Identifying metal coupon types with Visual Classification
The Challenge
A leading specialist in water
treatment needs to build a mobile
application to capture images of
metal coupons planted in RO
systems and detect their
composition. Further identification
of corrosion types per composition
will be executed. Tatras has been
given the responsibility to build a
ML solution based on image
understanding to automate the
process which currently is manual
and too much time consuming.
Hypothesis
- Historical image data is available to use.
- Image classifier to identify coupon types.
- Clustering per coupon type to understand corrosion groups based on visuals only.
- Assist human annotations with clustering.
Execution
- Tensorflow- ML model for classification.
- Fine tuning of pre trained model for better generalization.
- Implemented as a flow in DataIku platform.
- Jupyter-Notebooks
- Slack: for team planning and project management.
- JIRA- for project management and keeping track of progress.
Outcomes
- 95% metal coupons classified correctly.
- 7 days Vs 5 mins AUTOMATED COUPON AND CORROSION RECOGNITION WITHOUT WAITING FOR LAB ANALYSIS AND REPORT.
- Coupon types Identified.
- Corrosion type identified.
- Deployed as a flow in DataIku instance.
- Suggestive actions against corrosion types.