CN-121615488-B - Single event effect simulation and fault diagnosis system and method based on spaceborne computer

CN121615488BCN 121615488 BCN121615488 BCN 121615488BCN-121615488-B

Abstract

The invention discloses a single event effect simulation and fault diagnosis system and method based on a spaceborne computer, comprising a simulation host, wherein a simulation engine is arranged in a hardware environment of the simulation host, the simulation engine is respectively connected with a protocol monitor, a log module and a configuration and injection manager, and is provided with a peripheral model library, the peripheral model library is connected with an interrupt controller through a bus and an interface, the interrupt controller is in butt joint with the simulation engine, the protocol monitor is connected with the log module, the data set generator is connected with a diagnosis server, and a dynamic-attention-time sequence (DAT) hybrid neural network model is arranged in the diagnosis server. The system is used for carrying out fault type identification by adopting the DAT hybrid neural network based on general peripheral abstraction and discrete event semantics, and is suitable for on-orbit working condition simulation and diagnosis of multiple protocols and multiple peripherals.

Inventors

MU ZHONGCHENG
ZHAN YI
SONG ZIYANG
HUANG YIXIN

Assignees

上海交通大学

Dates

Publication Date: 20260508
Application Date: 20251201

Claims (8)

1. The single event effect simulation and fault diagnosis system based on the spaceborne computer is characterized by comprising a simulation host, wherein the simulation host is used for loading a peripheral model library and running system-level simulation based on discrete event semantics, a simulation engine is arranged in a hardware environment of the simulation host, and the simulation engine is respectively connected with a protocol monitor, a log module and a configuration and injection manager and is provided with the peripheral model library; The peripheral model library is connected with an interrupt controller through a bus and an interface, the interrupt controller is in butt joint with the simulation engine, and the interrupt controller is used for receiving interrupt requests generated by each peripheral in the peripheral model library, carrying out shielding and priority arbitration on the interrupt requests, vectorizing the arbitrated interrupt into discrete interrupt events, and outputting the discrete interrupt events to the simulation engine so as to participate in discrete event scheduling of the simulation engine; The protocol monitor and the log module are connected with a data set generator, the data set generator is connected with a diagnosis server, a dynamic-attention-time sequence hybrid neural network model is arranged in the diagnosis server, and the diagnosis server is used for identifying faults caused by a single event effect and outputting fault diagnosis results; the peripheral model library comprises a communication peripheral model and a control peripheral model, wherein the communication peripheral model and the control peripheral model describe interfaces, states, registers, functional mapping and time sequence semantics through peripheral abstract six-tuple respectively, and the formulas are as follows: ; Wherein, the Discrete signals and data ports read from the bus and pins are provided for the peripheral, Discrete signals and data ports driven externally by the peripheral devices, For each state of the internal finite state machine of the peripheral, In order to be able to read and write the register, Is in an event step, is in Input of Calculating new And (3) with Is used for the mapping of (a), Is protocol level temporal semantics; The peripheral model library is used for providing a register read-write callback, a time sequence event callback and an interrupt trigger interface and exposing a binding point with a system bus; the dynamic-attention-time sequence hybrid neural network model comprises a signal branch, a device embedding branch and a time embedding branch; the signal branch comprises a dynamic convolution layer I and a dynamic convolution layer II, wherein the back of the dynamic convolution layer I and the back of the dynamic convolution layer II are respectively connected with a maximum pooling layer I and a maximum pooling layer II, the back of the maximum pooling layer I and the back of the maximum pooling layer II are respectively connected with a random inactivation layer I and a random inactivation layer II, and the random inactivation layer I is connected with the dynamic convolution layer II; the device embedding branch comprises a device embedding mapping layer, a device input layer and a device processing layer, wherein the device embedding mapping layer is used for mapping a device identification vector into a device embedding vector with a preset dimension, and the input of the device embedding mapping layer is connected with the device input layer; the time embedding branch comprises a time embedding mapping layer, a time input layer and a time embedding processing layer, wherein the time embedding mapping layer is used for mapping the time scalar into a time embedding vector with a preset dimension, and the time input layer is connected with the time input layer in an input way; The attention aggregation module, the equipment embedding mapping layer and the time embedding mapping layer are all connected with a fusion discrimination layer, and the fusion discrimination layer is connected with an output layer; The dynamic-attention-time sequence hybrid neural network model is used for identifying weak-amplitude and sparse-segment fault characteristics caused by a single event effect, and the attention aggregation module strengthens weak signal response of the single event effect fault.
2. The single event effect simulation and fault diagnosis system based on the spaceborne computer of claim 1, wherein the simulation engine defines event types by adopting discrete event semantics and gives propagation delay, logic delay and interrupt constraint to each peripheral, and the formula is: ; Wherein, the Representing the delay of the combinational logic and register write to the observed output, Representing the link delay from the peripheral output to the opposite sampling point, Representing a timing verification reference period.
3. The single event effect simulation and fault diagnosis system based on-board computer of claim 1, wherein the configuration file of the configuration and injection manager comprises a target device, a target interface, a field type, a bit or bit segment range, an injection probability or occurrence interval, an injection time window, an injection level, a running round number and a random seed; The configuration and injection manager is used for executing fault injection caused by a single event effect according to a set probability in a set time window according to a designated bit or bit segment of the configuration file to the selected peripheral, and maintaining random seeds and round numbers.
4. The single event effect simulation and fault diagnosis system based on-board computer as set forth in claim 1, wherein the protocol monitor and log module are semantically in protocol level time Capturing discrete event data output by a simulation engine, and outputting the minimum full record of a four-tuple as a log, wherein the four-tuple formula is as follows: ; wherein the protocol monitor and log module output Is in combination with The discrete time scale of the alignment, For the identification of the instance of the peripheral, For an observed signal on the peripheral device, To map cores Unified tag set for decision The set of tags includes single event effect fault tags.
5. The single event effect simulation and fault diagnosis system based on-board computer of claim 1, wherein the data set generator converts the log from the protocol monitor and log module into a data set under a unified tag set, the unified tag set is set as: ; The data set generator realizes a cross-protocol mapping core, and the formula is as follows: ; Wherein, the In the case of a protocol or class of devices, Is bit or field semantics.
6. The satellite-borne computer-based single event effect simulation and fault diagnosis system according to claim 1, wherein the dynamic convolution layer I and the dynamic convolution layer II are both conditional convolution kernels, and weights are generated adaptively along with samples or working conditions so as to alleviate distribution drift across protocols and peripheral devices caused by single event effect faults.
7. The single event effect simulation and fault diagnosis system based on a satellite borne computer according to claim 1, further comprising an error evaluation component for quantifying a systematic error, wherein the formula is: ; Wherein, the In the event of a systematic error in the system, Is the standard caliber error of the steel wire rod, In order to align and window the errors, Errors are mapped for the device.
8. A single event effect simulation and fault diagnosis method based on a satellite-borne computer, based on the single event effect simulation and fault diagnosis system based on a satellite-borne computer as set forth in any one of claims 1 to 7, comprising the steps of: S1, setting parameters of the peripheral model library and the simulation engine on the simulation host, loading the peripheral model library, and binding the bus and the interrupt controller; S2, setting target equipment, a target interface, a bit or bit segment range, an injection probability or occurrence rate interval, an injection time window, an injection level, a running round number and a random seed of the configuration and injection manager, and starting a cross-protocol mapping core; S3, the simulation engine executes simulation, defines event types according to discrete event semantics, and meets the requirements of ; S4, starting the protocol monitor and the log module, deriving the minimum sufficient record, storing the configuration abstract and the random seed, using the data set generator to convert the log into a data set under a unified label set, and dividing the data set into a training set, a verification set and a test set; S5, training the dynamic-attention-time sequence hybrid neural network model in the diagnosis server, and outputting a single event effect fault diagnosis result.

Description

Single event effect simulation and fault diagnosis system and method based on spaceborne computer Technical Field The invention relates to the field of reliability of an on-orbit electronic system of a spacecraft, in particular to a single event effect simulation and fault diagnosis system and method based on a spaceborne computer. Background Single event effect under space radiation environment often causes problems of bit overturning, transient, latching and the like of an on-board computer (OBC), thereby causing abnormal gesture and task control, and possibly causing system reset or failure in extreme cases. The traditional anti-radiation means (redundancy, shielding, beam current and irradiation test) have high cost, long period and difficult coverage of on-orbit complex coupling behaviors, and the existing system-level simulation tool has defects in the aspects of protocol time sequence, peripheral modeling and consistency of a label system, so that reproducible, extensible and on-orbit consistent diagnostic data are difficult to obtain. On the fault diagnosis side, the method based on rules/thresholds and traditional machine learning has limited time sequence characteristic expression capability of crossing protocols and crossing peripherals, the existing one-dimensional CNN or CNN+ attention model can extract local modes, but has insufficient capture of long-term dependence and low signal noise rare faults, and the characteristic migration capability of crossing protocol domains is weak due to fixed convolution kernels, the pure RNN/BiLSTM method has the problems of gradient attenuation and low long sequence efficiency, and the quasi-trans former model has strong data dependence, high calculation cost and unstable training under the condition of unbalance of short samples and strong classes. In addition, many methods do not display aligned protocol level timing semantics and unified tag space, making it difficult to obtain stable, interpretable diagnostic representations in a multi-protocol/multi-peripheral scenario. Disclosure of Invention The invention aims to provide a single event effect simulation and fault diagnosis system and method based on a spaceborne computer, which can uniformly describe the state machine, the register and the time sequence relation of control and communication type peripherals by using six-tuple abstraction and Discrete Event Semantics (DES) outside a simulation side, realize uniform labels by using cross-protocol fault mapping of position perception on a labeling side, provide a DAT (Dynamic-Attention-technology) hybrid neural network on a diagnosis side, cooperatively model cross-protocol long time dependence and low signal noise rare faults by using Dynamic convolution, attention re-weighting and bidirectional time sequence coding, are suitable for on-orbit working condition simulation and diagnosis of multiple protocols and multiple peripherals, have obvious advantages on precision and overall balance, and reflect stability and robustness. In order to achieve the above purpose, the invention provides a single event effect simulation and fault diagnosis system based on a spaceborne computer, which comprises a simulation host, a configuration and injection manager, a simulation engine and a simulation system, wherein the simulation host is used for loading a peripheral model library and running system-level simulation based on discrete event semantics, the simulation engine is arranged in a hardware environment of the simulation host, is respectively connected with a protocol monitor and a log module, and is provided with the peripheral model library; The peripheral model library is connected with an interrupt controller through a bus and an interface, the interrupt controller is in butt joint with the simulation engine, and the interrupt controller is used for receiving interrupt requests generated by each peripheral in the peripheral model library, carrying out shielding and priority arbitration on the interrupt requests, vectorizing the arbitrated interrupt into discrete interrupt events, and outputting the discrete interrupt events to the simulation engine so as to participate in discrete event scheduling of the simulation engine; The protocol monitor and the log module are connected with a data set generator, the data set generator is connected with a diagnosis server, a dynamic-attention-time sequence hybrid neural network model is arranged in the diagnosis server, and the diagnosis server is used for identifying faults caused by single event effect and outputting fault diagnosis results. Preferably, the simulation engine defines event types by adopting discrete event semantics, and gives propagation delay, logic delay and interrupt constraint to each peripheral, and the formula is as follows: ; Wherein, the Representing the delay of the combinational logic and register write to the observed output,Representing the link delay from the peripheral output to the op