This dataset is related to a steel plate fault detection study where for seven different fault types in the plates, the corresponding visual attributes of the faulty plates are provided. Each data point represents a faulty plate with the color representing the fault type.

How to Use the Demo

This demo can display the data at various levels of refinement, so the analyst is able to choose the level that is most useful. Click Cluster to see the data points coarsened into fewer circles. Click Decluster to see more circles. At any level of refinement, circles can be selected with the Lasso Select or Box Select Tools. The histogram on the right will then display the selected data points in blue, in front of the distribution of the entire data set in grey for an attribute. To choose another attribute, select it from the drop down menu above the histogram.

When the distribution of the blue selected points differs from the distribution of the grey bars which represent the entire data set, this may be an opportunity to gain some insight. An example of an insight from this data set can be found below.


BiDViT Steel demo screen captureIn this data set, we selected an orange circle that represented a group of data points where the most common fault was “Bumps”. When we selected “Y Minimum” as the feature, we noticed that bumps occurred much more often when the measurement was between a certain range. The data set as a whole (shown by the grey bars) had a much different distribution. To confirm, we selected all the other orange circles and found that this pattern was repeated in all of them. This suggests that steel within this range is much more prone to this type of fault.


  Dark Green = Other Faults

  Orange = Bumps

  Purple = Dirtiness

 Pink = Stains

  Light Green = K_scratch

  Yellow = Z_Scratch

  Brown = Pastry