Ensuring the integrity of structural health monitoring data is increasingly pivotal as infrastructure systems age and degrade. A new study introduces an innovative data reconstruction method utilizing Variational Mode Decomposition (VMD) combined with the Sparrow Search Algorithm (SSA) optimized Gated Recurrent Unit (GRU), promising improvements for restoring incomplete datasets often compromised by sensor failures or transmission glitches.
Structural Health Monitoring (SHM) systems are integral for assessing the condition of engineering structures, providing valuable insights through continuous data collected via various sensors. With advancements fostering the deployment of large-span spatial structures, the reliance on accurate monitoring data is magnified. Data loss, whether arising from sensor malfunctions, intermittent signals, or environmental factors, can significantly hinder the process of modal identification and risk prediction.
Addressing this widespread challenge, researchers established this comprehensive reconstruction method, which begins with VMD to preprocess existing data. This process decomposes raw measurements from target sensors, breaking them down to Intrinsic Mode Functions (IMFs) and residuals, fostering easier data manipulation.
Once VMD has separated the data, the SSA enhances the GRU's effectiveness by optimizing its hyperparameters. The GRU model stands out for its ability to leverage data from multiple sensors to reconstruct these reduced forms accurately. The combination of SSA and GRU improves the model's performance without requiring tedious manual adjustments.
To validate the proposed method, the team utilized one month of actual monitoring data collected from structural projects, alongside public datasets. Results revealed substantial advancements: the new VMD + SSA + GRU model achieved reductions of 46.61% in Root Mean Square Error (RMSE) when compared to traditional GRU methods, which typically suffered from central convergence—resulting in reconstructed data skewing toward mean values.
Enhanced accuracy monitoring is achieved through the GRU's capacity for temporal data processing, facilitated by the optimization strategies provided by SSA. When subjected to various data loss rates ranging from 10% to 70%, the VMD + SSA + GRU model consistently outperformed both standalone GRU implementations and the VMD + GRU approach, highlighting its robustness.
Further corroboration came from testing this methodology against acceleration response datasets from the Hardanger Bridge, affirming its operational effectiveness across different structural environments. The researchers noted, “This model reconstructs the missing data trends exceedingly well, translating to high levels of consistency with real data.”
Such advances herald new possibilities not only for bridge monitoring but for all sectors reliant on structural safety assessments. The ability to recover data with minimized manual adjustment directly improves the dynamic and operational reliability of monitoring systems.
While the results are promising, the researchers are cautious about the SSA's potential to become trapped within local optima, which could impact overall efficiency. Future explorations may include the integration of more advanced optimization algorithms to fortify performance.
Given the increasing complexity and scale of structural monitoring systems around the globe, this innovative method could provide the necessary resilience against data loss, making sure safety assessments remain accurate and reliable.