SINPRO
This multi firm project (MFP) will investigate a novel decision support technology for assistance in the manufacturing- and production setting for plants in the context of the sintering process. The motivation is, that the outcome in product quality and the manufacturing efficiency can be increased by understanding the circumstances of all components and optimizing their interacting.
Within SINPRO, a huge amount of sensor data is gathered from a sinter production machine and analyzed to understand, which components of the production process affect the quality of the final product. With these findings a prediction model should be defined that uses the detected influencing factors of the whole process and predicts the quality and amount of production. On the top of that, the existing rules of the rule based expert system are investigated to understand current system-changing events and the resulting rules. With these findings, existing rules will be adapted and optimized, and new rules defined to gain a higher production and quality increase on the sinter production machine. Finally, the influencing factors, the prediction model and possible rules, as well as the production data, will be visualized to make the research outcome better understandable for the user.
The pursued results and targeted impact contain findings about the process and relationships in the sinter plant and the sinter process; a better representation of parameters in the sinter plant and the implementation of further analysis in an interactive way, forecasts and predictions for process variables and quality characteristics and a customized expert system and advanced decision support.
The novelty value and scientific relevance embrace the application of data analytics and visual analytics approaches in the area of a sinter plant in the steel industry; new scientific findings and contributions in the field of visual interactive prediction in the industrial sector; requirements and solution models for introducing data analytics in an industrial context; findings about the possible uses or the connection between data analytics and visual analytics; insights into the interplay and connection of data and rule-based decision support in the industrial environment.
Goals
The overall goal in this project is to optimize the production process of sinter material. This should be achieved by increasing the amount of the produced material from the sinter strand as well as improving the quality of the produced material. One of the most important factors to reach this goal is the optimization of the burn through point (BTP) of the material, which should be as close as possible to the end of the sinter strand.
Use Case 1: Understanding the influencing parameters for optimizing the harmonic diameter (DH).
Use Case 2: Optimizing the BTP towards the end of the sinter strand.
By applying the research methods described below, the implementation of the Use Cases should lead to fulfil the project goals.
Approach
The approach is to understand which factors affect the quality of the final sinter product. With these findings, the project goal should be achieved. The research methods and topics of interest of the approach contain a time series analysis and classification of existing data on production and quality; an identification of influencing variables for the identified classes; the creation of a prediction model for defined parameters; visual preparation of the data from the sinter plant; an user-specific representation of the visualizations; implementing interaction concepts for visual analysis of the data; an extension of the rule-based expert system with findings from the data analysis.
Expected and Achieved Results
The overall goal in this project is to optimize the production process of sinter material. This should be achieved by increasing the amount of the produced material from the sinter strand as well as improving the quality of the produced material. One of the most important factors to reach this goal is the optimization of the burn through point (BTP) of the material, which should be as close as possible to the end of the sinter strand.
We first defined a time model to have reference points for the analysis tools. We further used this model with the feature engineering and selection methods to identify the most relevant parameters for the production. These features are then applied to define a forecasting model to predict the harmonic diameter as a central quality parameter indicating the grain sizes distribution of the finished sinter. Due to the complexity of the model we developed and presented an approach for the increase of the explainability of the complex (black-box) forecasting model, enabling easier discovery of new insights and control strategies.
To visually assess temporal data and the relation between attributes inclusive the related correlation coefficient, we made use of two open-source visual analytics applications and extended their functionality. First, we took advantage of Ordino, an interactive rank-based web application, which is used for data-driven approaches to create, visualize, and explore rankings of items. Second, we added further functionality by using TourDino to calculate and visualize similarity measures. TourDino helped us in seeking relationships and patterns in data and provided an overview of the statistical significance of various attribute comparisons without losing the existing ranking.
Further effort has been put in defining a concept to extend the rule-based expert system. This concept dictates three steps:
(i) development of a prototype for a strand speed control; the purpose of this control is to keep the actual BTP around the BTP setpoint in an acceptable range and the speed as stable as possible
(ii) integration of the prediction models provided by WP 2 into the expert system
(iii) (optional) communication with WP3 for visualization: visually display intrinsic factors which have direct relationships to each other and are related to the strand speed control, and thus to the sinter productivity
The results we obtained so far have been submitted to AISTECH2020.
