A Simultaneous Fault Diagnosis Method Based on Cohesion Evaluation and Improved BP-MLL for Rotating Machinery

Table 2

Proposed fault diagnosis algorithm.

Training stage
1	Obtain original sampling signals from multichannel sensors as training data.
2	Compute feature parameters for all the channels to construct a high-dimensional feature vector.
3	Compute sensitivity weighting factor and sensitivity factor using equations (6) and (7) and Table 1. Select parameters of high sensitivity factor to construct the sensitive feature vector.
4	Use the feature vector as input vector and modify weights and bias by using equations (10), (13), and (14) until the trained model can meet the test requirement of high accuracy or the maximum number of training epochs has been reached
Diagnosis Stage
1	Construct the sensitive feature vector of new data.
2	Compute the outputs for each testing instance by epoch as shown in equations (11) and (10).
3	Use a preset threshold to classify each instance for fault diagnosis using equation (17)