Index
Fetal Re-Identification in Multiple Pregnancy Ultrasound Images Using Deep Learning
Natural Language Text Generation for Symbolic Descriptions Using Language Models
In today’s automated robotics industry, there is a growing need for efficient and automated methods for generating text descriptions for actuator & sensor variables, functions, and properties, which are collectively referred to as symbolic descriptions hereby. This is because symbolic descriptions are often used to document the functionality of robotic systems, model the functionality of the robots, and communicate the functionality of robotic systems to human operators. The current manual process of generating text descriptions for symbolic descriptions is time-consuming, labor-intensive, and inconsistent.
This research proposes to develop an automated text generation system for symbolic descriptions in robotics using open-source pre-trained language models like GPT-2 [1], LLaMA [2], MPT etc. The decision to finalize the model will be based on factors including, but not limited to, performance, suitability for the downstream task, cost of training and inference, and interpretability of the results produced by the models. The implemented model will be able to generate text descriptions in English and German and preserve the structure of the original text descriptions. The system will be implemented using a fine-tuning approach and trained on a dataset of symbolic descriptions and their corresponding text descriptions.
The expected outcomes of this research are:
- Fine-tune a language model for automated text generation for symbolic descriptions in robotics.
- Demonstrating its efficiency in generating accurate and contextually relevant text descriptions. Evaluating and analyzing the model’s performance using perplexity, ROUGE, and BLEU scores.
- A conceptual lifecycle consideration of the training pipeline, highlighting scalability, maintenance, adaptability, and reusability aspects.
- One of the constraints of this research is to interpret the reasoning behind the model. This is an active research field, with some established techniques like attention visualization [3], feature attribution [4], and potentially, Weights and Bias could be used for this purpose.
The proposed system is still under development, and several future work directions could be explored. These include:
- Expanding the system to support other languages.
- Improving the system’s ability to generate text descriptions that are consistent with the original text descriptions.
This research has the potential to significantly contribute to the field of automated text generation for robotics. The proposed system has the potential to be used in a variety of applications, such as documentation, modeling functionality, and communication. The research results will be of interest to the research community and practitioners in these domains.
References:
[1] Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language Models are Unsupervised Multitask Learners.
[2] Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., Rodriguez, A., Joulin, A., Grave, E., & Lample, G. (2023). LLaMA: Open and Efficient Foundation Language Models. ArXiv, abs/2302.13971.
[3] Yeh, C., Chen, Y., Wu, A., Chen, C., Vi’egas, F., & Wattenberg, M. (2023). AttentionViz: A Global View of Transformer Attention. ArXiv, abs/2305.03210.
[4] Zhou, Y., Booth, S., Ribeiro, M., & Shah, J.A. (2021). Do Feature Attribution Methods Correctly Attribute Features? ArXiv, abs/2104.14403.
AI for IT Operations (AIOps): Leveraging Large Language Models as Support for Process Management in Mainframes
AI-based generation of 2D vehicle geometries through Natural Language
Word Embeddings Applied to Alzheimer’s Disease
Development of a deep learning approach to detect faulty axial bearing components after assembly using acoustic signals
Bearings are used to support and guide relatively movable parts, especially those capable of rotational movement, in machines and devices, and to absorb and transmit external forces to a housing or other components [1]. In general, bearings are clustered into radial and axial bearings. Axial bearings have a disc-shaped flat design, which is particularly advantageous for pure axial loads, not radial loads. Axial bearings with needle-shaped rolling elements have a compact size and feature smooth operations for applications with high speeds and forces either too high for an axial ball bearing or too low for an axial roller bearing [2]. Axial needle bearings are primarily used in vehicle transmissions and allow a smooth and precise rotation of axles and shafts with axial impact. Axial needle bearings come (1) with an inner or an outer washer and a cage with rolling elements and (2) with both inner and outer washers and a cage with rolling elements. In the serial production line of axial needle bearings, these two or three individual components are assembled automatically in a rotary indexing machine.
Due to the geometry of the components the bearing is joined together by a rotational motion during the assembly. This can lead to deformation of one or more components, which hinders the freedom of movement of the components (clearance). In addition to deformed parts, components within the end of the tolerance limits can also cause problems with the clearance of the bearing. Currently, the clearance is tested using a “friction torque test station,” where the counter torque occurring during rotation of the bearing is measured with a torque sensor. Due to the lack of an established test procedure concerning the calibration of the testing station and insufficient resolution of the sensor for the measurement task, pseudo-defective goods are detected and ditched.
In this thesis, a new testing approach is considered to use deep learning methods for the classification of sound and defective axial needle bearings concerning the clearance of the components and other errors, like pressure marks, that cause acoustic abnormalities using acoustic or vibration data during the rotary motion of the bearing. Therefore, the signal can be converted to an image representation which can be processed by a convolutional neural network (CNN) architecture. Suitable pre-trained CNN architectures could be GoogleLeNet, ResNet-50, SqueezeNet, and Inception-Resnet v2, which should also be considered for faster development of a working neural network. By transfer learning, it is possible to use a pre-trained model that is already well suited for 2-D images and retrain it on a new target. The recorded signal data of the axial needle bearing during rotation can be converted to a 2-D time-frequency image, from which the CNN can automatically detect patterns to decide regarding the classification [3].
The main goal of this work is to implement a testing method using acoustic signals for detecting assembly errors in axial needle-bearing components. Different deep learning methods will be implemented and evaluated to achieve a high-performing, reliable, low cost and low-maintenance testing station. To achieve this goal, the following steps will be performed:
- Selection of an adequate test procedure concerning movement and signal recording regarding the clearance and noise development of the axial needle bearing (e.g., rotational movement of the components and proper acoustic or vibration recording)
- Selection of proper validation requirements regarding the deep learning application.
- Obtain a database with acoustic signals of assembled axial bearing components.
- Preprocess and label the samples in the dataset.
- Development, training, optimization, and comparison of different deep learning models and architectures for clearance testing of the axial needle-bearing components after the assembly.
- Evaluation of the applicability of the deep learning testing method in serial production.
Literature
- H. Wittel, D. Jannasch, J. Voßiek, C. Spura. „Roloff/Matek Maschinenelemente“. 23. Auflage. Wiesbaden, Deutschland: Springer Vieweg, 2017.
- Schaeffler. „Axial-Nadellager“. Schaeffler. Viewed 03. July 2023: https://medias.schaeffler.de/de/axial-needle-roller-bearings.
- B. U. Deveci, M. Celtikoglu, O. Albayrak, P. Unal, P. Kirci. “Transfer Learning Enable Bearing Fault Detection Methods Based on Image Representations of Single-Dimensional Signals”. In: Information Systems Frontiers, 2023. https://doi.org/10.1007/s10796-023-10371-z.
Photovoltaic Plant Inspection: Identifying Modules and their Defects in Electroluminesence Imagery
Deep learning-based detection of early biomarker in age-related macular degeneration in volume-merged high resolution optical coherence tomography
Eye diseases such as age-related macular degeneration (AMD) can cause significant visual impairment
and even vision loss, impacting patients’ quality of life. Early detection and diagnosis are crucial for
improving treatment outcomes. Optical coherence tomography (OCT) has become a vital medical
imaging technique for this purpose.
A promising biomarker for early AMD detection is the presence of basal laminar and basal linear
deposits in the outer retina1. However, conventional OCT devices often fail to detect these changes,
making them visible only in advanced stages or histological examinations.
In this thesis, an OCT prototype, which enables ultra-high-resolution imaging, is utilized to visualize
these deposits2. Moreover, motion correction and volume merging techniques were used to generate
consistent high-quality volumetric OCT data3. For accurate segmentation and thickness measurement
of basal laminar and basal linear deposits, deep learning algorithms will be investigated to provide a
data-driven automated approach with particular focus on the use of volumetric data and high-precision
results. For this purpose, a literature review of state-of-the-art approaches for retinal layer segmentation
will be conducted, followed by the development, training, and evaluation of a new problem-specific
designed deep learning approach and a comparison to existing baseline methods. A workflow utilizing
a high-performance computing cluster will be designed for efficient data-driven training.
The developed approach should be a step towards correlation analysis between the thickness and characteristics
of these deposits and the early stages of AMD solely using OCT images. A difficulty is the
particularly small size of this structure and its varying visibility, so that these characteristics must be
taken into account when working out the concept of the approach. The thesis aims to fully automate
the analysis of deposit thickness and visibility. Ultimately, such analysis could lead to better understanding
of the pathogenesis in AMD and in the long run improve treatment outcomes for patients.
1Sura, A. A., Chen, L., Messinger, J. D., Swain, T. A., McGwin, G., Freund, K. B., & Curcio, C. A. (2020). Measuring
the contributions of basal laminar deposit and Bruch’s membrane in age-related macular degeneration. Investigative
ophthalmology & visual science, 61(13), 19.
2Chen, S., Abu-Qamar, O., Kar, D., Messinger, J. D., Hwang, Y., Moult, E. M., … & Fujimoto, J. G. (2023).
Ultrahigh Resolution OCT Markers of Normal Aging and Early Age-related Macular Degeneration. Ophthalmology
Science, 3(3), 100277.
3Ploner, S., Chen, S., Won, J., Husvogt, L., Breininger, K., Schottenhamml, J., … & Maier, A. (2022, September). A
spatiotemporal model for precise and efficient fully-automatic 3D motion correction in OCT. In International Conference
on Medical Image Computing and Computer-Assisted Intervention (pp. 517-527). Cham: Springer Nature Switzerland.