3. Risk-based analysis
3.2. Risk-based Methodology
3.2.2. DRAM Methodology
Darmstadt Risk Analysis Methodology (DRAM) with its Darmstadt Risk Analysis Tool (DRAT) is mainly based on the FMEA-based risk analysis method developed in Darmstadt 15 years ago.
DRAM provides possibillty to model the cause-and-effect chain in different areas and to use this model by varying input parameters as well as modules of driver (road user) behaviour chain. In this research, DRAM is applied to construct driver behaviour chain of violating traffic regulations.
Figure 6. Structure of Darmstadt Risk Analysis Method DRAM In summary, working with DRAM methodology needs to determine:
- The cause-and-effect chain (from every possible causes to expected effects);
- Parameters described by risk values with their probability distributions; and
- The relationships among parameters. 3.2.2.2. Risk values and probability distributions
As mentioned ablove, risk is explained as the product of the probability and the extend of a possible damage. More generally, it can be described as the integral of this product for all possible extend values of damage.
Figure 7. Description of risk
[Bald 1991] It is important to mention, that the risk has to be related to time, place and referenced group (only one person or a more or less specific group of persons). Using risk values offers the opportunity to eleminate units of different variables (different terminologies). Therefore, it will be very much helpful when calculating the probability leading to different situations (normal situations or critical situations or harzards, etc.). Risk values help also in classifying different groups as well as summing up variables and their probability when needed. By regarding the risk of larger groups, it is possible to assess the total loss of the whole traffic system.
In methodology, interfaces and modules describe whole classes of states and actions, called ―situations‖ and ―developments‖. They are tainted with uncertainties resulting from the statistical nature and the influence of human behaviour in road traffic. It seems to be appropriate to describe most variables, representing the values of situations and the relations of developments, as probability distributions.
Additionally, the values of most situations and the relations of most developments depend on other variables (which themselves may be described by probability distributions and may be dependent on other variables), giving the need to describe them as a function of themselves. For that reason, the DRAM uses multidimensional probability distributions (see example in Figure 8), which means hierarchically organized sets of distributions.
damage
damage
p
R
(
)d
probability
for damage
damage
Risk 3
Risk 1
Risk 2
Figure 8. Example of a multidimensional probability distribution
[Bald 1991] The distributions themselves are given numerically, allowing to describe every form of distribution (not only the standard ones). The reason is, that the complex and non-linear relations are only rarely comparable to standard distributions and that the used data, which are currently used, are mostly empiric higher. Accuracy is only a question of the number of values (and dependent upon it, the computing power of the analysing machine)
3.2.2.3. Describing the System Systematically by Cause-and-Effect Chain
As mentioned from above, the DRAM is using probabilities to describe the states and situations of the systems. For all parameters, it is essential to obtain a reliable estimate of their value. Unfortunately, those estimates are based on relatively scarce information. In recent years, interest in the Bayesian approach to data analysis has increased significantly in many areas of application, including traffic safety. Traffic safety engineers are among the early adopters of Bayesian statistical tools for analysing crash data.
The Darmstadt Risk Analysis Method follows a modular approach by trying to describe the cause-and- effect-chain from the influencing parameters to possible damage with active and passive elements (―developments‖ and ―situations‖; see Figure 4). ―Development‖ and ―situation‖ are more general terms for ―action‖ or ―event‖ and ―state‖, which are often used to describe technical or organizational processes. These more general terms and thinking are necessary, because whole sets of actions/ states are analysed.
velocity size 1. dependence (e.g. enginepower) 2. dependence (e.g. roadcondition ) distribution dry wet icy low-powered (l-p) medium-powered(m-p) high-powered(h-p) p(v|l-p, dry) p(v|l-p, wet) p(v|l-p, icy) p(v|m-p, dry) p(v|m-p, wet) p(v|m-p, icy) p(v|h-p, dry) p(v|h-p, wet) p(v|h-p, icy) dry wet icy dry wet icy
analysis direction startingpoint of the analysis situation development time possible starting scenarios possible end scenarios
Figure 9. Describing a system with “situations” and “developments”
[Durth, Bald 1988] The use of a whole network of passive (interfaces) and, especially, active elements (modules) offers many advantages:
- very complex systems can be cut into independent modules, which may be analysed by different researcher groups (even from different disciplines) with different methods (e.g. theoretically, by simulator experiments, by observation of the process);
- every module may be improved and substituted as long as the interfaces and its basic functionality is not changed;
- every module may be refined by regarding it as a sub-process (Figure 10); it is also possible to analyse non-linear relations.
development development situation development situation situation situation situation
Figure 10. Refining the description by analysing an active element as a sub-process [Durth, Bald 1988] Every research group can concentrate on the modules they are especially interested in. For their analysis they can use generalized results of other groups. For the analysis of other research groups
they can give away a generalized description of their own findings. In the same manner, the whole cause-and-effect chain may be assembled from the findings of many research groups. This approach is very similar to a Finite-Element-Method (FEM), where stiffness data of many elements are combined systematically to one big system stiffness for analysing the whole system with knowledge only of the single elements.