(A) | PSG Construction. |
| Input: Sliced data , Coordinate of sensor . |
| Output: graph set . |
(1) | Obtain the normalized signal ; |
(2) | Calculate the feature data: , ; |
(3) | Separate feature data , ; |
(4) | Calculate the Euclidean distance: ; |
(5) | Obtain the k closest neighbors nodes set of node : if is k-th smallest; |
(6) | Establish the edge connections for every node; |
(7) | Embed , , as node features; |
(8) | Output graph set . |
(B) | PKG Construction. |
| Input: original feature matrix , training epoch M for the GPK. |
| Output: PKG GPK with high-level feature matrix F. |
(1) | Obtain the normalized signal ; |
(2) | Calculate the feature matrix: ; |
(3) | Calculate the Mahalanobis distance: ; |
(4) | Obtain the k closest neighbors node set of node : , if , is k-th smallest; |
(5) | Establish the edge connections for every node; |
(6) | Obtain original graph and original feature matrix ; |
(7) | for i = 1, 2, …, M: |
| Train the GCN model for M epochs: |
| and ; |
| end for |
(8) | Output the PKG . |
(C) | Fault diagnosis using MCGFF. |
| Input: and GPK. |
| Output: The health label Z. |
(1) | Divide the training set and testing set: Vtrain, Vtest; |
(2) | Train the GCN model; |
(3) | for V in Vtrain do: |
| MCGFF(V) ⟶ Z; |
| ⟶ CE loss; |
| Update with backward propagation; |
| end for |
(4) | Output the health label: MCGFF(Vtest) ⟶ Z. |