In industrial applications, data is widely utilized to enable cost savings through optimized production. However, in a technical research organization, experimental data is often available in large quantities yet not fully utilized. The work presents a case study of the development of a digital twin for a pilot-scale fibre process research environment (SUORA), located in Jyväskylä, Finland. Using the digital twin, discrete laboratory analyses could be partially replaced with real-time quality parameter estimations. To evaluate the replicability of the approach, multiple quality parameters were studied, including dryness (former, press and web), as well as web basis weight, density and thickness. To construct the digital twin, data spanning over five years was collected, consisting of process measurements and laboratory analysis results. Representative operating periods were selected by the process operators, and the respective data converted into a single dataset. Using the dataset, feature selection was carried out for each quality parameter separately, showing the most descriptive process measurements. With the identified features, gradient boosting models were trained, showing good agreement with test data. In addition, isolation forest models were trained to assess the quality of the input data. With this approach, the operator was able to estimate also the prediction quality. The digital twin was deployed in the pilot control room (Fig. 1, left), with input data from the process control system fed to the models at a constant interval (Fig. 1, right). As the main benefit, the digital twin enables better estimation of the current process state, allowing desired steady-state conditions to be reached more easily. During the future trials with the pilot environment further model validation and end-user experience evaluation will be conducted.
|Publication status||Published - 10 Feb 2021|
|MoE publication type||Not Eligible|
|Event||PaperWeek Canada - Virtual|
Duration: 8 Feb 2021 → 12 Feb 2021
|Abbreviated title||PWC 2021|
|Period||8/02/21 → 12/02/21|