The RIMcomb research project: Towards the application of building information modeling in Railway Equipment Engineering

This paper presents our research towards the utilization of Building Information Modeling (BIM) methods in Railway Equipment Engineering. While BIM is already applied in several railway infrastructure construction projects, so far, it is mainly used for construction engineering tasks, i.e. for modeling the alignment, bridges or tunnels. The RIMcomb research project targets the further application in scope of rail engineering subsections such as control and safety systems, train control systems, telecommunication systems, electric power systems, rail power supply or cable management. In this paper, we are going to give a general overview of the project and will discuss first findings of our approach to digitizing conventional technical drawings in order to translate their content into a machine-readable form. representing railway equipment. The paper ends with a summary and an outlook. 2 THE RIMCOMB PROJECT The research project “RIMcomb: Railway Information Modeling for the Equipment of Railway Infrastructure” was initiated by SIGNON Deutschland GmbH in 2016 in cooperation with Technical University of Munich and AEC3 Deutschland GmbH. The project is funded by the Bavarian Research Foundation and started in early 2017. The main focus of the research project is to develop and adapt new computer-supported methods for model-based collaboration between the different subsections of technical railway equipment in order to increase the efficiency of the planning process and the quality of the outcome. During the design, planning and construction of railway infrastructure and technical equipment, a multitude of domains experts are involved. Therefore, data exchange between these participants is an issue that needs to be addressed, as there a various specialized software tools available for different tasks that do not implement any common data standard. Also, most of railway construction projects involve the modernization or alteration of existing infrastructure and thus the industry has to rely on technical drawings from past decades that are not necessarily consistent with the real-world circumstances. As described in Section 2 we developed a method that allows the automatic recognition of plan symbols in technical drawings of railway infrastructure. Besides the use of this data described in Section 3, we also aim at comparing the generated data with realworld data in order to identify discrepancies. This use case may create a significant benefit for the railway companies as the manual comparison of the as-built drawings with real-world stock data requires a considerable amount of effort but is nonetheless necessary. Here, machine-learning and convolutional neural networks (as outlined in Section 2) will also be employed to process video files of railway tracks in order to identify objects in single frames for the mapping of objects such as signals, balises, switches or poles of overhead lines. Another aspect in the scope of the research project is the development of a method that allows the automated checking of technical rules and regulations. As such approaches have already been applied outside of the infrastructure domain (Preidel and Borrmann, 2016), we see a huge benefit in introducing them into the domain of railway equipment, as the amount of rules and regulations in this domain requires a large amount of manual work. 3 DITIALIZATION OF PLAN DATA One major topic in the research project is the digitalization of conventional drawings depicting railway equipment infrastructure. While most drawings are available digitally, the interpretation of these plans has to be undertaken manually. This is necessary when the accuracy of plans has to be compared to real-world circumstances or in case of stocktaking. Our approach aims at supporting this process in order to reduce the manual effort by automating at least parts of this image interpretation process. The first step towards this goal is the automatic recognition and highlighting of plan symbols on a given drawing and the subsequent storing of their count and location. 3.1 Theoretical background In a first step, three preexisting methods of image recognition are described. We evaluated those techniques in respect to the given problem. As none of those methods matched our requirements completely we also tested Convolutional Neural Networks, which are already widely used for image recognition, in respect of their ability to detect plan symbols. 3.1.1 Template Matching Template Matching is a well-known method for the searching of a template image in a larger image. This is made by sliding the template image over the input (larger) image and comparing them at every position. The result of this method is a grayscale image with a size of (W-w+1, H-h+1), where W and H are the width and the height of the input image, w and h are the width and the height of the template image. We investigated different comparison methods for each one of which there is a normalized version (Kaehler and Bradski, 2016). In this work two of these methods are used: Normalized Square Difference Matching Method: RR(xx,yy) = ∑ �TT(xx′,yy′) − II(xx + xx′,yy + xx′)� xx′,yy′ �∑ TT(xx′,yy′)2 ∙ ∑ II(xx + xx′,yy + yy′)2 xx′,yy′ xx′,yy′ Normalized Correlation Coefficient Matching Method: RR(xx,yy) = ∑ �TT′(xx′,yy′) ∙ II′(xx + xx′,yy + xx′)� xx′,yy′ �∑ TT′(xx′,yy′)2 ∙ ∑ II′(xx + xx′,yy + yy′)2 xx′,yy′ xx′,yy′ Here T is the template image, I the input image, R the result image and TT′(xx′,yy′) = TT(xx′,yy′) − 1/(ww ∙ h) ∙ � TT(xx′′,yy′′)