DefectDetection module

DefectDetection.annotate_plots(ax, defs)

Annotate charts with locations of defects

Parameters

  • ax (axis object) – plot axis

  • defs (dict) – defect parameters

Returns

None

DefectDetection.combine_features(feature_list)

Combine all features from different methods into one single dict

Parameters

feature_list (list) – list containing all entries of input features that need to be concatenated

Returns

features: feature dictionary containing all the input features

Return type

dict

DefectDetection.compute_features_grad(mat)

Calculates spatial and temporal gradients

Parameters

mat (ndarray) – raw data

Returns

features_grad: dictionary containing spatial and temporal gradient features

Return type

dict

DefectDetection.compute_features_sd(mat, t_stamps)

Calculates spatial features at every location and time stamp

Parameters

  • mat (ndarray) – raw data

  • t_stamps (list) – time stamps at which time domain features are calculated

Returns

features_sd: dictionary containing spatial domain features

Return type

dict

DefectDetection.compute_features_td(mat, t_stamps)

Calculating temporal features at every spatial location

Parameters

  • mat (ndarray) – raw data

  • t_stamps (list) – time stamps at which time domain features are calculated

Returns

features_td: dictionary containing time domain features

Return type

dict

DefectDetection.compute_features_wav(mat, t_stamps)

Calculates wavelet transformed features at every location

Parameters

  • mat (ndarray) – raw data

  • t_stamps (list) – time stamps at which wavelet features are calculated

Returns

features_wav: dictionary containing wavelet features

Return type

dict

DefectDetection.defect_detection_metrics(mat, mah, iso, s1_2d, s2_2d, defs, t_stamps, t, units, plot=True)

Quantification of defect detection, and plotting the results

True-Positive Rate (TPR), False-Positive Rate (FPR), Receiver Operating Curves (ROC) are calculated for the raw data, Mahalanobis distance and result of Isolation Forest method. In addition, Area Under Curve (AUC) is also calculated to quantify the performance. Often, performance in terms of higher TPR is desired at lower FPR. To aid this, TPR values are calculated at 2%, 5% and 10% FPR. Further, the results are presented graphically if needed.

Parameters

  • mat (ndarray) – raw data - 3D float array

  • mah (ndarray) – result of performing Mahalanobis distance - 3D float array

  • iso (ndarray) – result of performing Isolation Forest algorithm - 3D float array

  • s1_2d (ndarray) – 2D meshgrid representation of s1 axis

  • s2_2d (ndarray) – 2D meshgrid representation of s2 axis

  • defs (dict) – defect parameters

  • t_stamps (list) – time stamps at which features were calculated and where results are desired

  • t (list) – time coordinates

  • units (dict) – units of the different dimensions

  • plot (Bool) – Boolean to indicate if plots are needed to visualize

Returns

None

DefectDetection.define_defects(s1, s2, defs_coord, def_names)

Define coordinates of defects

Parameters

  • s1 (list) – spatial axis 1

  • s2 (list) – spatial axis 2

  • defs_coord (list) – list containing all defects - each defect contains a list of tuples containing the vertices of defect

  • def_names (dict) – dictionary containing the names of defects

Returns

defs: dictionary containing all the necessary parameters of all the defined defects

Return type

dict

DefectDetection.fit_isolationforest_model(features, t_stamps, pca_var)

Fit Isolation Forest model

Parameters

  • features (dict) – dictionary containing all input features

  • t_stamps (list) – time stamps at which features were calculated and where results are desired

  • pca_var (float) – contains the desired explained variance parameter, if less than 1.0, PCA will be performed

Returns

iso: result of Isolation Forest model over the data

Return type

ndarray

DefectDetection.main()

All the subroutines will be called from here

DefectDetection.mean_filter(mat, t, s1, s2, units, size, plot_sample)

performs mean filtering at each location

Parameters

  • mat (ndarray) – raw data

  • t (list) – time axis

  • s1 (list) – spatial axis 1

  • s2 (list) – spatial axis 2

  • units (dict) – units of the different dimensions

  • size (int) – number of elements to use in the mean filter. The higher, the more aggresive the filtering

  • plot_sample (Bool) – Boolean to indicate if time series plots are needed to compare raw and filtered data

Returns

filt_mat: mean filtered raw data based on kernel size

Return type

ndarray

DefectDetection.normalize_features(features, t_stamps)

Normalize features

Parameters

  • features (dict) – dictionary containing all input features

  • t_stamps (list) – time stamps at which features were calculated and where results are desired

Returns

features: dictionary containing all normalized features

Return type

dict

DefectDetection.outlier_mah(features, t_stamps, pca_var)

Mahalanobis distance to identify outliers

Parameters

  • features (dict) – dictionary containing all input features

  • t_stamps (list) – time stamps at which features were calculated and where results are desired

  • pca_var (float) – contains the desired explained variance parameter, if less than 1.0, PCA will be performed

Returns

mah: contains the result of computing Mahalanobis distance over the data

Return type

ndarray

DefectDetection.read_matlab_data(dataset, table)

Reads in raw matlab data using scipy IO modules

Parameters

  • dataset (ndarray) – name of the Matlab dataset

  • table (str) – name of table within Matlab

Returns

mat: matlab data that has been converted to numpy array

Return type

ndarray

DefectDetection.scale_frames(arr, t_stamps)

Scale frames between 0-1 for better interpretability

Parameters

  • arr (ndarray) – input array that needs to be scaled

  • t_stamps (list) – time stamps at which features were calculated and where results are desired

Returns

outarr: scaled array where the elements lie between 0-1

Return type

ndarray

DefectDetection.visualize_features(mat, features, s1_2d, s2_2d, feature, t_idx, t, units)

Visualize computed features

Parameters

  • mat (ndarray) – raw data

  • features (dict) – dictionary containing input features

  • s1_2d (ndarray) – 2D meshgrid representation of s1 axis

  • s2_2d (ndarray) – 2D meshgrid representation of s2 axis

  • feature (str) – desired feature that needs to be visualized

  • t_idx (int) – time index at which visualization is needed

  • units (dict) – units of the different dimensions

Returns

None

DefectDetection.visualize_spatial_data(mat, t, s1_2d, s2_2d, t_min_idx, t_max_idx, del_t_idx, units)

Visualize spatial slices of data at certain time stamps

Parameters

  • mat (ndarray) – raw data

  • t (list) – time axis

  • s1_2d (ndarray) – 2D meshgrid representation of s1 axis

  • s2_2d (ndarray) – 2D meshgrid representation of s2 axis

  • t_min_idx (int) – lower bound time index for visualization

  • t_max_idx (int) – upper bound time index for visualization

  • del_t_idx (int) – time index steps for visualization

  • units (dict) – units of the different dimensions

Returns

None

DefectDetection.visualize_time_series(mat, t, s1, s2, units)

Pick 4 random spatial coordinates and chart the time-series

Parameters

  • mat (ndarray) – raw data

  • t (list) – time axis

  • s1 (list) – spatial axis 1

  • s2 (list) – spatial axis 2

  • units (dict) – units of the different dimensions

Returns

None