Reliability Modeling And Performance Analysis Of AFDX Avionics Network

The robustness of avionics system largely depends on close interactions of sub-avionicssystems. Avionics network is obliged to provide these interactions that ensure high reliabilityand security of data transfer. Meanwhile, with emergence of new generation aircraftcharactered by more complex constructions and functions, especially with the heavy relianceof fly-by-wire and the associated avionics, next generation avionics network require higherthroughput and scalability, as well as lighter weight of construction and material. AFDX(Avionic Full-duplex Switched Ethernet) is the appropriate solution of this developingsituation. It is a standard that defines electrical and protocol specification for the exchange ofdata between avionics subsystems in ARINC664.AFDX combines concepts of asynchronous transfer mode (ATM). Each frame isstored-and-forwarded on switches that are connected by wirngs. The mechanism of sharingcommon communication and computation resources makes a challenge to verify the reliabilityof data transfer. This paper focuses on whether frames can be transferred to objective endsystem successfully and whether the transfer delays will exceed the deadlines required byavionic application sysrem.The main work and research results are summarized as follows: (1)Model the AFDX network using traditional fault tree theory with "loss of frame" asthe top event. Fault tree is constructed based on the AFDX network topology and networkentities' work mechanism. Passive loss and active loss are abstracted from AFDX model, thefomer is caused by malfunction of functional compenents and the latter is trigered by AFDXprotocal.The quantitative analysis is completed by boolean algebra.(2) Static Fault Tree is modified to capture sequence-dependent behavior in AFDX. Theredundance management algorithm and back-up of AFDX switches are represented byDynamic Fault Tree gate. Markov Chain is used to conduct quantitative analysis.(3) AFDX network worst case delay is obtained based on network calculus theory.Thewaiting delay is divided to external delay caused by other priority frames and internal delaycaused by frames in the same buffer. Each portion of delay is computed under the worst casescenarios. Moreover, a tighter upper-bound of delay on multi-hop network is obtained bymodelling the multi-server as one black-box server.(4) Delay distribution is analyzed in the worst-case delay frame. Based on AFDXdiff-serve architecture, the probability that frames of each critical-level flow miss its deadlineis obtained by statistical network calculus. Comparisons between probabilistic-constraineddelay and deternimistic delay are made.(5) The overall verify process is completed on an industrial-configurationed AFDXnetwork. As Virtual Links of different critical level have different safty requirments,probabilistic-constrained delay corresponding different flow is obtained for each VL.Reliability verify of AFDX network is accomplished by combination of reachability andusability.