Wear Pattern Analysis of Pneumatic Cylinders with Machine Learning

Siddhi Nitin GUNAJI

Diplomová práce

Abstract:

The thesis investigates the wear patterns of pneumatic cylinders at BMW's 02.20 plant in Dingolfing, Germany, focusing on stator assembly for electric engines. The study aims to model and predict cylinder wear patterns using historical operational data and advanced machine learning techniques, particularly Long Short-Term Memory (LSTM) networks. The research begins with a comprehensive review of current methodologies for wear pattern forecasting, identifying existing gaps that this work addresses. Data collection includes capturing operational parameters every millisecond, with rigorous preprocessing to ensure data quality. Challenges such as unscheduled maintenance, data volume, and the absence of station-specific temperature data are addressed throughout the study. Initial models using linear regression set a baseline, while LSTM models excel in capturing complex temporal dependencies, demonstrating superior predictive accuracy. Practical implications include improved maintenance scheduling to minimize downtime and reduce operational costs. Future research directions suggest exploring hybrid models, generalizing across different cylinders, and optimizing data aggregation intervals. This thesis highlights the potential of advanced data analytics to enhance operational efficiency and optimization in industrial settings.

Abstract:

The thesis investigates the wear patterns of pneumatic cylinders at BMW's 02.20 plant in Dingolfing, Germany, focusing on stator assembly for electric engines. The study aims to model and predict cylinder wear patterns using historical operational data and advanced machine learning techniques, particularly Long Short-Term Memory (LSTM) networks. The research begins with a comprehensive review of current methodologies for wear pattern forecasting, identifying existing gaps that this work addresses. Data collection includes capturing operational parameters every millisecond, with rigorous preprocessing to ensure data quality. Challenges such as unscheduled maintenance, data volume, and the absence of station-specific temperature data are addressed throughout the study. Initial models using linear regression set a baseline, while LSTM models excel in capturing complex temporal dependencies, demonstrating superior predictive accuracy. Practical implications include improved maintenance scheduling to minimize downtime and reduce operational costs. Future research directions suggest exploring hybrid models, generalizing across different cylinders, and optimizing data aggregation intervals. This thesis highlights the potential of advanced data analytics to enhance operational efficiency and optimization in industrial settings.

Keywords

predictive maintenance, pneumatic cylinders, electric engines, machine learning, linear regression, LSTM neural networks, data preprocessing, feature engineering, sensor signals, wear pattern analysis

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 20. 8. 2024

Obhajoba závěrečné práce

• Vedoucí: prof. Ing. Markus Mayer, Dr.

JIHOČESKÁ UNIVERZITA V ČESKÝCH BUDĚJOVICÍCH

Přírodovědecká fakulta

Magisterský studijní program / obor:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science