Falsification Algorithm for Hybrid System Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for verifying stochastic or probabilistic hybrid systems face challenges in determining the number of tests required to verify a model, particularly due to discontinuities in the search space, which makes it difficult to locate regions of interest for testing purposes, especially in hybrid systems where local minima can appear anywhere with small or non-smooth basins of attraction.
Innovation Solution
The approach involves representing Cyber-Physical Systems (CPS) models using Model Based Development languages like Ptolemy, LabVIEW, or Simulink/Stateflow, and employing Metric Temporal Logic (MTL) or Signal Temporal Logic (STL) to guide the falsification search towards promising regions by identifying and analyzing blocks with branching structures and discontinuities, thereby narrowing the search to improbable regions and improving verification performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional falsification algorithms are used for hybrid systems, then the search space can be explored, but the algorithm performance deteriorates due to discontinuities and small basins of attraction making it difficult to locate regions of interest
Solution Approach 1:
The patent performs preliminary extraction of discontinuity information from the hybrid system model before executing the falsification algorithm. By pre-identifying switching surfaces and discontinuous regions, the system prepares guidance data that directs the stochastic search algorithm toward critical areas, avoiding random exploration of the entire search space and reducing the number of tests required.
Solution Approach 2:
The patent implements feedback by using extracted discontinuity information to guide the falsification search process. The discontinuity extraction module provides feedback about switching surfaces and discontinuous regions to the falsification algorithm, which adjusts its search strategy accordingly, improving verification reliability while reducing the number of tests needed.
2Reliability
If the search space is explored without guidance, then all regions can be sampled, but the probability of detecting regions of interest becomes very small due to small or non-smooth basins of attraction
Solution Approach 1:
The patent introduces an intermediary discontinuity extraction module that acts as a mediator between the hybrid system model and the falsification algorithm. This intermediary extracts critical information about switching surfaces and discontinuous regions, transforming the raw model data into guided search directions, thereby increasing detection probability without sacrificing verification efficiency.
Solution Approach 2:
The patent changes the search parameters by using discontinuity information to bias the stochastic search toward regions with small or non-smooth basins of attraction. By modifying the search distribution based on extracted discontinuities, the system increases the probability of detecting critical regions while maintaining reasonable verification efficiency.
3Measurement precision
If gradient descent algorithms are used, then local minima can be optimized, but they fail in hybrid systems where local minima appear anywhere with very small or non-smooth basins of attraction
Solution Approach 1:
The patent replaces gradient descent algorithms with a stochastic falsification approach guided by discontinuity extraction. Instead of relying on gradient-based mechanics that fail in hybrid systems with discontinuities, the system uses stochastic sampling informed by extracted switching surfaces, making the approach applicable to hybrid systems while maintaining optimization precision in detecting violations.
Data Source
AI summary
One embodiment is a methodology for model verification. An embodiment obtaining, by a processor, a model for a system; identifying, by the processor, at least one block within the model that has a branching structure; identifying, by the processor, at least one model variable affecting a switching condition of the identified at least one block; generating, by the processor, an extended finite state machine modeling a switching behavior of the identified at least one block by using the at least one model variable; combining, by the processor, at least one output variable of the extended finite state machine with at least one of a first output port and a second output port of the system included in the model; and performing, by the processor, model verification and coverage of the model that utilizes outputs from the first output port and the second output port to verify the model.