Preliminaries
MBD is undoubtedly a powerful technique for precisely isolating a problem's origin(s). We need a special model for this reasoning though, and while MBD is complete with respect to this model, the entailed computations can become quite complex. That is, the diagnosis search space is exponential in the number of the health state variables that we have to introduce. The more health state variables we have, the more faults can be found, but the larger the search space is. Usually, we end up with more than one diagnosis matching the data, so that we have to choose one as a working hypothesis.
With SFL, we take a different approach and consider the involvement of components in failing and passing behavior. The reasoning then follows the idea that some component that is always involved in faulty behavior but never in correct behavior is very suspicious of being the source of the troubles (and vice versa). Since components are usually involved in both faulty and correct behavior, as well as the possibility that some faulty components cancel each other out, leading to correct behavior, many similarity coefficients, e.g. \cite{harrold2005,gemund2009}, for computing a component's suspiciousness have been proposed.
An intrinsic advantage of considering multiple executions in SFL by default is that fault masking (when multiple faults lead to correct output observations) has less effects on the reasoning, that is, if the set of observed behaviors is representative enough to contain also behavior without the masking effect. If the set is indeed representative and the faults always mask each other, then we are possibly facing an equivalent mutant so that we might want to consider the "faults" as implementation alternatives. For their computation, we consider the corresponding execution data about which component was involved in which behavior (stored in a matrix also referred to as spectrum), and whether some behavior is violating or complying with our expectations (the so-called error vector). Based on the components' suspiciousness values, we establish a ranking.