Application of randomized algorithms to the analysis and synthesis of model-based and data-driven fault diagnosis systems

• Applicant: Professor Dr.-Ing. Steven Xianchun Ding
• Subject Area: Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics
• Term: from 2014 to 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 242153335

• Final Report Year: 2018

Final Report Abstract

The overall goal of this project is to develop a framework for the assessment, analysis and design of model-based and data-driven diagnosis systems. To this end, the so-called randomized algorithms (RA) technique serves as a tool. Three major research objectives were formulated in the research proposal: • establishment of a framework for the analysis and design of fault detection (FD) systems using the RA and statistic learning theory, in which methods and algorithms should be developed – for the computation of false alarm rate (FAR) and missed detection rate (MDR) and for the optimization of FD systems; • development of fault models, which should allow the application of RA methods – to generate faulty data, and based on them, to isolate and identify the faults; • development of a RA-based platform for the FD performance assessment of observerbased and data-driven FD systems, including the estimation of FAR, MDR and detection time. A further project objective is to develop a software tool in the MATLAB environment to support the implementation of the randomized algorithms and the developed assessment, analysis and design algorithms. Finally, the achieved theoretical results should be demonstrated and illustrated by benchmark study on simulation and laboratory test beds. The major results achieved in this project are summarized as follows: • Probabilistic models for faults and model uncertainties have been developed. They build the basis for the framework of RA-aided assessment, analysis and synthesis of FD systems. • Definitions of FAR, fault detection rate (FDR), MDR and mean time to fault detection (MT2D) have been introduced in the probabilistic framework. Based on them, algorithms for the RA-aided computations and estimations of FAR, FDR, MT2D have been developed, which can be used for assessing FD performance of different types of FD systems and thus build the basis for FD system optimization. • Numerous optimization and iterative learning approaches and algorithms have been developed for – optimization of observer-based FD systems; threshold setting of different types of FD systems; multiple feature based FD systems. • A MATLAB toolbox has been developed, which serves as the software tool to support the implementation of the proposed algorithms, which are in general computationally involved. • Applications on different types of technical systems.

Publications

• (2019) Application of randomized algorithms to assessment and design of observer-based fault detection systems. Automatica 107 175–182
  Ding, Steven X.; Li, Linlin; Krüger, Minjia
  
• A fault detection approach for nonlinear systems based on data-driven realizations of fuzzy Kernel representations, IEEE Transactions on Fuzzy Systems, 99: 1-13, 2017
  L. Li, S. X. Ding, Y. Yang, K. Peng, and J. Qiu
  
• A probabilistic approach to robust fault detection for a class of nonlinear systems, IEEE Transactions on Industrial Electronics, 64: 3930-3939, 2017
  M. Zhong, L. Zhang, S. X. Ding and D. Zhou
  
• Randomized Algorithm Based Fault Detection System Design for Uncertain LTI Systems, Proc. of the 20th IFAC World Congress, Toulouse, France, 2017
  M. Krueger, T. Koenings, Y. Liu, S. X. Ding, J. Saijai and L. Li
  
• Randomized Algorithms Aided Analysis and Design of Model-Based Fault Detection Systems, Universität Duisburg-Essen, 2017
  M. Krüger
  
• Fault Detection for Non- Gaussian Processes Using Generalized Canonical Correlation Analysis and Randomized Algorithms, IEEE Transactions on Industrial Electronics, 65: 1559-1567, 2018
  Z. Chen, S. X. Ding, T. Peng, C. Yang, and W. Gui