Identify defects for fiber optic equipment manufacturer

Defect prediction for fiber optic equipment manufacturer

Client: One of the world’s largest manufacturers of fiber optic equipment and allied products

OPPORTUNITY: The client used lots of expensive components in their core processes which required effective utilization throughout the supply chain. The client was facing challenges in identifying defective products in the ongoing manufacturing process to reduce wastage.

SOLUTION:

SGA provided a comprehensive optimization solution using machine learning techniques, which helped to predict the defective components at various stages of the manufacturing process:

SGA’s DA team created a data set that had values of predictors in each step for every minute of the manufacturing process

The team added results from testing phase as a categorical variable with values Pass and Fail (for various factors including bending, flexing, torsion, impact resistance, and crush tests)

SGA trained the model on Spark to predict the outcome when the values of predictors deviate from expected values

SGA used random forest algorithm to get a robust model. The client then used the model on fibers in manufacturing process to predict failure

VALUE DELIVERED:

Reduced the number of defective fibers by10% with a 72% accuracy of prediction

Aided significant optimization of raw material usage, which would otherwise have been wasted

OPPORTUNITY

The client used lot of expensive components in their core processes which required effective utilization throughout the supply chain. The client was facing challenges in identifying defective products in the ongoing manufacturing process to reduce wastage.

SOLUTION

SG Analytics provided a comprehensive optimization solution using machine learning techniques, which helped to predict the defective components at various stages of the manufacturing process:

Our data analytics team created a data set that had values of predictors in each step for every minute of the manufacturing process.
The team added results from testing phase as a categorical variable with values Pass and Fail (for various factors including bending, flexing, torsion, impact resistance, and crush tests).
We trained the model on Spark to predict the outcome when the values of predictors deviate from expected values.
We used a random forest algorithm to get a robust model. The client then used the model on fibers in manufacturing process to predict failure.

VALUE DELIVERED

Reduced the number of defective fibers by 10% with a 72% accuracy of prediction.

Aided significant optimization of raw material usage, which would otherwise have been wasted.

Defect prediction for fiber optic equipment manufacturer