Application of cognitive software for specification and characterization of valuable side streams in metal industries
Maksimov, Pavel (2018)
Diplomityö
Maksimov, Pavel
2018
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2018053125051
https://urn.fi/URN:NBN:fi-fe2018053125051
Tiivistelmä
Currently, the rate of novel knowledge generation is ever increasing at an exponential rate, thus providing demand for new advanced data analysis methods. Therefore, large corporations along with smaller startups are trying to integrate new technologies for big data processing with the purpose of obtaining additional insights from seemingly useless datasets. At the same time, due to the recent widespread deployment of cloud-based technologies, software applications that require excessive computational capability can be accessed and operated from any device regardless of its processing power. In view of this, cloud-based software tools for information processing, that facilitate efficient data analysis without significant computational power consumption, have drawn increased attention from the modern scientific community. Thus far, these software utilities have found extensive application mostly in social studies or business and management areas, while their utilization for analysis of technical information has been relatively limited.
Within the limits of this work application of novel data processing software for metal industries related information analysis has been studied through the example of IBM Watson’s analytical platform. A series of datasets containing relevant information about aluminum and copper production processes has been collected and initially refined to facilitate further analysis. These datasets have been studied with the purpose of revealing
hidden interdependencies between processing parameters and determination of the key statistical drivers affecting main products quality. Special consideration has been given to the side stream of copper production, namely slag, and main processing parameters that were associated with the most significant impact on its properties. Predictive analysis of main slag and matte constituents’ content has been conducted in order to provide more insight into the process parameters and products properties. Additionally, a comparative analysis of the analytical platform with a conventional linear-regression based software application has been performed in order to assess and to highlight possibilities of Watson Analytics to group, classify and connect process related information.
The obtained results highlighted a crucial importance of quality and volume of the analyzed information set. More specifically, amount of available information had the most significant impact in terms of predictive analysis – increased quantity of observations resulted in generation of more detailed predictive models. Likewise, results of the product properties key drivers’ determination were also associated with increased statistical accuracy in case of analysis of larger datasets. Furthermore, the conducted study revealed significant interdependence between main matte and slag constituents’ concentrations. Among the most important processing parameters, cooling water temperature and oxygen concentration have been associated with the most substantial impact on the product properties.
Moreover, an additional study has been conducted by application of Watson Discovery service for analysis of a collection of scientific articles related to processing and reuse of waste and side streams in metal production industries. Data ingestion algorithms have been customized and the cognitive system has been trained to understand industry-specific terms and text context in order to facilitate natural language processing.
Within the limits of this work application of novel data processing software for metal industries related information analysis has been studied through the example of IBM Watson’s analytical platform. A series of datasets containing relevant information about aluminum and copper production processes has been collected and initially refined to facilitate further analysis. These datasets have been studied with the purpose of revealing
hidden interdependencies between processing parameters and determination of the key statistical drivers affecting main products quality. Special consideration has been given to the side stream of copper production, namely slag, and main processing parameters that were associated with the most significant impact on its properties. Predictive analysis of main slag and matte constituents’ content has been conducted in order to provide more insight into the process parameters and products properties. Additionally, a comparative analysis of the analytical platform with a conventional linear-regression based software application has been performed in order to assess and to highlight possibilities of Watson Analytics to group, classify and connect process related information.
The obtained results highlighted a crucial importance of quality and volume of the analyzed information set. More specifically, amount of available information had the most significant impact in terms of predictive analysis – increased quantity of observations resulted in generation of more detailed predictive models. Likewise, results of the product properties key drivers’ determination were also associated with increased statistical accuracy in case of analysis of larger datasets. Furthermore, the conducted study revealed significant interdependence between main matte and slag constituents’ concentrations. Among the most important processing parameters, cooling water temperature and oxygen concentration have been associated with the most substantial impact on the product properties.
Moreover, an additional study has been conducted by application of Watson Discovery service for analysis of a collection of scientific articles related to processing and reuse of waste and side streams in metal production industries. Data ingestion algorithms have been customized and the cognitive system has been trained to understand industry-specific terms and text context in order to facilitate natural language processing.