3。Change  detection

Having established a residual vector with desired properties (Gaussian and independent samples), change detection is a standard problem from de- tection theory [29, 30]。 The result is repeated here for convenience。

If at the sample k the g(k) is larger than the chosen threshold γ, where γ = ln(γ0) above, then the hypothesis H1 is accepted。

In an on-line implementation, a moving window of the latest M samples is considered and the test statistic for line j is, using Γj for the known direc- tion caused by fault on this line,

The vector-based hypothesis test problem is for-

mulated as:

H0 :   z(i) = µ0 + w(i) i = 1, 。 。 。 , k

H1 :    z(i) = µ1 + w(i) i = 0, 。 。 。 , k,

(15)

where w(i) is a Gaussian-distributed vector with

and the estimate of the magnitude of change is

independent increments and variance Q,  and  µ0

and µ1 are mean value vectors before and after the change, respectively。 A special case, which applies

here, is that the direction of change is known   but

the magnitude of change is unknown。 Let Γ be the known direction of change, and v an unknown scalar, then

H1 :   µ1 = µ0 + Γv。 (11)

The generalised likelihood ratio test (GLRT) on a sequence of k samples will decide H1 if

The detector in Eqs。 (15)-(16) are used in the

sequel for the processing of experiment data。

3。1。Alternative change detection schemes

CUSUM based change detection is efficient in detecting a change of known magnitude and the CUSUM approach was employed in [18] on the 

ition mooring application, and it was shown to give very convincing results for the known magnitude events, which are:  loss of a buoy, or a line   break-

wherevˆ1    is   the   maximum   likelihood   estimate

age。   This paper uses the vector based GLR    test,

(MLE) of µ1, and γ0 is a threshold selected to give a desired (low) probability of false alarms。

The MLE of v is found where the gradient of ln LG  is zero and as

as the GLR has the ability to detect a change of

unknown magnitude。 This implies that also partial loss of buoyancy can be detected。

Fault isolation is achieved in the vector-based ap- proach through the known direction of change for each particular fault。 Since a fault in mooring line j

will manifest itself in the vector r = [r1, r2, 。。。r5+j ]r。 The directions of change Γj  differ in one component

and isolation could be achieved in two ways。摘要：前期诊断和容错控制对浮式平台系泊系统保持船舶的位置和承受必要环境负荷情况下的安全操作至关重要。本文考虑通过矢量统计检测系泊线断裂和浮力因素的缺失两个关键故障的实时诊断。根据特定假警报概率和其上检测的变化统计量分布估计而设计了诊断设计审查和过程控制。结构可靠性指标针是对每个系泊线安全指标监测和各节点追踪算法的可靠性积分，并且容纳线断裂因素。在水域模型测试诊断和容错可行性控制的策略是不断变化的。

毕业论文关键词：故障诊断，容错控制，定位系泊，结构可靠性，检测变化，设定最佳追逐点

1 引言

如今越来越关注提高浮式平台的安全性和作业效率。在封闭的海岸作业往往需要海上的船舶控制在一定范围内，通过推进器定位的方法被称为DP系统。动力定位可以参考相关文献[1],[2],[3]。通过锚泊线系使船舶与海床的动态定位的方法被称为PM系统。PM系统的主要目标是保持船舶在一定范围内和通过保持船舶范围的方式防止锚链断线。关于推进器辅助PM系统的研究内容可参考相关文献[4],[5],[6],[7]，综合DP控制的可靠性问题参考文献[8]。本文将主要讨论的提议是：在早期的研究中对发生故障或锚链故障情况下保持安全控制能力的响应动作处理过程。