Problem Statement
Engine assembly is a complex
operation. Engine functioning
involves hundreds of moving parts
and thousands of fastening parts
which keep the moving parts in
place and firmly fastened for
continued functioning of the
engine. So, Visual inspection of
engine at every stage of assembly
is mandatory and red flagging of
wrongly assembled parts is
paramount.
Solution
DTC built a Deep Learning (DL)
based model trained on 1000’s of
images. This trained model accepts
images and identifies the part and
specifies the section of the engine
which has a wrongly assembled
part.
Tech stack
Tech stack
- Deep Learning algorithms
- Python, Pytorch, and TensorFlow
- Flask, Postman.
Solution Benefits
DTC’s DL based solution is able to accurately identify and localize the wrongly assembled part. The trained model is able perform with the following metrics
- Zero false negative i.e. every wrongly assembled part is always detected without fail by the model – since detection of wrongly assembled part is paramount, our trained model is able to deliver that requirement with flying colors.
- Very low false positive i.e. very few times good assembly is detected as wrongly assembled.
- The DL model reduced the error occurrence by 80% thus reducing rework time to 10% and subsequently increasing productivity by nearly twice the previous value
