Chapter 5 Online Risk Decision Support System
5.3 The Modeling of Data Analysis on DP System, Operator and Environment
The quantity analysis of the drive-off is based on the data about the possibility to fault (failure rate) of each component and activities in the basic event of the fault tree. For the online risk analysis, the inputting failure rate to the FTA is a real time dynamic data instead of historical data, such as constant experience data or experiment data from handbook.
There are three types of basic data in the basic event of fault tree according to data source, and they are environment data, real time dynamic human reliability data and DP system running data. The environment data and running data can be collected by sensors and DP system internal data transportation. Although there is no sensor
designed for DPO to measure their dynamic human reliability data, their intervention to DP system is measureable.
Figure 27 Generic RIF model for execution and control activities (Vinnem, 2011) The operation error is fault intervention to the DP system. The structure and relationship among human error and RIFs are shown as Figure 27. Level 1 includes RIFs with a theoretically and empirically justified influence on the failure type violations, mistakes or slips and lapses. Level 2 presents different aspects of management that influence the RIFs on Level 1(Vinnem, 2011). But there is no any sensor to measure these factors and to generate a real time data to input into ORDSS. Only after the intervention is done, can the action of intervention transform into a signal through the HMI of DP system, for example selecting an operation mode is same as a switch on/ off or digital signal and setting a new heading is same as input analog signal. Then the fault intervention is detected. This is a postpone detection, not predication and detection punctually. To solve the problem of predication and punctual detection, the solution is to press the button with confirmation activity. The design of DP operation system now is to select or deselect an operation mode by pressing the button two times. The aim is to avoid pressing the button by mistake. After the DPO input information into DP system, the ORDSS display the vessel response on the operation screen, and pick up this operation signal into Data Analysis mode to analysis. If the operation can cause drive-off, the operation will indicate as a reason of drive-off on the operation interface and to give a warning. The ORDSS has detected fault DPO operation. If the DPO confirm the information, ORDSS will issue
a drive-off alarm.
The Integrated Intelligent Model is introduced, and it is used to detect the process state, especially for fault, and behavior prediction (failure rate) according to process input and historical output. It combines the influence caused by the final results of process state (fault) detection and quantitative prediction, the failure rate of single component. Three Artificial Neural Networks (ANN) models are used for system identification of process characteristics in different process states. The whole model is developed basing on Fuzzy TS dynamic Nonlinear AutoRegressive with eXogenous input models (NARX models) (Tang, 2004). This intelligent mode has been tested by study-prediction case for supply chain process. With a four-year data study, it can predict the fifth-year operation data, and the rate of average relative error is 0.21. Therefore it can be used to calculate the online failure rate data by computation software and input into the FTA (Tang, 2004).
5.3.1 The Construction of Integrated Intelligent Model
There are five units in Integrated Intelligent Model. Each unit has a special function and sub-models. The construction of the model is shown as Figure 28.
Figure 28 Construction of Integrated Intelligent Model
This model consists of three modules. The Training module functions as building model by learning from historical data. The Working module functions as processing
diagnosis and prediction. The Optimization module functions as optimal model and model adaptability (Tang, 2004). The input data are the real time data and the historical data. And there are three outputs, the knowledge presentation decision making support, prediction value and the fault residual generator. The fault residual signal is generated to detect the fault event by the system.
5.3.2 Application in Failure Status Prediction
The relative system sensor can pick-up the real time data as input 1. And the history data can withdrawn from the Voyage Data Recorder, which is a equipment to record down all the component running data like the black box carried on aircraft. In addition, the failure rate that is calculated from MTTF can get from manufacture reliability manual. The history data is as input 2.
The mode is established from history data learning. And the Integrated Intelligent Model combining the variable input 1 and input 2 is described as a fuzzy TS NARX dynamic mode as Figure 29. It includes fuzzyfication, applying fuzzy operation, applying implication, aggregation, defuzzify and predication output Y. The Z11 & Z21
represent that Input 1& Input 2 output an error or fault state and The Z12 & Z22
represent that Input 1& Input 2 output a normal state. And wi is the fuzzy degree in
relation to different process states.
Figure 29 Fuzzy TS NARX dynamic Mode
formula
Y = aY + bY + c = a(w Z + w Z ) + b(w Z + w Z ) + c And a and b is the affection degree of input 1 and input 2 to the whole process of output. And through the historical data learning, the a, b, c and wi can result with a
constant value.