Abstract
The growing demand and new fields of application compel battery manufacturers to higher product quality. Thus, defect-free battery separators are a prerequisite for safe lithium-ion cells. Hence, typical production faults have to be reliably detected. Therefore, a machine vision system was developed [1]. Besides the detection of anomalies, a key element is the distinction between defect classes in order to distinguish non-quality related optical effects from faults using a machine learning approach for classification. Therefore, a method consisting of five phases is described. Starting with data analysis and preparation, expert knowledge is used to derive a database. On this basis models are generated by machine learning algorithms. The best model is chosen by a weighted evaluation of the target values using a preference matrix. Furthermore, the model gets evaluated by an analysis of the confusion matrix. Finally the implementation phase serves to transfer the results into the machine vision system and to monitor occurring process drifts. Last, the functionality of the method is proven within an application example.
Original language | English |
---|---|
Pages (from-to) | 585-590 |
Number of pages | 6 |
Journal | Procedia CIRP |
Volume | 57 |
DOIs | |
State | Published - 2016 |
Externally published | Yes |
Event | 49th CIRP Conference on Manufacturing Systems, CIRP-CMS 2016 - Stuttgart, Germany Duration: 25 May 2016 → 27 May 2016 |
Keywords
- Optical inspection
- battery separator
- classification
- data mining
- decision trees
- knowledge discovery in databases
- machine learning