Real-Time Structural Health Assessment of a Tension Rod Assembly Using Machine Learning 55 The methodology is depicted in Figure 3 below. Fig. 3 Methodology for calculating the structural health (H%) using a 1D CNN classification model. Results and discussions After training the network as detailed in Section 2.2, it was employed to monitor changes in the health percentage (H%) resulting from variations in rod tension. During testing, the beam was continuously excited using white noise, and H% was recalculated every 2 seconds as outlined in Section 2.3. The wing nut was adjusted by tightening and loosening it to different extents to simulate various damage levels. Figure 4 illustrates the H% calculated during multiple cycles of tightening and loosening. Initially, the wing nut was fully tightened, and the estimated health was nearly 100%, indicating a completely healthy structure. As the wing nut was gradually loosened, the H% dropped below 40%, indicating an increased likelihood of structural damage. Repeating this process multiple times caused the H% levels to fluctuate between high and low values, depending on the current tension level. This demonstrates the model’s accuracy and capability to reflect the real-time condition of the setup. Fig. 4 Real-time structural health percentage, H(%), calculated using the 1D CNN.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==