AUTOMATING MOTOR DATA COLLECTION WITH MACHINE LEARNING - Object Detection and Optical Character Recognition for Data Acquisition - Master’s thesis

Aug 5, 2026

Reliable data of installed electric motors is critical for maintaining productivity and reliability in industrial environments. ABB’s current on-site motor data collection relies on manual data entry, which is time-intensive and error-prone, limiting scalability and accuracy. This master’s thesis explores automating motor data collection for IEC standard low-voltage induction motors by integrating object detection and optical character recognition. It applies YOLOv11 for nameplate attribute detection and TrOCR for text transcription, achieving accurate extraction of key alphanumeric fields while testing generalization on unseen data.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages66

Reading Time2 h 46 min

AccessFree to download

Abstract

Motivation and problem

Proposed approach and models

Training, testing, and results

Limitations and future work

Practical impact

Ville Ruotsalainen AUTOMATING MOTOR DATA COLLECTION WITH MACHINE LEARNING Object Detection and Optical Character Recognition for Data Acquisition Master’s thesis Information Technology and Communication Sciences Examiners: Prof. Joni Kämäräinen MSc. Jarno Matilainen ... March 2025 i ABSTRACT Ville Ruotsalainen: Automating motor data collection with machine learning Master’s thesis Tampere University Signal Processing and Machine Learning March 2025 Reliable data of installed electric motors is critical for maintaining productivity and reliability in industrial environments. ABB’s current on site motor data collection processes rely on manual data entry, which is time-intensive and susceptible to errors, limiting scalability and accuracy. This thesis investigates the automation of motor data collection through the integration of advanced object detection and optical character recognition technologies, focusing on standard IEC low voltage induction motors. The research leverages state-of-the-art machine learning models: YOLOv11 for object detection and TrOCR for optical character recognition, to automate the identification and transcription of motor nameplate data. Trained with ABB’s annotated nameplate data YOLOv11 demonstrated strong performance in detecting key attributes such as type codes, product codes, and serial numbers. TrOCR, utilizing a transformer-based architecture, provided reliable text recognition, enabling accurate extraction of critical alphanumeric data. Combined, these models showcased significant potential to replace manual workflows, streamline data acquisition, and enhance ABB’s data management processes. Testing on unseen data confirmed the system’s ability to generalize effectively, although challenges such as class imbalance and variability in orientation of technical attributes remain. Future work could address these challenges by expanding the dataset to include diverse orientationsand training the models in cloud environment with better computing power. The findings suggest significant benefits for ABB, including reduced manual workload, improved data reliability, and enhanced scalability in managing large motor inventories. This research demonstrates the transformative potential of machine learning in industrial applications, providing a foundation for further advancements in digital asset management and predictive maintenance. By automating motor data collection, ABB can improve operational efficiency and service quality, contributing to a broader digital transformation in industrial maintenance workflows. Keywords: Data extraction, Machine Learning, Object detection, Optical Character Recognition (OCR) The originality of this thesis has been checked using the Turnitin OriginalityCheck service. ii TIIVISTELMÄ Ville Ruotsalainen: Moottorikartoituksen automatisoiminen koneoppimisen avulla Diplomityö Tampereen yliopisto Signaalinkäsittely ja koneoppiminen Maaliskuu 2025 Luotettava data asennetuista sähkömoottoreista on kriittistä tuottavuuden ja luotettavuudenylläpitämiseksi teollisissa ympäristöissä . ABB:n nykyinen moottorikartoitusprosessi perustuu manuaaliseen tietojen syöttöön, mikä on aikaa vievää, rajoittaen näin prosessin skaalautuvuutta jatarkkuutta. Tämä diplomityö tutkii moottorikartoituksen automatisointia hyödyntämällä objektintunnistusta sekä optista merkintunnistusta kuvista, keskittyen IEC-standardiin perustuviin pienjännitteisiin oikosulkumoottoreihin. Tutkimuksessa hyödynnetään koneoppimismalleja: YOLOv11 ja TrOCR moottorien arvokilpitietojen tunnistuksen ja litteroinnin automatisoimiseksi. ABB:n moottorikilpidatalla koulutettuna YOLOv11 osoitti vahvaa suorituskykyä teknisten tietojen, kuten tyyppikoodien, tuotekoodien ja sarjanumeroiden, tunnistamisessa. TrOCR, joka perustuu transformer arkkitehtuuriin, tarjosi luotettavan tekstintunnistuksen mahdollistaen kriittisendatan lukemisen. Yhdessä nämä mallit osoittivat merkittävää potentiaal

FAQs

What problem does the thesis address in motor data collection?

Installed motor data is essential for industrial reliability, but ABB’s current workflow uses manual entry, which is slow, error-prone, and hard to scale. The thesis targets automation to improve accuracy and efficiency.

Which machine learning models are used for nameplate data extraction?

YOLOv11 performs object detection to localize key nameplate attributes, while TrOCR performs optical character recognition to transcribe the alphanumeric text. Together they enable automated identification and reading of motor nameplate data.

How did the system perform when tested on unseen data?

Testing on previously unseen data confirmed effective generalization. The results also highlighted remaining challenges such as class imbalance and variability in the orientation of technical attributes.

Aug 5, 2026

Upload by Evangeline