Algorithmen für den Erfahrungstransfer zwischen Experten und Einsteigern bei der Bedienung komplexer Systeme am Beispiel von Utility-Tunnelbohrmaschinen
Der optimale Betrieb komplexer Systeme erfordert umfangreiche Erfahrungen von qualifizierten Betreibern, deren Gewinnung lange dauern kann. Solche Betriebserfahrungen lassen sich unter Umständen nur schwer formalisieren und folglich nur schwer an Neulinge weitergeben. Die Bedeutung und Schwierigkeit eines solchen Erfahrungstransfers ist besonders groß bei Systemen, die unter hoher Unsicherheit und hohen Anforderungen an die Betriebsleistung betrieben werden, wie z.B. bei Tunnelbohrmaschinen.
Abstract
In tunnel construction projects, delays entail high costs. Thus, tunnel boring machine (TBM) operators aim for high advance rates without compromising safety. This is a difficult mission in uncertain subterranean environments and often, experienced operators are better able to find optimal control parameters. The recently published paper by the chair of intelligent maintenance systems presents an intelligent decision support system that provides recommendations on the control parameters. To do so, it learns from past projects how the controls in various geological conditions impacted the boring efficiency. Then, given the control parameters set by an operator and a rough knowledge of the geological conditions, it can recommend incremental control changes toward better efficiency, leading the operator step-by-step toward a better set of control parameters.
The proposed framework comprises three steps. In the first step, the needed data sources, past TBM operational data and geological condition data, are pre-processed and cleansed. The researchers also propose an optimality score to evaluate TBM operation performance, taking into account the advance rate and the working pressure safety. The optimality scores provide an information to the operators on how optimal their current set of control parameters is with respect to the current geological conditions. In the second step, a deep learning model learns from historic data a mapping between geological conditions, control parameters, and the optimality score. It then recommends incremental control parameter adjustments in order to improve the optimality score. In the third step, the researchers propose a credibility score that evaluates the confidence in the model recommendation and helps the decision makers to develop trust in the algorithm. This process analyses the performance of the model on historic data in the neighborhood of the current conditions. Absence of similar past conditions or poor performance will lead to a low credibility score while proven performance on similar past data will rise the recommendations’ credibility. A dashboard can finally present the recommendations and their credibility measures to the operators.
The proposed approach is evaluated on a real micro-tunneling project. The case study comprised a project of six micro-tunnels (with a small diameter of about half a meter). The six micro-tunnels had a length of 688 meters each. The researchers collaborated with Herrenknecht who provided the data and the domain expertise for this project and also with the rock mechanics, engineering geology, and underground construction group at MIT.
The proposed framework demonstrates great promise for future applications.
An article of our work was also published on our department website, click here to read
Der Lehrstuhl für Intelligente Instandhaltungssysteme fokussiert sich auf die Entwicklung intelligenter Algorithmen, um die Leistung, Zuverlässigkeit und Verfügbarkeit komplexer Industrieanlagen zu verbessern und die Instandhaltung kosteneffizienter zu gestalten. Weitere Informationen finden Sie auf unserer Webseite