Fault detection, isolation and quantification from Gaussian residuals with application to structural damage diagnosis

Döhler, Michael; Mevel, Laurent; Zhang, Qinghua

doi:10.1016/j.arcontrol.2016.08.002

Cited by 37 publications

(47 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For further comparison, the very recent approach about residual evaluation presented in is also employed. According to the local approach, a new residual designed as

ζ (k) = \frac{1}{\sqrt{N}} \sum_{j = k - N + 1}^{k} r (k)

is assumed to ensure

ζ (k) \sim scriptN (0, R)

in the fault free case where R is an estimated covariance matrix.…”

Section: Simulation Studymentioning

confidence: 99%

“…It is noted that most of the above works focus on residual generation, while the second step, residual evaluation, is by far less studied, as has been stated in and . However, residual evaluation is an essential part and an appropriate EF is critical to the FD performance.…”

Section: Introductionmentioning

confidence: 99%

“…However, residual evaluation is an essential part and an appropriate EF is critical to the FD performance. Works focus on residual evaluation are mainly about FDI design for stochastic systems, see , etc. A notable advantage of these schemes is that prior knowledge about plant models and probability distributions of concerned signals is fully used, thus makes it suitable for small or incipient FD.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A New Residual Evaluation Function Based Fault Diagnosis for a Kind of Nonlinear Systems

Hai

Zhong

Liu

2018

Asian Journal of Control

View full text Add to dashboard Cite

In this paper, a new residual evaluation function (EF) is proposed for the H i ∕H ∞ fault detection filter (FDF). Residuals are projected into the direction of fault, then energy of this projection is adopted as the new EF. Compared with the traditional EF, improvement in FD ability of the new EF is demonstrated rigorously, and a recursive way to calculate it is also proposed under the assumption that fault is slowly time varying. Furthermore, application of this new EF for fault isolation (FI) is discussed, which finally leads to a systematic FDI scheme for a kind of nonlinear systems. Simulations with a nonlinear unmanned aerial vehicle (UAV) model validate the proposed method.new,i and J f new,i , the accumulated terms in J r new,i are not quadric forms and their signs may change from step to step, thus it is reasonable to assume that J r new,i is much smaller than J d new,i and J f new,i . That is, J new,i is roughly approximated asThen, since J f new, ≥ J f new,i holds true, it is expected to isolate the fault correctly by J new, when J d new,i ≈ J d new, ≪ J f new, . That is, J new,i and J new, in the fault free case are roughly equal. Compared with the assumption that residual is dominated by fault (r ≈ Ω f ), this conditional is milder and more realistic.However, undesirable situations still exist, that is, J new,i and J new, differs from each other significantly in the fault free case. Without loss of generality, it is assumed that the direction of r 0 is much closer to Ω i , i ≠ , which result in J new,i ≫ J new, in the fault free case. To deal with this situation, the following normalized EF is proposed.where (J 0 new,i ) is an empirical mean of J new,i obtained from historical normal data. Then we havēSince the direction of r 0 is close to Ω i but far from Ω , directions of Ω i and Ω are not close to each other, thus J f new,i is very small when f occurs. Therefore, it is expected thatJ new,i will not increase obviously after f occurs, whileJ new, increase significantly after f occurs. Finally, it is expected to haveJ new, >J new,i . When J new,i and J new, are close to each other in the fault free case, J new,i ≈ 1 + J f new,i (J 0 new,i ) ≤J new, ≈ 1 + J f new, ance instrument, Beihang University, Beijing, China. His research interests are model-based fault diagnosis and applications.Maiying Zhong received the Ph.D. degree in control theory and control engineering from Northeastern University,

show abstract

“…For further comparison, the very recent approach about residual evaluation presented in is also employed. According to the local approach, a new residual designed as

ζ (k) = \frac{1}{\sqrt{N}} \sum_{j = k - N + 1}^{k} r (k)

is assumed to ensure

ζ (k) \sim scriptN (0, R)

in the fault free case where R is an estimated covariance matrix.…”

Section: Simulation Studymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A New Residual Evaluation Function Based Fault Diagnosis for a Kind of Nonlinear Systems

Hai

Zhong

Liu

2018

Asian Journal of Control

View full text Add to dashboard Cite

show abstract

“…element stiffness) of the monitored system, where θ 0 corresponds to the (healthy) reference state. Then, the non-centrality parameter of the test statistic distribution is given by δ T F δ, where δ = √ N (θ −θ 0 ), N is the measurement length and F is the Fisher information of θ contained in the computed residual vector [9]. Thus, the statistical distribution of the test statistic is known for a given system degradation (linked to D) in the monitored system (linked to F ).…”

Section: Updating the Structural System Reliability With Damage De-tementioning

confidence: 99%

Damage Detection and Deteriorating Structural Systems

Long¹,

Thöns²,

Döhler³

2017

Structural Health Monitoring 2017

Self Cite

View full text Add to dashboard Cite

This paper addresses the quantification of the value of damage detection system and algorithm information on the basis of Value of Information (VoI) analysis to enhance the benefit of damage detection information by providing the basis for its optimization before it is performed and implemented. The approach of the quantification the value of damage detection information builds upon the Bayesian decision theory facilitating the utilization of damage detection performance models, which describe the information and its precision on structural system level, facilitating actions to ensure the structural integrity and facilitating to describe the structural system performance and its functionality throughout the service life. The structural system performance is described with its functionality, its deterioration and its behavior under extreme loading. The structural system reliability given the damage detection information is determined utilizing Bayesian updating. The damage detection performance is described with the probability of indication for different component and system damage states taking into account type 1 and type 2 errors. The value of damage detection information is then calculated as the difference between the expected benefits and risks utilizing the damage detection information or not. With an application example of the developed approach based on a deteriorating Pratt truss system, the value of damage detection information is determined, demonstrating the potential of risk reduction and expected cost reduction.

show abstract

“…Then, damage indicators are defined with respect to the elements of the FE model, without updating it. For example, damage is located by a statistical analysis of a data‐driven subspace‐based damage residual with respect to model‐based sensitivities or by analyzing stress fields that are computed from data‐driven load vectors and the FE model …”

Section: Introductionmentioning

confidence: 99%