Transformer networks for univariate time series prediction in predictive process control

This study aims to evaluate the effectiveness of using a transformer network for univariate time series forecasting in the context of predictive process control. According to established research, univariate time series forecasting can serve as an effective method for predicting process quality, especially in situations where data fusion or sampling is not possible. These models can effectively incorporate implicit process knowledge, making them more effective at predicting process quality. Throughout the study, no hypotheses were found to be disadvantageous in regards to the fundamental research topic. The model results showed that the process capability index had a Mean Squared Error of 0.0125, and the quality characteristics had a Mean Squared Error of 0.0339. These findings suggest that the transformer network used in this study was effective at predicting process quality. In addition, it underscores the importance of predictive process control in maintaining the stability of manufacturing processes. By providing process managers with the information they need to make informed decisions and intervene before production failures occur, predictive process control can reduce production costs, improve resource utilization, and enhance the interdependencies between employees and machines. However, embedding forecasting applications in the production process requires a deep understanding of the underlying processes, the equipment used, and the individuals involved. As such, a commitment to continuous improvement is essential to ensuring the success of these applications.