Sterlite Minimizes Optical Fiber Manufacturing Cost Using Machine Learning

“The superior data processing capability of MATLAB enabled us to optimize data coming in from various locations; consider data from other external applications and sources, in different formats; and work on it to develop an app that runs at the click of a button on the shop floor.”

Key Outcomes

  • Reduced fiberglass costs by 8%
  • Enabled 24 x 7 automated runtime, eliminating 14+ full-time equivalents (FTEs)
  • Built end-to-end lightweight app for operators on the shop floor

Sterlite Technologies Limited specializes in optical fiber and cables, hyper-scale network design, and deployment and network software.

In the optical fiber manufacturing process, the glass material starts a controlled drip to create a very thin strand of optical fiber. This process involves many variables and can result in costly scrap.

Sterlite’s analytics team built a machine learning model to reduce scrap numbers as much as possible. First, optical testing machines measure a set of parameters of the glass rods. MATLAB® combines this dataset with the geometric properties of the glass cylinder to calculate the theoretical profile of the fiber if it were produced from the resulting preform. Then another engineering software converts this fiber profile into a dataset that can be sent to the machine learning model, which matches the rods with the cylinder. These complex steps are packaged using MATLAB Compiler™ into a lightweight app for operators on the shop floor.