MoULDyS implements the ideas presented in the paper by Bineet Ghosh and Étienne André, Offline and Online Monitoring of Scattered Uncertain Logs Using Uncertain Linear Dynamical Systems (published in FORTE 2022).

Offline Monitoring

The main idea of offline monitoring is to recreate the missing samples. Consider two consecutive logs k and k +1 with time stamps t and t+5. In this case, the main objective is to check if an unsafe behavior has happened between time steps t and t+5. This is achieved by computing (over-approximate) reachable sets from time step t, with the initial set as per the log k, up-to time step t + 5 respectively. If the over-approximate reachable set intersects with the unsafe set, the system is flagged as possibly unsafe; if not, the system is definitely safe. It further proposes a heuristic to reduce the number of false alarms.

Online Monitoring

The online monitoring framework, too, uses reachable sets at discrete steps to detect if there is a possible violation of safety. If there is a violation, it triggers and sample at that time step.

• MoULDyS can be used to recreate the results in the paper by Bineet Ghosh and Étienne André, Offline and Online Monitoring of Scattered Uncertain Logs Using Uncertain Linear Dynamical Systems (published in FORTE 2022).

• It can be done by using simple easy-to-use python scripts. Check out the details here.

• MoULDyS provides the following two case studies.

Anesthesia

An automated Anesthesia delivery model using the drug propofol is used. The system models the metabolization of the drug by the body, and the depth of hypnosis. The system models the metabolization of the drug by the body, and the depth of hypnosis. The state variables encode the various concentration levels---that must be within a certain limit at all times—modeling the metabolization of the drug and the depth of hypnosis. Note that a higher concentration level would mean that the patient remains unconscious for a longer period of time, while a lower concentration level would mean that patient remains conscious during the surgery—which can be traumatic. Clearly, monitoring of this system for safety (i.e., the concentration levels are within a safe bound at all times) is crucial. However, it is not possible for a busy human doctor or a practitioner to manually monitor all the patients at all times. Thus, MoULDyS enables them to perform an automated monitoring of the patients without compromising on safety and save their time.

Adaptive Cruise Control (ACC)

Most modern day cars are equipped with adaptive cruise control, which behaves like an ordinary cruise control when there is no car in its sight, and maintains a safe distance when there is a car in its sight. In the event of a crash, the only data available to analyze the crash is the stored logs. Offline Monitoring provides an automated way to detect the cause of the crash and who was at fault—these questions are particularly useful to police and insurance companies. Similarly, consider a vehicle driving in a highway with a vehicle in its sight. The ACC unit will have to continuously read sensor values to track several parameters, such as acceleration of the lead vehicle, braking force, etc.—causing an wastage of energy. In these cases, deploying online monitoring on the vehicle ACC will ensure that the sensor values are only read when there is a potential unsafe behavior—thus saving a lot energy. Thus, MoULDyS enables the cars to perform an automated monitoring.