A study on remaining useful life estimation for predictive maintenance of a robot cell

Abstract

With the introduction of digitization, a vast amount of data is available for industries, which can be used for their future sustainability and competitive advantage. The amount of data generated by the modern machines exceeds the capacity of manual analysis. At the same time, the improvement in computational power and advancements in the field of Artificial Intelligence (AI) provides insightful analysis, which can enable data driven decision making in numerous fields such as manufacturing, automobile, construction and food processing, and so on. This thesis project aims to propose a study on a Remaining Useful Life (RUL) estimation for predictive maintenance by using of a data-driven approach. This thesis analyzes the present maintenance practices that are being followed in Volvo Cars Torslanda and introduce data-driven based maintenance planning to optimize the available resources. For this reason, the suitable predictive maintenance strategies are examined for gluing workstations in the automobile production factory. The purpose of the project is to transfer from time-based maintenance to predict equipment failure and RUL using state-of-the-art machine learning algorithms. In order to perform a data driven approach for predictive maintenance, the CRoss Industry Standard Process for data mining (CRISP-DM) is followed as a reference methodology in this thesis. Different data sources are analyzed and the most relevant sources for the study are selected. An exploratory data analysis is done with the available data from specified workstation and the most suitable parameters are selected for the study. Due to data from different sources, many data pre-processing techniques are utilized in order to merge and make the data suitable for the Machine Learning (ML) algorithms used for prediction. Two of the most common ML algorithms such as Auto Regressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) are utilized for the prediction and LSTM model provides a better prediction on the test data. In conclusion, this thesis provides a set of recommendations for the company, which would enable them to conduct future predictive maintenance projects and to help Volvo Cars in their advancements with Smart maintenance road map.