification actions by a control module closes the loop so that predicted unsafe or deteriorated responses lead to timely interventions, reducing required downtime, lowering the probability of unexpected failures and improving allocation of resources.The digital twin is maintained in the processing system memory as a physics-based computational model derived from structural design data such as building information model (BIM) and finite element model (FEM) representations. Structurally, the digital twin thus comprises a geometric model coupled to one or more computational models that represent material constitutive behaviour. State variables stored in the digital twin include field quantities (for example nodal displacements, element stresses), parameter vectors (for example element stiffness values, damping coefficients) and explicit defect descriptors (for example crack location coordinates, crack width in millimetres, corrosion indices). A term "state variables" in the present disclosure refers to numerical model quantities stored in the digital twin that represent mechanical behaviour, for example nodal displacements, element stresses, stiffness parameters, modal properties and explicit defect descriptors such as crack geometry. The integration of measurement data into the digital twin is performed by updating the state variables. In practice, integration can be realized by model updating routines executed by the processing circuitry, for example by adjusting local stiffness matrices where measured strain or modal deviations indicate degradation, by inserting discrete damage elements where imaging identifies cracks and by updating boundary or loading conditions to reflect measured environmental or operational conditions. This way an updated digital twin is obtained. For terminology: When the digital twin is updated then it is an updated digital twin. After the update the updated digital twin can be considered to be the digital twin which is used for simulations etc and for further updating it when the system is used again (or next round of iterations).

[0025] Damage detection logic executes on the processing circuitry and consumes both acquired measurement data and the updated digital twinstate. Functionally, the damage detection logic implements a set of analytic procedures and algorithms such as image processing and computer vision for visible defects, signal-processing and anomaly detection for time-series sensor data, and residual-based model comparisons where the measured response is compared to the (updated) digital twin's predicted response for the same loading. When the measured response deviates from the predicted response beyond defined thresholds, the damage detection logic identifies candidate damage locations and signals. For imaging sensors, quantitative defect parameters are extracted (for example defect dimension in millimetres and temporal growth rates in mm / day) using segmentation and calibration against known scales or point-cloud registration with the BIM. For distributed strain or modal monitoring, the logic identifies local stiffness reductions or modal frequency shifts consistent with damage and localizes them in the digital twin coordinates. The damage detection logic can be implemented using deterministic algorithms or machine learning models trained on labelled damage examples as described in the embodiments; where machine learning (ML) is used, training data comprise historical sensor signatures, simulated damaged model responses and annotated images.

[0026] Determining damage characteristics using the updated digital twin. Examples of damage characteristics include spatial location (coordinates in the model reference frame), geometric descriptors (crack length and width, area), severity metrics (for example local reduction factor of Young's modulus), and temporal growth parameters (for example rate of crack width increase). The processing circuitry computes these characteristics by fusing outputs of the damage detection logic with geometric information from the digital twin.The simulation module uses the updated digital twin and the determined damage characteristics, to perform physics-based simulations of structural response under predefined load conditions. Structurally, the simulation module operates on the same model representation as the digital twin (for example a finite element mesh whose elements have been modified to reflect damage characteristics) and executes structural analyses appropriate to the load cases of interest, such as static linear or nonlinear analyses, dynamic modal and transient analyses, or timehistory simulations under environmental and operational loading scenarios. The simulation module returns structural response parameters. By explicitly including the damage characteristics in the model, the simulations predict responses that account for reduced local capacities and altered load paths, thereby reducing conservatism and enabling accurate assessment of whether operational limits will be exceeded.

[0027] The control module is functionally connected to the simulation module and to interfaces for human or automated actors. The control module evaluates simulated structural responses against predefined operational limits and decision thresholds and, when predetermined criteria are met, automatically initiates maintenance and / or modification actions. Examples of such actions, include generating maintenance work orders, scheduling inspections, commanding deployment of inspection robots, initiating remote alerts to asset managers, ordering materials, or triggering local safety measures such as temporary load restrictions. The mapping from simulated response to actions may be rule-based (for example if predicted maximum crack width exceeds 1.0 mm within 30 days, then schedule corrective injection) or implemented by an autonomous agent which applies predefined decision logic and learned policies. The control module records actions, provides feedback to theprocessing system and can accept subsequent measurement data to close the loop; the system thus supports iterative adaptation where postintervention sensor data are used to validate and update the digital twin and the decision logic.

[0028] In one example of a network of sensors (strain gauges, accelerometers, environmental sensors and imaging devices) streams data for preprocessing and initial analysis. The processing system maintains a digital twin together with an FEM capable of simulating mechanical behaviour. Measurement data are assimilated to update state variables such as element stiffness and crack geometry; for visible defects the system extracts quantitative defect parameters (defect dimension and temporal defect growth rate) from camera images or LiDAR. point clouds and projects these onto the digital twin geometry. Damage detection logic that combines image processing, anomaly detection on time-series signals and model residual checks identifies damage and quantifies its characteristics. The simulation module runs a suite of simulations under predefined loading scenarios at present and at future time points (for prognostic assessment), returning responses and enabling the control module to determine whether and when to initiate maintenance. The method therefore provides a complete sensing-to-action chain in which measurement data, physics-based modelling and automated decisionmaking cooperate to deliver timely, quantitatively justified maintenance and modification actions.

[0029] Optionally, the determining of the damage characteristics of the detected structural damage and the simulating of the at least one structural response are performed by an autonomous software agent executed by the processing circuitry, the autonomous software agent being configured to perform said determining and simulating without user input, based onpredefined rules and / or learning models, the measurement data and the digital twin.

[0030] This arrangement enables the autonomous software agent to use predefined rules and learning models together with the measurement data and the digital twin to determine damage characteristics and run simulations without user input, thereby reducing decision latency, enabling continuous closed-loop sensing-to-action operation, improving consistency of determinations and timing of initiated maintenance actions, and freeing operators from routine decision tasks. A term "autonomous software agent" in the present disclosure refers to an executable software entity configured to perform damage characteristic determination and simulation tasks without user input by applying predefined rules and / or learned policies to measurement data and the digital twin.

[0031] Optionally, the plurality of sensors comprises a first set of sensors physically deployed on or within the structure at predetermined locations to measure mechanical response parameters of the structure, and / or a second set of sensors arranged externally to the structure and configured to acquire observation data representing at least one of: geometric, surface, or material characteristics of the structure by remote sensing.

[0032] The first set of sensors, physically deployed on or within the structure, provides direct measurements of mechanical response parameters that can be mapped onto the digital twin state variables to improve localisation and temporal sensitivity of damage detection, while the second set, arranged externally for remote sensing of geometric, surface or material characteristics, supplies observation data that enables extraction of quantitative defect parameters and model updates, therebyimproving simulation accuracy and decision timing. Example of the first set of sensors include accelerometers and strain gauges configured to measure axial or surface strain. Examples of the second set of sensors comprise imaging sensors such as digital cameras or LiDAR. Additionally, the first or second type of sensors can be environmental sensors for temperature, humidity and pressure measurements. These sensors supply time-stamped measurement data to the processing system via a sensing hub and communication layers so that the processing circuitry receives synchronized streams of measurement data. Measurement data can be prepossessed (filtering, averaging etc) or used as it is.

[0033] Optionally, the digital twin comprises at least one computational model selected from a building information model defining geometric and material properties of the structure, and a finite element model configured to simulate mechanical behaviour of the structure under load.

[0034] A building information model of geometric and material properties with and / or a finite element model can be used to simulate mechanical behaviour under load increases digital twin fidelity, enabling mapping of measurements to the model and more accurate simulations that reflect detected damage characteristics, improving response prediction and timing of maintenance.

[0035] Optionally, the second set of sensors comprises at least one imaging sensor configured to acquire image data representing at least one visible defect of the structure, and wherein the method further comprises: extracting, from the image data, quantitative defect parameters including at least a defect dimension and a temporal defect growth parameter; updating the digital twin based on the extracted defect parameters, performing, using the digital twin and the simulation module, a pluralityof simulations of structural response under predefined load conditions at different future time points;

[0036] and automatically determining a time window for initiating a maintenance and / or modification process, wherein the time window is determined as a time period in which a simulated structural response parameter remains within a predefined operational limit while exceeding a predefined deterioration threshold.

[0037] Extracting the defect dimension and temporal defect growth parameter from image data enables the digital twin to hold explicit, image-derived defect descriptors that feed the simulation module and permit prognostic assessment. Updating the digital twin with those quantified defect parameters causes the plurality of simulations at different future time points to reflect progressive local capacity reductions and altered load paths, yielding time-varying structural response predictions. By automatically determining the time window defined by a simulated response remaining within an operational limit while exceeding a deterioration threshold provides an objective, actionable trigger for scheduling maintenance and / or modification, reducing unnecessary early interventions and preventing late unexpected failures.

[0038] Optionally, the predefined operational limit and the predefined deterioration threshold are automatically adapted based on feedback obtained from at least one of subsequent measurement data acquired from the plurality of sensors after initiation of the maintenance and / or modification process, and a deviation between a simulated structural response and a measured structural response of the structure, the adaptation being performed by updating parameter values used by the digital twin for subsequent simulations.Automatically adapting the predefined operational limit and the predefined deterioration threshold based on subsequent measurement data and on deviations between simulated and measured structural responses provides feedback-driven calibration of decision criteria and digital twin parameters, thereby reducing model error and uncertainty, improving alignment between simulated and actual responses for subsequent simulations, and enabling more timely, accurate initiation of maintenance or modification actions.

[0039] Optionally, calculating, by the processing circuitry, a life cycle assessment and life cycle cost for maintenance and / or modification action using the digital twin and the simulated at least one structural response, wherein the life cycle management comprises environmental impact metrics derived from at least one of: estimated material quantities, construction processes, or operational effects, and the life cycle cost comprises at least one of: initial intervention costs, predicted maintenance costs, or projected service-life extension, wherein the at least one maintenance and / or modification action is selected, prioritised, or scheduled based on a comparison of the calculated life cycle assessment and life cycle cost values against predefined environmental and economic performance criteria.

[0040] The life cycle assessment and life cycle cost calculated by the processing circuitry using the digital twin and the simulated structural response provide environmental impact metrics (for example CO2 emissions, embodied energy) and economic metrics (initial intervention costs, predicted maintenance costs, projected service-life extension) that, when compared against predefined environmental and economic performance criteria, enable objective selection, prioritisation and scheduling of maintenance and modification actions. Using the digital twin andsimulated responses improves estimates of estimated material quantities, construction processes and operational effects, thereby reducing uncertainty in LCA / LCC outputs and enabling trade-off analysis that minimises environmental impact while optimising long-term cost and service-life. A term "Life Cycle Assessment (LCA)" in the present disclosure refers to an environmental impact evaluation method that quantifies metrics such as CO2 emissions and embodied energy associated with candidate maintenance or retrofitting actions over their life cycle. A term "Life Cycle Costing (LCC)" in the present disclosure refers to the economic analysis of maintenance or retrofitting alternatives that accounts for initial intervention costs, predicted maintenance costs and projected service-life extension using discounted cash-flow or equivalent techniques.

[0041] In a second aspect, the present disclosure provides a processing system for managing structural maintenance and / or modification of a structure, comprising processing circuitry, and a memory, operatively coupled to the processing circuitry, storing instructions which, when executed by the processing circuitry, cause the processing system to:

[0042] (I) acquire measurement data representing operational conditions of the structure from a plurality of sensors;

[0043] (II) maintain, in the memory, a digital twin representing the structure as a physics-based computational model derived from structural design data;

[0044] (III) integrate the acquired measurement data into the digital twin to update state variables of the digital twin representing mechanical behaviour of the structure to obtain an updated digital twin;

[0045] (IV) process the measurement data and the updated digital twin using damage detection analytics to detect structural damage;(V) determine damage characteristics wherein the damage characteristics comprise a location and a severity of the detected structural damage within the updated digital twin;

[0046] (VI) simulate, using the digital twin and a simulation module, at least one structural response of the structure under predefined load conditions taking into account the determined damage characteristics; and

[0047] (VII) automatically initiate, by a control module, at least one maintenance and / or modification action based on the simulated structural response.

[0048] The combination of a plurality of sensors delivering time-stamped measurement data, a physics-based digital twin held in memory, damage detection analytics operating on both raw measurements and updated model state, simulation of structural response that explicitly includes determined damage characteristics, and an automated control module produces synergistic technical effects that exceed the sum of the individual features. Continuous sensor acquisition and synchronous ingestion into the digital twin reduce model error by ensuring state variables track the actual structure rather than only the as-designed condition, which lowers uncertainty in subsequent analyses. High-resolution sensing and image-based analytics enable earlier detection of defects allowing corrective action before gross capacity loss. Encoding damage characteristics explicitly in the digital twin (for example crack geometry, location coordinates in the model reference frame, and a local stiffness reduction percentage) causes the simulation module to predict altered load paths and local stress concentrations accurately. The control module that evaluates these physics-based predictions against predefined operational limits thereby closes a sensing-to-action loop that reduces downtime, lowers the probability of unexpected failures and enables prioritized allocation of maintenance resources.Optionally, the instructions cause the determining of the damage characteristics of the detected structural damage and the simulating of the at least one structural response to be performed autonomously by an autonomous software agent executed by the processing circuitry, the autonomous software agent being configured to perform said determining and simulating without user input, based on predefined rules and / or learning models, the measurement data and the digital twin.

[0049] This arrangement enables the autonomous software agent to use predefined rules and learning models together with the measurement data and the digital twin to determine damage characteristics and run simulations without user input, thereby reducing decision latency, enabling continuous closed-loop sensing-to-action operation, improving consistency of determinations and timing of initiated maintenance actions, and freeing operators from routine decision tasks.

[0050] Optionally, the plurality of sensors comprises a first set of sensors physically deployed on or within the structure and / or a second set of sensors arranged externally to the structure and configured for remote sensing. The first set provides direct mechanical-response measurements mappable to digital twin state variables for accurate localisation and temporal sensitivity, and the second set supplies remote geometric / surface / material observations for extracting quantitative defect parameters; together they reduce model error and improve damage detection and simulation accuracy.

[0051] Optionally, the digital twin comprises a building information model and a finite element model. Combining a building information model and a finiteelement model increases digital twin fidelity and enables more accurate mapping of measurements and simulations.

[0052] Optionally, the second set of sensors comprises at least one imaging sensor configured to acquire image data representing at least one visible defect of the structure, and the instructions cause the processing system to extract quantitative defect parameters, update the digital twin, perform simulations at different future time points using the simulation module, and determine a time window for initiating maintenance and / or modification based on predefined operational and deterioration thresholds. Using an imaging sensor to acquire image data of visible defects enables extraction of quantitative defect parameters, for example defect dimension and temporal growth, that when used to update the digital twin permit the simulation module to run prognostic simulations at multiple future time points, yielding time-varying structural response predictions that allow the control module to determine an objectively defined time window for initiating maintenance based on the predefined operational and deterioration thresholds.

[0053] Optionally, the structure is a concrete structure, and wherein the instructions cause the processing system to identify cracking in the concrete structure, predict crack initiation and / or evolution using the digital twin under predefined load conditions, and initiate a predictive corrective action when a measured or predicted crack width exceeds a predefined crack threshold.

[0054] Technically, applying these features to a concrete structure enables the processing system to localise and quantify cracking within the digital twin and to run prognostic simulations under predefined load conditions so that predicted or measured crack width thresholds trigger automatedpredictive corrective action, thereby reducing probability of unexpected failure, enabling earlier interventions and improving allocation of maintenance resources.

[0055] According to a third aspect, the present disclosure provides a computer program comprising instructions which, when executed by processing circuitry of a processing system, cause the processing system to perform a method. The computer program is executed in the processing circuitry to perform above discussed method. The processing circuitry can be local, server based or arranged as cloud. The computer program is stored in the processing system memory as a set of software routines and data structures that the processing circuitry loads and executes.

[0056] EXAMPLE USE CASE

[0057] A concrete road bridge girder fitted with surface strain gauges and an imaging camera continuously streams measurements to the processing system where a BIM / FEM digital twin is updated; the damage detection logic extracts a 0.8 mm crack width and growth rate from images and strain anomalies and the simulation module predicts that, under service traffic loading, peak deflection will exceed the predefined operational limit within three months; the control module therefore automatically schedules an epoxy injection repair, issues a work order and temporarily reduces allowable axle loads, and the autonomous agent executes the determination, simulation and initiation steps without user input while recording the action for subsequent life-cycle cost and environmental assessment.OTHER. CONSIDERATIONS

[0058] There are issues such as inefficient maintenance practices, lack of accurate damage detection, overlooking environmental impact, economic inefficiency in maintenance decisions, complexity in selecting retrofitting techniques, and poor user interaction in the existing structural maintenance and retrofitting technology. Therefore, in order to solve the above problems, the present disclosure provides a system and method for optimizing structural maintenance and retrofitting using predictive analytics, digital twins, and user-friendly interfaces.

[0059] The present disclosure provides a comprehensive system and method for optimizing structural maintenance and retrofitting, and has the following beneficial effects:

[0060] The present disclosure enhances maintenance efficiency by employing automated predictive maintenance using sensors, Al, and digital twins to continuously monitor structural health, reducing the need for manual inspections and minimizing downtime through proactive maintenance.

[0061] The present disclosure ensures accurate damage detection and assessment by utilizing advanced sensor technology and cloud analytics to precisely locate damage and assess its severity, leading to more effective maintenance actions.

[0062] The present disclosure promotes environmental sustainability by incorporating Life Cycle Assessment (LCA) to evaluate the environmental impact of maintenance actions, including CO2 emissions, carbon footprint, and embodied energy.

[0063] The present disclosure optimizes economic outcomes by using Life CycleCosting (LCC) to identify cost-effective solutions that balance immediate expenses with long-term savings, addressing the issue of economic inefficiency in maintenance decisions.

[0064] The present invention simplifies decision-making for stakeholders by integrating a digital twin of the structure for detailed analysis and simulation, and providing Al-driven recommendations on suitable retrofitting techniques, addressing the complexity and variability of structural conditions and external factors.

[0065] The present invention improves user engagement and interaction by implementing a GAI (general artificial intelligence) chatbot that facilitates natural language communication between end-users and the system, making it easy for users to understand recommendations and make informed decisions.

[0066] Example steps:

[0067] The purpose of the disclosure is to overcome the drawbacks in the existing technology and provide a method for optimizing structural maintenance and retrofitting, which includes:

[0068] Step 1, continuously monitoring the structural health using sensors, Al, and digital twins for automated predictive maintenance;

[0069] Step 2, precisely locating and assessing the severity of structural damage using advanced sensor technology and cloud analytics;Step 3, judge if maintenance or retrofitting is required based on the damage assessment, if yes, return to Step 1 for continuous monitoring, if no, proceed to Step 4;

[0070] Step 4, optimizing maintenance and retrofitting decisions based on Life Cycle Assessment (LCA), Life Cycle Costing (LCC), and Al-driven recommendations.

[0071] Step 1 includes:

[0072] Step 101, deploying a network of smart sensors, such as strain gauges, accelerometers, and environmental sensors, on the structure to collect real-time data on structural conditions, environmental factors, and potential damage indicators;

[0073] Step 102, integrating the sensor data with a digital twin, a virtual replica of the structure, to create a comprehensive digital representation of the structure's current state;

[0074] Step 103, applying machine learning algorithms and predictive analytics to the sensor data and digital twin to continuously monitor the structural health, identify potential issues, and trigger proactive maintenance alerts.

[0075] Step 2 includes:

[0076] Step 201, analyzing the sensor data and digital twin data using advanced cloud-based analytics and machine learning models to precisely locate and characterize structural damage, such as cracks, deformations, or material degradation;

[0077] Step 202, assessing the severity of the detected damage by considering factors like location, extent, and potential impact on structural integrity; Step 203, generating detailed damage reports and visualizations, including

[0078] 3D representations and severity ratings, to aid in decision-making.

[0079] Step 4 includes:Step 401, evaluating the environmental impact of potential maintenance and retrofitting actions using Life Cycle Assessment (LCA) techniques, considering factors like CO2 emissions, carbon footprint, and embodied energy;

[0080] Step 402, performing Life Cycle Costing (LCC) analysis to identify the most cost-effective maintenance and retrofitting solutions, balancing immediate expenses with long-term savings and operational costs;

[0081] Step 403, utilizing Al-driven recommendation engines to suggest suitable retrofitting techniques based on the damage assessment, environmental impact analysis, and economic optimization;

[0082] Step 404, integrating a GAI chatbot interface to facilitate natural language communication between end-users and the system, enabling users to understand recommendations, provide feedback, and make informed decisions.

[0083] Description of some workflow cases

[0084] Case 1 example

[0085] A system for optimizing structural maintenance and retrofitting comprises a network of smart sensors deployed on a structure, a digital twin of the structure, and a cloud-based analytics platform. The system continuously monitors the structural health using automated predictive maintenance techniques.

[0086] Step 101: A network of smart sensors, including strain gauges, accelerometers, and environmental sensors, is deployed on the structure to collect real-time data on structural conditions, environmental factors, and potential damage indicators. The strain gauges measure deformations and strains within a range of 0.001% to 5%, with a resolution of 0.0001%. The accelerometers detect vibrations andaccelerations in the range of 0.001 g to 10 g, with a sampling rate of 1 kHz to 10 kHz. The environmental sensors measure temperature from -40°C to 85°C with an accuracy of ±0.1°C, humidity from 0% to 100% with an accuracy of ±2%, and atmospheric pressure from 300 hPa to 1100 hPa with an accuracy of ±0.1 hPa.

[0087] Step 102: The sensor data is integrated with a digital twin, a virtual replica of the structure, to create a comprehensive digital representation of the structure's current state. The digital twin is a 3D model of the structure, built using computer-aided design (CAD) software and incorporating material properties, structural specifications, and environmental conditions. The digital twin is updated in real-time with the sensor data, enabling a dynamic representation of the structure's behavior. BIM (Building Information Modelling) model will be utilized as a virtual replica of the physical structure.

[0088] Step 103: Machine learning algorithms and predictive analytics are applied to the sensor data and digital twin to continuously monitor the structural health, identify potential issues, and trigger proactive maintenance alerts. The algorithms use techniques such as time-series analysis, pattern recognition, and anomaly detection to identify deviations from expected behavior. When potential issues are detected, the system generates alerts and maintenance recommendations, which are communicated to the stakeholders through a user interface or a GAI chatbot.

[0089] Step 201: The sensor data and digital twin data are analyzed using advanced cloud-based analytics and machine learning models to precisely locate and characterize structural damage, such as cracks, deformations, or material degradation. The analysis employs techniques like imageprocessing, finite element analysis, and machine learning-based damage detection algorithms.

[0090] The system can detect cracks with a minimum width of 0.1 mm and deformations as small as 0.01% of the structural dimensions.

[0091] Step 202: The severity of the detected damage is assessed by considering factors like location, extent, and potential impact on structural integrity. The system uses structural analysis models and failure criteria to evaluate the remaining load-bearing capacity and the risk of failure. The damage severity is classified into levels ranging from minor (no immediate action required) to critical (immediate intervention necessary).

[0092] Step 203: Detailed damage reports and visualizations are generated, including 3D representations and severity ratings, to aid in decisionmaking. The reports include information such as the location and dimensions of the damage, the estimated remaining service life, and recommended maintenance or retrofitting actions.

[0093] Step 401: The environmental impact of potential maintenance and retrofitting actions is evaluated using Life Cycle Assessment (LCA) techniques, considering factors like CO2 emissions, carbon footprint, and embodied energy. The LCA analysis considers the entire life cycle of the maintenance or retrofitting process, from material extraction and manufacturing to transportation, installation, and end-of-life disposal.

[0094] Step 402: Life Cycle Costing (LCC) analysis is performed to identify the most cost-effective maintenance and retrofitting solutions, balancing immediate expenses with long-term savings and operational costs. The LCC analysis considers factors such as material costs, labor costs, energyconsumption, and maintenance requirements over the expected service life of the structure.

[0095] Step 403: Al-driven recommendation engines are utilized to suggest suitable retrofitting techniques based on the damage assessment, environmental impact analysis, and economic optimization. The recommendation engines use machine learning models trained on historical data and expert knowledge to identify the most appropriate retrofitting techniques for the specific damage scenario.

[0096] Step 404: A GAI chatbot interface is integrated to facilitate natural language communication between end-users and the system, enabling users to understand recommendations, provide feedback, and make informed decisions. The chatbot uses natural language processing and generation techniques to interpret user queries and provide clear and concise responses, including explanations of the recommended actions and their rationale.

[0097] Case 2 example:

[0098] A method for optimizing structural maintenance and retrofitting comprises the following steps:

[0099] Step 101: Deploying a wireless sensor network consisting of strain gauges, accelerometers, and environmental sensors on the structure. The strain gauges have a measurement range of 0.005% to 3% and an accuracy of ±0.1%. The accelerometers have a measurement range of 0.01 g to 8 g and a sampling rate of 2 kHz to 8 kHz. The environmental sensors measure temperature from -20°C to 60°C with an accuracy of ±0.5°C, humidity from 10% to 90% with an accuracy of ±3%, and atmospheric pressure from 500 hPa to 1000 hPa with an accuracy of ±0.2 hPa.Step 102: Creating a digital twin of the structure using finite element modeling software and incorporating the structural design specifications, material properties, and environmental conditions. The digital twin is a 3D model that can simulate the structural behavior under various loading scenarios and environmental conditions.

[0100] Step 103: Integrating the sensor data with the digital twin and applying machine learning algorithms, such as support vector machines and neural networks, to continuously monitor the structural health. The algorithms are trained on historical data and simulations to identify patterns and anomalies that may indicate potential structural issues.

[0101] Step 201: Analyzing the sensor data and digital twin data using cloudbased image processing techniques and machine learning models to detect and locate structural damage, such as cracks, deformations, and material degradation. The system can detect cracks with a minimum width of 0.2 mm and deformations as small as 0.02% of the structural dimensions.

[0102] Step 202: Assessing the severity of the detected damage by performing finite element analysis and structural reliability analysis on the digital twin. The analysis considers factors such as the location and extent of the damage, the applied loads, and the material properties. The damage severity is classified into levels ranging from minor (no immediate action required) to critical (immediate intervention necessary).

[0103] Step 203: Generating detailed damage reports and visualizations, including 3D representations of the damage location and extent, as well as severity ratings and recommended maintenance or retrofitting actions.Step 401: Conducting Life Cycle Assessment (LCA) using industrystandard software and databases to evaluate the environmental impact of potential maintenance and retrofitting actions. The LCA considers factors such as CO2 emissions, energy consumption, and waste generation throughout the entire life cycle of the maintenance or retrofitting process.

[0104] Step 402: Performing Life Cycle Costing (LCC) analysis using discounted cash flow techniques to identify the most cost-effective maintenance and retrofitting solutions. The LCC analysis considers factors such as material costs, labor costs, energy costs, and maintenance requirements over the expected service life of the structure, with a discount rate of 5% to 10%.

[0105] Step 403: Utilizing Al-driven recommendation engines based on decision tree algorithms and expert systems to suggest suitable retrofitting techniques based on the damage assessment, environmental impact analysis, and economic optimization. The recommendation engines are trained on a database of historical retrofitting projects and expert knowledge.

[0106] Step 404: Implementing a GAI chatbot interface using natural language processing and generation techniques, such as transformer-based language models, to facilitate natural language communication between end-users and the system. The chatbot can interpret user queries, provide explanations of the recommended actions, and assist in decisionmaking.

[0107] Case 3 example:A system for optimizing structural maintenance and retrofitting comprises the following components:

[0108] 1. A network of smart sensors deployed on the structure, including:

[0109] - Fiber optic strain sensors with a measurement range of 0.01% to 2% and an accuracy of ±0.05%

[0110] - Triaxial accelerometers with a measurement range of 0.005 g to 5 g and a sampling rate of 5 kHz to 20 kHz

[0111] - Environmental sensors measuring temperature from -30°C to 70°C with an accuracy of ±0.3°C, humidity from 5% to 95% with an accuracy of ±2%, and atmospheric pressure from 400 hPa to 1200 hPa with an accuracy of ±0.1 hPa

[0112] 2. A digital twin of the structure, created using computational fluid dynamics (CFD) and finite element analysis (FEA) software, incorporating structural design specifications, material properties, and environmental conditions.

[0113] 3. A cloud-based analytics platform with the following components:

[0114] - Machine learning algorithms, such as random forests and deep neural networks, for continuous structural health monitoring and predictive maintenance

[0115] - Image processing and computer vision algorithms for damage detection and localization

[0116] - Structural analysis and simulation tools for assessing damage severity and remaining service life

[0117] - Life Cycle Assessment (LCA) and Life Cycle Costing (LCC) modules for environmental impact and economic optimization analysis

[0118] - Al-driven recommendation engines based on reinforcement learning and

[0119] multi-objective optimization techniques for suggesting retrofitting techniques- A GAI chatbot interface using natural language processing and generation techniques, such as transformer-based language models and dialogue management systems, for user interaction and decision support The system operates as follows:

[0120] Step 101: The network of smart sensors continuously collects real-time data on structural conditions, environmental factors, and potential damage indicators.

[0121] Step 102: The sensor data is integrated with the digital twin, enabling a dynamic representation of the structure's behavior and simulating various loading scenarios and environmental conditions.

[0122] Step 103: The cloud-based analytics platform applies machine learning algorithms and predictive analytics to the sensor data and digital twin to continuously monitor the structural health, identify potential issues, and trigger proactive maintenance alerts.

[0123] Step 201: The analytics platform uses image processing and computer vision algorithms to analyze the sensor data and digital twin data, precisely locating and characterizing structural damage, such as cracks, deformations, or material degradation.

[0124] Step 202: The platform performs structural analysis and simulations on the digital twin, considering the detected damage, to assess the severity of the damage and its potential impact on structural integrity.

[0125] Step 203: Detailed damage reports and visualizations, including 3D representations and severity ratings, are generated to aid in decisionmaking and communicate recommended maintenance or retrofitting actions.

[0126] Step 401: The LCA module evaluates the environmental impact of potential maintenance and retrofitting actions, considering factors like CO2 emissions, carbon footprint, and embodied energy throughout the entire life cycle.Step 402: The LCC module performs economic optimization analysis using discounted cash flow techniques to identify the most cost-effective maintenance and retrofitting solutions, balancing immediate expenses with long-term savings and operational costs.

[0127] Step 403: The Al-driven recommendation engines suggest suitable retrofitting techniques based on the damage assessment, environmental impact analysis, and economic optimization, using reinforcement learning and multi-objective optimization techniques.

[0128] Step 404: The GAI chatbot interface facilitates natural language communication between end-users and the system, enabling users to understand recommendations, provide feedback, and make informed decisions through a conversational interface. While the GAI chatbot interface enhances user interaction through natural language communication, its capabilities are limited to reactive and predefined conversational workflows.

[0129] In contrast, "intelligent agents" go beyond by proactively learning user preferences, autonomously adapting to dynamic scenarios, and executing complex tasks without direct user input, offering superior operational efficiency and decision-making support.

[0130] DETAILED DESCRIPTION OF DRAWINGS

[0131] FIG. 1 illustrates high level method steps, in accordance with an embodiment of the present disclosure. FIG 1. presents an example flow 100 of the approach, integrating sensors 120 , loT, edge computing, Al, and digital twin 130 technology to assess damage location, severity, andprovide retrofit recommendations based on environmental and economic impacts.

[0132] FIG. 1 illustrates an exemplary embodiment of a digital-twin-based predictive maintenance system for structural maintenance and modification. The system comprises three main operational stages (1-3) arranged in a hierarchical data-processing architecture.

[0133] Stage 1 - Data Acquisition and Preprocessing

[0134] At a first stage (1), a plurality of sensors are deployed on or around a structure. These sensors may include mechanical, environmental, imaging, or other monitoring devices configured to acquire operational and condition-related data.

[0135] The sensing hub (120) aggregates sensor outputs and transfers the data to a data-processing layer comprising:

[0136] • Data processing modules

[0137] • loT infrastructure

[0138] • Edge computing units

[0139] • Cloud analytics components

[0140] The data are filtered, normalized, and structured for further analysis. Preprocessing may include signal conditioning, feature extraction, anomaly filtering, and timestamp synchronization; and images of the damages are analyzed to characterize damage location and severity. Within this framework:

[0141] 1. Location of Damage is classified (e.g., local, element-level, building-level).

[0142] 2. Severity of Damage is categorized (e.g., high, medium, low).Processed data are then provided to the predictive maintenance framework.

[0143] Stage 2 - Predictive Maintenance Framework (Al & Digital Twin) Stage (2) corresponds to a predictive maintenance framework integrating:

[0144] • Artificial intelligence (Al) models

[0145] • A digital twin of the monitored structure

[0146] In this stage, the location and severity of the detected damage are mapped onto the virtual representation of the structure within the digital twin, which may comprise a Building Information Model (BIM) and / or a Finite Element Model (FEM).

[0147] Multiple structural response scenarios are then simulated using the digital twin in order to identify critical regions or "hot spots" within the structure. Based on the simulated scenarios and the identified damage characteristics, the Al model predicts the environmental and economic impacts associated with the required retrofit, including life cycle assessment (LCA) and life cycle cost (LCC) parameters.

[0148] The structured data from Stage 1 is injected into the digital twin model to:

[0149] • Update structural state variables,

[0150] • Simulate structural response under predefined loading conditions, • Estimate degradation trends.

[0151] The Al module depicted in FIG. 1 is described in further detail in FIG. 3. Outputs of Stage 2 include:

[0152] • Damage classification,

[0153] • Predicted remaining useful life (R.UL),

[0154] • Environmental and economic impact indicators (LCA and LCC), • A retrofit class determination (high, medium, low).These outputs are provided to Stage 3.

[0155] Stage 3 - Customized Agent for Inspection and Retrofit

[0156] Stage (3) comprises a customized autonomous agent configured to generate inspection and retrofit recommendations using generative artificial intelligence and large language models (GAI-LLM).

[0157] This agent:

[0158] • Receives outputs from Stage 2,

[0159] • Evaluates environmental and economic indicators,

[0160] • Determines retrofit priority and scheduling,

[0161] • Generates maintenance or retrofit recommendations,

[0162] • Supports early warning functionality.

[0163] The detailed structure and functional outputs of Stage 3 are illustrated in FIG. 2.

[0164] FIG. 2 illustrates expert system advantages, in accordance with an embodiment of the present disclosure. FIG 2. presents an example of the advantages of the expert system, highlighting features like structural health monitoring, remaining useful life prediction, retrofit schemes, and economic and environmental assessments. FIG. 2 further illustrates the architecture and functional outputs of the automated expert system agent.

[0165] The agent interfaces bidirectionally with an asset owner or management system and comprises the following decision layers:

[0166] 1. Structural Health Monitoring (SHM)

[0167] • Sensor data handling

[0168] • Damage tracking

[0169] 2. Remaining Useful Life (RUL) EstimationElement-level assessment

[0170] Structure-level assessment

[0171] 3. Suggested Retrofit Scheme

[0172] • Maintenance timing (when)

[0173] • Maintenance type

[0174] • Duration estimation

[0175] 4. Environmental Impact (LCA)

[0176] • CO2emissions

[0177] • Embodied energy

[0178] 5. Economic Impact (LCC)

[0179] • Initial intervention cost

[0180] • Life-cycle maintenance cost

[0181] 6. Early Warning System

[0182] • Triggered when predicted thresholds are exceeded.

[0183] The agent integrates technical outputs from the Al and digital twin layer (Stage 2 in FIG. 1) and converts them into structured decision outputs suitable for inspection scheduling and retrofit planning.

[0184] Thus, FIG. 2 represents the operational implementation of Stage (3) in FIG. 1.

[0185] FIG. 3 represents a machine learning model, in accordance with an embodiment of the present disclosure. FIG 3. presents an example of a machine learning model for retrofitting reinforced concrete structures, detailing the inputs (damage location, severity) and outputs (economic and environmental impacts, retrofit class), along with contributing factors such as deterioration extent and safety considerations. Indeed the FIG.

[0186] 3 illustrates the internal structure of the Al component shown in FIG. 1. The Al module receives structured inputs including:

[0187] • Location of damage• Severity of damage

[0188] • Retrofitting schemes

[0189] • Empirical datasets

[0190] • Industry databases

[0191] • Literature reviews

[0192] Additional contextual factors include:

[0193] • Extent of deterioration

[0194] • Performance indicators

[0195] • Structural type

[0196] • Applicable codes of practice

[0197] • Expert knowledge

[0198] • Industrial standards

[0199] The Al model processes these inputs through machine-learning or rulebased computational models.

[0200] The outputs of the Al module include:

[0201] • Environmental and economic impacts (LCA and LCC),

[0202] • Retrofit class classification (high, medium, low).

[0203] These outputs are transferred back to the predictive maintenance framework (Stage 2 in FIG. 1), and subsequently to the autonomous agent (Stage 3 in FIG. 1).

[0204] Figures FIGI, FIG2, and FIG 3 have thus functional relationships as follows. The three figures represent a layered architecture:

[0205] • FIG. 1 provides the system-level overview.

[0206] • FIG. 3 details the internal Al engine used in Stage 2 of FIG. 1. • FIG. 2 details the autonomous expert agent corresponding to Stage 3 of FIG. 1.

[0207] The data flow is as follows:Sensors — Preprocessing Al & Digital Twin (FIG. 3) — Impact & Retrofit Classification — Autonomous Agent (FIG. 2) — Maintenance / Retrofit Action.

[0208] This architecture forms a closed-loop predictive maintenance and modification system in which:

[0209] 1. Sensor data continuously update the digital twin,

[0210] 2. Al evaluates structural condition and impacts,

[0211] 3. The autonomous agent generates optimized intervention decisions, 4. The system may trigger corrective actions or early warnings.

[0212] FIG. 4A depicts low severity damage, in accordance with an embodiment of the present disclosure. A crack feature 450A is presented. The figure shows a narrow surface crack running along a soffit or beam face; the crack is shallow, exhibits no concrete spalling and does not expose reinforcement, and is therefore classified in the low severity category. In the drawing the low severity crack is intended to illustrate a defect whose measured defect dimension and temporal defect growth rate are below deterioration thresholds used by the severity assessment module and the machine learning classifier, and which therefore would typically trigger monitoring and scheduled minor maintenance rather than immediate structural intervention.

[0213] FIG. 4B depicts medium severity damage, in accordance with an embodiment of the present disclosure. Figure shows medium crack and spall features and corroded region shows multiple interconnected cracks 450B accompanied by localized material loss and exposed or corroded reinforcement indicated by corroded region; these visual features result in a medium severity classification. Structurally, the medium severity features correspond to quantified defect parameters such as larger crack widths and measurable spalled area that when projected onto the digitaltwin geometry produce element-level severity metrics used by the simulation module to predict reduced local capacity and to schedule corrective retrofit within a defined time window.

[0214] FIG. 4C depicts high severity damage, in accordance with an embodiment of the present disclosure. A wide crack and extensive spalling shows pronounced cracking 450B with substantial concrete loss and likely reinforcement exposure or section loss; these conditions are assigned a high severity classification in the decision logic. The high severity features illustrated in FIG. 4C are representative of defects whose extracted quantitative parameters exceed operational safety thresholds and therefore cause the predictive maintenance framework and control module to prioritise immediate intervention actions, such as emergency restrictions, rapid repair scheduling or detailed structural assessment.

Claims

CLAIMS1. A computer-implemented method (100), executed by processing circuitry of a processing system, for managing structural maintenance and / or modification a structure, the method comprising:(i) acquiring measurement data representing operational conditions of the structure from a plurality of sensors (120);(ii) maintaining, in a memory of the processing system, a digital twin (130) representing the structure as a physics-based computational model derived from structural design data;(iii) integrating the acquired measurement data into the digital twin to update state variables of the digital twin representing mechanical behaviour of the structure to obtain an updated digital twin;(iv) processing, by the processing circuitry, the measurement data and the updated digital twin using damage detection logic to detect structural damage;(v) determining, by the processing circuitry, damage characteristics of the detected structural damage within the updated digital twin;(vi) simulating, using the updated digital twin and a simulation module, at least one structural response of the structure under predefined load conditions taking into account the determined damage characteristics; and(vii) automatically initiating, by a control module, at least one maintenance and / or modification action based on the simulated structural response.

2. The method according to claim 1, wherein the determining of the damage characteristics of the detected structural damage and the simulating of the at least one structural response are performed by an autonomous software agent executed by the processing circuitry, the autonomous software agent being configured to perform said determining and simulating without user input, based on predefined rules and / or learning models, the measurement data and the digital twin.

3. The method according to any of claims 1 or 2, wherein the plurality of sensors comprises a first set of sensors physically deployed on or within the structure at predetermined locations to measure mechanical response parameters of the structure, and / or a second set of sensors arranged externally to the structure and configured to acquire observation data representing at least one of: geometric, surface, or material characteristics of the structure by remote sensing.

4. The method according to any of claims 1-3, wherein the digital twin comprises at least one computational model selected from a building information model defining geometric and material properties of the structure, and a finite element model configured to simulate mechanical behaviour of the structure under load.

5. The method according to any of claims 3-4, wherein the second set of sensors comprises at least one imaging sensor configured to acquire image data representing at least one visible defect of the structure, and wherein the method further comprises:extracting, from the image data, quantitative defect parameters including at least a defect dimension and a temporal defect growth parameter; updating the digital twin based on the extracted defect parameters, performing, using the digital twin and the simulation module, a plurality of simulations of structural response under predefined load conditions at different future time points; andautomatically determining a time window for initiating a maintenance and / or modification process, wherein the time window is determined as a time period in which a simulated structural response parameter remains within a predefined operational limit while exceeding a predefined deterioration threshold.

6. The method according to claim 5, wherein the predefined operational limit and the predefined deterioration threshold are automatically adapted based on feedback obtained from at least one of subsequent measurement data acquired from the plurality of sensors after initiation of the maintenance and / or modification process, and a deviation between a simulated structural response and a measured structural response of the structure, the adaptation being performed by updating parameter values used by the digital twin for subsequent simulations.

7. The method according to any of the preceding claims, wherein the method further comprisescalculating, by the processing circuitry, a life cycle assessment and life cycle cost for maintenance and / or modification action using the digital twin and the simulated at least one structural response, wherein the life cycle management comprises environmental impact metrics derived from at least one of: estimated material quantities, construction processes, or operational effects, and the life cycle cost comprises at leastone of: initial intervention costs, predicted maintenance costs, or projected service-life extension,wherein the at least one maintenance and / or modification action is selected, prioritised, or scheduled based on a comparison of the calculated life cycle assessment and life cycle cost values against predefined environmental and economic performance criteria.

8. A processing system for managing structural maintenance and / or modification a structure, comprisingprocessing circuitry, and a memory, operatively coupled to the processing circuitry, storing instructions which, when executed by the processing circuitry, cause the processing system to:(I) acquire measurement data representing operational conditions of the structure from a plurality of sensors;(II) maintain, in the memory, a digital twin representing the structure as a physics-based computational model derived from structural design data;(III) integrate the acquired measurement data into the digital twin to update state variables of the digital twin representing mechanical behaviour of the structure to obtain an updated digital twin;(IV) process the measurement data and the updated digital twin using damage detection analytics to detect structural damage;(V) determine damage characteristics wherein the damage characteristics comprise a location and a severity of the detected structural damage within the updated digital twin;(VI) simulate, using the digital twin and a simulation module, at least one structural response of the structure under predefined load conditions taking into account the determined damage characteristics and(VII) automatically initiate, by a control module, at least one maintenance and / or modification action based on the simulated structural response.

9. The processing system according to claim 8, wherein the instructions cause the determining of the damage characteristics of the detected structural damage and the simulating of the at least one structural response to be performed autonomously by an autonomous software agent executed by the processing circuitry, the autonomous software agent being configured to perform said determining and simulating without user input, based on predefined rules and / or learning models, the measurement data and the digital twin.

10. The processing system according to any of claims 8 or 9, wherein the plurality of sensors comprises a first set of sensors physically deployed on or within the structure and / or a second set of sensors arranged externally to the structure and configured for remote sensing.

11. The processing system according to any of claims 8-10, wherein the digital twin comprises a building information model and a finite element model.

12. The processing system according to any of claims 10-11, wherein the second set of sensors comprises at least one imaging sensor configured to acquire image data representing at least one visible defect of the structure, and wherein the instructions cause the processing system toextract quantitative defect parameters,update the digital twin,perform simulations at different future time points using the simulation module, anddetermine a time window for initiating maintenance and / or modification based on predefined operational and deterioration thresholds.

13. The processing system according to any of claims 8-12, wherein the structure is a concrete structure, and wherein the instructions cause the processing system to identify cracking in the concrete structure, predict crack initiation and / or evolution using the digital twin under predefined load conditions, and initiate a predictive corrective action when a measured or predicted crack width exceeds a predefined crack threshold.

14. A computer program comprising instructions which, when executed by processing circuitry of a processing system, cause the processing system to perform a method according to any of the claims 1-7.

Images