Distilling the Knowledge of Multiscale Densely Connected Deep Networks in Mechanical Intelligent Diagnosis

Algorithm 1

Process of MSDC-NET (or CNN) for fault classification

Input: Sample X, Sample category, Test sample

Step1: The training process of the MSDC- NET model:

 1) Processing sample input data:

  Unify the sample data X into the same segmentation method, input type and input size, and finally we get the input as (x, y).

 2) Train the MSDC-NET model:

  Train MSDC-NET through the processed training data, and the objective function (6) as:

Step2: The training process of the CNN model:

 1) Processing sample input data:

 2) Train the CNN model:

  Train CNN through the processed training data, and the objective function (10) as

Step3: Fault classification

 1) Output of MSDC-NET (or CNN):

  Input the test data sample into the pre-trained model (MSDC-NET or CNN) to get the output .

 2) Category determination

  After obtaining the output of the model (MSDC-NET or CNN), use Formula (7) to predict the category of the test sample:

Output: Predict the category of input x