Modeling and Predicting an Industrial Process Using a Neural Network and Automation Data

Production optimization and prevention of faults and unplanned production halts are areas of particular interest in industry. Predictive analysis is commonly implemented with data analytics and machine learning techniques. Usually, the usage of such tools requires knowledge of the machine learning theory and the subject to be studied, e.g. a pumping process. This paper presents a case study on modeling of a pumping process using stored automation data. The model is trained to predict the performance percentage of the process with minimal background knowledge of the process and data analytics. The proposed model is built with IBM SPSS Modeler, a data analysis tool not usually used in real-time industrial predictive analysis as it is not often considered the best tool when working with time series data. The model is deployed in a cloud service to implement a real-time, visualized predictive analysis system. The case study shows that Modeler can be used for data analysis, modeling, and production purposes. Depending on the case, Modeler can provide an alternative tool compared with typical machine learning tools, as models built with Modeler can be deployed into a cloud service for production use. The findings indicate that industrial automation data are a valuable resource, and data analysis can be conducted on various platforms and tools.