In the manufacturing industry, automated optical inspection aims to improve the detection and classification of anomalies by utilizing artificial intelligence and computer vision techniques to enhance quality control processes and minimize production defects. However, this automated system faces significant challenges, particularly regarding the detection of anomalies due to predominance of normal instances over defected ones. Addressing this imbalance is crucial for effective real-time anomaly detection particularly in images captured by Airbag Sensors among other automotive parts. Earlier contributions in domain-specific fields commonly relied on traditional computer vision methods, while recent systems are increasingly using deep learning techniques. Utilizing various data augmentation techniques ensures a more balanced representation of anomalies in the dataset, thereby enhancing the accuracy of the detection process. Moreover, it also enhances the robustness and generalization of the anomaly detection model by exposing it to a more diverse range of instances during training. Such work has not been carried out to augment Airbag Sensor images for analysis through a deep learner. Accordingly, this paper introduces a framework that employs data augmentation techniques for Convolutional Neural Networks (CNNs). The proposed system, based on data augmentation and CNN, significantly improves the performance for anomaly detection in Airbag Sensor images with a classification accuracy on the unaugmented dataset being 53 % which improves to 90% with augmentation.

Reusing and managing existing knowledge from available documents is crucial for success in the factory planning domain. By leveraging Artificial Intelligence (AI) and Question Answering (QA) systems, users can query a document corpus through a chat-based application and receive precise answers. The recent advancements in Large Language Models (LLMs) and their linguistic capabilities present new opportunities for such applications. Utilizing the methodology of Retrieval Augmented Generation (RAG), document sections are provided to the LLM based on user queries. However, existing RAG implementations that use vector databases as document repositories face limitations when answering questions that extend beyond the text content of the documents. To address this issue, this paper proposes a concept to enhance RAG systems by integrating a Knowledge Graph (KG) constructed from the document structures.

This paper introduces a novel methodological approach to transform a traditional model-centric machine learning pipeline into a morphological box. Utilizing a taxonomy development method, we iteratively refine a morphological box to address the complexity inherent in selecting and adjusting components within machine learning pipelines. Our method leverages a generic active learning process tailored for quality control in manufacturing, serving as a practical example. We demonstrate that decomposing the machine learning pipeline into distinct morphological box dimensions with meta char-acteristics significantly enhances decision-making clarity by reducing option complexity. This transformation is further supported by defining universal attributes-Cost, Time, Avail-ability, and Complexity-that cater to users with varying machine learning expertise. Future work will focus on empirical validation and the development of software tools to facilitate the practical application of morphological boxes in diverse machine learning pipeline contexts.

The presence of class imbalance, denoting a dis-proportionate distribution of class instances in a dataset, has emerged as a significant challenge in the era of Deep Learning (DL) where models crave abundance in data. This issue is pervasive in various real-world applications, where certain classes exhibit limited data representation. This problem is frequently encountered when dealing with image data, which exhibits an imbalanced distribution, with one class significantly outnumbering the others. Failing to address class imbalance introduces bias in machine learning and deep learning models, favoring the majority classes and leading to subpar performance for the minority classes. This research specifically delves into the recurrent problem in the context of image data, that is “class imbalance”. The research comprehensively explores the existing challenges based on data pre-processing, algorithmic techniques, hybrid methodologies, and state-of-the-art solutions.

Virtual commissioning is evolving to become an indispensable part included in modern planning processes. New challenges related to recent technologies like the usage of information models in Industry 4.0 implementations make it necessary to test the behaviour in production-related communication beforehand. Different data aquisition methods have to be evaluated concerning their benefits, risks and the environment it has to be implemented in. Information model management and aggregation can help to fulfill these tasks. This concept paper focuses on architectural challenges related to the implementation of information models in brown- as well as green-field approaches and the overall management system behind.

Information models are defined as a comprehensive semantic description of data within a production system. These systems underlay a constant change. Therefore, the corresponding information models are also subject to continuous evolution. This paper shows approaches for the versioning compliant design of information models and their implementation as well as support strategies to assist such changes.

The objective of this paper is to present a hybrid methodology of analysing acoustic signals arising in industrial processes through comparisons of known numerical techniques such as clustering. Apart from data acquisition and pre-processing, the other essential component of using acoustics is to design an analysis methodology, that can culminate in practical applications. This paper applies Gaussian Mixture Models and Self-Organising Maps to cluster pre-processed AE hits obtained from acoustic sensors in the form of tensile, shear and mixed modes of compression on a material. For an in-depth analysis, custom features such as high peak regions, low peak regions, strongly hit and weakly hit signals are introduced to compare with the clusters formed. The results show that for small AE signals that are obtained or extracted after events detection, a time-domain based clustering can be applied and used for isolating similarities and distinctions among the signals belonging to the same group.

Information exchange across different departments in factories must be structured and verified for effective production processes. However, any such exchange across networks to clients must be compliant to company standards so that uniformity and security in data dispensation is maintained. Open Platform Communications Unified Architecture (OPC UA) solutions provide users with the flexibility of discharging information through network based protocols. This paper aims to build a compliance testing methodology for external vendors of an organisation who are interconnected with the OPC UA Server-Client Protocols. The structure is built on a host of UA clients that test UA specifications and information models from these vendors. This enables an organisational entity, for example, a manufacturer, to provide its clients, in this case external vendors such as suppliers, with the flexibility of conforming to prescribed standards in a testing phase on an automated digital platform rather than with older methods such as data transfer though databases or documents. An implementation of this approach is presented using example servers created for this purpose. It follows that external vendors can perform compliance testing and possibly improve their standards to match the company standards through remote testing after proper authentication measures provided by OPC UA.

Virtual Commissioning is an important part of modern design engineering approaches. Early simulation models of newly developed production structures can help to elaborate and evaluate different ideas concerning new processes. The machine communication is an important part of this evaluation. Currently the focus is on different approaches like CPS, IIoT and the overall crosslinking in Industry 4.0 implementations. In many of these cases the functional static behaviour of the communication interface is checked in later phases of the development process and not in a dynamical aspect, although it is useful to understand it early to draw conclusions for the proceeding project. This paper shows an early simple approach for checking the dynamical interoperability between the different automation levels in a production environment by adapting known Virtual Commissioning methods to a broader view.

A production system consists of multiple production entities. To enable a manufacturing process, these entities have to exchange information. Information models offer the possibility to standardize the data exchange between the entities. In addition to a common communication protocol, concepts that integrate and aggregate the different entities into a common address space, have to be developed. Another important issue is the historicization of information, for example, to improve the production process or for legal reasons. Graph databases enable semantic relations between objects for representation and storage of information. Therefore, it is a promising concept to be used in the production context. This paper presents a concept for the aggregation and historicization of production entities into a graph database.

Processes such as welding involve consumption of huge amounts of energy leading to generation of significant electrical data consisting of current and voltage signals. The added task is to inspect the quality of welding using such data as early as possible to identify defects in producing welded parts or equipment. From the perspective of machine learning, this paper presents a data mining approach to analyse small sampled amounts of electrical signals to identify welding inconsistencies using conventional methods such as clustering algorithms, time-series and multi-label classifiers. Using unlabelled and discrete signals, an attempt is made to build a process profile on the welding robots with the use of comparison measures such as Jaccard's metric. To monitor such a mechanism, a simulation application based on IIoT standard Open Platform Communication (OPC UA) is developed to present the analysis over secure network servers to clients. The application setup presents a basic monitoring system for welding processes using available technologies like machine learning algorithms and OPC UA.