Aircraft component sorting with machine learning.

Designed and built by:

Marc Dudley

Laurens Gosseling

Dennis V.d Berg

Marcel Koop

The problem for the client was: experienced staff do not like sorting aircraft components, contracted workers get bored and therefore do not sort them accurately enough, and third party contracted outsourcing lead times are too long. The client wanted to explore a way to bring the process back under their control.

This project focused on the process of classifying aircraft engine bolts (Components). The process is carried out during disassembly and reassembling of the aircraft engines and is critical in order to follow the regulations set down by the Civil Aviation Authority (CAA). When fully stripped each engine requires the classification and separation of around 3000 bolts before it can be reassembled.

The solution was to concept a machine that utilised deep learning algorithms combined with mechatronics to load, detect and sort the components. This approached allowed for a reduced cycle time through mechatronic automation replacing human physical action, and by replacing the cognitive element of recognition through deep neural network predictions.

The eventual outcome:

Cycle time of 4 seconds for individually fed components.

An accuracy of 99.47 percent across 10 bolt types.

A proof-of-concept that can be easily programmed in retraining on different components, have thresholds easily set and more easily scaled across similar sites.