Machine learning algorithms for failure prediction and yield improvement during electric resistance welded tube manufacturing

Vinay Sanjay Gujre, Rahul Anand

Journal of Experimental & Theoretical Artificial Intelligence · 2019 · 11 citations · 7 references

Concepts

Abstract

An electric-resistance-welded (ERW) tube is one of the leading contributor products for automotive manufacturers. Such tubes tend to fail due to weak weld quality that may be the result of several parameters, such as inappropriate selection of weld power, positions of forming rolls and fin pass rolls, mill operating speed and many more. This paper presents an application of classifier algorithm to predict weak welded tube during real production process. Real-time data with 86 independent variables are trained using logistic regression as classifier. A logistic regression model is validated on real-time test data with capability of predicting failure 82.55% of the time (often termed as sensitivity). Operating ranges for mill parameters are estimated with the help of model results. Significant yield improvement has been observed after deployment of model recommended operating ranges as recipe in manufacturing unit. Working of newly developed human machine interface (HMI) has been explained. This exercise has been performed for three different product segments being manufactured in Tubes Division, Tata Steel Limited. This work demonstrates an application of artificial intelligence in manufacturing plant for quality improvement.

References

7