ME
|
DIGITALTHREAD
100
80
60
40
20
0
Models accuracy &recall (%)performance heatmap
Training Test
97.00 95.00
DEF Model (Images +Time-Series, Custom Loss)-Recall_Pristine
DEF Model (Images +Time-Series, Custom Loss)-Accuracy
95.00 90.00
DEF Model (Images +Time-Series, Custom Loss)-Recall_UnderExt 100.00 100.00
DEF Model (Images +Time-Series, Custom Loss)-Recall_Warping 96.00 100.00
DEF Model (Images +Time-Series, CCE Loss)-Accuracy 88.00 87.00
DEF Model (Images +Time-Series, CCE Loss)-Recall_Pristine 86.00 85.00
DEF Model (Images +Time-Series, CCE Loss)-Recall_UnderExt 87.00 85.00
DEF Model (Images +Time-Series, CCE Loss)-Recall_Warping 85.00 84.00
CNN (Best Processed Images Model)-Accuracy 91.75 88.26
CNN (Best Processed Images Model)-Recall_Pristine 88.00 81.00
CNN (Best Processed Images Model)-Recall_UnderExt 85.00 81.00
CNN (Best Processed Images Model)-Recall_Warping 92.00 89.00
CNN (Best Raw Images Model)-Accuracy 90.26 89.35
CNN (Best Raw Images Model)-Recall_Pristine 95.00 90.00
CNN (Best Raw Images Model)-Recall_UnderExt 72.00 87.00
CNN (Best Raw Images Model)-Recall_Warping 65.00 57.00
FFNN (Best Time-Series Model)-Accuracy 73.50 72.07
FFNN (Best Time-Series Model)-Recall_Pristine 99.50 92.00
FFNN (Best Time-Series Model)-Recall_UnderExt 14.00 12.00
FFNN (Best Time-Series Model)-Recall_Warping 77.00 12.00
Training Test
3
2
1
Models loss performance heatmap
CNN (Best Processed Images Model) 3.45 2.37
CNN (Best Raw Images Model) 0.21 0.26
DEF Model (Images +Time-Series, CCE Loss) 1.56 1.23
DEF Model (Images +Time-Series, Custom Loss) 0.06 0.06
FFNN (Best Time-Series Model) 0.79 0.59
Figure 9. DEF model multi-defect results compared with the best single-modality models: (a) accuracy and recall heatmap across training and test
datasets (b) loss heatmap across the same datasets.
46
M AT E R I A L S E V A L U AT I O N • J U LY 2 0 2 6
Model
and
metric
Model