Aero Engine Gas-Path Fault Diagnose Based on Multimodal Deep Neural Networks
Algorithm.
Diagnose Method Based on Multimodal Deep Neural Networks.
Input: Aero engine gas-path fault datasets of fault modes.
Output: Diagnosis result of sample sequence in test set
1 Preprocess the input dataset: clean, normalize by formula (4), label, merge and shuffle it. Divide it into training set ,validation set and test set .
2 For M in {A: CNN, B: BPNN}:
3 Repeat:
4 Adjust one hyper-parameter, including layers, learning rate, etc.
5 For i <maxiterations:
6 Calculate the value of cross-entropy loss.
7 Update parameters by BP algorithm: .
8 End For
9 Until: Performance of M will no longer be improved.
10 Give the diagnosis scores of and , by M before its SoftMax process.
11 End For
12 Choose one scores information to be adjust, and fix the other =1.
13 Repeat
14 Adjust of the selected scores information.
15 Execute steps given in 20-23 on validation set .
16 Calculate of its initial probability assignment by formula (14).
17 Until: Find the continuous range of where diagnosis result for are best.
18 Select a value for with proper standard deviation in that range.
19 For M in {A: CNN, B: BPNN}:
20 With the selected , compute basic probability assignments by formula (5) to (10):
21 End For
22 Fusion the multimodal evidence bodies by formula ((11) and ((12):
23 According to the comprehensive information after fusion, give decision by rule:
