Figure 5
The scatter plot is titled “Model Prediction Comparison”. The horizontal axis is labeled “True Fault Severity” and ranges from 0.0 to 1.0 in increments of 0.2 units. The vertical axis is labeled “Predicted Fault Severity” and ranges from 0.0 to 1.0 in increments of 0.2 units. A legend in the upper left identifies two data series: “Black-box A I” and “Physics-Informed A I”. A dashed diagonal reference line runs from (0.0, 0.0) to (1.0, 1.0). For true fault severity 0.0, the “Black-box A I” points are clustered tightly near (0.0, 0.0) with values between negative 0.01 and 0.01. The “Physics-Informed A I” points are slightly more spread, ranging approximately from negative 0.03 to 0.07. For true fault severity 0.5, the “Black-box A I” points range approximately from (0.5, 0.42) to (0.5, 0.58), showing moderate spread around the diagonal. The “Physics-Informed A I” points range approximately from (0.5, 0.48) to (0.5, 0.59), showing tighter clustering closer to the reference line. For true fault severity 1.0, the “Black-box A I” points range approximately from (1.0, 0.98) to (1.0, 1.08), slightly exceeding the perfect prediction line at the upper end. The “Physics-Informed A I” points range approximately from (1.0, 0.88) to (1.0, 1.02), showing slightly wider spread below the diagonal. Note: All numerical data values are approximated.Comparison of true versus predicted fault severity for the black-box and physics-informed models. The physics-informed model exhibits predictions more closely aligned with the ideal one-to-one relationship, with reduced scatter across severity levels, indicating improved physical consistency and regularisation compared to the unconstrained black-box model