Enhancing Parts Flow Data Quality in Serial Production Lines: Algorithms and Computational Implementation

With the advent of the Industry 4.0 era, the manufacturing industry is implementing a range of novel technologies on the factory floor, leading to the generation of substantial quantities of production data. However, the development of analytics tools capable of processing these data and extracting valuable information for decision-making and production control lags behind. In addition, a noticeable amount of raw data collected from the factory floor is prone to errors, especially in small- and medium-sized manufacturing plants, and their processing often requires a laborious data cleaning process due to the limitations of the sensors and the noisy environment of the manufacturing facilities. This presents a challenge in utilizing factory floor production data effectively. This paper addresses the challenge by focusing on the parts flow data, which reflects the number of parts in each buffer as a function of time in a production system. In particular, we study the parts flow data in discrete-time serial production line models, assuming that the data are subject to random noise, and develop effective and robust algorithms that can effectively detect and correct errors in these data. To improve the computational efficiency for complex cases (longer lines, higher error rates, etc.), a decomposition-based approach is used to parallelize the computation procedure at implementation. Numerical experiments demonstrate that the proposed methods can enhance data quality by more than 40% and improve the accuracy of system performance metrics estimation by over 50% using corrected data. These improvements can facilitate more reliable process monitoring and production control in manufacturing environments.