ACI has implemented a Production Quality Optimization solution to predict phosphorous impurity for a steel manufacturer located in Ohio for improved operational and production efficiency​

A Ohio-based Steel Manufacturing Company

Goals/ Challenges

Goals

  • To predict the amount of phosphorus content in the steel making process based on input production parameters
  • To improve the then existing predictive accuracy from 75% to above 85%
  • To reduce the waiting time for phosphorus content measurement by providing real-time predictions

Challenges

  • Improving the prediction accuracy from 75% to 85%
  • Handling in-memory data generated through combinations for predictive cost function optimization using advanced algorithms
  • Pre-processing and feature engineering for identifying the optimal features for predictive accuracy and reducing bias in data

Our Solution/ Approach

Solution

  • Established a causal relationship between phosphorous content and the input parameters through correlation analysis
  • Residual and error metrics analysis for iterative cross-validations for identifying best-fit algorithms for a cluster 
  • Model suitability analysis and monitoring of performance through dynamic train-test cross-validations

Approach

  • Variable selection through causal-correlation analysis between variables and selecting the most significant parameters for analysis
  • Dynamic selection of machine learning algorithms to predict phosphorus content, depending on the cluster behavior in the most recent input data period
  • Performed cross-validations across train-test and post-predictive evaluation data to assess the results through continuous monitoring of model output performance

Download Full Case Study