Simon Kamm

Abstract

In many application domains data from different sources are increasingly available to thoroughly monitor and describe a system or device. Especially within the industrial automation domain, heterogeneous data and its analysis gain a lot of attention from research and industry, since it has the potential to improve or enable tasks like diagnostics, predictive maintenance, and condition monitoring. For data analysis, machine learning based approaches are mostly used in recent literature, as these algorithms allow us to learn complex correlations within the data. To analyze even heterogeneous data and gain benefits from it in an application, data from different sources need to be integrated, stored, and managed to apply machine learning algorithms. In a setting with heterogeneous data sources, the analysis algorithms should also be able to handle data source failures or newly added data sources. In addition, existing knowledge should be used to improve the machine learning based analysis or its training process. To find existing approaches for the machine learning based analysis of heterogeneous data in the industrial automation domain, this paper presents the result of a systematic literature review. The publications were reviewed, evaluated, and discussed concerning five requirements that are derived in this paper. We identified promising solutions and approaches and outlined open research challenges, which are not yet covered sufficiently in the literature.

Abstract

With the increasing amount of available and connected data sources, industrial automation applications such as condition monitoring of a production machine can be improved by considering various data. To gain insights from this data and make it useable, heterogeneous data has to be analyzed intensively. Limited machine learning approaches exist in industrial automation and manufacturing for analyzing data acquired from multiple sources. In this paper, first, a suitable concept for handling heterogeneous data from integration to analysis is presented as well as a multi-layer architecture for the concept’s realization. The architecture encapsulates functionalities into the different layers and allows easy extendability and modifiability. Afterwards, a context modeling approach for managing heterogeneous data and existing approaches and algorithms for analyzing this data robustly and dynamically analyzing it are presented.

Abstract

The demand for reconfigurations of production systems is increasing, driven by shorter innovation and product life cycles and economic volatility. Another trend in the domain of industrial automation is the emergence of cyber-physical production systems, which offer promising potentials, for example, self-organization capabilities. A suitable cyber-physical production system architecture that incorporates knowledge modelling and management concerns, plus a reconfiguration management methodology, is crucial for realizing self-organized reconfiguration management. In this paper, first reference architectures, architectural patterns, and basic principles, as well as knowledge modelling and management approaches, are discussed in general. Afterwards, these are examined concerning the reconfiguration management use case focusing on UML/XML-based and ontology-based approaches. A novel approach comprising a multi-agent system and the MAPE-K concept for reconfiguration management is presented. In addition, the approach contains a service-oriented architecture for a deterministic plant control within a layered architecture. The knowledge modelling is realized through a UML information model, which can be integrated into the system utilizing XML files. Furthermore, the provided tool support is described. It enables a user to describe system components in an effort-reduced manner and conform to the schema defined by the information model and its restrictions via a GUI.

Abstract

This paper presents an approach to combine multiple temperature-sensitive electrical parameters to improve the accuracy and precision of the junction temperature estimation of power transistors using the example of a silicon-carbide power MOSFET. Switching delays and the threshold voltage of the power transistor during turn-on and -off of a silicon-carbide power transistor are used as temperature-sensitive electrical parameters for the online junction temperature measurements. In order to improve the accuracy, a shallow fully-connected neural network is used as the means to combine the four measurements in one switching cycle of the transistor. The maximum measurement error of the junction temperature of the power transistor is reduced approximately 10-fold from 8.98 K to 0.92 K.

Abstract

Current industrial automation systems are facing increasing dynamics. Thus, acquiring and managing heterogeneous data within the Digital Twin to enable decision making is necessary although challenging. Knowledge Graphs unify and relate data, enabling the derivation of new insights. In this contribution, an approach for context-enriched modeling of cyber-physical production systems is proposed, in order to realize a Knowledge Graph enhanced intelligent Digital Twin further considering the context. Therefore, the modeling approach of the Knowledge Graph considers context and serves as a base for the graph embeddings to gain further knowledge about the production system. This knowledge is used within the architecture of the intelligent Digital Twin. The resulting benefits, i.e. diverse manifestations of an improved decision making, are highlighted by the use case of self-organized reconfiguration management.

Abstract

Failure analysis is essential for improving the reliability and manufacturability of electronic devices. With the time-domain reflectometry method, failures can be analyzed non-destructively. The method enables the detection, location, and characterization of hard interconnection failures (open or shorts) as well as of soft interconnection failures, which can give an outlook on imminent hard failures. Generating measurement data from real failed devices is costly since failed devices need to be selected and the measurements need to be performed and prepared. In contrast, simulation models are often available where all possible kinds of failures can be created. Therefore, we propose simulation to real transfer learning for the failure analysis on time-domain reflectometry data. A deep learning model shall first be trained on time-domain reflectometry simulation data and then be transferred to measurement data of a power transistor. We investigate different possibilities of transfer and evaluate the performance of a SiC power transistor.

Abstract

Electronic devices are one of the key factors for recent advances in smart production systems or automotive. Reliability and robustness are key issues. To further increase this reliability, occurring failures in an electronic device has to be investigated in post-production failure analysis processes. One recent technique to detect and locate failures in electronic components is Time-Domain Reflectometry. This method offers the chance to detect several kinds of failures (e.g. a hard or soft failure) and localize the failure nondestructively. In theory, this can be determined following defined physical formulas. Nevertheless, the received signals are not perfect and mixed with noise from the measurement device or disturbed by nonoptimal material properties. In addition, complex architectures of devices are hard to model based on analytical models. Thus, these models solely are not sufficient for the failure analysis process. For this reason, a hybrid modeling approach is proposed, using a Machine Learning model in combination with physical models to detect and characterize the failure and its exact position. The Machine Learning model will be trained with simulated Time-Domain Reflectometry data.

Abstract

Im Sinne der Qualitätssicherung sollen zukünftig produzierte Halbleiterbauelemente den steigenden Anforderungen an Zuverlässigkeit, Konnektivität und Automatisierung standhalten. Dazu ist ein Fehleranalyseprozess notwendig. Um die Effizienz zu steigern und die Fehleranfälligkeit zu reduzieren, soll der Fehleranalyseprozesses mit dazugehöriger Datenanalyse automatisiert werden. Dafür muss zunächst ein Konzept für die automatisierte Integration heterogener Daten aus unterschiedlichen Datenquellen auf Basis aktueller Daten Engineering Methoden entwickelt werden. Die Daten aus unterschiedlichen Quellen (z.B. Produktionsdaten, Datenblätter, Analyseverfahren) müssen eindeutig einem Bauelement zugeordnet werden können und neue Daten oder Analyseergebnisse diesem hinzugefügt werden können. In diesem Beitrag wird für die automatisierte Datenintegration des Fehleranalyseprozesses von Halbleiterbauelementen ein Konzept mithilfe von Ontologien und Graphen vorgestellt und dessen Mehrwerte anhand eines Anwendungsfalls mit einer prototypischen Umsetzung diskutiert. Dafür wird ein hybrider Ontologie-Ansatz genutzt, um die heterogenen Daten in einem Knowledge Graphen zu verknüpfen. Der vorgestellte Ansatz ermöglicht sowohl eine lokale als auch eine globale Datenanalyse. Dadurch können sowohl bisher entwickelte Datenanalyse-Ansätze genutzt werden, als euch neue Algorithmen aufgebaut werden.

Abstract

The utilization of deep learning in the field of industrial automation is hindered by two factors: The amount and diversity of training data needed as well as the need to continuously retrain as the use case changes over time. Both problems can be addressed by industrial deep transfer learning allowing for the performant, continuous and potentially distributed training on small, dispersed datasets. As a specific example, a dual memory algorithm for computer vision problems is developed and evaluated. It shows the potential for state-of-the-art performance while being trained only on fractions of the complete ImageNet dataset at multiple locations at once.

Abstract

Ensuring and improving the reliability of electronic devices requires post-production failure analysis processes. One of the techniques to perform failure analysis for electronic devices is Time-Domain Reflectometry. With this method, failures can be detected, located, and characterized non-destructively. It enables not only the detection of hard interconnection failures (such as an open or a short) but also the detection and characterization of soft failures. For Time-Domain Reflectometry, known physical equations can be applied to model the ideal behavior of a signal for different failures. However, since electronic device architectures and thus the data become more and more complex and measurements are disturbed by noise or nonoptimal material properties, these models solely are not sufficient for failure analysis. Therefore we propose hybrid modeling, where machine learning models (e.g. convolutional neural networks) work together with physical models to detect, locate, classify and characterize the failure. The first approach is shown on simulated microstrip line data and then transferred to SiC transistors for evaluation.

Abstract

With the rise of the Internet of Things and Industry 4.0, the number of digital devices and their produced data increases tremendously. Due to the heterogeneity of devices, the generated data is mostly heterogeneous and unstructured. This challenges established approaches for knowledge discovery, which typically consume structured data from one source. The paper first describes aspects of data heterogeneity and their relevance for Industry 4.0 systems. Following, the upcoming challenges for different steps inside the knowledge discovery process for Industry 4.0 systems, such as for data integration and data mining, are discussed. Additionally, it mentions approaches to tackle them.

Abstract

Cyber-Physical Systems, characterized by networking capabilities and digital representations, offer many promising potentials for industrial automation. In an attempt to further enrich the system's digital representation by incorporating interdisciplinary models and considering a continuous and synchronized representation of it within the cyber layer, the concept of the Digital Twin emerged, enabling system monitoring, virtual commissioning, failure diagnosis and simulations by managing the Cyber-Physical Systems data along its lifecycle. To add further intelligence into the Digital Twin, the architecture of the intelligent Digital Twin was proposed. Nevertheless, managing and relating the complex and dynamic digital models as well as the heterogeneous data of the intelligent Digital Twin present open challenges. Due to their inherent extensibility and adaptability as well as their semantic expressiveness, Knowledge Graphs are a suitable concept to overcome these challenges and enable reasoning to gain new insights. Prominent applications of Knowledge Graphs are recommendation systems and exploratory search within the semantic web. However, there seems to be a lacking yet potential applicability for Knowledge Graphs in the industrial domain. Therefore, this contribution proposes a Knowledge Graph enhanced architecture of the intelligent Digital Twin, offering capabilities, which are internal linking and referencing, knowledge completion, error detection, collective reasoning and semantic querying. Based on the proposed concept, potential application fields for Knowledge Graph enhanced intelligent Digital Twin are addressed.

Abstract

In power electronic applications, transistors are a vital component. They are, however, susceptible to failures due to degradation of the interconnections and the chip itself. This paper presents a non-destructive approach for failure detection and location in power electronic devices using time-domain reflectometry. The proposed measurement and data generation method is applied to a silicon-carbide power transistor where several characteristics (R, L, C, open, short) and the location of the failure is simulated and characterized. Moreover, the method is also used to find the intrinsic properties of the transistor such as parasitic inductance and capacitance. The data generated is mapped to physical equations, however, the reflected signal of the time-domain reflectometry can be noisy due to multiple discontinuities in the transmission path. Therefore, the simulation and measurement data can be used to train hybrid machine learning models for parameter extraction which automates the failure analysis in Industry4.0 processes to ensure a smart and reliable manufacturing process.

Abstract

In this paper, a novel light-weight incremental class learning algorithm for live image recognition is presented. It features a dual memory architecture and is capable of learning formerly unknown classes as well as conducting its learning across multiple instances at multiple locations without storing any images. In addition to tests on the ImageNet dataset, a prototype based upon a Raspberry Pi and a webcam is used for further evaluation: The proposed algorithm successfully allows for the performant execution of image classification tasks while learning new classes at several sites simultaneously, thereby enabling its application to various industry use cases, e.g. predictive maintenance or self-optimization.