The Initial Steps on the Road to Digitalisation of Impol
Thursday 23rd May 14:00 - 14:15
In this work, the following examples of digitalisation in Impol Aluminium Industry has been presented: (i) the use of the artificial intelligence in modelling of production processes, (ii) the use of self-learning algorithms in development of improved properties of the end-products, and (iii) the use of predictive artificial intelligence in quality control and certification.
The first example is modelling of the processing path for the production of the extruded semis with required mechanical properties. For that purpose, the powerful and cost-effective algorithm has been developed and learned for (i) extracting structured data (process parameters, concentrations of alloying elements, and mechanical properties), (ii) finding the correlations between the individual processing paths and the end-product properties, and (iii) performing the predictions on the composition of new alloys and the processing parameters for matching the required mechanical properties.
The end-product performances improvement achieved by the use of the self-learning algorithm is illustrated by developing more intergranular corrosion-resistance wrought aluminium alloys. The purpose of the self-learning algorithm, developed for that particular case, was to compare and correlate the results of corrosion resistance measured by two different methods. In the first set of experiments, the corrosion resistance was determined by metallographic measurement of the intergranular corrosion. In the second, the corrosion resistance numerical data were exported directly from an electrochemical cell. The entire process has resulted in the accumulation of the appropriate filtered and structured data for high-quality data-driven predictions of the most stress-corrosion resistive compositions of wrought aluminium alloys.
The use of predictive artificial intelligence in quality control and certification was exampled by a new method of data-driven inspection of inclusions in wrought aluminium alloys. Promising approach for fast inspection of inclusions in wrought aluminium alloys is the optical emission spectroscopy (OES). However, in order to separate the peaks corresponding to particular inclusions from the peaks obtained from various microstructural features in the matrix, an advanced filtering of the OES spectrum is necessary. The methodology developed is based on big-data-driven predictions of whether the on-line analysing sample is good or bad. By following a machine-learning process, an algorithm was developed enabling the on-line division of the samples into good and bad, based on criteria received from the casting house.
Finally, the initial efforts towards e-verifying the compliance of international standards and customer requirements, specified on certificates of quality, with internal processing parameters and work-order activities, will be also discussed.
Speaker: Dr Varužan Kevorkijan, Managing Director at Impol Group
Varužan Kevorkijan is the doctor of the materials science with more than 25 years of experience in aluminium industry and industrial R&D.
In Impol Aluminium Group, as the managing Director of Quality and Research, he is currently fully responsible for the quality control and certification.
His industrial and academicals R&D activities are focused on the advanced aluminium recycling technologies, the development of new, recycling friendly aluminium alloys, the fabrication and characterization of aluminium-based composites, the stochastic and cognitive modelling of advanced materials, the development of advanced continuous casting technologies, the surface engineering of light metals and composites, the development of novel forming technologies and the advanced materials characterisation.
He is an author of more than 300 scientific and technical papers and the member of numerous international professional societies.
Abstract - Data-driven Industrial Modelling of Future Wrought Aluminium Alloys
In the successful development and commercialization of future wrought aluminium alloys and the corresponding end-products, it is necessary to understand and to be able to predict how the required combination of properties is affected by the different available processing paths, i.e., the chemical composition of the alloy and the main processing parameters.
An industrial tool developed within the Impol Aluminium Group for such a modelling of wrought aluminium alloys is OPTIAl – the cognitive computing algorithm for correlating the properties of wrought aluminium alloys, the chemical composition and the processing parameters. As will be reported, the computing methodology is based on the Bayesian inference, which makes the algorithm extremely fast and efficient. The inductive learning of the algorithm was performed by applying the experimentally confirmed equivalency of different technological paths, able to provide the same combination of properties. In the first step, by practicing the data mining, a data matrix was created, consisting of the results of standard, room-temperature tensile tests and the corresponding technological paths for different production lots of the AA 6110 alloy. Next, the most probable technological path-property correlations were identified. Finally, various standard and some non-standard alloy compositions, derived from the alloy AA 6110, and the processing parameters were cognitively inducted to provide the desired combination of properties.
The validation of the above-described methodology was performed through the regular production of a limited number of cognitively computed alloys and their detailed characterization. It was found that the achieved predictions are with high accuracy, sufficient for most industrial applications.