No momento, este caso está disponível somente em inglês.
AlemanhaIndústria de transformação2019 – em andamentoSiemens
AI quality prediction at the Amberg electronics plant
Siemens' Amberg plant predicted solder defects from process data and avoided an additional X-ray machine
Problema
The Siemens Electronics Works in Amberg makes about 17 million Simatic control products a year in more than 1,000 variants. Automated X-ray inspection of solder joints on circuit boards had become the bottleneck of the line, and a second X-ray machine would have cost about €500,000.
Abordagem
Plant engineers trained a model on process parameters from the soldering line. Running on an edge device on the shop floor, it predicts whether a board's solder joints are free of faults and so whether the board needs the X-ray test at all. The plant uses similar models elsewhere, for example to warn of a milling-spindle failure 12 to 36 hours in advance.
Números principais
Lição
The model selected units for the slow inspection step, avoiding an equipment purchase. Siemens did not quantify the reduction in X-ray inspections; the 40% reduction in testing and downtime from prediction models reported by ROI in 2019 has not been independently verified.
Categorias
- Porte
- Grande empresa
- Função empresarial
- Produção e qualidade
- Tipo de inovação
- Inovação de processos
- Tecnologia de IA
- Análise preditiva e ML
- Forma de adoção
- Desenvolvimento interno
- Evidência
- Declaração da própria empresa
Fontes
Resumos das fontes
Siemens reports a quality level of 99.9990% at Amberg and annual output of about 17 million Simatic products in more than 1,000 variants. It describes replacing a planned €500,000 X-ray machine with an AI model, trained on process data and run on an edge device, that predicts whether solder joints are fault-free. It also describes a spindle model that warns 12 to 36 hours before a failure, without giving overall savings.
The consultancy ROI ran the Industry 4.0 Award that Amberg won in 2018. ROI reports a production quality of 99.999% and output of up to 16 million products a year, and says prediction models already flag process deviations early and have cut testing and downtime by 40%. It describes AI-based pattern recognition as a next step.
Os detalhes completos estão disponíveis nas fontes originais.
