Periodic Reporting for period 1 - ObSerVation (Optimization of Seismic Structural Health Monitoring Systems Based on Value of Information Analysis)
Reporting period: 2023-11-01 to 2026-04-30
Summary of the context and overall objectives of the project
Observation (Optimization of Seismic Structural Health Monitoring Systems Based on Value of Information Analysis) addresses the need for more reliable, data-informed management of structures and infrastructure systems exposed to seismic hazard. Aging bridges and buildings, limited maintenance resources, and the high social and economic consequences of earthquake damage make it essential to improve the use of monitoring information in engineering and emergency decisions. Although Structural Health Monitoring systems can provide valuable data on structural condition and seismic response, their use is still often limited by uncertainty about their actual benefit for decision-making.
The overall objective of the project is to develop probabilistic, Bayesian and decision-oriented methodologies for seismic risk assessment, structural monitoring and seismic emergency management. The project aims to improve the prediction of seismic demand, structural response and damage, and to quantify how additional information from monitoring systems can support better management actions before and after earthquakes. The research combined probabilistic seismic hazard analysis, ground-motion modeling, structural reliability, Bayesian inference, transfer learning, Value of Information, and Structural Health Monitoring.
The pathway to impact is based on linking advanced scientific models to practical decisions faced by infrastructure owners, engineers and civil-protection authorities. Rather than evaluating monitoring systems only in terms of instrumentation density or data quality, the project assesses their value according to the decisions they support. This is particularly important for transportation networks and public buildings, where post-earthquake decisions such as keeping a structure open, closing it, or prioritizing inspections have direct implications for safety and connectivity.
A central motivation of the project is also the need to better account for seismic sequences. Existing engineering models often focus on mainshocks, although aftershocks may strongly affect structures that have already been damaged. ObSerVation therefore develops mainshock–aftershock modelling approaches to improve seismic demand and damage assessment for damaged systems.
The expected impacts are scientific, technological, economic and societal. Scientifically, the project advances the integration of monitoring data, ground-motion modelling, Bayesian learning and decision analysis. Technologically, it supports the development of decision-support tools for monitoring-system design and seismic risk management. Economically, it contributes to a more efficient allocation of monitoring, inspection and maintenance resources. Societally, it supports safer and more resilient infrastructure and buildings in earthquake-prone regions, helping reduce the consequences of structural failure, service disruption and unnecessary closures.
The overall objective of the project is to develop probabilistic, Bayesian and decision-oriented methodologies for seismic risk assessment, structural monitoring and seismic emergency management. The project aims to improve the prediction of seismic demand, structural response and damage, and to quantify how additional information from monitoring systems can support better management actions before and after earthquakes. The research combined probabilistic seismic hazard analysis, ground-motion modeling, structural reliability, Bayesian inference, transfer learning, Value of Information, and Structural Health Monitoring.
The pathway to impact is based on linking advanced scientific models to practical decisions faced by infrastructure owners, engineers and civil-protection authorities. Rather than evaluating monitoring systems only in terms of instrumentation density or data quality, the project assesses their value according to the decisions they support. This is particularly important for transportation networks and public buildings, where post-earthquake decisions such as keeping a structure open, closing it, or prioritizing inspections have direct implications for safety and connectivity.
A central motivation of the project is also the need to better account for seismic sequences. Existing engineering models often focus on mainshocks, although aftershocks may strongly affect structures that have already been damaged. ObSerVation therefore develops mainshock–aftershock modelling approaches to improve seismic demand and damage assessment for damaged systems.
The expected impacts are scientific, technological, economic and societal. Scientifically, the project advances the integration of monitoring data, ground-motion modelling, Bayesian learning and decision analysis. Technologically, it supports the development of decision-support tools for monitoring-system design and seismic risk management. Economically, it contributes to a more efficient allocation of monitoring, inspection and maintenance resources. Societally, it supports safer and more resilient infrastructure and buildings in earthquake-prone regions, helping reduce the consequences of structural failure, service disruption and unnecessary closures.
Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far
The work performed during the fellowship produced a coherent set of probabilistic and decision-oriented methodologies for seismic structural health monitoring, seismic demand modeling, monitored-building assessment, and damage modeling under mainshock–aftershock sequences.
The first major activity concerned the Value of Information provided by monitoring systems for seismic emergency management. Methodologies based on Value of Information and Relative Entropy were developed and applied to quantify the benefit of monitoring data in decision-making contexts. The work included prior and pre-posterior analyses, probabilistic modeling of seismic demand and structural capacity, representation of damage states, and comparison of expected management costs with and without monitoring information. This research was applied especially to bridge systems and transportation networks, where post-earthquake management decisions can have important safety and economic consequences.
Within this area, a methodology was developed for the optimal placement of strong-ground-motion sensors. Seismic demand was modeled through an event-based hazard approach using Monte Carlo simulation and ground-motion prediction equations, while the value of information was used as the optimization criterion. The framework accounted for uncertainty in future earthquakes, spatial correlation of intensity measures, and the relationship between demand estimates and bridge-network reliability.
A related activity examined how assumptions about seismic sources affect the value of information from ground-motion sensors. Different source representations, including area-source and fault-source models, were compared to assess their influence on seismic demand, expected damage and optimal monitoring configurations. The results showed that monitoring-system design is sensitive to the adopted hazard representation.
A second research line investigated the influence of Ground Motion Model selection on the benefit derived from monitoring data. Several candidate models applicable to Central Italy were assessed using statistical goodness-of-fit measures and then propagated into Value of Information and Relative Entropy calculations for an exemplary bridge under seismic emergency-management conditions. This work showed that ground-motion model selection is not only a hazard-modelling issue, but also a decision-support issue, because it directly affects estimated demand, reliability and the value assigned to monitoring data.
Another major achievement was the development of aftershock-specific ground-motion models. A Bayesian structured random-field formulation was developed starting from established NGA-West2 models and introducing corrective functions to capture aftershock-specific effects. The model coefficients were parameterized as smooth functions of spectral period, ensuring spectrally continuous median predictions. A hierarchical Bayesian structure was used to separate sequence-level, event-level and record-level variability, while spike-and-slab priors supported predictor selection and reduced overfitting. A dedicated aftershock dataset was created from the NGA-West2 database through sequence identification, record selection, metadata quality control and spectral usability checks.
A further activity focused on developing mainshock–aftershock ground-motion models for Italy using the ITACA dataset. Italian strong-motion records were processed to distinguish mainshock and aftershock observations within a common hierarchical Bayesian framework. Separate regression coefficients were used for mainshock and aftershock records, while shared sequence-level dependence was preserved. This provides a step toward Italy-specific mainshock–aftershock ground-motion models based on recorded data.
The project also developed building-specific probabilistic seismic demand models for monitored existing reinforced-concrete buildings using sparse structural health monitoring data. A hybrid Bayesian transfer-learning framework was formulated to combine nonlinear time-history analyses of representative archetype buildings with sparse observations from monitored buildings. Source-domain models were built from nonlinear analyses of non-ductile infilled reinforced-concrete archetypes subjected to mainshock–aftershock sequences. The target-domain layer then used monitoring data to recalibrate the low-intensity part of the demand model while preserving the nonlinear response pattern inherited from the archetypes. The methodology was applied to real monitored buildings.
A final major activity concerned probabilistic damage-index modeling for reinforced-concrete frame structures with various stories under mainshock–aftershock sequences. Structural damage was quantified using the Park–Ang damage index, considering both story-specific and global damage measures. A two-stage modeling strategy was developed. First, estimating mainshock damage, and then modeling the aftershock-induced damage increment conditional on the damage state left by the mainshock. This made it possible to represent damage accumulation as a path-dependent process. A multivariate Bayesian formulation was also introduced to model story-level damage while accounting for vertical dependence among stories, censoring, nonlinear predictors and uncertainty.
The first major activity concerned the Value of Information provided by monitoring systems for seismic emergency management. Methodologies based on Value of Information and Relative Entropy were developed and applied to quantify the benefit of monitoring data in decision-making contexts. The work included prior and pre-posterior analyses, probabilistic modeling of seismic demand and structural capacity, representation of damage states, and comparison of expected management costs with and without monitoring information. This research was applied especially to bridge systems and transportation networks, where post-earthquake management decisions can have important safety and economic consequences.
Within this area, a methodology was developed for the optimal placement of strong-ground-motion sensors. Seismic demand was modeled through an event-based hazard approach using Monte Carlo simulation and ground-motion prediction equations, while the value of information was used as the optimization criterion. The framework accounted for uncertainty in future earthquakes, spatial correlation of intensity measures, and the relationship between demand estimates and bridge-network reliability.
A related activity examined how assumptions about seismic sources affect the value of information from ground-motion sensors. Different source representations, including area-source and fault-source models, were compared to assess their influence on seismic demand, expected damage and optimal monitoring configurations. The results showed that monitoring-system design is sensitive to the adopted hazard representation.
A second research line investigated the influence of Ground Motion Model selection on the benefit derived from monitoring data. Several candidate models applicable to Central Italy were assessed using statistical goodness-of-fit measures and then propagated into Value of Information and Relative Entropy calculations for an exemplary bridge under seismic emergency-management conditions. This work showed that ground-motion model selection is not only a hazard-modelling issue, but also a decision-support issue, because it directly affects estimated demand, reliability and the value assigned to monitoring data.
Another major achievement was the development of aftershock-specific ground-motion models. A Bayesian structured random-field formulation was developed starting from established NGA-West2 models and introducing corrective functions to capture aftershock-specific effects. The model coefficients were parameterized as smooth functions of spectral period, ensuring spectrally continuous median predictions. A hierarchical Bayesian structure was used to separate sequence-level, event-level and record-level variability, while spike-and-slab priors supported predictor selection and reduced overfitting. A dedicated aftershock dataset was created from the NGA-West2 database through sequence identification, record selection, metadata quality control and spectral usability checks.
A further activity focused on developing mainshock–aftershock ground-motion models for Italy using the ITACA dataset. Italian strong-motion records were processed to distinguish mainshock and aftershock observations within a common hierarchical Bayesian framework. Separate regression coefficients were used for mainshock and aftershock records, while shared sequence-level dependence was preserved. This provides a step toward Italy-specific mainshock–aftershock ground-motion models based on recorded data.
The project also developed building-specific probabilistic seismic demand models for monitored existing reinforced-concrete buildings using sparse structural health monitoring data. A hybrid Bayesian transfer-learning framework was formulated to combine nonlinear time-history analyses of representative archetype buildings with sparse observations from monitored buildings. Source-domain models were built from nonlinear analyses of non-ductile infilled reinforced-concrete archetypes subjected to mainshock–aftershock sequences. The target-domain layer then used monitoring data to recalibrate the low-intensity part of the demand model while preserving the nonlinear response pattern inherited from the archetypes. The methodology was applied to real monitored buildings.
A final major activity concerned probabilistic damage-index modeling for reinforced-concrete frame structures with various stories under mainshock–aftershock sequences. Structural damage was quantified using the Park–Ang damage index, considering both story-specific and global damage measures. A two-stage modeling strategy was developed. First, estimating mainshock damage, and then modeling the aftershock-induced damage increment conditional on the damage state left by the mainshock. This made it possible to represent damage accumulation as a path-dependent process. A multivariate Bayesian formulation was also introduced to model story-level damage while accounting for vertical dependence among stories, censoring, nonlinear predictors and uncertainty.
Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)
The project produced several results beyond the state of the art in seismic structural health monitoring, ground-motion modelling and risk-informed decision support.
First, it demonstrated that the usefulness of monitoring data in seismic emergency management can be quantified rigorously within a probabilistic decision framework. Instead of treating monitoring as an assumed benefit, the developed methods quantify how much monitoring information improves decisions, reduces uncertainty and decreases expected management costs. This provides a rational basis for justifying the installation and optimization of monitoring systems.
Second, the project advanced strong-ground-motion sensor placement by using Value of Information as the design criterion. This goes beyond traditional approaches based mainly on coverage, engineering judgement or signal characteristics. Sensor configurations were evaluated according to their expected contribution to post-earthquake decisions. The work also showed that optimal layouts depend on hazard representation, spatial correlation and source-model assumptions, linking seismic hazard modelling directly with monitoring-system design.
Third, the project showed that the choice of Ground Motion Model can substantially affect the estimated benefit of monitoring data. In the Central Italy bridge case study, models with better statistical fit to recorded ground motions produced higher Value of Information and Relative Entropy. The results establish a direct link between ground-motion model quality and the decision value of monitoring information. The close relationship observed between Relative Entropy and Value of Information also suggests that entropy-based indicators may serve as computationally efficient screening tools in future studies.
Fourth, the development of aftershock-specific ground-motion models represents an advance beyond conventional mainshock-based modelling. The Bayesian structured random-field models provide spectrally continuous median predictions and a coherent representation of uncertainty for aftershock ground motions. The results confirmed that using mainshock-only models may bias estimates of seismic demand for already damaged structures, while aftershock-specific models provide a more suitable basis for post-mainshock risk assessment.
Fifth, the Italy-specific mainshock–aftershock model based on ITACA data advances regional seismic demand modelling by explicitly distinguishing mainshock and aftershock records while accounting for sequence dependence. This provides an operational framework for sequence-dependent spectral acceleration modelling and future applications in Italian seismic hazard and risk assessment.
Sixth, the hybrid Bayesian transfer-learning framework for monitored reinforced-concrete buildings addresses a key practical limitation of structural health monitoring. Monitored buildings often provide only sparse low-amplitude observations, while nonlinear simulations provide broader but generic information. The developed framework combines these sources in a controlled way, allowing monitoring data to anchor simulation-informed priors to building-specific behaviour without overfitting. This creates a practical bridge between population-level seismic demand modelling and building-specific assessment.
Finally, the probabilistic damage-index models under mainshock–aftershock sequences advance damage modelling by explicitly representing path dependence. Instead of modelling total sequence damage as a single response, the framework separates mainshock damage from the aftershock-induced damage increment and conditions the latter on the damage already accumulated. The story-level Bayesian formulation further advances the state of the art by accounting for vertical dependence, nonlinear effects, censoring and uncertainty in a unified probabilistic framework.
First, it demonstrated that the usefulness of monitoring data in seismic emergency management can be quantified rigorously within a probabilistic decision framework. Instead of treating monitoring as an assumed benefit, the developed methods quantify how much monitoring information improves decisions, reduces uncertainty and decreases expected management costs. This provides a rational basis for justifying the installation and optimization of monitoring systems.
Second, the project advanced strong-ground-motion sensor placement by using Value of Information as the design criterion. This goes beyond traditional approaches based mainly on coverage, engineering judgement or signal characteristics. Sensor configurations were evaluated according to their expected contribution to post-earthquake decisions. The work also showed that optimal layouts depend on hazard representation, spatial correlation and source-model assumptions, linking seismic hazard modelling directly with monitoring-system design.
Third, the project showed that the choice of Ground Motion Model can substantially affect the estimated benefit of monitoring data. In the Central Italy bridge case study, models with better statistical fit to recorded ground motions produced higher Value of Information and Relative Entropy. The results establish a direct link between ground-motion model quality and the decision value of monitoring information. The close relationship observed between Relative Entropy and Value of Information also suggests that entropy-based indicators may serve as computationally efficient screening tools in future studies.
Fourth, the development of aftershock-specific ground-motion models represents an advance beyond conventional mainshock-based modelling. The Bayesian structured random-field models provide spectrally continuous median predictions and a coherent representation of uncertainty for aftershock ground motions. The results confirmed that using mainshock-only models may bias estimates of seismic demand for already damaged structures, while aftershock-specific models provide a more suitable basis for post-mainshock risk assessment.
Fifth, the Italy-specific mainshock–aftershock model based on ITACA data advances regional seismic demand modelling by explicitly distinguishing mainshock and aftershock records while accounting for sequence dependence. This provides an operational framework for sequence-dependent spectral acceleration modelling and future applications in Italian seismic hazard and risk assessment.
Sixth, the hybrid Bayesian transfer-learning framework for monitored reinforced-concrete buildings addresses a key practical limitation of structural health monitoring. Monitored buildings often provide only sparse low-amplitude observations, while nonlinear simulations provide broader but generic information. The developed framework combines these sources in a controlled way, allowing monitoring data to anchor simulation-informed priors to building-specific behaviour without overfitting. This creates a practical bridge between population-level seismic demand modelling and building-specific assessment.
Finally, the probabilistic damage-index models under mainshock–aftershock sequences advance damage modelling by explicitly representing path dependence. Instead of modelling total sequence damage as a single response, the framework separates mainshock damage from the aftershock-induced damage increment and conditions the latter on the damage already accumulated. The story-level Bayesian formulation further advances the state of the art by accounting for vertical dependence, nonlinear effects, censoring and uncertainty in a unified probabilistic framework.