Self-adaptive system-level fault self-diagnosis method
Through adaptive system-level fault self-diagnosis method, network topology properties and PMC model are used to dynamically identify fault nodes, which solves the problem of insufficient fault identification accuracy in large-scale multi-processor systems and realizes efficient and accurate fault diagnosis.
Patent Information
- Application Number
- CN202510758139.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-10-03
AI Technical Summary
In large-scale multi-processor systems, existing technologies have insufficient fault identification accuracy in complex network environments, resulting in heavy and redundant research work and difficulty in quickly diagnosing and locating faults.
An adaptive system-level fault self-diagnosis method is adopted. By determining the network topology properties, node interconnection rules and extra-branch connectivity, the PMC model is used for diagnostic testing. Combined with breadth-first search and random generation of fault nodes, the fault nodes are dynamically identified and a composite diagnostic standard is constructed.
It significantly improves the accuracy and efficiency of fault identification in complex network environments, reduces the misjudgment rate, and can identify faulty nodes with an accuracy of more than 99.75%.
Smart Images
Figure CN120743591A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault self-diagnosis methods, and in particular to an adaptive system-level fault self-diagnosis method. Background Art
[0002] Large-scale multiprocessor systems consist of interconnected processors, each with its own local memory. Information is communicated through one or more intermediate processors exchanging messages. The reliability of a multiprocessor system depends heavily on the network topology of its interconnections. As the number of processors and physical connections rapidly increases in any multiprocessor system based on network interconnections, processor and link failures are inevitable. Rapidly diagnosing, locating, detecting, and isolating these failures not only improves system robustness and reliability, but also provides a strong foundation for building a secure and reliable network environment.
[0003] Fault diagnosis technology for interconnected networks can be used to solve the compatibility problem between application communication modes and network topology structures. It is an important technical means to improve the communication performance of large-scale parallel applications. However, as the scale of high-performance parallel systems increases, existing diagnostic schemes consider only a single condition and require studying different networks one by one, resulting in heavy and redundant research work and insufficient fault identification accuracy in complex network environments. Summary of the Invention
[0004] The purpose of the present invention is to provide an adaptive system-level fault self-diagnosis method, which can improve the fault identification accuracy in complex network environments.
[0005] To achieve the above objectives, the present invention provides an adaptive system-level fault self-diagnosis method, comprising:
[0006] Based on the topological properties of the network, the number of nodes in the network, the interconnection rules between nodes and the degree of each node are determined, and the extra-branch connectivity is determined;
[0007] determining an extra-branch condition diagnosticity of the interconnection network based on the extra-branch connectivity;
[0008] Taking the extra-branch conditional diagnostic degree as the fault bound, randomly generate the threshold number of fault nodes;
[0009] Find the connected branches generated by the faulty node;
[0010] According to the relationship between the threshold of fault nodes and the size of connected branches, the PMC model is used to perform diagnostic tests and output the fault set.
[0011] The specific steps of determining the number of nodes in the network, the interconnection rules between nodes, and the degree of each node based on the topological properties of the network and determining the extra-branch connectivity include:
[0012] Remove different vertex subsets from the network and observe the connectivity of the network after removal. If the connected components formed reach a preset threshold and the number of nodes contained in each connected component reaches a preset threshold, then the vertex subset is used as an additional branch point cut set;
[0013] All extra-branch point cut sets that meet the conditions are screened, and the extra-branch point cut set with the smallest cardinality is taken as the extra-branch connectivity.
[0014] In the step of randomly generating the threshold number of fault nodes by taking the extra-branch conditional diagnostic degree as the fault bound:
[0015] A random function is used to randomly select a node as a fault node.
[0016] Among them, in the step of finding the connected branches generated by the faulty node:
[0017] The breadth-first search method is used to find the connected branches generated by the faulty node.
[0018] Specifically, the PMC model is used to perform diagnostic testing based on the relationship between the threshold of the fault node and the size of the connected branches. The specific steps for outputting the fault set include:
[0019] Perform the first round of diagnostic testing on the two nodes j and k participating in the mutual test. If σ(j,k)=0 and σ(k,j)=1, the PMC model determines that node j is faulty. The identified faulty node is placed in the fault set F1, where j is the test node and k is the tested node.
[0020] For the remaining unidentified connected branches, a second round of diagnostic test is performed. i , if |C i |≥f-|F1|, then C i The nodes in C are identified as fault-free nodes. i The surrounding neighboring nodes are identified as faulty nodes, and the identified faulty nodes are further placed in the fault set F1, and the fault set F1 is updated, where f represents the additional-branch condition diagnostic degree;
[0021] Among the branches that were not judged in the second round of diagnostic testing, select a connected branch C k , if |N(C k )|≥f-|F1|, then C k The nodes in C are identified as faulty nodes and k The nodes in the fault set F1 are continued to be put into the fault set F1, and the fault set F1 is updated again, where N(C k ) means C ksurrounding neighbor nodes;
[0022] The diagnostic test is completed and the fault set F1 is output.
[0023] This adaptive system-level fault self-diagnosis method integrates additional diagnostics and branch diagnostics to create a composite diagnostic standard that more closely reflects actual network fault characteristics. This method dynamically identifies faulty nodes through breadth-first search and incorporates a statistical verification strategy involving 1,000 repeated tests, significantly reducing the false positive rate. This method can improve the accuracy and efficiency of fault identification in complex network environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art.
[0025] Figure 1 It is a flow chart of the adaptive system-level fault self-diagnosis method of the present invention.
[0026] Figure 2 It is a fault test diagram of the PMC test model.
[0027] Figure 3 It is a schematic diagram of the situation where fault sets F1 and F2 can be distinguished under the PMC test model.
[0028] Figure 4 The present invention is a flow chart for determining the number of nodes in the network, the interconnection rules between nodes, the degree of each node, and the extra-branch connectivity based on the topological properties of the network.
[0029] Figure 5 The present invention uses a PMC model to perform diagnostic testing on the relationship between the threshold of a fault node and the size of a connected branch, and outputs a flow chart of a fault set. DETAILED DESCRIPTION
[0030] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be understood as limiting the present invention.
[0031] See also Figure 1-Figure 5 The present invention provides an adaptive system-level fault self-diagnosis method, comprising:
[0032] S100 determines the number of nodes in the network, the interconnection rules between nodes, and the degree of each node based on the topological properties of the network, and determines the extra-branch connectivity;
[0033] The specific steps include:
[0034] S101 removes different vertex subsets from the network and observes the connectivity of the network after the removal. If the connected components formed reach a preset threshold and the number of nodes contained in each connected component reaches a preset threshold, the vertex subset is used as an additional-branch point cut set;
[0035] S102 screens all extra-branch point cut sets that meet the conditions, and uses the extra-branch point cut set with the smallest cardinality as the extra-branch connectivity;
[0036] In an embodiment of the present invention, based on the topological properties of the network, key information such as the number of nodes in the network, the interconnection rules between nodes, and the degree of each node is determined, and then according to the definition of h additional r branch connectivity κ (h additional r branch connectivity means that after removing a fault set F, the remaining network has at least r branches, and each branch has at least h nodes. Among them, h emphasizes the number of branches in the remaining network; r emphasizes the size of each branch. When a certain number of fault nodes are removed from the network, it is hoped that the number of branches in the remaining network is as small as possible. On the other hand, in order to ensure the connectivity of the remaining network, it is of course hoped that these branches are as large as possible. Therefore, after removing certain nodes, the number of branches and the size of the largest branch can be used as a measure of network fault tolerance), an h additional r branch point cut set (extra-branch point cut set) that meets specific conditions is found. That is, try to remove different vertex subsets, observe the connectivity of the network after removal, and determine whether r connected components are formed, and each connected component contains at least h+1 nodes. Finally, screen all h additional r branch point cut sets that meet the conditions and find the additional branch point cut set with the smallest cardinality, which is the h additional r branch connectivity κ.
[0037] S200 determines the extra-branch condition diagnosticity of the interconnection network based on the extra-branch connectivity;
[0038] In the embodiment of the present invention, the h additional r branch connectivity κ is used as a bridge, and the extreme graph construction method and proof by contradiction are used to determine the h additional r branch conditional diagnostic degree f of the interconnected network. Specifically, the determination of the h additional r branch conditional diagnostic degree f requires the use of the PMC model, which is a test-based diagnostic model. Under this model, it is agreed that the probability of a test result of 0 when a fault-free processor tests a fault-free processor is 1; and the probability of a test result of 1 when a fault-free processor tests a faulty processor is 1 (see Figure 2); however, if the fault handler is used as a test node, the final test results will be unreliable regardless of whether the tested node is faulty or not. When some nodes in a network fail, the primary task is to consider whether the remaining network can operate normally. The h-extra r-branch connectivity parameter considers not only the number of branches in the remaining network but also the size of each branch. These two parameters are mutually constrained; for different topologies, changes in one inevitably lead to changes in the other. The basic idea is to fix one quantity and analyze the rate of change of the other. Therefore, for the PMC test model, this method proposes a new diagnostic parameter: h-extra r-branch diagnosticity. Building on branch connectivity, we investigate h-extra r-branch connectivity and further determine the h-extra r-branch diagnosticity of the network under the PMC model. The conclusion is reached by proving lower and upper bounds. The basic idea is to prove the upper bound using the extreme graph construction method. The key is to construct an independent substructure of size r+1 such that its outer neighbor set is as small as possible, subject to the constraint that its size does not exceed its branch connectivity. To prove the lower bound, we use the proof by contradiction to obtain two indistinguishable fault sets, and then prove that the intersection of these two indistinguishable fault sets is a cut set and calculate the upper bound of its size. Then, we use the structural properties to obtain a lower bound for the cut set. Due to the contradiction between the upper and lower bounds, the problem is proved. The details are as follows:
[0039] First, in the PMC test model, the two possible fault sets F1 and F2 can be distinguished and must meet Figure 3 The specific solutions are as follows:
[0040] Case 1: Node x∈F1\F2, node
[0041] If F1 is the fault set and F2 is the non-fault set, then x is the faulty node and y is the non-faulty node, Figure 2 It can be seen that y tests x, and the test result is 1; if F1 is a fault-free set and F2 is a fault set, then x and y are both fault-free nodes, and Figure 2 It can be seen that y tests x and the test result is 0; F1 and F2 can be distinguished by different symptoms.
[0042] Case 2: Node x∈F2\F1, node
[0043] If F1 is a non-fault set and F2 is a fault set, then x is a faulty node and y is a non-faulty node, Figure 2 It can be seen that y tests x, and the test result is 1; if F1 is the fault set and F2 is the fault-free set, then x and y are both fault-free nodes, and Figure 2 It can be seen that y tests x and the test result is 0; F1 and F2 can be distinguished by different symptoms.
[0044] Therefore, for the PMC model, if we want to determine the lower bound of the branch diagnosticity of the triangular interconnected network, we first assume that there is no Figure 3 The edges connecting the two structures, at this time, the nodes outside F1∪F2 can only be connected to the nodes within F1∩F2, then F1∩F2 is a cut set of the entire network. Considering whether there is a node in the symmetric difference between F1 and F2, a contradiction is deduced.
[0045] S300 uses the extra-branch conditional diagnostic degree as the fault bound and randomly generates the threshold number of fault nodes;
[0046] In this step, a random function is used to randomly select a node as the fault node.
[0047] In an embodiment of the present invention, a random function is used to randomly select a node as a fault node; with the help of the PMC test model, if the test node is a fault-free point and the tested node is also a fault-free node, the test result is 0; if the tested node is a faulty node, the test result is 1; however, if the fault point is used as the test node, the test result of the tested node is randomly 0 or 1 regardless of its state. Based on the bidirectional test results between nodes, an n×n 0-1 matrix is constructed to generate a symptom matrix.
[0048] S400 searches for connected branches generated by the faulty node;
[0049] In this step, the breadth-first search (BFS) method is used to find the connected branches generated by the faulty node.
[0050] S500 uses the PMC model to perform diagnostic tests based on the threshold of the fault node and the size relationship of the connected branches, and outputs the fault set;
[0051] The specific steps include:
[0052] S501 performs the first round of diagnostic testing. For two nodes j and k participating in the mutual test, if σ(j, k) = 0 and σ(k, j) = 1, node j is determined to be faulty through the PMC model and the identified faulty node is placed in the fault set F1.
[0053] S502 performs the second round of diagnostic testing. For the remaining unidentified connected branches, any one branch C is selected. i , if |C i |≥f-|F1|, then C i The nodes in C are identified as fault-free nodes. i The neighboring nodes are identified as faulty nodes, and the identified faulty nodes are added to the fault set F1, and the fault set F1 is updated;
[0054] S503 performs the third round of diagnostic testing, and selects a branch C from the branches that were not determined in the second round of diagnostic testing. k , if |N(C k )|≥f-|F1|, then C k The nodes in C are identified as faulty nodes and k The nodes in the fault set F1 are continued to be put into the fault set F1, and the fault set F1 is updated again, where N(C k ) means C k surrounding neighbor nodes;
[0055] S504 The diagnostic test is completed and the fault set F1 is output.
[0056] This method utilizes a Python simulation environment and conducts simulation experiments on a high-performance computer. Faulty nodes are randomly assigned to the network. Simulations are then conducted using the proposed diagnostic method to determine the method's runtime. The experimental results and data are analyzed to verify the method's accuracy. The proposed method is compared with other diagnostic strategies in terms of operational efficiency and fault tolerance to identify deficiencies and continuously optimize the network's fault diagnosis strategy. This approach undergoes 1000 iterations of experiments using experimental parameters such as accuracy, recall, false positive rate, and true negative rate. The experiments demonstrate that the accuracy of faulty node identification (ACCR) for the proposed three-round adaptive fault diagnosis strategy is consistently above 99.75%. This demonstrates that the proposed adaptive fault diagnosis method is optimal. This method offers a flexible diagnostic scope, allowing for both local and global diagnosis. It can be applied when the maximum number of diagnosable faulty nodes within the diagnostic scope (i.e., the conditional diagnostic degree) is known.
[0057] This adaptive system-level fault self-diagnosis method combines h-extra diagnostics (requiring each branch in the remaining network to contain at least h+1 nodes) with r-branch diagnostics (requiring at least r connected branches in the post-fault network), creating a composite diagnostic standard that more closely reflects the characteristics of actual network faults. This method dynamically identifies faulty nodes through breadth-first search and incorporates a statistical verification strategy involving 1,000 repeated tests, significantly reducing the false positive rate and improving the accuracy and efficiency of fault identification in complex network environments.
[0058] The above disclosure is merely one or more preferred embodiments of the present application and is not intended to limit the scope of the present application. A person skilled in the art will understand that all or part of the processes of the above embodiments and equivalent changes made in accordance with the claims of the present application are still within the scope of the present application.
Claims
1. An adaptive system-level fault self-diagnosis method, characterized in that: include: Based on the topological properties of the network, the number of nodes in the network, the interconnection rules between nodes and the degree of each node are determined, and the extra-branch connectivity is determined; determining an extra-branch condition diagnosticity of the interconnection network based on the extra-branch connectivity; Taking the extra-branch conditional diagnostic degree as the fault bound, randomly generate the threshold number of fault nodes; Find the connected branches generated by the faulty node; According to the relationship between the threshold of fault nodes and the size of connected branches, the PMC model is used to perform diagnostic tests and output the fault set.
2. The adaptive system-level fault self-diagnosis method according to claim 1, characterized in that: Based on the topological properties of the network, the number of nodes in the network, the interconnection rules between nodes, and the degree of each node are determined. The specific steps for determining the extra-branch connectivity include: Remove different vertex subsets from the network and observe the connectivity of the network after removal. If the connected components formed reach a preset threshold and the number of nodes contained in each connected component reaches a preset threshold, then the vertex subset is used as an additional branch point cut set; All extra-branch point cut sets that meet the conditions are screened, and the extra-branch point cut set with the smallest cardinality is taken as the extra-branch connectivity.
3. The adaptive system-level fault self-diagnosis method according to claim 2, characterized in that: In the step of randomly generating the threshold number of fault nodes by taking the extra-branch condition diagnostic degree as the fault bound: A random function is used to randomly select a node as a fault node.
4. The adaptive system-level fault self-diagnosis method according to claim 3, wherein: In the step of finding the connected branches generated by the faulty node: The breadth-first search method is used to find the connected branches generated by the faulty node.
5. The adaptive system-level fault self-diagnosis method according to claim 4, characterized in that: Based on the relationship between the threshold of the fault node and the size of the connected branches, the PMC model is used to perform diagnostic testing. The specific steps of outputting the fault set include: Perform the first round of diagnostic testing on the two nodes j and k participating in the mutual test. If σ(j,k)=0 and σ(k,j)=1, the PMC model determines that node j is faulty. The identified faulty node is placed in the fault set F1, where j is the test node and k is the tested node. For the remaining unidentified connected branches, a second round of diagnostic test is performed. i , if |C i |≥f-|F1|, then C i The nodes in C are identified as fault-free nodes. i The surrounding neighboring nodes are identified as faulty nodes, and the identified faulty nodes are further placed in the fault set F1, and the fault set F1 is updated, where f represents the additional-branch condition diagnostic degree; Among the branches that were not judged in the second round of diagnostic testing, select a connected branch C k , if |N(C k )|≥f-|F1|, then C k The nodes in C are identified as faulty nodes and k The nodes in the fault set F1 are continued to be put into the fault set F1, and the fault set F1 is updated again, where N(C k ) means C k surrounding neighbor nodes; The diagnostic test is completed and the fault set F1 is output.