PSI - Issue 82

A.T. Andreasen et al. / Procedia Structural Integrity 82 (2026) 146–152

151

6

A.T. Andreasen et al. / Structural Integrity Procedia 00 (2026) 000–000

Additionally, the local gauge placed close to the weld is also compared based on the experimentally obtained strains and the predicted strains based on the estimated loads. In this case, the measured strains from the local gauge have not been used in the load estimation framework LLSM approach. Therefore, the comparison will be more realistic, as the loads are not optimized for acquiring the local strains. The comparison of the predicted and measured strains is shown in Fig. 6. The predicted strain { /0 } shows larger discrepancies at the peaks compared to the predicted strain from the global gauges; but it can be seen in the zoomed-in signal in Fig. 6(b) that the general trend during the test is captured.

Fig. 6. (a) Full signal of local strain for test 2; ( b) Zoomed in signal of local strain for test 2. As seen from Fig. 6. The predicted and measured strains are, similarly to the comparison with the global gauge, showing the same trends. Examining Fig. 6(b), it can however be seen that the predicted strain is less accurate than for the global gauge in Fig. 5, but the overall trends are predicted relatively accurately. Many potential reasons for the predicted strain being less accurate exist. Firstly, the local strain gauge is sensitive to placement and orientation near the weld. If the FE model and real-life strain gauge is not placed identically, it will result in discrepancies. Secondly, the global gauges were directly used for determining the loading, thus it is expected that the determined loads will predict very accurate strains. Furthermore, the FE model may not be completely accurate compared to the real lifting arm. The digital twin of the lifting arm was developed based on a simple test setup in (Larsen et al. (2024)), but in this paper the lifting arm was mounted directly in the real machine, which potentially decreases the accuracy of the FE model. However, as seen from Fig. 6, the framework determines the loading to be very accurate, even when considering the potential sources of errors. For all tests, the average root mean square error (RMSE) was calculated for the experimental data and the data from the FE model for the 13 global gauges. The RMSE can be seen from Table 1. The RMSE is calculated based on the pseudo stresses defined by = ⋅ . As observed from the table, the relative errors between the actual measured strains and the predicted strains are low. The RMSE calculated for the single signal in Fig. 5 is 3.97, thus indicating a very good agreement between the predicted strains and the measured strains. As seen from the various transport tests, the mean RMSE for all transport tests is 4.47, with two outliers of 8.63 and 6.55. However, overall, the RMSE for all transport tests was similar. Based on the loading predicted using the newly developed framework, it is possible to determine the stresses and strains anywhere on the lifting arm using the digital twin. This makes it possible to evaluate the fatigue performance of all welded locations, even those that are impossible to measure physically using e.g. strain gauges.

Table 1. Experimental tests performed with the agricultural mower and corresponding RMSE.

Test

Description

RMSE

Test

Description

RMSE

6 7 8 9

Transport test 5 Transport test 6 Transport test 7 Transport test 8 Transport test 9

2.71 5.22 4.19 3.73 3.32 4.76

1 2 3 4 5

Work mode test Transport test 1 Transport test 2 Transport test 3 Transport test 4

7.39 8.63 6.55 3.40 2.47

10

Average