A methodology for building a data-enclosing tunnel for automated online-feedback in simulator training

Abstract

Extensive research confirms that feedback is the key to an effective training. However, in many domains, human trainers, who can provide feedback to trainees, are considered not only a costly but also a scarce resource. For trainees to be more independent and undergo self-training and unbiased support, effective automated feedback is highly recommended. We resort to elements from the theory of data mining to devise a data-driven automated feedback system. The data-enclosing tunnel is a novel concept that may be used to detect deviations from correct operation paths and be the base for automated feedback. Two case studies demonstrate the viability of this methodology and its usefulness in industrial simulation scenarios. Case study 1 focuses on the increase of oil production, whilst case study 2 focuses on the decrease of gas production. The data-enclosing tunnel is validated and compared with three other assessment methods. These methods are simpler versions of the data-enclosing tunnel method, as they are three variants of a baseline approach Data Enclosing Band (DEB), namely DEB1, DEB2, DEB3. The methods accuracy is determined by calculating how precisely they can classify new data. The data-enclosing tunnel yielded the highest accuracy, 94.3%, compared to 81.4%, 62.9%, and 70% for DEB1, DEB2, DEB3 respectively.