Improve Quality and Efficiency of Textile Process using Data-driven Machine Learning in Industry 4.0

The capabilities of self-awareness, self-prediction, and self-maintenance are important for textile factory in Industry 4.0. One of the most important issues is to intellectualize the way of setting operation parameters as the cyber-physical system (CPS), instead of using traditional trial and error method. To achieve these goals, this paper focuses on the relationship between key operation parameter and defect for machine learning to design an operation parameters recommender system (OPRS) in the textile industry. From the perspective of data science, this paper integrates historic manufacturing process data, such as machine operation parameters from warping, sizing, beaming and weaving process, and management experience data, such as textile inspection results from quality control section. Then, the regression models are applied to predict the textile operation parameters. This research also uses the classification models to predict the quality of textile. Based on the ten-fold cross-validation testing, experimental results show that our model can achieve 90.8% accuracy on quality level prediction and the best regression model for predicting weaving operation parameters can reduce the mean square error (MSE) to 0.01%. By combining the above two models, proposed OPRS can provide a completed analysis data of operation parameters. It provides good performance when comparing with previous stochastic methods. As the proposed OPRS can support technician setting operation parameters more precisely even for a new type of yarn, it can help to fix the tech skills gap in the textile manufacturing process.