CLOI-NET: Class segmentation of industrial facilities’ point cloud datasets

Agapaki, Eva; Brilakis, Ioannis

doi:10.1016/j.aei.2020.101121

Cited by 46 publications

(24 citation statements)

References 131 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The flexible, modular handling of the object recognition framework must also be provided [16]. The most manual work is expected on the segmentation of complex scenes which promises the highest rationalization potential, although no support from public directories can be expected [17]. A successful segmentation anticipated a reduction in data volume of point clouds with a factor of 1.000 and more between the rough point cloud and the segmented point cloud focused on segments for further processing can be achieved.…”

Section: Process Definitionmentioning

confidence: 99%

Automated Generation of a Digital Twin of a Process Plant by Using 3D Scan and Artificial Intelligence

Grau¹,

Korol²,

Lützenberger³

et al. 2021

Advances in Transdisciplinary Engineering

View full text Add to dashboard Cite

The simulation of production processes using a Digital Twin is a promising tool for predictive planning, analysis of existing systems or process-parallel monitoring. In the process industry, the concept of Digital Twin provides significant support for process optimization. The generation of the Digital Twin of an already existing plant is a major challenge – in particular for small and medium-sized enterprises. In this sense, the twinning of the existing physical environment has got a particular importance due to high effort. Shape segmentation from unstructured (e.g. point cloud data) is a core step of the digital twinning process for industrial facilities. This is an inherent issue of Product Lifecycle Management how to acquire data of existing goods. The practice of Digital Twin is described based on object recognition by using methods of Machine Learning. The exploration of the pipeline semantics presents a particular challenge. The highly automated procedure for the generation of Digital Twin is described based on a use case of a biogas plant. Commercial deployment, pitfalls, drawbacks and potential for further developments are further explored.

show abstract

Section: Process Definitionmentioning

confidence: 99%

Automated Generation of a Digital Twin of a Process Plant by Using 3D Scan and Artificial Intelligence

Grau¹,

Korol²,

Lützenberger³

et al. 2021

Advances in Transdisciplinary Engineering

View full text Add to dashboard Cite

show abstract

“…We showed in our previous work (Agapaki and Brilakis 2020;) that geometric modeling currently consists of three main steps: (a) primitive shape detection, (b) semantic classification of detected shapes and (c) fitting. We evaluated in (Agapaki 3 Agapaki, December 29, 2020 et al.…”

Section: Introductionmentioning

confidence: 99%

“…However, taking an example of a small petrochemical plant with 240,687 objects and 53,834 pipes, 2,382 manual labor hours are still needed to model these cylinders ). In summary, we have shown in (Agapaki and Brilakis 2020;Agapaki and Nahangi 2020) that the state-of-the-art 3D modeling practice has three main limitations: (a) the modelers fit standardized structural steel shapes after segmenting the structural elements manually or roughly selecting regions of interest using clipping polygons, (b) the modelers define parameters which determine cylinder detection and (c) EdgeWise does not directly assign class or instance labels per point. Rather it fits 3D solid standardized shapes and from them instance labels can be inferred.…”

Section: Introductionmentioning

confidence: 99%

“…We argue that cost reduction of gDT generation will be achieved by automating the following steps: (a) class segmentation, (b) instance segmentation and (c) fitting. We addressed class segmentation with our CLOI-NET framework estimating 70% of labor hour savings in that task 4 Agapaki, December 29, 2020 (Agapaki and Brilakis 2020). Point cloud fitting to other geometric representations is a solved problem as proved by .…”

Section: Introductionmentioning

confidence: 99%

“…We first elaborate on the state-of-research instance segmentation methods in the background section, which is followed by our proposed methodology. We then evaluate our methodology on a new point cloud dataset named "CLOI" that was introduced in our previous work (Agapaki and Brilakis 2020). The results, achievements and limitations of our method are extensively discussed, and the final section follows with the conclusions, contributions and future directions for research.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Instance Segmentation of Industrial Point Cloud Data

Agapaki¹,

Brilakis²

2020

Preprint

Self Cite

View full text Add to dashboard Cite

The challenge that this paper addresses is how to efficiently minimize the cost and manual labour for automatically generating object oriented geometric Digital Twins (gDTs) of industrial facilities, so that the benefits provide even more value compared to the initial investment to generate these models. Our previous work achieved the current state-of-the-art class segmentation performance (75% average accuracy per point and average AUC 90% in the CLOI dataset classes) as presented in (Agapaki and Brilakis 2020) and directly produces labelled point clusters of the most important to model objects (CLOI classes) from laser scanned industrial data. CLOI stands for Cshapes, L-shapes, O-shapes, I-shapes and their combinations. However, the problem of automated segmentation of individual instances that can then be used to fit geometric shapes remains unsolved.We argue that the use of instance segmentation algorithms has the theoretical potential to provide the output needed for the generation of gDTs. We solve instance segmentation in this paper through (a) using a CLOI-Instance graph connectivity algorithm that segments the point clusters of an object class into instances and (b) boundary segmentation of points that improves step (a). Our method was tested on the CLOI benchmark dataset (Agapaki et al. 2019) and segmented instances with 76.25% average precision and 70% average recall per point among all classes. This proved that it is the first to automatically segment industrial point cloud shapes with no prior knowledge other than the class point label and is the bedrock for efficient gDT generation in cluttered industrial point clouds.

show abstract