Seunghee Park

Continuous monitoring of the prestressed members of a bridge under construction using the free cantilever method (FCM) is crucial for ensuring bridge safety. Temperature-sensitive sensors require special considerations as they may misinterpret the signal and tension. Moreover, the unnecessary and inappropriate use of features obtained from the sensor signal can deteriorate the efficiency of the signal and, therefore, tension analysis. This study proposes a tension estimation method using an embedded elastomagnetic (EM) sensor with a temperature-compensation technique. Changes in the signal due to the tension in the temporary steel rods were analyzed using a full-scale test, and the sensor data were acquired for 15 months via the field application. The temperature effect on the signal could be removed by subtracting the tension from the signal using the thermistor data, reducing the error by 91.99% when considering permeability. Additionally, linear regression (LR) and machine learning (ML) algorithms were adopted to predict the tension. Furthermore, the performances of both algorithms were compared using mean absolute error (MAE) and R2. For the prediction using each feature in magnetic hysteresis, LR surpassed ML and the permeability exhibited the highest prediction performance. Meanwhile, predictions using multiple features were attempted to investigate the applicability of ML. Two cases of prediction were performed using ML: on using all the features and the other using three features excluding coercivity, which showed poor relevance to tension. As a result, the performance of the tension prediction was improved significantly compared to the results obtained by LR. In summary, the obtained results have demonstrated that the utilization of selective features of data with temperature compensation techniques could enhance predictive power.