← All projects
Project 06 of 07
23 March to 28 June 2026
Digital
twin
Testing and Information Processing Technology of Machinery
Individual report
*
Digital twin
*
Vibration
*
Fault diagnosis
*
Hold-out testing
I built a digital twin of a healthy rotating machine. It predicts the vibration level for any speed and load. When the real machine drifts away from the twin, the gap is the warning.
+
01
0.039 mm/s
twin error (RMSE)
+
02
96.0%
of faulty windows flagged at 2.8% false alarms
+
03
94.0%
against 81.7% for a fixed threshold
+
04
330 runs
1,980 windows
What
I did
01
Modelled healthy velocity RMS against speed and load on a rig with a 6204 bearing.
02
Used the residual as the fault signal.
03
Added a physics rule cascade to name the fault, with macro-F1 of 0.927.
04
Compared four standard classifiers, which scored between 98.3% and 99.2%.
+ Honest result
Where it fell short
The twin features added no significant gain over the plain classifiers. Bearing defect detection was the weakest class at 85.0%. I report both.
Fig. Line drawing of the setup