Detect Steel Plate Defects with ML
2.8 هزار بار بازدید -
3 سال پیش
-
Steel plate defects are extracted
Steel plate defects are extracted from photos of several faulty steel plates with surface imperfections. Image analysis revealed 27 different features to describe the steel fault. A total of 6 unique types of faults are categorized, with a final category of "other faults" for any type of fault that does not fit into the other specific 6 categories.
Case Study Source Code: https://apmonitor.com/pds/index.php/M...
Machine Learning Course: https://apmonitor.com/pds
There are 27 features that are used to predict the steel faults. These features are extracted from steel plate samples. Computer vision can automatically extract some of this information from images or manually extracted with a user inspecting each plate defect or photo of the steel plate.
There are 7 types of steel plate defects that are labelled with a 1 if present and 0 if not present with One-Hot Encoding. A unique aspect of this data set is that the labels are imbalanced, meaning that there is a large difference in the number of specific defects.
In cases where there is a large imbalance, a strategy is to synthetically generate new data sets to improve the balance. That strategy is not taken in this case study but could be an option to improve the classification training. The fault label is not stored as a sequential value (e.g. fault label as 0-6) but is One-hot encoded to translate the fault label into a binary representation (0 or 1) for each fault. There are 7 additional columns of data with 1 if the fault exists and 0 if the fault does not exist.
Generate summary statistics with a profiling report to statistically characterize the data. Use box plots to identify any outliers in the data. Remove any outliers from the data set. Generate a pair plot and correlation matrix. What factors are highly correlated to the steel faults? Generate a bar chart that shows the imbalance of faults in the data.
Case Study Source Code: https://apmonitor.com/pds/index.php/M...
Machine Learning Course: https://apmonitor.com/pds
There are 27 features that are used to predict the steel faults. These features are extracted from steel plate samples. Computer vision can automatically extract some of this information from images or manually extracted with a user inspecting each plate defect or photo of the steel plate.
There are 7 types of steel plate defects that are labelled with a 1 if present and 0 if not present with One-Hot Encoding. A unique aspect of this data set is that the labels are imbalanced, meaning that there is a large difference in the number of specific defects.
In cases where there is a large imbalance, a strategy is to synthetically generate new data sets to improve the balance. That strategy is not taken in this case study but could be an option to improve the classification training. The fault label is not stored as a sequential value (e.g. fault label as 0-6) but is One-hot encoded to translate the fault label into a binary representation (0 or 1) for each fault. There are 7 additional columns of data with 1 if the fault exists and 0 if the fault does not exist.
Generate summary statistics with a profiling report to statistically characterize the data. Use box plots to identify any outliers in the data. Remove any outliers from the data set. Generate a pair plot and correlation matrix. What factors are highly correlated to the steel faults? Generate a bar chart that shows the imbalance of faults in the data.
3 سال پیش
در تاریخ 1400/11/08 منتشر شده
است.
2,886
بـار بازدید شده