Many production steps are automated in the plant of an international automotive supplier; This also applies to the machining of injection nozzles with the help of spindles. There are almost 20 of these cutting machines for this production step. Although these machines all do the same thing, use the same tools, have the same settings, use the same material from the same supplier, etc., it does happen that - sometimes - one and - sometimes - the other machine produces poorer quality.
First, sensors were attached to the machining equipment to measure vibration, temperature, ..... The machine data was recorded over a period of time, including the quality assessment of whether the component is of good quality. 11 quality indicators were evaluated, such as roughness, accuracy, profile depth, ...
Artificial intelligence analytics was applied to the data in order to identify reliable patterns that lead to poor quality.
Prediction algorithms predicted the values of the corresponding quality indicators about halfway through the respective lead and support spindle machining.
Self-learning artificial intelligence solutions were trained on part of the data. Hidden and multi-layered data patterns were discovered in order to be able to automatically make stable predictions as to which value the respective quality indicator will achieve.
The trained algorithms were applied to unknown data, i.e. data that was not used for training. This verifies whether the data patterns could really be correctly identified.
In the event of process changes, the self-learning algorithm automatically adjusts the forecast model so that no data scientist has to intervene.
For sustainable quality optimization, it is important that the AI learns and adapts itself to changes in the event of changes in processes, products, influences, ... independently - i.e. without the assistance of a data scientist. This is where self-learning algorithms with dynamic and adaptive learning come into play.
Managing Director IS Predict GmbH