EFFICIENT DATA GENERATION FOR MACHINE LEARNING: INTEGRATING DESIGN OF EXPERIMENT AND DATA QUALITY

Kurzfassung

A good detection algorithm and its underlying model rely fundamentally on a high-quality data foundation. However, the immediate question arises: What constitutes a high-quality data set? While large volumes of operational data can comprehensively reflect realistic system, newly designed systems lack access to such data during early life cycle phases. Consequently, data synthesis during the design phase becomes increasingly important, particularly when preparing experiments for subsequent machine learning and deep learning applications. In the context of machine health systems, experimental data acquisition is often costly, time-consuming, or infeasible due to safety risks and system constraints. Therefore, the number of physical experiments must be minimized, while ensuring that the generated data remains informative and representative for downstream modelling tasks. This paper lays the groundwork for data synthesis and optimal data set design, leveraging structured and statistical methodologies. The discipline of Design of Experiments (DoE) has been well established in academic and industrial domains as an efficient approach to systematically explore input–output relationships under constrained experimental budgets. By integrating DoE principles with simulation-based data generation and machine learning requirements, the proposed approach enables the creation of informative, diverse, and statistically robust data sets, even in the absence of extensive real-world observations. The concept of data quality is well established and standardized. Data quality follows the principle of fitness for use and ISO/IEC 25012 defines data quality as “degree to which the characteristics of data satisfy stated and implied needs when used under specified conditions”. Many methods are available for assessing the quality of data sets which are being integrated. This method is demonstrated for the use-case of acoustic fault detection of unmanned aerial vehicle propellers.