Algorithm 1: The overlap algorithm to generate training samples.

Input: Given one long signal $Sig$ witha length of $l$, whose fault type is $F$; the training sample’s number $N$; and the training sample’s length $w$. Output: N training samples {$x_i,y_i$}. 1. calculating the overlap step $k=⎣\frac{l-w}{N-1}⎦$; defined each sample’s starting index $s=0$ and ending index $e=w$; defined $Input$and $Label$ matrix to save training features $x_i$ and corresponding fault type $y_i$. 2. For i in 1 to $l$:{ 3.  $Input\left[i\right]=Sig\left[s:e\right]$ 4.  $Label\left[i\right]=F$ 5.  $s=s+k$ 6.  $e=s+w$} 7. End for 8. Return $Input$ and $Label$