Posts

Classification

Image
  Neural and Non-neural Techniques for Classification  Summary of Classification models Classifier Classification Accuracy (%) Test Set Comments Shallow NN 99.9% Number of neurons: 12, Architecture: 5In|5out , ,Preprocessor: PCA LSTM 98.9% Number of neurons: 33, Architecture: 5In|5out, Preprocessor: PCA, Activation: Softmax Fine Tree 98.8% Max. Number of Splits = 100, Split Criterion: Gini’s Diversity index Medium Tree 98.7% Max. Number of Splits = 20, Split Criterion: Gini’s Diversity index Coarse Tree 60.1% Max. Number of Splits = 4, Split Criterion: Gini’s Diversity index Linear Discriminant 40.9% Full Covariance Structure Quadratic Discriminant 63.6% Full Covariance Structure Linear SVM ...
Post a Comment
Read more

Independent Component Analysis (ICA)

Image
15 Statistical Features (SF) calculated on the Extended Park quantity (ip), then performing ICA gives the 2D and 3D plots shown below: 2D plots  No load  75% load 100% load 3D plots  No load 75% load 100% load
Post a Comment
Read more

Exploratory analysis: Principal Component Analysis (PCA)

Image
PCA of the three phase current signatures (i_a, i_b, and i_c) 2D plots   and 3D plots   Pareto Charts 15 Statistical Features (SF) calculated on the Parks quantity (id and iq), then perform PCA gives the 2D and 3D plots shown below: 2D PCA Plots                                                                                                                 (a)                                                             (b)                                  ...
Post a Comment
Read more

Fast Fourier Transform Analysis on: Parks and Extended Parks Quantity

Image
  Extended Park Vector Approach and 15 Statistical Features The EPVA method works by converting the three-phase currents (i_sa, i_sb, i_sc)  into direct and quadrature axis (id and iq)    and it finds the modulus id + jiq (ip) of     as follows [1]: Note: Sample length = 100 samples           Overlap = 65 samples   Fast Fourier Transform (FFT) analysis: Characteristic Fault Frequencies (CFFs) 0.3% fault at no load and 100% load (a) 0.7% fault at no load and 100% load (b) Figure 1: Frequency response at no load and 100% load condition, (a)  for 0.3% severity and (b) 0.7% severity   1.05% fault at no load and 100% load (a) 2.1% fault at no load and 100% load (b) Zoom view of the Parks quantity (c) Figure 2: Frequency response at no load and 100% load condition, (a)  for 1.05% severity, (b) 2.1% severity and (c) zoom view of id, and iq . Refere...
Post a Comment
Read more

Method followed in the project

Image
   Methodology The proposed SITF diagnosis scheme are: a)        Gathering of data The collection of data is for the three-phase current signatures for the Healthy and Faulty IM which contains SITF. The data records for the faulty IM at different loads with different severity level. Furthermore, data samples to be analyzed are extracted from simulation. b)        Data  pre-processing Data cleaning and smoothing is done at this stage which is necessary to ensure that the dataset is free of any unlogged values. Figure  1 : Data Pre-Processing block scheme c)        Exploratory analysis Data cleaning, exploration and dimensionality reduction of datasets is done in this stage. It is the way of reducing higher dimensions dataset into lesser dimensions dataset which ensures that is provides similar information. Principle component analysis (PCA), and pareto charts techniques will be u...
Post a Comment
Read more