RedUsa
The power of the predictive maintenance lies on providing immediate assistance in situations where human judgment disregards the reactions times or when human beings do not possess the required skills. This is considered as highly important particularly in uncertain conditions where making a poor selection might cause high consequences for the production process. This MFP will investigate methods and tools to identify factors that might affects the quality of production and in turn to allow maintenance to be planned before the failure occurs. This project is motivated by the fact that while producing aluminum plates in Austria Metall GmbH (AMAG) and in the affiliated companies (with regard to § 189 a UGB), there might be not-metallic indications in the produced items caused by unknown factors. To be appropriate, these indications have an enormous effect on the quality of the produced plate. In order to tackle this issue, this project should provide methods that can be used to identify the influencing factors and to reveal relationships among these parameters and production quality. Furthermore, the gained insights should be applied to forecast the production. Finally, a visual analytics tool should be provided, which shows the end user (engineers from AMAG CAST) the influencing parameters visually and allows to interact with them.
The data being used in this project are production and quality data.
Goals
The goal of this project is to define visual methods, which can reveal the relationships among production parameters and the production quality. Sensors at various production steps deliver a stream of production data, which is time-dependent and typically, high-dimensional. While the data is continuously captured, its preprocessing and analysis are a challenge, due to the size and heterogeneity of data. The main challenge, however, is to map the time-dependent production data to the run length of the cast aluminum. An interactive visualization tool should therefore provide means to visually inspect the possible influences of production parameters on product quality and promote a better understanding of parameters in production. On the top of that, the tool should be defined in a way that it can be used by the users (engineers from AMAG CAST) that have little or no expert knowledge in visualizations but possess the required domain knowledge about production aluminum plates.
Even with an interactive visual analysis tool that offers users several functions for exploring their data, the users can still be overwhelmed by the huge amount of data and may have difficulties to identify critical patterns in their data sets. With our visual analysis tool, we also want to provide various methods for the detection of specific ultrasonic patterns and thus try to help users to identify possible critical process deviations in production.
Approach
There exist several state-of the art algorithms that can be used to analyse data and identify the influencing process parameter. This, however, requires an extensive literature review to analyse which methods better applies to industrial data. Thus, within the scope of this project, we investigate different algorithms to detect the factors that might influence the produced aluminum plate and to forecast the production.
Note that the data we used within the scope of this project is collected by an ultrasonic device, used to scan the produced aluminum plates. For the visual analytics tool, however, there exist powerful visualization libraries that provide different interactive 2D visualizations. These visualizations provide a good base to support user to visually navigate through the data and explore them to gain insights and draw important conclusions.
Our next goal in this project concerns the identification of meaningful patterns in process data. The existing research covers a broad spectrum of pattern recognition methodologies that can be potentially applied to elicit patterns in data collected from industrial production. Hence, in this paper, we further analyse the applicability of different methods for recognition of specific ultrasonic patterns which may indicate critical process deviations in aluminum production.
Expected and Achieved Results
During a parallel aluminum cast, each batch results in several ingots via a casting pit. We developed a visual analytics tool (ADAM: Aluminum production Data Analysis and Monitoring) which includes scatter plots, showing the front and the top view of ingots, linked with three frequency histograms which provide information about the number of indications in length, width, and thickness of cast ingot. ADAM has been successfully presented at the poster session in EuroVis2019.
Further effort in this project has been put to define a classification model to classify the ingots into “good” and “bad” quality regarding the not-metallic indications they have. Moreover, we defined a prototype of a glyph-based visualization to scale multidimensional data and methods (currently, the classification model) to reveal the relationships among production parameters and the production quality. Yet, batches and ingots have a different distribution of indications in length, width, and thickness. It is important to group similar batches and ingots in order to investigate the influence of production parameters in a more precise manner. To do this, we integrated interactive pattern search in our tool and allowed the user to search for ingots with similar distribution compared to a selected ingot.
