CORDIS bietet Links zu öffentlichen Ergebnissen und Veröffentlichungen von HORIZONT-Projekten.
Links zu Ergebnissen und Veröffentlichungen von RP7-Projekten sowie Links zu einigen Typen spezifischer Ergebnisse wie Datensätzen und Software werden dynamisch von OpenAIRE abgerufen.
Leistungen
The rolling force calculation is the basis for the control of the geometric quantities of steel flat products and also affects technological properties. The accuracy of analytic rolling force models used in process automation of hot strip mills is limited regarding the effect of the various alloy components and phase transitions in modern high-performance materials.
Within Neuroll the adaptation and optimisation of the analytic rolling force model by means of neural networks was developed further. The Neuroll work demonstrated that accurate modelling of rolling force is possible for an extremely wide product mix, and for 2-phase rolling.
Applications of 2-phase rolling is required for low carbon qualities and for special high alloyed stainless steels. Phase transitions show substantially different characteristics for different steel grades. In recent years, such rolling has been carried out to a large extend at partner TKS. Rolling models are currently being extended to those cases. Improving quality and range of validity of rolling force models in the described way meets the continuously rising requirements of steel customers concerning geometrical accuracy and material properties.
In the beginning of the project a general concept of SOM-based data analysis was developed. A thorough analysis of the rolling process data was performed in accordance with the concept. The analysed data consisted of both the on-line measurements of the process variables and the one strip statistics (averages, standard deviations etc.) calculated from the measurements. Several new data visualisation and interpretation methods was introduced. Also SOM-based process monitoring was enhanced. Two novel approaches for monitoring was developed and a broad-brush monitoring tool for research purposes was implemented. The basis for the data analysis and monitoring tools developed and implemented in HUT is the SOM Toolbox introduced above. In fact most of the tools are included in the toolbox as contributed functions. Hence, these data analysis and monitoring methods are freely available and suitable for diverse application areas.
This is a model containing six Artificial Neural Networks (ANN) or six statistical formulas for prediction of strip width in the roughing stand of a hot strip mill. The models are fitted to data from Finite Element Method (FEM) calculations. The models can consider different shape of strip edges as positive or negative bulges. The so-called "dogbone" profile formation at edge rolling is considered.
The task of the Short-Stroke-Control (SSC) is to achieve the best possible width constancy of the steel strip at the strip head end and the strip tail end. The SSC therefore calculates displacement curves for the hydraulic position control of the edger rolls near the strip head and tail end. In many cases, today's SSC's cannot find an optimal edger displacement curves for an optimal width accuracy at the head and tail end of the strip and commissioning is very time consuming, because an SSC usually takes a lot of individual solutions for each mill-type.
Within Neuroll, a robust, adaptable rougher width model and a neural network based Short-Stroke Controller was developed. The new features of the SSC are: robust controller commissioning procedure even at strongly disturbed width measurements, consideration of various rolling-technological constraints in each individual pass and pass-schedule independence for an optimal width result at a high variety of roughing mill configurations and pass plans. The SSC-System including the required commissioning tools was programmed and prepared for inclusion into a process control system. It was tested in simulations and was successfully piloted (without on-line adaptation) by Siemens AG at a European hot wide strip mill.
A hybrid model containing two Artificial Neural Networks (ANN) and an elastic roll deflection model for prediction of strip profile and flatness in the finishing stands of a hot strip mill. The ANNs are fitted to data from more advanced strip rolling simulation calculations. The models can consider the influence of strip tension and changes in relative strip profile.
SOM Toolbox is a software library for Matlab 5 implementing the Self-Organizing Map (SOM) algorithm. It implements basic pre-processing, training and visualisation routines for making SOMs. In addition, the package includes several other clustering and projection algorithms as well as some more advanced visualisation as contributed functions. Hence, the toolbox provides an efficient data-driven, adaptive, and intelligent approach for mining and visualisation of large, high-dimensional data sets. Moreover, the package is free, so it is readily applicable for various other data analysis applications where large amount of numerical data is available.
A steel strip body width model for rolling in a roughing mill was developed. It is the basis for strip body width process control as well as for SSC. In order to obtain a general model, data from steel mills with different roughing train configurations and pass schedules have been used (including partner Rautaruukki�s data). The main advantage of the model is that it is robustly adaptable. Model terms were selected from an analysis of a wide range of experiments. Although partly selected experimentally, each model term is physically plausible and allows for extrapolation.
The model was implemented at Rautaruukki Raahe rolling mill and is in operation within the existing roughing mill automation system. Control accuracy was much less than modelling accuracy. In fact, the model will only reveal its full qualities when employed in an up-to-date automation system.
The objective of the basic tasks concerning data analysis was to find non-linear relationships and correlations between chosen process data in the high dimensional steel mill operation data space and use them as a base for modelling, monitoring, fault diagnostics and novelty detection. We developed a data analysis tool which includes a couple of new advanced methods: various clustering methods, a dedicated data sorting algorithm and a method for checking noisy process data for plausibility. Data analysis is an important pre-condition for the training of neural networks and data based modelling. Additionally data analysis can be used for monitoring the health stage of an industrial process.
Suche nach OpenAIRE-Daten ...
Bei der Suche nach OpenAIRE-Daten ist ein Fehler aufgetreten
Es liegen keine Ergebnisse vor