Deep Generative Networks for Nondestructive Cylinder Liner Inspection in Large Internal Combustion Engines

Digitalization offers a variety of promising tools for improving large internal combustion engine technology. This also includes the inspection of important engine components such as cylinder liners. Modern concepts for condition monitoring of the inner surfaces of cylinder liners are often based on indirect methods such as lubricating oil or vibration condition monitoring. However, a position-based inspection of roughness and lubrication properties of the liner surface is currently not possible during operation, nor is it feasible during engine standstill. For large engines in particular, the evaluation of surface properties currently requires disassembly and cutting of the inspected liner, followed by a sophisticated microscopic surface depth measurement. Although this process provides a high-resolution three-dimensional surface model, such measurement methods are destructive and costly. The goal of the research presented here is to develop a simpler and nondestructive method for generating reasonable 3D models of the inner surfaces of cylinder liners in large engines for stationary power generation. A deep learning framework is proposed that allows prediction of surface texture depth from RGB images that can be collected with a handheld microscope. The proposed method is trained on a self-built database of liner surfaces that contains over 2400 RGB images and 1200 depth measurements from 190 cylinder liners with a representative variance of accumulated operating hours taken from large gas engines. The use of convolutional neural networks and adversarial learning techniques makes possible the reliable prediction of surface texture depth in the micrometer range. These textures are comprehensively evaluated using standard measures of surface condition. In addition, recent advances in the field of uncertainty quantification are deployed to detect quality deficiencies in the input images in advance, thereby strengthening the reliability of the approach. The observed success of the proposed method suggests its great potential for nondestructive liner inspection during engine servicing.

• Record URL:
  • https://doi.org/10.4271/2023-01-0066
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/01487191
• Supplemental Notes:
  • Abstract reprinted with permission of SAE International.
• Authors:
  • Angermann, Christoph
  • Laubichler, Christian
  • Kiesling, Constantin
  • Dreier, Florian
  • Haltmeier, Markus
  • Jonsson, Steinbjörn
• Conference:
  • WCX SAE World Congress Experience
  • Location: Detroit Michigan, United States
  • Date: 2023-4-18 to 2023-4-20
• Publication Date: 2023-4-11

Language

• English

Media Info

• Media Type: Web
• Features: References;
• Pagination: pp 841-852
• Serial:
  • SAE Technical Paper
  • Volume: 6
  • Issue Number: 2
  • Publisher: Society of Automotive Engineers (SAE)
  • ISSN: 0148-7191
  • EISSN: 2688-3627
  • Serial URL: http://papers.sae.org/

Subject/Index Terms

• TRT Terms: Engine components; Engine cylinders; Gasoline engines; Lubricating oils; Machine learning; Neural networks; Properties of materials; Scale models; Tribology
• Subject Areas: Highways; Vehicles and Equipment;

Filing Info

• Accession Number: 01879563
• Record Type: Publication
• Source Agency: SAE International
• Report/Paper Numbers: 2023-01-0066
• Files: TRIS, SAE
• Created Date: Apr 19 2023 4:34PM