Automated Model Re-Training Framework Based on Expert Feedback

Thesis Type Master
Thesis Status
Number of Students
Thesis Supervisor

INNIO Digital has deployed more than 70 models for anomaly detection and preventive maintenance on their cloud-based platform myPlant, using Machine Learning methods on time series data gathered from millions of sensors in the Industrial Internet of Things. The goal of this thesis is to 

  • Conceptualize a framework for automated computations of accuracy scores of issue detection algorithms based on various inputs (e.g., detected issues with feedback labels from experts)
  • Design a system for evaluating the accuracy scores and triggering re-training workflows when necessary, with subsequent validations of the impact of the re-training (improvement of model accuracy)
  • Demonstrate feasibility by implementing an end-to-end example using an existing model with a given
    re-training workflow specification.