Fatigue life prediction of bending polymer films using random forest

Bending Fatigue Analysis of Various Polymer Films by Real-Time Monitoring of the Radius of Curvature and Heat Generation

“…Figures 9b and S7 show MAPE and R 2 of each prediction compared with the results predicted by the previous model. 28 It should be noted that the RF model developed in this study resulted in a better prediction with a lower MAPE and a higher R 2 for all data compared to the previous model. We successfully improved the prediction accuracy of the fatigue life using the RF model that employed the influential variable (curvature), a quantitative indicator analyzed with the specially designed bending tester.…”

“…Six types of polymer films were employed as common and practical substrates in flexible electronics: polyethylene naphthalate (PEN, Teonex Q51, Toyobo Co., Ltd., Japan), polyethylene terephthalate (PET; Lumirror S10, Toray Industries Inc., Japan), PVC (107-16601 and 107-16602, KOKUGO Co., Ltd., Japan), PC (Lupilon NF-2000, Mitsubishi Gas Chemical Inc., Japan), cycloolefin (COP, ZeonorFilm ZF-14, ZEON Corp., Japan), and PS (Santoclear, Mitsubishi Chemical Corp., Japan). − The direction of the polymer main chain orientation was identified by polarized optical microscopy (POM; BX50, Olympus Corp., Japan) and attenuated total reflectance infrared spectroscopy (ATR-IR; FTIR-6600, JASCO Corp., Japan), as shown in the previous report . The polymer films with various thicknesses (50–300 μm) were used for the bending fatigue test after being cut into dimensions of 40 mm length and 30 mm width according to the polymer main chain orientation.…”

“…1−5 The direction of the polymer main chain orientation was identified by polarized optical microscopy (POM; BX50, Olympus Corp., Japan) and attenuated total reflectance infrared spectroscopy (ATR-IR; FTIR-6600, JASCO Corp., Japan), as shown in the previous report. 28 The polymer films with various thicknesses (50−300 μm) were used for the bending fatigue test after being cut into dimensions of 40 mm length and 30 mm width according to the polymer main chain orientation.…”

“…Before training an RF model, input variables were selected based on the Pearson correlation, as shown in Figure S5, according to the previous report. 28 We used the yield stress, toughness, thickness, ν/2, and 1/r (curvature) as input variables to improve the prediction accuracy. The difference in these variables from our previous model is that θ/2 was replaced with 1/r.…”

“…For the prediction of fatigue life, machine learning becomes a powerful choice because of its efficiency and accuracy. − In our previous work, the bending fatigue life of various polymer films was predicted by a machine learning model that employed only test conditions, such as film dimensions and fundamental mechanical properties, including toughness and yield stress, as input variables . Inputting additional effective variables into the machine learning model can improve the prediction accuracy of the fatigue life under cyclic bending.…”