Using motor current for detecting handling errors in electronic pipettes
Publication date
2026-06-16
Document type
Konferenzbeitrag
Author
Organisational unit
Scopus ID
Conference
International Symposium on Artificial Intelligence for Instrumentation and Measurement (AI4IM 2026) ; Amalfi, Italy ; May 21–23, 2026
Publisher
IEEE
Book title
2026 IEEE International Symposium on Artificial Intelligence for Instrumentation and Measurement (AI4IM)
Part of the university bibliography
✅
Language
English
Keyword
artificial intelligence
classification
electronic pipettes
error detection
liquid handling
motor current
Abstract
Micropipettes are essential tools in laboratory workflows, with pipetting errors contributing significantly to laboratory errors and costs in research, industry and diagnostics. Electronic pipettes offer better precision compared to manual pipettes, but do not eliminate operator mistakes. Existing error detection methods in liquid handling such as pressure sensors or computer vision are mostly limited to automated liquid handling and are less suitable to hand-held pipettes due to added complexity and cost. We propose an AI-based approach that analyzes motor current time series during pipetting to detect errors. We collected a dataset of 2902 pipetting processes, which is publicly available, by using five electronic pipettes under controlled laboratory conditions including correct, air ingestion and total error scenarios. We trained convolutional neural networks (InceptionTime) on the motor current data, and achieved over 90% accuracy for both multiclass and binary error detection on known pipettes. Generalization to unknown pipettes remains challenging, with accuracy dropping to about 42% (multiclass) and 60% (binary), indicating significant interpipette variability. Fine-tuning with a small number of samples from a new pipette rapidly improves classification performance, reaching ∼85% accuracy with 100 samples. In conclusion, we highlight the feasibility of motor current based error detection for electronic pipettes, and the need and potential for further improvement by collecting larger datasets. The presented dataset is made publicly available to encourage further research into error detection for laboratory pipetting.
Version
Published version
Access right on openHSU
Metadata only access