Optimalizace operací s Big Data v nadnárodní společnosti

This thesis addresses the problem of optimizing the processing of large volumes of manufacturing data in an industrial environment. The main objective is to design and experimentally validate methods that reduce the volume of stored and transmitted data while enabling effective identification of anomalies in the manufacturing process. The work focuses on two main areas: the analysis of scalar process data and the processing of curve-based data representing torque curves. For scalar data, a statistical approach to anomaly detection based on robust methods, specifically the median and the median absolute deviation, is employed. For curve-based data, a compression method is first proposed to reduce data volume while preserving informational value, followed by the application of a convolutional autoencoder model to detect anomalous patterns based on reconstruction error. The proposed methods are implemented in the Python programming environment and validated on real-world data from a production line. The results indicate that the use of statistical methods and machine learning approaches enables the identification of abnormal behavior in the manufacturing process while simultaneously reducing the volume of processed data. The thesis also includes an indicative evaluation of the economic benefits of some of the proposed solutions in an industrial context.