Malvern Panalytical is the manufacturer of diffractometers. Diffractometers are used to measure particle sizes and chemical composition. This project focuses on the sensor chip inside the machine. In the assembly of a machine, chips are tested in three different stages: first when the chips arrive on a wafer, secondly when a chip is mounted on chipboard and finally when the chipboard is mounted in a sensor module. The chips value is increased by a factor ten, for every assembly phase it proceeds.
The objective of this study investigates whether it is possible to create a model that uses the historical test data to label defect digital analogue converters on chips automatically. Automatically labelling should lead to a decrease in manual labour and fewer defect that proceed to a next assembly phase.
The study researches two approaches. One approach is using a cluster algorithms and the other by combining different statistical filters to classify defect chips.
The statistical model can consistently find more defect chips as that the human operator would. This model is capable of reducing the number of defect chips that normally would continue in the assembly process by 10 per cent. From this study also became clear that the current tests process should be more standardised, and that the current way of labelling is not specific enough. Further improvement may be to build model trained with all test data instead of only the partial test data.