Key to exceeding expert and traditional performance is a deep convolutional network which can map a sequence of signal to tool wear label along with PHM 2010 dataset similar with previous dataset of its kind.
The network takes as input a time-series of secondary-sampled sensor signal, and outputs a sequence of responding tool wear. The original signal is sampled at 50KHz. We arrive at an architecture which is 32 layers of convolution followed by a fully connected layer.
To make the optimization of such a deep model tractable, we use residual connections and batch-normalization. The depth increases both the non-linearity of the computation as well as the size of the context window for each regression task.
Each channel of signal is corresponding to:
1: Force (N) in X dimension
2: Force (N) in Y dimension
3: Force (N) in Z dimension
4: Vibration (g) in X dimension
5: Vibration (g) in Y dimension
6: Vibration (g) in Z dimension
7: AE-RMS (V)
The PHM data is sampled at a frequency of 50000Hz and have 8GB size. In this machining condition, the spindle speed of the cutter was 10400 RPM; feed rate was 1555 mm/min; Y depth of cut (radial) was 0.125 mm; Z depth of cut (axial) was 0.2 mm.
You may find more details about this experiment at HERE
Although we have no data about test data, according to the score on known dataset, loss of this model is ~1000 times less than the 1st prize winner, which shows a great accuracy of CNN.