Position innerhalb des Seitenbaumes

Institut für Automatisierungstechnik und Softwaresysteme
Institut
Team
Ding, Sheng

Sheng Ding

M.Sc.

Akademischer Mitarbeiter
Institut für Automatisierungstechnik und Softwaresysteme

Kontakt

+49 711 685 69198
+4971168567302
E-Mail

Pfaffenwaldring 47
70550 Stuttgart
Raum: 3.252

Publikationen

Zeitschriften und Konferenzen:

2021
1. Y. Ma, A. Morozov, und S. Ding, „Anomaly Detection for Cyber Physical Systems using Transformers“, The International Mechanical Engineering Congress and Exposition, IMECE, 2021.
  - Zusammenfassung
  - BibTeX
  Zusammenfassung
  Safety and reliability are two critical factors of modern Cyber-Physical Systems (CPS). However, the increasing structural and behavioral complexity of modern automation systems significantly increases the possibility of system errors and failures, which can easily lead to economic loss or even hazardous events. Anomaly Detection (AD) techniques provide a potential solution to this problem, and conventional methods, e.g., Autoregressive Integrated Moving Average model (ARIMA), are no longer the best choice for anomaly detection for modern complex CPS. Recently, Deep Learning (DL) and Machine Learning (ML) anomaly detection methods became more popular, and numerous practical applications have been presented in many industrial scenarios. Most of the modern DL-based anomaly detection methods use the prediction approach and LSTM architecture. The Transformer is a new neural network architecture that outperforms LSTM in natural language processing. In this paper, we show that the Transformer-based deep learning model, which has received much attention recently, can be applied to the anomaly detection of industrial automation systems. Specifically, we collect time-series data from a system of two industrial robots using a Simulink model. Then, we feed these data into our Transformer-based model and train it to be a time-series data predictor. The paper presents the experimental results that show the comparison of precision and speed of a Long-Short Time Memory (LSTM) predictor and our Transformer-based predictor.
  BibTeX
  @article{ma2021anomaly, abstract = {Safety and reliability are two critical factors of modern Cyber-Physical Systems (CPS). However, the increasing structural and behavioral complexity of modern automation systems significantly increases the possibility of system errors and failures, which can easily lead to economic loss or even hazardous events. Anomaly Detection (AD) techniques provide a potential solution to this problem, and conventional methods, e.g., Autoregressive Integrated Moving Average model (ARIMA), are no longer the best choice for anomaly detection for modern complex CPS. Recently, Deep Learning (DL) and Machine Learning (ML) anomaly detection methods became more popular, and numerous practical applications have been presented in many industrial scenarios. Most of the modern DL-based anomaly detection methods use the prediction approach and LSTM architecture. The Transformer is a new neural network architecture that outperforms LSTM in natural language processing. In this paper, we show that the Transformer-based deep learning model, which has received much attention recently, can be applied to the anomaly detection of industrial automation systems. Specifically, we collect time-series data from a system of two industrial robots using a Simulink model. Then, we feed these data into our Transformer-based model and train it to be a time-series data predictor. The paper presents the experimental results that show the comparison of precision and speed of a Long-Short Time Memory (LSTM) predictor and our Transformer-based predictor.}, author = {Ma, Y. and Morozov, A. and Ding, Sheng}, journal = {The International Mechanical Engineering Congress and Exposition, IMECE}, title = {Anomaly Detection for Cyber Physical Systems using Transformers}, year = 2021 }
2. S. Ding, A. Morozov, T. Fabarisov, und S. Vock, „KrakenBox: Deep Learning based Error Detector for Industrial Cyber-Physical Systems“, in The International Mechanical Engineering Congress and Exposition, IMECE, 2021.
  - Zusammenfassung
  - BibTeX
  Zusammenfassung
  Online error detection helps to reduce the risk of failure of safety-critical systems. However, due to the increasing complexity of modern Cyber-Physical Systems and the sophisticated interaction of their heterogeneous components, it becomes harder to apply traditional error detection methods. Nowadays, the popularity of Deep Learning-based error detection snowballs. DL-based methods achieved significant progress along with better results. This paper introduces the KrakenBox, a deep learning-based error detector for industrial Cyber-Physical Systems (CPS). It provides conceptual and technical details of the KrakenBox hardware, software, and a case study. The KrakenBox hardware is based on NVIDIA Jetson AGX Xavier, designed to empower the deep learning-based application and the extended alarm module. The KrakenBox software consists of several programs capable of collecting, processing, storing, and analyzing time-series data. The KrakenBox can be connected to the networked automation system either via Ethernet or wirelessly. The paper presents the KrakenBox architecture and results of experiments that allow the evaluation of the error detection performance for varying error magnitude. The results of these experiments demonstrate that the KrakenBox is able to improve the safety of a networked automation system.
  BibTeX
  @conference{noauthororeditor2021krakenbox, abstract = {Online error detection helps to reduce the risk of failure of safety-critical systems. However, due to the increasing complexity of modern Cyber-Physical Systems and the sophisticated interaction of their heterogeneous components, it becomes harder to apply traditional error detection methods. Nowadays, the popularity of Deep Learning-based error detection snowballs. DL-based methods achieved significant progress along with better results. This paper introduces the KrakenBox, a deep learning-based error detector for industrial Cyber-Physical Systems (CPS). It provides conceptual and technical details of the KrakenBox hardware, software, and a case study. The KrakenBox hardware is based on NVIDIA Jetson AGX Xavier, designed to empower the deep learning-based application and the extended alarm module. The KrakenBox software consists of several programs capable of collecting, processing, storing, and analyzing time-series data. The KrakenBox can be connected to the networked automation system either via Ethernet or wirelessly. The paper presents the KrakenBox architecture and results of experiments that allow the evaluation of the error detection performance for varying error magnitude. The results of these experiments demonstrate that the KrakenBox is able to improve the safety of a networked automation system.}, author = {Ding, Sheng and Morozov, A. and Fabarisov, T. and Vock, S.}, booktitle = {The International Mechanical Engineering Congress and Exposition, IMECE}, title = {KrakenBox: Deep Learning based Error Detector for Industrial Cyber-Physical Systems}, year = 2021 }
2020
1. S. Ding, A. Morozov, S. Vock, M. Weyrich, und K. Janschek, „Model-Based Error Detection for Industrial Automation Systems Using LSTM Networks“, In: International Symposium on Model-Based Safety and Assessment, pp. 212-226. Springer, Cham, 2020.
  Zusammenfassung
  The increasing complexity of modern automation systems leads to inevitable faults. At the same time, structural variability and untrivial interaction of the sophisticated components makes it harder and harder to apply traditional fault detection methods. Consequently, the popularity of Deep Learning (DL) fault detection methods grows. Model-based system design tools such as Simulink allow the development of executable system models. Besides the design flexibility, these models can provide the training data for DL-based error detectors. This paper describes the application of an LSTM-based error detector for a system of two industrial robotic manipulators. A detailed Simulink model provides the training data for an LSTM predictor. Error detection is achieved via intelligent processing of the residual between the original signal and the LSTM prediction using two methods. The first method is based on the non-parametric dynamic thresholding. The second method exploits the Gaussian distribution of the residual. The paper presents the results of extensive model-based fault injection experiments that allow the comparison of these methods and the evaluation of the error detection performance for varying error magnitude.
  BibTeX
  @article{ding2020modelbased, abstract = {The increasing complexity of modern automation systems leads to inevitable faults. At the same time, structural variability and untrivial interaction of the sophisticated components makes it harder and harder to apply traditional fault detection methods. Consequently, the popularity of Deep Learning (DL) fault detection methods grows. Model-based system design tools such as Simulink allow the development of executable system models. Besides the design flexibility, these models can provide the training data for DL-based error detectors. This paper describes the application of an LSTM-based error detector for a system of two industrial robotic manipulators. A detailed Simulink model provides the training data for an LSTM predictor. Error detection is achieved via intelligent processing of the residual between the original signal and the LSTM prediction using two methods. The first method is based on the non-parametric dynamic thresholding. The second method exploits the Gaussian distribution of the residual. The paper presents the results of extensive model-based fault injection experiments that allow the comparison of these methods and the evaluation of the error detection performance for varying error magnitude.}, author = {Ding, Sheng and Morozov, Andrey and Vock, Silvia and Weyrich, Michael and Janschek, Klaus}, journal = {In: International Symposium on Model-Based Safety and Assessment, pp. 212-226. Springer, Cham}, title = {Model-Based Error Detection for Industrial Automation Systems Using LSTM Networks}, url = {https://link.springer.com/chapter/10.1007/978-3-030-58920-2_14}, year = 2020 }
  Link
  https://link.springer.com/chapter/10.1007/978-3-030-58920-2_14
2019
1. K. Ding, S. Ding, A. Morozov, T. Fabarisov, und K. Janschek, „On-line error detection and mitigation for time-series data of cyber-physical systems using deep learning based methods“, in 2019 15th European Dependable Computing Conference (EDCC) (pp. 7-14). IEEE, 2019.
  Zusammenfassung
  A cyber-physical system consists of sensors, micro-controller, networks, and actuators that interact with each other, generate a substantial amount of data, and form extremely complex system operational profiles. These heterogeneous components are subject to errors, e.g. spikes, off-sets, or delays, that may result in system failures. As the complexity of modern systems increases, it becomes a challenge to apply traditional fault detection and isolation methods to such complex systems. Deep learning based methods have surpassed traditional methods in terms of performance as the data size and complexity increase. The signals of cyber-physical systems are mainly time-series data. In this paper, we propose a new on-line error detection and mitigation approach for common sensor, computing hardware, and network errors of cyber-physical systems using deep learning based methods. More specifically, we train a Long Short-Term Memory (LSTM) network as a single step prediction model for the detection and mitigation of errors, like spikes, or offsets. In order to detect the long-duration errors that show no sharp change (a sudden drop or rise) between two successive data samples when errors occurred, e.g. network delays, we train an LSTM encoder-decoder as a multi-step prediction model. We also introduce the on-line error mitigation approach. Automatic recovery is achieved by replacing the detected errors with the predicted values. Finally, we demonstrate on-line error detection and mitigation capabilities of the trained single step and multi-step predictors using representative case studies.
  BibTeX
  @conference{ding2019online, abstract = {A cyber-physical system consists of sensors, micro-controller, networks, and actuators that interact with each other, generate a substantial amount of data, and form extremely complex system operational profiles. These heterogeneous components are subject to errors, e.g. spikes, off-sets, or delays, that may result in system failures. As the complexity of modern systems increases, it becomes a challenge to apply traditional fault detection and isolation methods to such complex systems. Deep learning based methods have surpassed traditional methods in terms of performance as the data size and complexity increase. The signals of cyber-physical systems are mainly time-series data. In this paper, we propose a new on-line error detection and mitigation approach for common sensor, computing hardware, and network errors of cyber-physical systems using deep learning based methods. More specifically, we train a Long Short-Term Memory (LSTM) network as a single step prediction model for the detection and mitigation of errors, like spikes, or offsets. In order to detect the long-duration errors that show no sharp change (a sudden drop or rise) between two successive data samples when errors occurred, e.g. network delays, we train an LSTM encoder-decoder as a multi-step prediction model. We also introduce the on-line error mitigation approach. Automatic recovery is achieved by replacing the detected errors with the predicted values. Finally, we demonstrate on-line error detection and mitigation capabilities of the trained single step and multi-step predictors using representative case studies.}, author = {Ding, K. and Ding, Sheng and Morozov, A. and Fabarisov, T. and Janschek, K.}, booktitle = {2019 15th European Dependable Computing Conference (EDCC) (pp. 7-14). IEEE}, doi = {https://ieeexplore.ieee.org/abstract/document/8893390}, title = {On-line error detection and mitigation for time-series data of cyber-physical systems using deep learning based methods}, url = {https://ieeexplore.ieee.org/abstract/document/8893390}, year = 2019 }
  Link
  https://ieeexplore.ieee.org/abstract/document/8893390

Sheng Ding ist Doktorand und akademischer Mitarbeiter an der Universität Stuttgart, Institut für Automatisierungstechnik und Softwaresysteme. Sein Forschungsschwerpunkt sind Deep Learning Fehlererkennung und Zuverlässigkeitanalyse von Cyber-Physischen Systemen.

Forschung

Forschungsschwerpunkt:

Deep Learning Fehlererkennung und Zuverlässigkeitanalyse von Cyber-Physischen Systemen.

Beschreibung:

Die Erkennung von Anomalien ist ein bekanntes Konzept, das in einer Vielzahl von Bereichen, einschließlich der Systemtechnik, angewendet wird, um Fehler zu erkennen bzw. zu vermeiden. Angesichts der zunehmenden Verbesserung von Deep Learning (DL)-Ansätze wird die Anwendbarkeit von DL-Techniken zur Fehlererkennung erforscht. In diesem Forschungsvorhaben wird daran gearbeitet, Zeitreihendatenverarbeitung und häufige CPS-Fehler zu konzentrieren. Dabei werden zunächst die Trainingsdaten mithilfe der Kombination von vorhandenen und eigenen Fehlerinjektionsmethoden und –werkzeugen generiert. Hierzu werden Daten mittels Simulation im Bereich vernetzte Automatisierungssysteme und Fehler mittels Injektion mit entsprechendem Label erzeugt. Damit die Methoden in die reale Welt umgesetzt werden können, werden reale Fallstudien betrachtet und z. B. dabei gemessene Signale mit der Simulation verglichen.

Die überwiegende Mehrheit der vorhandenen Forschung zur Erkennung von Anomalien berücksichtigt keine Echtzeit. Dies ist jedoch sehr wichtig, um ein CPS-System zu schützen. Einerseits soll daher Online-Echtzeit Messung und Datenverkehr vom Host-System dokumentiert und evaluiert werden. Andererseits werden Online-Echtzeit-Antwort durch Vergleich der Berechnungskosten verschiedener DL-Modelle erforscht.

In dem Vorhaben werden viele Arten von Methoden bzw. DL-Architechturen verglichen und integriert. Geeignete DL-basierte Fehlererkennungsansätze wie Vorhersage, Klassifizierung, und Rekonstruktion werden verglichen. Das Training und die Evaluation der DL-Modelle erfolgen offline mit dem gesammelten Datensätzen und umfasst drei Phasen. Erstens, verschiedene Netzwerktypen, z.B MLP, CNN, LSTM / GRU, Autoen-Codierer, und Transformatoren, sowie deren Struktur- und Hyper-Parameter werden untersucht und identifiziert. Zweitens wird die Leistung mit Genauigkeit und Geschwindigkeit verschiedener Modelle als Wissensbasis dokumentiert. Drittens, um die Genauigkeit zu verbessern, wird die dynamische Rekonfiguration erforscht. Ziel ist, eine DL-basierte Error Detektion mit Echtzeit-Reaktion und hoher Genauigkeit zu entwickeln. Zur Demonstration des Konzepts wird die implementierte Methode online getestet. Die Ergebnisse zeigen, dass die Fehler in Echtzeit erfolgreich erkannt werden können.

Forschungsportal:

Google Scholar: https://scholar.google.de/citations?hl=en&user=jnvhuokAAAAJ

ResearchGate: https://www.researchgate.net/profile/Sheng-Ding-12