Sensor network health self-evaluation and reconstruction method and system applied to intelligent ship

Abstract

The application provides a sensor network health self-evaluation and reconstruction method and system applied to an intelligent ship, relates to the technical field of intelligent ship sensor network management, and first collects a set of original response signals of each sensor node in the sensor network; then analyzes the original response signal set to obtain signal drift trend characteristics and generate inter-node response difference characteristics; generates a sensor network health state topology graph based on the signal drift trend characteristics and the inter-node response difference characteristics, which contains the health state degradation level of each node and the availability identification of the inter-node function replacement path; determines a network topology reconstruction strategy for the set of failed nodes according to the sensor network health state topology graph; and finally converts the network topology reconstruction strategy into a set of reconstruction execution instructions and distributes them to the sink node to trigger the network topology reconfiguration operation, thereby improving the reliability and adaptability of the intelligent ship sensor network.

Description

Technical Field

[0001] This application relates to the field of intelligent ship sensor network management technology, and more specifically, to a method and system for self-assessment and reconstruction of sensor network health applied to intelligent ships. Background Technology

[0002] Sensor networks are critical infrastructure for acquiring information about the structural status of a ship, and their stable and reliable operation is essential for ensuring safe navigation. Currently, traditional sensor network management methods primarily focus on the initial deployment of sensor nodes and routine data collection, lacking effective means for assessing and maintaining the health status of the sensor network. On the one hand, most existing technologies only focus on individual performance indicators of sensor nodes, such as simple data accuracy checks, failing to comprehensively consider the interrelationships between sensor nodes and the impact of the entire network topology on health status, making it difficult to comprehensively and accurately assess the overall health of the sensor network. On the other hand, when node failures occur in the sensor network, existing methods often lack flexible and effective reconstruction strategies, typically only allowing for simple node replacement or network repair, unable to dynamically adjust according to the actual health status and functional requirements of the network. This leads to problems such as data acquisition interruptions and inaccurate information when facing complex and ever-changing ship operating environments, thus affecting the safe operation and decision-making of the ship. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method and system for self-assessment and reconstruction of the health of sensor networks applied to intelligent ships.

[0004] According to a first aspect of this application, a method for self-assessment and reconstruction of the health of a sensor network applied to intelligent ships is provided, the method comprising: The raw response signal set of each sensor node in the sensor network deployed in the intelligent ship hull structure monitoring area is collected. The raw response signal set contains multi-dimensional physical quantity sensing data streams with acquisition time stamps. The original response signal set is subjected to signal drift situation analysis to obtain the signal drift situation characteristics of each sensor node, and the response dissimilarity characteristics between nodes are generated based on the signal amplitude deviation of different sensor nodes in the original response signal set under the same acquisition time mark. Based on the signal drift characteristics and the inter-node response dissimilarity characteristics, a sensor network health status topology map representing the overall health status of the sensor network is generated through a preset node health status topology deduction logic. The sensor network health status topology map includes the health status degradation level of each sensor node and the availability identifier of functional alternative paths between nodes. Based on the set of failed nodes whose health status degradation level exceeds a preset degradation threshold in the sensor network health status topology diagram and the availability identifier of the functional alternative paths between the nodes, a network topology reconstruction strategy is determined for the set of failed nodes. The network topology reconfiguration strategy is converted into a set of reconfiguration execution instructions that includes routing table update instructions and node working parameter adjustment instructions. The set of reconfiguration execution instructions is then distributed to the aggregation node of the sensor network to trigger a network topology reconfiguration operation.

[0005] According to a second aspect of this application, a sensor network health self-assessment and reconstruction system for intelligent ships is provided. The sensor network health self-assessment and reconstruction system for intelligent ships includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the sensor network health self-assessment and reconstruction system for intelligent ships implements the aforementioned sensor network health self-assessment and reconstruction method for intelligent ships.

[0006] Based on any of the above aspects, the technical effect of this application is as follows: By collecting the raw response signal sets of each sensor node within the sensor network, a comprehensive multi-dimensional physical quantity sensing data stream with acquisition time stamps is obtained. Signal drift state analysis is performed on the raw response signal sets to accurately obtain the signal drift state characteristics of each sensor node. Simultaneously, based on the signal amplitude deviation of different sensor nodes under the same acquisition time stamp, inter-node response dissimilarity characteristics are generated. Based on these characteristics, a sensor network health status topology map is generated, presenting the health status degradation level of each sensor node and the availability of functional alternative paths between nodes. Based on this sensor network health status topology map, a network topology reconstruction strategy for the set of failed nodes is determined and converted into a set of reconstruction execution instructions, which are distributed to the aggregation node to trigger network topology reconfiguration. This allows for dynamic adjustment of the network structure according to the real-time health status of the network, ensuring stable and reliable operation of the sensor network even when node failures occur, effectively improving the robustness and adaptability of the intelligent ship sensor network. Attached Figure Description

[0007] Figure 1 A flowchart illustrating the sensor network health self-assessment and reconstruction method for intelligent ships provided in this application embodiment is shown. Figure 2 This paper illustrates a schematic diagram of the component structure of a sensor network health self-assessment and reconstruction system for intelligent ships provided in an embodiment of this application. Detailed Implementation

[0008] Figure 1This paper illustrates a flowchart of a sensor network health self-assessment and reconstruction method for intelligent ships provided in an embodiment of this application. The detailed steps include: This embodiment uses an intelligent ship hull structure health monitoring sensor network as an application scenario to describe the technical solution of the present invention in detail. In this embodiment, the sensor network consists of multiple sensor nodes deployed in key monitoring areas of the ship's hull structure. These sensor nodes include vibration sensors, stress sensors, and temperature sensors, and each sensor node is connected to a convergence node via a wireless communication protocol. The sensor network covers key stress-bearing parts of the ship's hull structure and is used for real-time monitoring of the hull structure's health status. All sensor data acquisition and processing are conducted with the authorization of the ship operator, and the data is used only for ship structural safety assessment and maintenance decisions.

[0009] Step S110: Collect the raw response signal set of each sensor node in the sensor network deployed in the intelligent ship hull structure monitoring area. The raw response signal set contains multi-dimensional physical quantity sensing data streams with acquisition time stamps.

[0010] The aggregation node sends data acquisition commands to each sensor node in the sensor network at a fixed polling period P. A star network topology is established between the aggregation node and each sensor node via a low-power wireless communication protocol based on the IEEE 802.15.4 standard. The aggregation node maintains a list of sensor node addresses. At the beginning of each polling period, the aggregation node unicasts a data acquisition request frame to each sensor node in the address list order. The header of the data acquisition request frame includes the target sensor node address field and the request sequence number field. Upon receiving the data acquisition request frame, the sensor node encapsulates all the sensor sampling data stored in its data buffer since the time it was polled in the previous polling period into a data response frame. The payload of the data response frame includes a sensor node identification field, a data acquisition time stamp sequence field, and a multi-dimensional physical quantity sensing data stream field. The multi-dimensional physical quantity sensing data stream field contains subfields for multiple physical quantity dimensions. Each subfield stores the sequence of sampled values ​​corresponding to that physical quantity type. The physical quantity types include vibration acceleration amplitude, structural stress value, and temperature value. Each acquisition timestamp corresponds to a set of multi-dimensional physical quantity sensing data sample values. The acquisition timestamp is the timestamp value recorded by the real-time clock inside the sensor node at the sampling moment. After sending all data acquisition request frames in this round, the aggregation node starts a receive timeout timer. Before the timeout timer expires, it continues to receive data response frames returned by each sensor node and summarizes the payload content of all received data response frames into a raw response signal set.

[0011] Step S120: Perform signal drift situation analysis on the original response signal set to obtain the signal drift situation characteristics of each sensor node, and generate the inter-node response dissimilarity characteristics based on the signal amplitude deviation of different sensor nodes in the original response signal set under the same acquisition time mark.

[0012] Signal drift situation analysis is performed separately for each physical quantity dimension of each sensor node. The aim is to separate the baseline drift component reflecting the slow changes in the sensor node's own sensing characteristics from long-term acquired multi-dimensional physical quantity data, and to identify signal drift situation features characterizing sensor node performance degradation. The inter-node response dissimilarity feature is based on the redundant design principle that multiple sensor nodes within the same physical monitoring sub-region should produce similar response amplitudes to the same physical stimulus. The consistency of sensor node operating status is evaluated by quantifying the degree of deviation between the response amplitudes of different sensor nodes.

[0013] Step S121: Arrange the multi-dimensional physical quantity sensing data stream of each sensor node in the original response signal set according to the acquisition time mark to generate a multi-dimensional time-series signal array for each sensor node. Perform signal baseline drift separation processing on the physical quantity sensing data stream of each dimension in the multi-dimensional time-series signal array to extract the signal baseline drift component and signal transient fluctuation component corresponding to the physical quantity of that dimension.

[0014] For the d-th dimension physical quantity sensing data stream of the i-th sensor node, the aggregation node extracts all sampled data of the sensor node in that physical quantity dimension from the original response signal set, arranges them in ascending order according to the acquisition time stamp, and forms a one-dimensional vector Xid of length T. Xid is the time-series vector of the physical quantity sensing data stream of that sensor node in that physical quantity dimension. The signal baseline drift separation processing adopts a scattered smoothing decomposition algorithm based on local weighted regression to decompose Xid into a slowly changing signal baseline drift component Bid and a rapidly changing signal transient fluctuation component Rid. The decomposition process is as follows: for each time index t in Xid, take all data points in a time neighborhood window with a time window radius of W as the center, and assign a Gaussian weight value w(k)=exp(-(tk-tt)2 / (2σ2)) to each data point in the window according to the time distance of the data point from the center time point, where σ is the Gaussian kernel bandwidth parameter. We fit a local straight line to the data points within the window using weighted least squares. The estimated value of this local straight line at the center time index t is Bid(t), and Rid(t) = Xid(t) - Bid(t). We perform the above operation for all time indices to obtain the complete Bid and Rid vectors.

[0015] Step S122: Based on the trend direction change sequence and trend inflection point density of the signal baseline drift component within the continuous acquisition time-marked interval, construct a signal drift situation trajectory line describing the long-term evolution trend of the sensor node sensing characteristics. Input the trend direction change sequence in the signal drift situation trajectory line into a preset trend direction change detector to obtain the change slope amplitude and the trend duration after the change corresponding to the trend direction change moment in the signal drift situation trajectory line. Encapsulate the change slope amplitude and the trend duration after the change into the signal drift situation feature.

[0016] The signal baseline drift component vector Bid forms a one-dimensional curve on the time axis, which is the signal drift trajectory. A discrete sampling sequence is taken along the signal drift trajectory at a fixed time step S. For each adjacent sampling point, the first-order difference value ΔBid(k) = Bid(tk+1) - Bid(tk) is calculated. If ΔBid(k) > 0, the trend direction between the adjacent sampling points is marked as Ddir(k) = 1 (rising); if ΔBid(k) < 0, Ddir(k) = -1 (falling); if ΔBid(k) = 0, Ddir(k) = 0 (flat). An interval of consecutive sampling points with the same Ddir value is divided into a trend segment. When the Ddir values ​​of two adjacent trend segments are different, it is a trend inflection point. The trend inflection point density ρturn = Nturn / Ttotal, where Nturn is the total number of trend inflection points and Ttotal is the total time span of the signal drift trajectory.

[0017] A pre-defined trend direction abrupt change detector calculates the abrupt change slope magnitude Kjump = |slope_after - slope_before| for each trend inflection point, where slope_before is the slope of the trend segment preceding the inflection point, and slope_after is the slope of the trend segment following the inflection point. Kjump is compared with a pre-defined abrupt change slope threshold Kth. If Kjump > Kth, the trend inflection point is determined to be a trend direction abrupt change point, and its abrupt change time tjump, abrupt change slope magnitude Kjump, and the duration of the trend after the abrupt change Djump are recorded. The Kjump and Djump values ​​of all trend direction abrupt change points across all physical quantity dimensions of the sensor node are arranged in order of physical quantity dimension and encapsulated into a feature vector Fdrift. Fdrift is the signal drift state feature vector.

[0018] Step S123: Extract the signal amplitude output by sensor nodes deployed in the same physical monitoring sub-area under the same acquisition time mark from the original response signal set, and calculate the absolute deviation value of the signal amplitude between each pair of sensor nodes. Perform statistical distribution analysis on the absolute deviation values ​​of the signal amplitude between all pairs of sensor nodes under each acquisition time mark, and extract the deviation median, which represents the central tendency of the deviation value distribution, and the deviation interquartile range parameter, which represents the dispersion of the deviation value distribution.

[0019] There are Na sensor nodes within the same physical monitoring sub-region. For each acquisition time marker t, the signal amplitude Xi(t) of the main monitored physical quantity dimension is extracted from each sensor node. The absolute deviation value Dij(t) = |Xi(t) - Xj(t)| between any two different sensor node signal amplitudes is calculated, where i and j traverse from 1 to Na and i ≠ j, resulting in a total of Npair = Na × (Na-1) / 2 absolute deviation values. These absolute deviation values ​​are arranged in ascending order as a sequence Dordered(t). The median deviation Dmed(t) = Dordered(t)[ceil(Npair × 0.5)], where ceil represents rounding up. The interquartile range parameter Diqr(t) = Dordered(t)[ceil(Npair × 0.75)] - Dordered(t)[ceil(Npair × 0.25)].

[0020] Step S124: Construct a deviation distribution evolution time series based on the median deviation and the interquartile range of deviation under multiple consecutive acquisition time marks, and perform fluctuation pattern clustering processing on the deviation distribution evolution time series. Divide acquisition time mark segments with similar deviation distribution evolution trends into the same deviation fluctuation pattern cluster. Calculate the mode value of the absolute deviation values ​​of the signal amplitude between all pairs of sensor nodes under all acquisition time marks in each deviation fluctuation pattern cluster, and use the mode value as the reference value of the inter-node response dissimilarity corresponding to the deviation fluctuation pattern cluster.

[0021] M two-dimensional vectors Vpair(t) = [Dmed(t), Diqr(t)] are formed by combining the median deviation Dmed(t) and the interquartile range parameter Diqr(t) of the deviations from M consecutive acquisition time markers. These vectors are arranged in ascending order of the acquisition time markers to form a time series of deviation distribution evolution. A hierarchical agglomerative clustering algorithm based on dynamic time warping distance is applied to this time series of deviation distribution evolution. Initially, each two-dimensional vector is treated as an independent cluster. In each iteration, the dynamic time warping distance between all pairs of clusters is calculated, and the two clusters with the smallest distance are merged. This iteration is repeated until the dynamic time warping distance between any two clusters is greater than the clustering stopping threshold Dstop. For each cluster, the absolute deviation values ​​Dij(t) of all Npair groups of sensor nodes under all time markers within the cluster are summarized into a set. The frequency of each deviation value in this set is counted, and the deviation value with the highest frequency is taken as the mode Dmode. Dmode is the benchmark value for the inter-node response dissimilarity of this cluster.

[0022] Step S125: Based on the benchmark value of inter-node response dissimilarity and the slope change direction of the time series of deviation distribution evolution corresponding to the deviation fluctuation mode cluster, generate an inter-node response dissimilarity feature vector indexed by the time boundary of the deviation fluctuation mode cluster. Each element in the inter-node response dissimilarity feature vector contains a dissimilarity amplitude component and a dissimilarity change trend component.

[0023] For each deviation fluctuation mode cluster, Dmode is taken as the dissimilarity amplitude component A. Linear regressions are performed on Dmed(t) and Diqr(t) for the time series segment of the deviation distribution evolution corresponding to this cluster, yielding the slopes βmed and βiqr of the two regression lines. The value rule for the dissimilarity trend component is: if (βmed+βiqr) / 2>0, then Tdir=1; if (βmed+βiqr) / 2<0, then Tdir=-1; if (βmed+βiqr) / 2=0, then Tdir=0. The start and end acquisition time markers of this cluster are used as time boundary indices, and [A, Tdir] is encapsulated as a feature element under this index. The feature elements of all deviation fluctuation mode clusters are arranged in chronological order according to the time boundaries to form the inter-node response dissimilarity feature vector.

[0024] Step S130: Based on the signal drift characteristics and the inter-node response dissimilarity characteristics, a sensor network health status topology map representing the overall health status of the sensor network is generated through a preset node health status topology deduction logic. The sensor network health status topology map includes the health status degradation level of each sensor node and the availability identifier of functional alternative paths between nodes.

[0025] Step S131: Obtain the spatial deployment coordinates and the initial adjacency topology of each sensor node in the sensor network. The initial adjacency topology includes the communication link connection relationship and the communication link quality parameters between adjacent sensor nodes.

[0026] The spatial deployment coordinates of the sensor nodes are the installation position coordinates (Xi, Yi, Zi) of each sensor node in the hull three-dimensional coordinate system extracted from the sensor layout diagram of the ship design drawing. The initial adjacency topology is parsed from the network configuration file of the sensor network and includes an Na×Na adjacency relationship matrix A and an Na×Na communication link quality parameter matrix Q. In the adjacency relationship matrix A, Aij = 1 indicates that there is a direct wireless communication link between sensor node i and sensor node j, and Aij = 0 indicates that there is no such link. Each element in the communication link quality parameter matrix Q stores two communication link quality parameters: the received signal strength indication value RSSIij and the packet error rate value PERij. These parameters are used for the subsequent calculation of the functional coupling strength coefficient.

[0027] Step S132: According to the mutation slope amplitude and the duration of the trend after mutation included in the signal drift trend characteristics, perform normalization processing respectively. After mapping them to the same dimensionless scoring scale, calculate the drift risk score of each sensor node by weighted calculation, and compare the drift risk score with the preset multi-level degradation threshold interval to determine the initial health state degradation level of each sensor node.

[0028] Extract the mutation slope amplitude values of all physical quantity dimensions from the signal drift trend feature vector Fdrift of the sensor node, and take the maximum value Kmax_node as the representative mutation slope amplitude of the sensor node. Perform normalization processing: Knorm = Kmax_node / Kref, where Kref is the slope reference value pre-calibrated according to the physical characteristics of the sensor type to which the sensor node belongs. Extract all the durations of the trends after mutation from Fdrift, and take the maximum value Dmax_node. Perform normalization processing: Dnorm = Dmax_node / Dref, where Dref is the duration reference value calibrated according to the designed life of the sensor node. The drift risk score Sdrift = α×Knorm + β×Dnorm, where α and β are preset weight coefficients and α + β = 1. The preset multi-level degradation threshold interval is defined by four thresholds T1 < T2 < T3 < T4, dividing the scoring range into five intervals: Sdrift < T1 is the healthy level, T1 ≤ Sdrift < T2 is the sub-healthy level, T2 ≤ Sdrift < T3 is the mild degradation level, T3 ≤ Sdrift < T4 is the moderate degradation level, and Sdrift ≥ T4 is the severe degradation level. Determine the initial health state degradation level of the sensor node according to the interval in which Sdrift falls.

[0029] Step S133: Extract the inter-node response dissimilarity benchmark value corresponding to the most recent deviation fluctuation mode cluster from the inter-node response dissimilarity feature vector, and construct an inter-node dissimilarity matrix that measures the consistency of the current response between different sensor nodes.

[0030] The inter-node response dissimilarity baseline value, Dmode_recent, is taken from the inter-node response dissimilarity feature vector, with the latest time boundary index value. The inter-node dissimilarity matrix, Dinter, is of size Na×Na. The assignment rule for matrix elements Dinter(i,j) is as follows: if sensor node i and sensor node j belong to the same physical monitoring sub-region and a dissimilarity baseline value exists in the sensor node pair corresponding to Dmode_recent, then Dinter(i,j) takes that baseline value; otherwise, Dinter(i,j) takes the arithmetic mean of all known dissimilarity baseline values ​​as the fill value. The smaller the value of Dinter(i,j), the more consistent the responses of the two sensor nodes are.

[0031] Step S134: Normalize the magnitude components of the dissimilarity of each pair of sensor nodes in the dissimilarity matrix and the communication link quality parameters of the pair of sensor nodes in the initial adjacency topology. After mapping the two to the same evaluation scale, perform fusion weighting to generate the inter-node functional coupling strength coefficient that reflects the functional correlation between nodes.

[0032] Normalization of the dissimilarity magnitude component: Dnorm(i,j)=Dinter(i,j) / Dmax, where Dmax is the maximum value of all elements in the dissimilarity matrix between nodes. The communication link quality parameters include the received signal strength indication value RSSIij and the packet error rate value PERij, which are normalized as follows: Rnorm(i,j)=RSSIij / RSSImax, where RSSImax is the theoretical maximum received signal strength indication value; Pnorm(i,j)=1-PERij, where PERij ranges from 0 to 1. The overall normalized value of the communication link quality is Qnorm(i,j)=(Rnorm(i,j)+Pnorm(i,j)) / 2. The inter-node functional coupling strength coefficient Cfunc(i,j)=λ1×(1-Dnorm(i,j))+λ2×Qnorm(i,j), where λ1 and λ2 are preset weighting coefficients and λ1+λ2=1. The more consistent the responses and the higher the communication quality between two sensor nodes, the greater their functional coupling strength coefficient, indicating that one sensor node is more capable of replacing the function of the other sensor node.

[0033] Step S135: Traverse all sensor nodes in the sensor network, take the currently traversed sensor node as the target node, and filter out the adjacent sensor nodes whose inter-node functional coupling strength coefficient with the target node exceeds the preset coupling threshold Cth, thus forming a set of functional replacement candidate nodes for the target node.

[0034] For the target node, iterate through all sensor nodes j in the sensor network except for target. If Cfunc(target, j) > Cth and the current initial health status degradation level of sensor node j is healthy or sub-healthy, then add sensor node j to the set of candidate nodes for functional replacement of target, Calternate(target). Cth is a preset coupling threshold; the larger the value, the more stringent the functional requirements for the replacement node.

[0035] Step S136: For each candidate node in the set of candidate nodes for functional replacement of the target node, based on the initial health status degradation level of the candidate node and the functional coupling strength coefficient between the candidate node and the target node, evaluate the functional transfer loss parameter when the candidate node carries the sensing function of the target node, integrate the functional transfer loss parameters of all candidate nodes in the set of candidate nodes for functional replacement of the target node, generate an ordered list of functional replacement paths for the target node in ascending order of functional transfer loss parameters, and assign an availability identifier to each functional replacement path in the ordered list of functional replacement paths.

[0036] The function transfer loss parameter Lk for candidate node k is calculated as Lk = γ × GradeCost(Hk) + δ × (1 - Cfunc(target, k)), where Hk is the initial health degradation level of candidate node k, and GradeCost(Hk) is the preset cost value for the level based on the initial health degradation level Hk, with the lowest cost value for the healthy level, followed by the sub-healthy level, and so on. γ and δ are preset weighting coefficients, and γ + δ = 1. A smaller Lk value indicates a smaller function transfer loss when replacing the target node with candidate node k. All candidate nodes are arranged in ascending order of Lk value to form an ordered list of function replacement paths. For each path, if its Lk value is less than the preset maximum allowable loss value Lmax, the path is marked as available; otherwise, it is marked as unavailable.

[0037] Step S137: Combine the initial health status degradation level of each sensor node with the ordered list of alternative functional paths for each sensor node to obtain a set of node-level health status information containing the health status degradation levels of all sensor nodes and the availability identifiers of alternative functional paths between nodes. Map the set of node-level health status information onto a two-dimensional planar grid using the spatial deployment coordinates of the sensor network, and connect the corresponding node pairs through the availability identifiers of alternative functional paths between nodes to generate the health status topology map of the sensor network.

[0038] The two-dimensional planar mesh uses the ship's deck-view projection plane as the reference plane, with a mesh resolution of Lgrid. The spatial deployment coordinates of each sensor node are projected onto the plane and mapped to the nearest mesh cell. A sensor node symbol is drawn at this mesh cell, and the initial health degradation level of the node is labeled. For each functional alternative path with available identifiers, a dashed arrow connects the corresponding target sensor node and a candidate sensor node in the topology diagram, with the arrow pointing towards the candidate sensor node. The function transfer loss parameter Lk value for this path is labeled on the dashed arrow. All sensor node symbols, health degradation level labels, and functional alternative path connecting lines and labels together constitute the sensor network health topology diagram.

[0039] Step S140: Based on the set of failed nodes whose health status degradation level exceeds a preset degradation threshold in the sensor network health status topology map and the availability identifier of the functional alternative paths between the nodes, determine the network topology reconstruction strategy for the set of failed nodes.

[0040] Step S141: Traverse all sensor nodes in the sensor network health status topology diagram, mark sensor nodes whose health status degradation level exceeds the preset degradation threshold as failed nodes, and mark sensor nodes whose health status degradation level does not exceed the preset degradation threshold as healthy nodes.

[0041] The preset degradation threshold is set to mild degradation. Sensor nodes with an initial health degradation level of moderate or severe degradation are marked as failed nodes and included in the failed node set. Sensor nodes with an initial health degradation level of healthy, sub-healthy, or mild degradation are marked as healthy nodes and included in the healthy node set.

[0042] Step S142: Take each failed node in the set of failed nodes as a node to be reconstructed, extract the healthy nodes pointed to by the functional alternative paths with availability identifiers connected to the node to be reconstructed from the sensor network health status topology map, and form a list of available alternative nodes for the node to be reconstructed.

[0043] In the sensor network health status topology diagram, starting from the node to be reconstructed, trace along the direction of the dashed arrow of the functional alternative path, filter out the paths whose availability is marked as available and whose end sensor nodes belong to the set of healthy nodes, and collect the end healthy nodes of the above paths into the list of available alternative nodes Aavail.

[0044] Step S143: Calculate the number of inter-node functional alternative paths currently carried by each healthy node in the list of available alternative nodes, calculate its theoretical maximum carrying capacity score based on the initial communication link quality parameters of the healthy node, and reduce its theoretical maximum carrying capacity score according to the number of inter-node functional alternative paths already carried to obtain the remaining functional carrying capacity value of the healthy node.

[0045] The number of functional alternative paths currently carried by a healthy node m, Ncur(m), represents the number of functional alternative paths for which the healthy node has been used as an alternative node in the sensor network health status topology. The theoretical maximum carrying capacity score is Cmax(m) = Qnorm_avg(m) × Cref, where Qnorm_avg(m) = (RSSIm_avg / RSSImax + (1-PERm_avg)) / 2, RSSIm_avg and PERm_avg are the average received signal strength indication value and average packet error rate value between the healthy node and all adjacent sensor nodes, respectively, and Cref is the maximum number of alternative paths that a standard sensor node can carry. The remaining functional carrying capacity value Crem(m) = max(0, Cmax(m) - Ncur(m)).

[0046] Step S144: Obtain the set of neighboring nodes of the node to be reconstructed in the sensor network health status topology graph, and select the available replacement node with the largest sum of the functional coupling strength coefficients between the node and the healthy node in the set of neighboring nodes from the list of available replacement nodes as the main replacement node.

[0047] The set of neighboring nodes Nneighbor of the node to be reconstructed is the intersection of the set of sensor nodes and the set of healthy nodes that have direct communication links with the node to be reconstructed in the initial adjacency topology. For each available replacement node p in the list of available replacement nodes Aavail, calculate F(p) = sum(Cfunc(p, q)), where q iterates through all healthy nodes in the set of neighboring nodes Nneighbor. The available replacement node with the largest F(p) value is selected as the primary replacement node Pmain.

[0048] Step S145: Based on the spatial deployment coordinates of the primary replacement node, calculate the communication link redirection path for each healthy node in the set of adjacent nodes of the node to be reconstructed received by the primary replacement node, and generate an updated communication link routing relationship.

[0049] For each healthy node q in the neighboring node set Nneighbor, the Dijkstra's shortest path algorithm is used to calculate the optimal communication path from healthy node q to the primary replacement node Pmain on the RSSI value matrix of the communication link quality parameter matrix Q. This optimal communication path is a sequence of nodes connected in series, with the starting point being healthy node q and the ending point being the primary replacement node Pmain. The updated communication link routing relationship modifies the next-hop address field in the routing table entry for healthy node q, which originally used the failed node as the next-hop destination address, to the network address of the next-hop sensor node of healthy node q in this optimal communication path.

[0050] Step S146: Adapt the operating parameters of the main replacement node. The adaptation adjustment includes increasing the signal acquisition frequency of the main replacement node and expanding the signal sensing threshold range of the main replacement node, thereby obtaining the operating parameter adjustment instruction of the main replacement node.

[0051] The original signal acquisition frequency of the primary replacement node is f0. The new signal acquisition frequency after adaptation adjustment is fnew = f0 × (1 + Nfail / Ntotal), where Nfail is the number of all failed nodes currently replaced by the primary replacement node, and Ntotal is the total number of sensor nodes that the primary replacement node originally needed to acquire. The original signal sensing threshold range of the primary replacement node is [θL0, θH0], and the new signal sensing threshold range after adaptation adjustment is [θL0 × η, θH0], where η is an expansion coefficient less than 1, used to extend the lower limit of sensing to a lower range to adapt to the increased range of physical quantities required during replacement monitoring. The instruction for adjusting the working parameters of the primary replacement node includes updated values ​​for the signal acquisition frequency parameter and the signal sensing threshold range parameter.

[0052] Step S147: For the remaining available alternative nodes in the list of available alternative nodes other than the primary alternative node, group the set of adjacent nodes of the node to be reconstructed according to the remaining functional carrying capacity value of the remaining available alternative nodes, and assign a corresponding backup alternative node to each group of adjacent nodes to generate an auxiliary routing path.

[0053] All candidate nodes in the available alternative node list, excluding the primary alternative node Pmain, are sorted in descending order of their remaining functional capacity values ​​to obtain the backup alternative node sequence. A greedy allocation algorithm is used to group the set of adjacent nodes Nneighbor of the node to be reconstructed: Each backup alternative node is sequentially selected from the backup alternative node sequence, and its remaining functional capacity value is used as the upper limit of the maximum number of adjacent nodes that the backup alternative node can support. For the current backup alternative node, several adjacent nodes with the largest inter-node functional coupling strength coefficient Cfunc value are selected from the unassigned adjacent nodes in Nneighbor. The number of these adjacent nodes is not greater than the remaining functional capacity value of the backup alternative node. These adjacent nodes are grouped together, and the current backup alternative node is designated as the endpoint of its auxiliary route for this group. This process is repeated until all adjacent nodes in Nneighbor are assigned to a group. For each group of adjacent nodes, the shortest communication path from each adjacent node in the group to the designated backup alternative node in that group is calculated. This shortest communication path is the auxiliary route path for that adjacent node. The secondary routing path serves as a backup to the primary routing path and is automatically switched on and activated by the aggregation node when the communication link between the primary and secondary nodes is interrupted.

[0054] Step S148: Combine the communication link redirection path, the auxiliary routing path, and the working parameter adjustment instruction to form a single node reconstruction strategy for the node to be reconstructed.

[0055] The single-node reconstruction strategy is a structure containing a field for the node to be reconstructed, a field for the primary replacement node, a field for the communication link redirection path list, a field for the auxiliary routing path list, and a field for the working parameter adjustment instruction. The communication link redirection path list field stores the optimal communication paths from each neighboring node to the primary replacement node generated in step S145. The auxiliary routing path list field stores the auxiliary routing paths from each neighboring node to the corresponding backup replacement node generated in step S147. The working parameter adjustment instruction field stores the working parameter adjustment instructions for the primary replacement node generated in step S146. After filling the above field values ​​into the structure, it is serialized into a binary data block, which serves as the single-node reconstruction strategy for the node to be reconstructed.

[0056] Step S149: Summarize the single-node reconstruction strategies of each failed node in the set of all failed nodes, and resolve the conflicts of single-node reconstruction strategies that share the same replacement node. Merge conflicting routing entries and uniformly adjust the working parameter adjustment instructions of the shared replacement node to obtain a network topology reconstruction strategy covering the entire set of failed nodes.

[0057] Iterate through the single-node reconstruction strategies of all failed nodes in the set of failed nodes, and check whether the single-node reconstruction strategies of different failed nodes contain the same primary replacement node or backup replacement node. For multiple single-node reconstruction strategies sharing the same replacement node, the conflict resolution process includes: checking whether there are channel conflicts using the same communication channel time slots in the communication link redirection paths of these strategies; if so, reallocating the communication time slots of the conflicting paths in a time-division multiplexing manner and recording their respective communication time slot offsets; merging duplicate routing table entries and removing mapping records from identical destination addresses to next-hop addresses. For the operating parameter adjustment instructions of the shared replacement node, the signal acquisition frequency of the shared replacement node is the sum of the frequencies required by each single-node reconstruction strategy, the signal sensing threshold range is the union of the ranges required by each strategy (the lower limit is the minimum of all required lower limits, and the upper limit is the maximum of all required upper limits), the signal acquisition frequency is the sum of the frequency increments required by all failed nodes plus the original acquisition frequency, the lower limit of the signal sensing threshold range is the minimum of all lower limits, and the upper limit is the maximum of all upper limits. The routing relationships and working parameters after resolution constitute the network topology reconstruction strategy.

[0058] Step S150: Convert the network topology reconfiguration strategy into a set of reconfiguration execution instructions that includes routing table update instructions and node working parameter adjustment instructions, and distribute the set of reconfiguration execution instructions to the aggregation node of the sensor network to trigger the network topology reconfiguration operation.

[0059] Step S151: Extract all descriptive information for communication link redirection paths and auxiliary routing paths from the network topology reconstruction strategy, convert the descriptive information into a next-hop routing mapping table record from the perspective of the aggregation node, compress and encode the next-hop routing mapping table record, and generate an incremental routing table update instruction for updating the internal routing forwarding table of the aggregation node. The incremental routing table update instruction includes an operation type identifier, a destination address field, and a next-hop address field.

[0060] The description information for communication link redirection paths and auxiliary routing paths is in the form of node sequences. For each node sequence, starting from the sink node, the target sensor node address that the sink node needs to forward and the corresponding next-hop sensor node address are extracted. All extracted (target address, next-hop address) pairs are paired and organized into a next-hop routing mapping table record. The next-hop routing mapping table record is compressed and encoded, using dictionary encoding to compress recurring sensor node addresses, replacing the complete address string with a short integer index. The encoded incremental routing table update instruction is a structured binary message containing an operation type identifier field (value indicating whether a routing table entry is added or overwritten), a target address field, and a next-hop address field, with multiple records arranged sequentially. This incremental routing table update instruction is used to add or overwrite routing table entries in the sink node's internal routing forwarding table.

[0061] Step S152: Extract all working parameter adjustment instructions for the primary replacement node and other designated replacement nodes from the network topology reconstruction strategy. Group the working parameter adjustment instructions according to the target node identifier to obtain a node working parameter adjustment instruction group corresponding to each replacement node. Add an instruction validity period timestamp and instruction version number to each working parameter adjustment instruction in the node working parameter adjustment instruction group to generate a traceable parameter adjustment instruction with timeliness and version control functions.

[0062] The working parameter adjustment instructions are extracted from the working parameter adjustment instruction field of the primary replacement node in the network topology reconstruction strategy. These instructions are grouped according to the sensor node identifier of the target node (i.e., each replacement node), with all working parameter adjustment instructions belonging to the same replacement node grouped together. For each working parameter adjustment instruction in a group, an instruction validity period timestamp is appended. This timestamp equals the current system time plus a preset instruction validity duration, indicating when the instruction was valid. Instructions that expire and are not executed will be automatically discarded by the aggregation node. An instruction version number is also appended, a monotonically increasing integer value that increments each time a new adjustment instruction is generated for the working parameters of the replacement node. The instruction validity period timestamp and the instruction version number together constitute the timeliness and version control metadata of the traceable parameter adjustment instructions, which is embedded in the original working parameter adjustment instruction message to form traceable parameter adjustment instructions.

[0063] Step S153: Aggregate the incremental routing table update instruction and the traceable parameter adjustment instruction by instruction type to form the set of reconstruction execution instructions indexed by the instruction target node.

[0064] The reconstructed execution instruction set adopts a key-value storage structure. The key is the sensor node identifier of the instruction target node, and the value is the instruction list corresponding to that sensor node. The instruction list contains two types of instructions: if the sensor node is the aggregation node itself, its instruction list contains incremental routing table update instructions; if the sensor node is a replacement node, its instruction list contains traceable parameter adjustment instructions. The key-value storage structure allows for quick retrieval of all instructions corresponding to each instruction target node.

[0065] Step S154: Search for currently active aggregation nodes in the sensor network, establish an instruction delivery channel with the aggregation nodes, and split and deliver the reconstruction execution instruction set to the corresponding aggregation nodes according to the instruction target node identifier.

[0066] Aggregator nodes periodically broadcast active status beacon frames to the sensor network. These frames contain the aggregation node identifier and current load status information. Aggregator nodes with loads below a preset load threshold are selected from the received active status beacon frames as targets for instruction delivery. A reliable data transmission channel is established between the aggregation nodes, and instructions are delivered using a transmission control protocol. Based on the aggregation node associated with the target node identifier of each instruction in the reconstructed execution instruction set, the instructions are split and distributed to the corresponding aggregation nodes. Each aggregation node only receives instructions related to the sensor nodes it is responsible for.

[0067] Step S155: After each aggregation node receives the set of reconstruction execution instructions, it parses the incremental routing table update instructions through the communication protocol stack between the aggregation node and the instruction target node, and writes new routing table entries or overwrites old routing table entries in the routing forwarding table inside the aggregation node.

[0068] Each aggregation node maintains a routing table. Each record in the routing table contains a destination address field, a next-hop address field, and a routing table entry status field. The aggregation node parses incremental routing table update commands and determines the operation type based on the identifier: if the operation type is "add," a new record is added to the routing table, the destination address and next-hop address fields are filled in, and the routing table entry status field is set to valid; if the operation type is "overwrite," an existing record with the same destination address field as the command is found in the routing table, its next-hop address field is updated to the next-hop address field specified in the command, and the routing table entry status field remains valid. After the routing table update is complete, the aggregation node returns a routing table update confirmation message to the control terminal.

[0069] Step S156: After each aggregation node completes the routing table entry update, the traceable parameter adjustment instruction is transmitted through the aggregation node to the corresponding instruction target node. The instruction target node parses the traceable parameter adjustment instruction and modifies its own signal acquisition frequency parameter and signal sensing threshold range parameter. After the instruction target node completes the modification of the working parameters, the instruction target node reports a parameter effectiveness confirmation message to the aggregation node. After the aggregation node summarizes the parameter effectiveness confirmation messages of all instruction target nodes, it broadcasts a topology update completion notification to the sensor network.

[0070] The aggregation node checks the validity period timestamp of the traceable parameter adjustment command. If the current system time is earlier than the validity period timestamp, the command is valid and is passed through the communication protocol stack to the command target node. If the validity period timestamp has expired, the command is invalid, discarded, and an expiration notification is generated and sent to the network topology reconfiguration policy initiator. Upon receiving the traceable parameter adjustment command, the command target node parses the updated values ​​for the signal acquisition frequency parameter and the signal sensing threshold range parameter, writes them into its configuration register, overwriting the original parameter values. After writing the parameters, the command target node performs a self-check to confirm that the new parameters have taken effect, and then reports a parameter effectiveness confirmation message to the aggregation node. This message includes the sensor node's identifier, the effective parameter type, and the effectiveness timestamp. The aggregation node collects parameter effectiveness confirmation messages from all command target nodes. Once all confirmation messages have been collected, the aggregation node generates a topology update completion notification message and broadcasts it to all sensor nodes in the sensor network, informing them that the network topology reconfiguration operation is complete. Each sensor node switches to the new network topology state upon receiving the topology update completion notification.

[0071] Step S210: Collect the real-time voltage and current signals of the power supply branches of each sensor node in the sensor network, and generate the instantaneous power consumption time-series tracking sequence of each sensor node based on the real-time voltage and current signals. Perform power mode decomposition processing on the instantaneous power consumption time-series tracking sequence, extract the power consumption transient pulse component that presents a pulse shape and the power consumption static offset component that presents a stable offset shape in the power mode decomposition processing, and combine the power consumption transient pulse component and the power consumption static offset component to obtain the power consumption feature spectrum of the sensor node.

[0072] A shunt resistor is connected in series on the power supply branch of each sensor node. A real-time current signal I is obtained by measuring the voltage across the shunt resistor using a differential amplifier. Simultaneously, a real-time voltage signal U is obtained by directly measuring between the positive and negative terminals of the power supply. Multiplying U and I at the same moment yields the instantaneous power consumption value for that moment. Arranging the instantaneous power consumption values ​​in chronological order constitutes an instantaneous power consumption time-series tracking sequence. Power consumption mode decomposition processing employs a wavelet transform multi-resolution analysis algorithm, selecting the Daubechies wavelet basis function to perform multi-level decomposition of the sequence, obtaining wavelet coefficients at different frequency scales. Reconstructing the high-frequency coefficient layer yields a transient power consumption pulse component exhibiting a pulse shape, reflecting the instantaneous spike characteristics of the sensor node during sudden high-power events such as wireless transmission and analog-to-digital conversion. Reconstructing the low-frequency coefficient layer yields a static power consumption offset component exhibiting a stable offset shape, reflecting the slow drift of the baseline power consumption of the sensor node due to power supply aging and temperature changes. The peak amplitude, pulse width, and pulse repetition frequency of the power transient pulse component are combined with the offset slope and offset direction of the power static offset component to form a feature vector, which is the power characteristic spectrum Fp.

[0073] Step S220: Perform spectral deviation analysis on the power consumption characteristic spectrum and the preset historical power consumption benchmark spectrum of sensor nodes to obtain the power supply status change score value of each sensor node. Use the power supply status change score value as a health influence factor, which is a supplementary dimension alongside the health status degradation level. Then, add the health influence factor to the node attributes of the corresponding sensor node in the sensor network health status topology diagram.

[0074] The historical power consumption baseline spectrum of the sensor node is the power consumption characteristic spectrum reference value Fb collected and stored by the node during the factory calibration period according to step S210. The spectrum deviation analysis uses a joint measurement method of cosine similarity and Euclidean distance to calculate the cosine similarity and Euclidean distance between Fp and Fb. The cosine similarity value and Euclidean distance value are normalized and then weighted and summed to obtain the power supply status anomaly score value Sp = α × (1 - cosθ) / 2 + β × (d / dmax), where α and β are preset weight coefficients, d is the Euclidean distance, and dmax is the normalized reference value for the Euclidean distance. Sp is written as a health impact factor into the node attribute supplementary field of the grid cell where the node is located in the sensor network health status topology diagram, and displayed as an auxiliary annotation next to the node symbol.

[0075] Step S230: Obtain the abnormal code sequence returned by the internal hardware self-test status register of each sensor node in the sensor network, and generate a node hardware vulnerability indicator function based on the frequency and timing distribution of the setting of specific error bit masks in the abnormal code sequence.

[0076] Each sensor node's microcontroller contains a hardware self-test status register. This register automatically executes a built-in self-test program upon power-on reset and writes the result in bitmask form. Sensor nodes periodically send their current register value to the aggregation node via status reporting messages. The aggregation node stores all abnormal code sequences from each sensor node within a continuous reporting period. For each sensor node, the setting frequency *fr* of each error bitmask is obtained by dividing the total number of times each error bitmask is set by the total number of reports. Simultaneously, the time interval sequence between two consecutive set points of each error bitmask is recorded, and the mean *μt* and standard deviation *σt* of the interval sequence are calculated as time-series distribution parameters. The node hardware vulnerability indicator function is defined as Hv(b) = fr(b) × exp(-σt / μt), where *b* is the error bitmask index; a larger function value indicates a more severe hardware vulnerability.

[0077] Step S240: Receive the spectral energy distribution data sent by the environmental vibration sensor deployed on the hull of the smart ship, and use the cumulative vibration energy in the spectral energy distribution data that matches the inherent resonant frequency band of the sensor node as the long-term physical damage induction parameter of the node.

[0078] Environmental vibration sensors are installed on the main deck structure of the ship's hull. They collect vibration acceleration signals of the hull structure at a fixed frequency and perform Fast Fourier Transform (FFT) to output vibration energy distribution data for each frequency band. The inherent resonant frequency band of the sensor node is obtained from the technical specifications provided by the sensor node manufacturer. This band [fl, fh] represents the frequency range [fl, fh] where the sensor node's outer shell and internal circuit board resonate and amplify under mechanical vibration excitation. The vibration energy values ​​of all frequency bands falling within the [fl, fh] interval in the spectral energy distribution data are summed to obtain the cumulative vibration energy Ev. Multiplying Ev by the cumulative operating time Twork of the sensor node yields the long-term physical damage induction parameter Dphy = Ev × Twork, where Dphy represents the degree of physical damage suffered by the sensor node due to long-term cumulative hull vibration.

[0079] Step S250: Construct a multi-source anomaly signal fusion model that combines the coupled power supply status anomaly score, node hardware hidden danger indicator function, and node long-term physical damage induced parameter. Analyze each anomaly signal independently using the multi-source anomaly signal fusion model and output the corresponding node health status deterioration correction factor.

[0080] The multi-source anomaly signal fusion model is a parallel structure of three independent analysis channels. The first channel receives Sp and maps it to a power supply degradation correction factor Δ1 using a preset power supply anomaly mapping function. The second channel receives Hv, takes the maximum value of Hv(b) from all error bitmasks as the representative value, and maps it to a hardware degradation correction factor Δ2 using a preset hardware vulnerability mapping function. The third channel receives Dphy and maps it to a physical degradation correction factor Δ3 using a preset physical damage mapping function. All three mapping functions are piecewise linear mapping functions; the larger the input value, the larger the output correction factor value.

[0081] Step S260: Based on the multiple node health status degradation correction factors obtained for each node, a comprehensive health status degradation coefficient is obtained through weighted fusion; based on the comprehensive health status degradation coefficient, the interval boundary of the multi-level degradation threshold interval used to determine the initial health status degradation level in the node health status topology deduction logic is adjusted; based on the adjusted multi-level degradation threshold interval, the revised health status degradation level of each sensor node is recalculated, and the sensor network health status topology is updated according to the revised health status degradation level.

[0082] The overall health status degradation coefficient Δtotal = w1×Δ1 + w2×Δ2 + w3×Δ3, where w1, w2, and w3 are preset weighting coefficients. Subtracting Δtotal from the preset multi-level degradation threshold interval boundary values ​​T1 to T4 in step S132 yields the adjusted new threshold boundaries T1' to T4'. The drift risk score Sdrift of each sensor node calculated in step S132 is re-compared with T1' to T4' to obtain the revised health status degradation level. The revised health status degradation level replaces the original level label of the corresponding node in the sensor network health status topology diagram, completing the topology diagram update.

[0083] Step S310: Obtain sensor node failure event logs generated by the sensor network during historical operating cycles. The sensor node failure event logs include node failure timestamps and spatial deployment coordinates of the failed nodes.

[0084] Step S320: Based on the node failure timestamp and the spatial deployment coordinates of the failure nodes, draw a scatter plot of the spatiotemporal distribution of failure events on the time series axis and the three-dimensional model space of the ship structure. Perform spatiotemporal density clustering analysis on the scatter plot of the spatiotemporal distribution of failure events to extract spatiotemporal hotspot regions where the cluster density of failure events exceeds the clustering threshold. Perform common damage cause tracing analysis on the failure nodes contained in each spatiotemporal hotspot region. By comparing the temporal clustering of the trend direction change moment in the signal drift characteristics of the failure nodes in the spatiotemporal hotspot region, determine the spatiotemporal propagation direction vector of the failure propagation path.

[0085] Spatiotemporal density clustering analysis employs a spatiotemporal scanning statistical algorithm. A cylindrical scanning window is constructed with each failure event as its center, a temporal radius of Rt, and a spatial radius of Rs. The ratio of the number of failure events within the window to the expected number is used as the cluster density. The clustering threshold is set as the critical value for the log-likelihood ratio significance test. Regions with cluster densities exceeding the threshold are identified as spatiotemporal hotspots. For each failure node within a hotspot, the abrupt change times in the signal drift characteristics of each node are extracted. After sorting by time, the spatial direction connecting the earliest and latest abrupt change times is taken as the spatiotemporal propagation direction vector of the failure propagation path.

[0086] Step S330: Construct a directed topology graph of failure propagation with the spatiotemporal propagation direction vector as the edge and the sensor node as the node. Calculate the causal source strength value of node failure along the propagation direction in the directed topology graph of failure propagation, and extract the set of key root cause nodes that trigger cascading failures based on the causal source strength value. Collect the microenvironmental parameters of the physical installation location of each key root cause node in the set of key root cause nodes. The microenvironmental parameters include the stress value of the installation base, the local environmental corrosion rate, and the impedance fluctuation amplitude of the connecting cable.

[0087] In the directed topology graph, nodes are sensor nodes. If the angle between the line connecting node A to node B and the spatiotemporal propagation direction vector is less than a preset angle threshold, a directed edge from A to B is established. The causal tracing strength value is the difference between the in-degree and out-degree of a node; the larger the difference, the more likely it is to be a root cause node. Nodes with causal tracing strength values ​​exceeding a preset tracing threshold constitute the set of critical root cause nodes.

[0088] Step S340: Compare each of the microenvironment parameters with a preset threshold to obtain the independent accelerated failure level of each parameter; combine the independent accelerated failure levels of each parameter and obtain the estimated remaining effective service time of the key root cause node through a preset failure level-remaining time mapping model.

[0089] The threshold for the stress value of the mounting base is the yield stress of the sensor mounting base material multiplied by a safety factor. The actual measured value is divided by this threshold to obtain the stress ratio. The stress ratio is divided into multiple intervals, each corresponding to an independent accelerated failure level. The threshold for the local environmental corrosion rate is the maximum annual average corrosion rate allowed by the hull structure design. The actual corrosion rate is divided by this threshold to obtain the corrosion rate ratio, and the accelerated corrosion failure level is determined according to the same interval division rules. The threshold for the impedance fluctuation amplitude of the connecting cable is a fixed percentage of the nominal impedance value of the sensor signal cable at the factory. The actual impedance fluctuation value is divided by this threshold to obtain the impedance fluctuation ratio, and the impedance accelerated failure level is determined according to the same rules. The comprehensive accelerated failure level is obtained by weighting and summing the three independent accelerated failure levels according to preset weights and then rounding. The preset failure level-remaining service time mapping model is a two-dimensional mapping table stored in the aggregation node. The first column is the comprehensive accelerated failure level, and the second column is the estimated remaining effective service time. The corresponding estimated remaining effective service time is obtained by looking up the table based on the comprehensive accelerated failure level.

[0090] Step S350: Replace the preset failure time array of sensor nodes belonging to the set of critical root cause nodes in the sensor network health status topology diagram with the estimated remaining effective service time of the critical root cause nodes.

[0091] The failure time preset array is a pre-defined field in the attributes of each sensor node in the sensor network health status topology diagram, used to record the expected failure time of the sensor node. The original default value in this field is replaced by the estimated remaining effective service time of the key root cause node obtained in step S340. After the replacement, the health status degradation prediction of the sensor node is more accurate.

[0092] Step S360: Before generating the network topology reconstruction strategy, nodes in the key root cause node set whose estimated remaining effective service time is lower than the preset task cycle are marked as potential failure nodes in advance, and the potential failure nodes are added to the failure node set.

[0093] The preset mission period is the total planned duration of the current voyage performed by the intelligent ship, from departure to return. The estimated remaining effective service time of each critical root cause node is compared with the preset mission period. If the estimated remaining effective service time is less than the preset mission period, it indicates that the critical root cause node is highly likely to fail within the current mission period. Therefore, it is marked as a node to be expected to fail in advance, and its sensor node identifier is added to the set of failed nodes.

[0094] Step S370: Take the proposed failure node and the original failure node in the set of failure nodes as input objects of the network topology reconstruction strategy to generate a forward-looking network topology reconstruction strategy that simultaneously covers both failure nodes and proposed failure nodes.

[0095] The expanded set of failed nodes from step S360 is used as input for step S140. Following the processing flow of steps S141 to S149, a unified network topology reconstruction strategy is generated for both the original failed nodes and the nodes slated for failure. This forward-looking network topology reconstruction strategy considers nodes that are about to fail during reconstruction, avoiding the need to trigger reconstruction again due to the actual failure of nodes slated for failure during navigation, thus improving the long-term stability of the network topology.

[0096] Step S410: After the aggregation node executes the set of reconstruction execution instructions, it receives steady-state response signal samples reported by each sensor node under the new topology of the sensor network, and obtains the response energy spectrum distribution of the new topology by performing frequency domain energy spectrum analysis on the steady-state response signal samples.

[0097] After the topology reconfiguration operation in step S150 is completed at the aggregation node, a preset stabilization convergence time is waited for all sensor nodes to stabilize under the new topology. Once all sensor nodes are operating stably, the aggregation node sends a steady-state response signal sampling command to all sensor nodes. Upon receiving the sampling command, the sensor nodes collect a fixed-duration sensing signal and report it to the aggregation node. The aggregation node performs a Fast Fourier Transform on the collected steady-state response signal samples, converting the time-domain signal into a frequency-domain representation to obtain the energy amplitude distribution of each frequency component, thus forming the response energy spectrum distribution of the new topology.

[0098] Step S420: Extract the spatial coverage of the physical monitoring sub-region covered by the set of failed nodes from the sensor network health status topology map, and calculate the signal sensing coverage of the physical monitoring sub-region by the new topology based on the spatial deployment coordinates of the sensor nodes that assume the alternative function in the new topology and their preset effective physical sensing radius.

[0099] The physical monitoring sub-region covered by the set of failed nodes consists of multiple polygonal monitoring sub-regions formed by clustering of each failed node under the original topology. The effective physical sensing radius is the maximum spatial distance at which a sensor node can accurately sense physical quantities under standard operating conditions. A sensing circle is generated on the ship's structural plane, centered on the spatial deployment coordinates of each sensor node undertaking the replacement function and with its effective physical sensing radius as the radius. The union of the sensing circles of all replacement nodes within the same physical monitoring sub-region is taken as the signal sensing coverage area of ​​that sub-region.

[0100] Step S430: Perform coverage difference boundary extraction processing on the signal sensing coverage area and the historical signal sensing coverage area before the failure of the set of failed nodes to obtain spatial boundary contour data of the sensing hole area. Based on the spatial boundary contour data of the sensing hole area, locate the sensing hole frequency band energy attenuation curve corresponding to the spatial coordinates of the sensing hole area in the response energy spectrum distribution, and extract the attenuation slope and cutoff frequency of the sensing hole frequency band energy attenuation curve.

[0101] The historical signal sensing coverage area is the sensing coverage area generated by the original position coordinates and original effective physical sensing radius of the failed node before it failed. A spatial difference operation is performed between the historical coverage area and the signal sensing coverage area; the spatial region located within the historical coverage area but not belonging to the signal sensing coverage area is the sensing void region. Edge contour extraction is performed on the sensing void region to obtain spatial boundary contour data. This spatial boundary contour data is mapped to the spatial index of the response energy spectrum distribution, and the energy values ​​of each frequency at the corresponding spatial location are extracted to construct the energy attenuation curve of the sensing void frequency band. The attenuation slope is the linear fitting slope of the curve in the descending segment of the logarithmic coordinate system, and the cutoff frequency is the frequency value corresponding to the energy value on the curve attenuating to a fixed percentage of the peak energy.

[0102] Step S440: Call the preset signal reconstruction compensation algorithm library, select the corresponding signal reconstruction compensation algorithm according to the attenuation slope and cutoff frequency, and use the signal reconstruction compensation algorithm to perform inverse filtering compensation processing on the signal in the sensing hole frequency band to generate a signal compensation dataset for the sensing hole region.

[0103] The preset signal reconstruction compensation algorithm library is a collection of multiple algorithm modules, each corresponding to a combination of attenuation slope intervals and cutoff frequency intervals. Based on the attenuation slope and cutoff frequency extracted in step S430, a matching interval combination is retrieved from the algorithm library, and the corresponding algorithm module is selected. Inverse filtering compensation processing constructs an inverse filter transfer function based on the attenuation characteristics, performing frequency domain compensation amplification on the signal in the sensing hole frequency band. The compensated frequency domain signal is then subjected to inverse Fourier transform to obtain the compensated time domain signal dataset.

[0104] Step S450: Align the signal compensation dataset with the spatial boundary contour data of the sensing cavity region using spatial coordinates and fuse the data to generate a complete monitoring area signal field distribution map without gaps in the sensing cavity.

[0105] The signal compensation dataset contains signal values ​​obtained after inverse filtering compensation of spatial coordinate points within the sensing void area. The spatial boundary contour data of the sensing void area is a closed polygonal boundary. The spatial coordinate alignment operation matches the spatial coordinates of each data point in the signal compensation dataset with the grid coordinates of the sensing void area, establishing a mapping relationship from spatial coordinates to signal values. The data fusion operation is based on the actual measured signal data within the signal sensing coverage area of ​​the new topology. At the spatial location of the sensing void area, the compensated signal values ​​from the signal compensation dataset are used to fill and replace the data. The two are smoothly transitioned at the spatial boundary using a weighted gradual fusion method. The weight of the compensated signal value inside the boundary decreases with increasing distance, while the weight of the measured signal value outside the boundary increases with decreasing distance. After fusion, the sensing void is filled, generating a continuous signal field distribution map covering the entire physical monitoring sub-area.

[0106] Step S460: Obtain the current working condition task sequence of the intelligent ship, and parse the key structural safety monitoring requirements from the working condition task sequence. The key structural safety monitoring requirements include the lower limit requirement of monitoring point density and the mandatory monitoring area range of the specified structural parts.

[0107] The operational task sequence is arranged and distributed to the aggregation node by the ship's central control system based on the current navigation mission. The operational task sequence is a structured message containing a task number field, a task type field, and a task parameter field. The critical structural safety monitoring requirements are parsed from the task parameter field. The minimum monitoring point density requirement is the minimum number of sensors covered per square meter. The mandatory monitoring area for specified structural parts is the set of spatial coordinates of the structural parts in the hull 3D model that require key monitoring.

[0108] Step S470: Perform a demand matching analysis between the signal field distribution map of the complete monitoring area of ​​the seamless sensing cavity and the key structural safety monitoring requirements, identify the mandatory monitoring area range of the specified structural parts where the signal sensing capability cannot meet the lower limit requirement of the monitoring point density, and when there is a mandatory monitoring area range of the specified structural parts that does not meet the requirements, generate a mobile sensing unit scheduling instruction and send the mobile sensing unit scheduling instruction to the inspection equipment equipped with mobile sensors, and schedule the inspection equipment to move to the mandatory monitoring area range of the specified structural parts to enhance the local monitoring capability.

[0109] The demand matching degree analysis uses the mandatory monitoring area of ​​a specified structural part as the analysis domain. In the signal field distribution map of the complete monitoring area, the actual sensor node coverage number of each grid cell within this analysis domain is counted and compared grid-by-grid with the lower limit requirement for monitoring point density. Grid cells with a coverage number lower than the lower limit requirement are marked as non-compliant units. The spatial union of all non-compliant units constitutes the area that does not meet the demand. The mobile sensing unit scheduling instruction includes the spatial boundary coordinate sequence of the non-compliant area and the scheduling execution priority, and is sent to the inspection equipment by the aggregation node via wireless communication. After receiving the instruction, the inspection equipment calculates the optimal movement path using the shortest path planning algorithm based on its current positioning coordinates and the spatial boundary coordinates of the non-compliant area. After moving along the path to the designated area, it activates the mobile sensors to perform supplementary monitoring.

[0110] Step S510: Monitor the communication protocol data stream of the sensor network in real time, and capture the health status beacon frames that each sensor node periodically uploads to the aggregation node from the communication protocol data stream. The health status beacon frames carry the sensor node identifier and the current health status degradation level.

[0111] Health status beacon frames are maintenance frames generated by each sensor node at a fixed beacon period and sent to the aggregation node. The frame header of the beacon frame contains a beacon type identifier field, and the frame body contains a sensor node identifier field and a current health status degradation level field. The aggregation node receives the communication protocol data stream at the physical layer, parses it layer by layer through the media access control layer and the network layer, and then captures all health status beacon frames at the application layer by frame type filtering.

[0112] Step S520: Arrange the health status degradation levels carried by the health status beacon frames received in multiple consecutive cycles in a temporal sequence to construct a node health degradation rate sequence, and calculate the transition step size value of the health status degradation level between adjacent cycles in the node health degradation rate sequence.

[0113] For the i-th sensor node, the health status degradation levels reported by the sensor node in each period are arranged into a discrete time series according to the time sequence of the beacon period. Let the health status degradation levels be integer coded values: healthy level is coded as 1, sub-healthy level as 2, mild degradation level as 3, moderate degradation level as 4, and severe degradation level as 5. The transition step size from the k-th period to the (k+1)-th period is Δk = Hk+1 - Hk. A positive transition step size indicates an increase in degradation level (i.e., deterioration of health status), a negative transition step size indicates a decrease in degradation level (i.e., recovery of health status), and zero transition step size indicates no change in degradation level.

[0114] Step S530: Perform abnormal transition pattern recognition processing on the transition step value. When the transition step value exceeds the preset upper limit of the normal transition step threshold, generate a health status deterioration alarm message for the sensor node.

[0115] The upper limit of the normal transition step size threshold is set as the maximum allowable increase in the health status degradation level per unit time. The transition step size value is compared with the upper limit of the normal transition step size threshold; if it exceeds the upper limit, the sensor node is determined to have experienced a sudden health status deterioration event. The health status deterioration sudden event alarm message includes a sensor node identifier field, a sudden event occurrence cycle index field, a transition step size value field, and an alarm timestamp field. The aggregation node pushes this alarm message to the ship monitoring and management terminal.

[0116] Step S540: Intercept the regular monitoring data messages sent by sensor nodes to the aggregation node within the normal transition step size threshold range, extract the multi-dimensional physical quantity sensing data stream from the application layer payload of the regular monitoring data message, and perform information entropy estimation processing on the multi-dimensional physical quantity sensing data stream to obtain the sensing data information entropy value.

[0117] Routine monitoring data messages are sensor data messages periodically sent by sensor nodes under normal operating conditions without triggering abnormal alarms. Information entropy is calculated for each dimension of the multi-dimensional physical quantity sensing data stream. The data value interval of each dimension is equally divided into several discrete data bins, and the frequency of data points appearing in each data bin is used as the probability value. The information entropy is calculated as Hentropy = -sum(pm × log2(pm)), where pm is the probability value of the m-th data bin, and log2 is the logarithm to the base 2. The average of the information entropies of all physical quantity dimensions is taken as the perceived data information entropy value.

[0118] Step S550: Perform correlation analysis between the information entropy value of the sensing data and the signal reliability score calculated based on the signal drift situation characteristics of the corresponding sensor node. Evaluate the information contribution of the sensor node's sensing data based on the results of the correlation analysis, and mark sensor nodes with information contribution values ​​lower than a preset contribution threshold as inefficient nodes.

[0119] The signal reliability score Rreliability = exp(-Knorm × Dnorm), where Knorm is the maximum value of the normalized mutation slope amplitude and Dnorm is the maximum value of the normalized trend duration. Association analysis substitutes the perceived data information entropy value Hentropy and the signal reliability score Rreliability into the information contribution evaluation function to calculate the information contribution Icontribute = Hentropy × Rreliability. The preset contribution threshold Imin is the lower limit of the contribution. If Icontribute < Imin, the sensor node is marked as an inefficient node.

[0120] Step S560: Before generating the network topology reconstruction strategy, compare and remove duplicates from the inefficient node set and the failed node set, and construct a comprehensive set of nodes to be optimized that includes failed nodes and inefficient nodes.

[0121] Perform a union operation on the inefficient node set marked in step S550 and the failed node set marked in step S141. For sensor nodes that appear in both sets, only keep one record, and the elements of the two sets after duplicate removal are combined to form a comprehensive set of nodes to be optimized.

[0122] Step S570: Obtain the node betweenness centrality metric value of each node to be optimized in the sensor network health state topology graph of the comprehensive set of nodes to be optimized, and preferentially select the nodes to be optimized with a node betweenness centrality metric value lower than the preset centrality threshold, and set them as the preferred replacement nodes.

[0123] The calculation formula for the node betweenness centrality metric value Cb(n) is Cb(n) = sum(σst(n) / σst) summed over all pairs of sensor nodes (s, t), where σst is the total number of all shortest functional replacement paths from sensor node s to sensor node t in the sensor network health state topology graph, σst(n) is the number of paths passing through sensor node n among these shortest paths, and s and t traverse all sensor nodes and s ≠ t ≠ n. The higher the calculated Cb(n) value, the more important the hub role of sensor node n in the functional replacement path network. Sort the Cb values of all nodes to be optimized in the comprehensive set of nodes to be optimized from smallest to largest, and select the nodes to be optimized within the preset selection ratio range after sorting as the preferred replacement nodes. The Cb values of the preferred replacement nodes are relatively low, and replacing or putting these nodes to sleep has less impact on the topological connectivity of the entire functional replacement path network.

[0124] Step S580: Add a node sleep flag bit to the working parameter adjustment instruction for the preferred replacement nodes, and through the setting operation of the node sleep flag bit, make the preferred replacement nodes enter the low-power sleep state after receiving the working parameter adjustment instruction.

[0125] For each priority replacement node, a dedicated operating parameter adjustment instruction carrying the node sleep flag is generated. The node sleep flag is a bit in the control field of this instruction message; setting this bit from 0 to 1 indicates sleep is enabled. The node sleep flag is combined and packaged with the traceable parameter adjustment instruction in step S152. Upon receiving this instruction, the priority replacement node parses the sleep flag as 1 and executes the sleep procedure: switching its own sensor operating mode to a low-power sleep state, stopping all sensor signal acquisition and wireless communication transmission, and retaining only the timed wake-up circuit to automatically resume operation after a preset sleep duration.

[0126] Step S590: Add the topology vacancy information after the priority replacement node leaves the network to the network topology reconstruction strategy, and adjust the communication link routing relationship in the network topology reconstruction strategy so that the communication links of the remaining active nodes completely bypass the priority replacement node in the dormant state.

[0127] After a priority replacement node leaves the network, its spatial position in the original network topology becomes a topological vacancy. Obtain the adjacency list of each priority replacement node, and set all element values ​​of the corresponding row and column in the adjacency matrix A to 0 for each priority replacement node. Re-execute steps S144 to S149 to generate adjusted communication link routing relationships based on the updated adjacency matrix. In the adjusted routing relationships, all communication paths no longer pass through the dormant priority replacement node.

[0128] For example, the method may further include: step S610: monitoring the hull structure vibration response signal generated by the interaction between the hull structure and seawater wave load during the intelligent ship's navigation at sea, and separating the wave flutter component corresponding to the total vibration mode of the hull beam from the hull structure vibration response signal.

[0129] The hull structure vibration response signal was collected by vibration acceleration sensors installed on the port and starboard sides of the midship section. The wave flutter component is the first few orders of the total bending vibration modal response of the hull beams when encountering waves of a specific wavelength, excited by the impact force of high-frequency waves. Spectral analysis was performed on the hull structure vibration response signal to identify the spectral peaks corresponding to the natural frequencies of the lower-order bending vibrations of the hull beams. The narrowband signal with concentrated energy at the spectral peak frequency is the wave flutter component.

[0130] Step S620: Collect hull structure vibration response signal samples containing the wave flutter component through the acceleration sensor nodes installed in the hull structure monitoring area within the sensor network, and perform signal source correlation decoupling processing on the hull structure vibration response signal samples and the original response signal set.

[0131] The signal source correlation decoupling process uses the ship structure vibration response signal sample as a reference signal, and combines it with the multi-dimensional physical quantity sensing data streams from each sensor node in the original response signal set to form a multi-channel hybrid signal matrix. Independent component analysis is performed using a mutual information-based independence criterion, with the maximization of mutual information between the reference signal and each independent component as the discrimination criterion, to separate the independent source components representing ship wave flutter. The decoupled signal matrix is ​​obtained by subtracting the product of the column corresponding to the ship wave flutter source component in the hybrid matrix and its corresponding mixing coefficient from the hybrid signal matrix.

[0132] Step S630: Based on the result of the signal source correlation decoupling process, remove the vibration interference components transmitted from the ship structure vibration response signal to each sensor node from the original response signal set to obtain a pure sensor sensing data stream. Re-execute the signal drift situation analysis process on the pure sensor sensing data stream to generate a corrected signal drift situation feature, and calculate the anti-vibration interference drift stability index of each sensor node based on the corrected signal drift situation feature.

[0133] After decoupling in step S620, the row vector corresponding to each sensor node in the signal matrix is ​​extracted as the pure sensor sensing data stream for that sensor node. The signal drift state analysis is then re-executed on this pure sensor sensing data stream according to steps S121 to S122 to generate corrected signal drift state characteristics. The vibration interference resistance drift stability index is the ratio of the normalized abrupt change slope amplitude in the signal drift state characteristics before and after correction; a smaller ratio indicates a smaller impact of vibration interference on the sensor node drift state.

[0134] Step S640: Extract sensor nodes whose health status degradation level is in the tail interval and whose vibration interference drift stability index is in the head interval in the sensor network health status topology diagram. Mark these sensor nodes as vibration-sensitive false alarm nodes, remove the failure mark of the vibration-sensitive false alarm nodes from the set of failed nodes, and restore the vibration-sensitive false alarm nodes to the set of healthy nodes. At the same time, backtrack the historical version of the sensor network health status topology diagram and roll back the health status degradation level of the vibration-sensitive false alarm nodes to the historical health status level.

[0135] The tail interval of the health status degradation level ranking refers to the interval that falls within a preset tail proportion after sorting by health status degradation level from best to worst. The head interval of the vibration disturbance drift stability index ranking refers to the interval that falls within a preset head proportion after sorting by stability index from high to low. Sensor nodes that simultaneously meet both conditions indicate that their high degradation level is due to spurious drift caused by wave flutter. The node is removed from the failed node set and added to the healthy node set. The historical version database of the sensor network health status topology is retrieved to show the node's health status degradation level at a historical moment without vibration disturbance, and the current degradation level is modified to that historical value.

[0136] Step S650: Obtain the wave spectrum parameters of the sea area on the future navigation path provided by the shipborne meteorological sensing equipment, predict the future vibration intensity envelope of the ship structure vibration response signal based on the wave spectrum parameters, compare the future vibration intensity envelope with a preset vibration interference intensity threshold, identify the ship structure area where the vibration intensity exceeds the threshold, and map the ship structure area as the potential vibration interference range to the sensor network. Within the potential vibration interference range, pre-enhance the anti-vibration interference signal acquisition mode of the affected sensor nodes. The anti-vibration interference signal acquisition mode includes shortening the signal sampling interval of the sensor nodes and enabling the signal adaptive filtering function.

[0137] The wave spectrum parameters of the sea area include parameters such as significant wave height and spectral peak period. These parameters are input into the ship's structural vibration response transfer function model to calculate the predicted vibration acceleration curves for each structural component during future navigation. The envelope of the absolute values ​​of the vibration acceleration at each moment on the predicted curve is used as the future vibration intensity envelope. The vibration disturbance intensity threshold is the maximum allowable environmental vibration acceleration value when the sensor nodes are operating normally. Mesh cells on the 3D ship model whose predicted vibration intensity exceeds the threshold are mapped to the set of sensor nodes within their spatial coverage area. For affected sensor nodes, the original signal sampling interval is reduced to half its original value, and the adaptive digital filter built into the sensor node is remotely enabled through the aggregation node.

[0138] Step S660: Based on the estimated power consumption increment of the sensor node after the anti-vibration interference signal acquisition mode is enabled, a temporary power consumption scheduling instruction is generated to allocate network energy resources to adapt to this power consumption increment, and the temporary power consumption scheduling instruction is distributed to the affected sensor nodes through the aggregation node.

[0139] The estimated power consumption increment is the sum of the increased signal sampling frequency and the additional power consumption generated by filter operations after enabling the anti-vibration interference signal acquisition mode. The temporary power consumption scheduling instruction includes the affected sensor node identifier, the estimated power consumption increment value, and the power consumption scheduling execution period. The aggregation node sends the temporary power consumption scheduling instruction to the affected sensor nodes, and the sensor nodes adjust their own energy management strategies according to the power consumption scheduling execution period after receiving the instruction.

[0140] Step S710: Designate some sensor nodes in the sensor network as proxy sensing anchors, and control the proxy sensing anchors to transmit active excitation signals with known waveform parameters to the hull structure monitoring area. After the active excitation signals propagate in the hull structure medium, they are received by the adjacent sensor nodes as structural response transmission signals.

[0141] Several healthy nodes closest to the spatial deployment coordinates of the failed nodes are selected from the set of healthy nodes as proxy sensing anchors. The active excitation signal with known waveform parameters is a linear frequency modulated (LFM) signal, whose instantaneous frequency increases linearly from low frequency to high frequency within one excitation cycle, with a fixed signal amplitude. The convergence node sends an excitation enable command to the proxy sensing anchors. The piezoelectric excitation module built into the proxy sensing anchor converts the LFM signal into mechanical vibration, which is coupled to the ship's structural medium through the mounting base. The active excitation signal propagates along the ship's steel structure in the form of stress waves. When it encounters discontinuous interfaces such as cracks, corrosion, or material delamination within the structure, it is reflected and transmitted. The transmitted stress wave is received by the neighboring sensor nodes around the proxy sensing anchor through their own piezoelectric sensing elements. The received signal is the structural response transmission signal.

[0142] Step S720: From the set of original response signals of the neighboring sensor nodes that receive the transmission signal of the structure response, separate the arrival waveform of the transmission signal that has the same waveform parameter characteristics as the active excitation signal, and record the arrival time and waveform energy attenuation rate of the arrival waveform of the transmission signal.

[0143] The raw response signal received from the neighboring sensor node is subjected to matched filtering, with the linear frequency modulated waveform of the active excitation signal used as a reference template. The output of the matched filter is the cross-correlation function between the raw response signal and the reference template; the time corresponding to the peak value of the cross-correlation function is the arrival time of the transmitted signal waveform. The ratio of the peak amplitude of the cross-correlation function to the peak amplitude of the autocorrelation of the reference template is normalized, logarithmed, and divided by the propagation path length to obtain the waveform energy attenuation rate.

[0144] Step S730: Using the time difference between the transmission time of the active excitation signal and the arrival time of the transmitted signal waveform, multiply by the stress wave propagation velocity constant in the hull structure medium to obtain the equivalent propagation distance calculation value between the proxy sensing anchor point and the adjacent sensor node. Compare the equivalent propagation distance calculation value with the straight-line distance between the spatial deployment coordinates of the proxy sensing anchor point and the adjacent sensor node. When the deviation between the equivalent propagation distance calculation value and the straight-line distance exceeds the preset propagation path abnormal deviation threshold, mark this pair of proxy sensing anchor points and adjacent sensor nodes as a propagation path abnormal node pair.

[0145] The transmission time of the active excitation signal is recorded by the agent sensing anchor and reported to the aggregation node via the communication link. The equivalent propagation distance is calculated as (arrival time - transmission time) × Vstress, where Vstress is the calibrated propagation velocity constant of the longitudinal stress wave in the ship's structural steel. The straight-line distance is the Euclidean distance between the spatial deployment coordinates of the agent sensing anchor and the spatial deployment coordinates of the adjacent sensor nodes. The propagation path anomaly deviation threshold is a preset percentage value of the straight-line distance. If the deviation between the two distances exceeds this threshold, it indicates that the stress wave encountered an abnormal region within the structure during propagation, causing the propagation path to be non-straight.

[0146] Step S740: Collect the structural physical attribute information of the ship structure area traversed by the straight path between the abnormal node pairs of the propagation path. The structural physical attribute information includes structural plate thickness parameters, structural weld distribution density, and structural coating peeling degree index.

[0147] Structural plate thickness parameters were obtained from the plate thickness distribution database of the ship's structural design drawings. The structural weld distribution density was calculated by dividing the sum of the lengths of all welds within the area traversed by the propagation path by the area of ​​that region. The structural coating peeling degree index was obtained by taking high-resolution surface images of the area traversed by the propagation path and inputting them into a convolutional neural network recognition model to classify the degree of peeling.

[0148] Step S750: Using the structural physical property information as input, call the preset structural micro-damage inversion model, and use the structural micro-damage inversion model to invert and calculate the structural medium state between the abnormal node pairs of the propagation path, to obtain the damage type inference result and damage degree classification result of the structural micro-damage between the propagation path, and map the damage type inference result and damage degree classification result to the hull structure spatial coordinate interval corresponding to the abnormal node pairs of the propagation path, and generate a structural micro-damage spatial distribution marking map on the three-dimensional model space of the hull structure.

[0149] The pre-set structural micro-damage inversion model is a trained multilayer perceptron neural network model. The model input layer receives structural plate thickness parameters, weld distribution density, and coating peeling degree indices. After feedforward propagation through three hidden layers, the output layer outputs the damage type probability distribution and damage severity rating. Damage types include four categories: no damage, microcracks, corrosion pits, and interlayer peeling. The damage severity rating is divided into several discrete levels. The output results are mapped onto the hull structure's 3D model mesh elements corresponding to the spatial coordinate intervals of the anomalous node pairs along the propagation path to generate color-coded markers.

[0150] Step S760: Extract the damage severity rating results of the locations of the structural micro-damages that coincide with the spatial deployment coordinates of each sensor node from the spatial distribution map of the structural micro-damages, and convert the damage severity rating results into the structural integrity attenuation factor of the sensor node mounting base.

[0151] The conversion of damage severity rating results into structural integrity attenuation factor is accomplished through a preset mapping table. The mapping table defines the structural integrity attenuation factor value corresponding to each damage type at each damage severity level, with values ​​ranging from zero to one. The smaller the value, the more severe the damage to the mounting base structure.

[0152] Step S770: The structural integrity attenuation factor is used as an external disturbance input and injected into the node health state topology deduction logic. Together with the signal drift situation characteristics and the inter-node response dissimilarity characteristics, it participates in the generation process of the sensor network health state topology map to obtain an enhanced sensor network health state topology map that incorporates the influence of structural micro-damage.

[0153] In step S132, when calculating the drift risk score, the original scoring formula is modified to Sdrift_mod = Sdrift × (1 + η × (1 - ζ)), where ζ is the structural integrity attenuation factor and η is the disturbance amplification factor. The smaller the value of ζ, the greater the increase of Sdrift_mod relative to the original Sdrift after multiplying by the disturbance factor, and the more likely the health status degradation level of the sensor node is to be assessed as a higher level of degradation.

[0154] Step S810: Extract the functional alternative path network composed of functional alternative paths with availability identifiers from the sensor network health status topology graph, abstract the functional alternative path network into a sensor function dependency graph composed of sensing function nodes and functional alternative path connection edges, perform network motif enumeration analysis on the sensor function dependency graph, identify the recurring local subgraph structure patterns in the sensor function dependency graph, and perform motif matching between the identified local subgraph structure patterns and a preset sensor function vulnerability motif library to find vulnerable motif instances that successfully match the sensor function vulnerability motif library.

[0155] In the sensor function dependency graph, sensor function nodes are sensor nodes, and alternative function path connecting edges are alternative function paths with availability indicators. The network motif enumeration analysis employs a traversal statistical method, enumerating all triangular subgraph structures in the sensor function dependency graph that consist of three sensor function nodes and three alternative function path connecting edges, and counting the frequency of each type of triangular subgraph in the sensor function dependency graph. A pre-defined sensor function vulnerability motif library contains several known vulnerability motif topological structure patterns. The identified local subgraph structure patterns are matched with the connection relationships of each motif in the vulnerability motif library using graph isomorphism matching; successfully matched patterns are identified as vulnerability motif instances.

[0156] Step S820: Extract the set of sensor nodes contained in each vulnerable phantom instance, and perform cascade failure simulation and deduction processing on the set of sensor nodes. The cascade failure simulation and deduction processing simulates the failure of any sensor node in the set of sensor nodes and the propagation of the failure effect along the functional substitution path of the sensing function dependency graph, and calculates the cumulative number of failed nodes when the failure propagation terminates in each simulation.

[0157] The cascading failure simulation process is as follows: For each sensor node in the sensor node set of the vulnerable phantom instance, its state is set to failed and it is removed from the sensor function dependency graph. Check if any of the remaining sensor function nodes have lost all alternative functional paths due to the failure of this node. If so, set that sensor function node to failed as well and continue propagating the failure. Repeat this process until no new failed sensor function nodes are generated, and record the final cumulative number of failed nodes.

[0158] Step S830: Mark the initial failure node corresponding to the simulation scenario in which the cumulative number of failure nodes exceeds the preset cascading runaway threshold during multiple cascading failure simulation and deduction processes as a cascading failure trigger node, and summarize all cascading failure trigger nodes into a cascading failure trigger node list. For each cascading failure trigger node in the cascading failure trigger node list, calculate its cascading influence fan-out degree in the sensing function dependency graph. The cascading influence fan-out degree is defined as the maximum cumulative number of failure nodes that can be triggered by a single failure of the cascading failure trigger node.

[0159] For each cascading failure trigger node, the cumulative number of failed nodes in multiple simulated scenarios with that node as the initial failure node has been recorded in the cascading failure simulation and deduction process in step S820. The cascading effect fan-out degree is the maximum value of the cumulative number of failed nodes for that node in all simulated scenarios. The larger this value, the more widespread the chain failure reaction will be triggered once the node fails, and the stronger the cascading destructive power of the node in the functional alternative path network.

[0160] Step S840: During the network topology reconstruction strategy generation process, the cascading impact fan-out degree is used as the priority ranking criterion for reconstruction of newly added nodes, so that the failed nodes with large cascading impact fan-out degrees are given priority in the allocation of functional alternative paths and the planning of communication link redirection paths.

[0161] Before processing the node to be reconstructed in step S142, all failed nodes in the failed node set are sorted in descending order of their cascading effect fan-out degree. Failed nodes with the same cascading effect fan-out degree are then sorted in descending order of their health status degradation level from worst to best, as a secondary sorting key. Steps S142 to S148 are then executed sequentially for each failed node in the sorted order. Failed nodes with larger cascading effect fan-out degrees are given priority in allocating replacement nodes and redirection paths, ensuring that reconstruction resources are primarily concentrated on eliminating potential cascading failure propagation risks.

[0162] Step S850: When allocating functional alternative paths, for failed nodes whose cascading effect fan-out degree exceeds a preset fan-out threshold, it is prohibited to point their functional alternative paths to candidate alternative nodes that have functional dependencies on other sensor nodes in the vulnerable phantom instance, and candidate alternative nodes that violate this prohibition rule are removed.

[0163] The preset fan-out threshold is taken as the upper quartile value of the fan-out degree of the cascaded influence of all sensor nodes in the sensor function dependency graph. For failed nodes whose cascaded influence fan-out degree exceeds this threshold, each candidate node in the set of functional replacement candidate nodes generated in step S135 is traversed to check whether the candidate node has a functional dependency relationship with other sensor nodes identified in step S810 that belong to the same vulnerable phantom instance as the failed node, where the functional coupling strength coefficient Cfunc value calculated in step S134 exceeds the preset functional dependency threshold within the phantom. If such a relationship exists, the candidate node is removed from the set of functional replacement candidate nodes. If the set of functional replacement candidate nodes is empty after removal, the fan-out threshold is temporarily lowered by one level and re-filtered until the set of functional replacement candidate nodes is not empty.

[0164] Step S860: When planning the communication link redirection path, for failed nodes whose cascading impact fan-out degree exceeds the preset fan-out threshold, their communication link redirection path must be forwarded through relay nodes that do not belong to the same vulnerable motif instance, and a demodified communication link routing relationship is generated. The demodified communication link routing relationship is written into the routing table update instruction in the network topology reconstruction strategy, so that the functional alternative path network in the reconstructed sensor network health status topology graph structure no longer contains the identified vulnerable motif instance.

[0165] For the communication link redirection path generated in step S145, check whether the path contains sensor nodes belonging to the same vulnerable motif instance as the failed node as relay nodes. If so, mark the aforementioned relay nodes as prohibited relay nodes. Re-execute the Dijkstra shortest path algorithm, setting the weights of all outgoing and incoming edges of the prohibited relay nodes to infinity in the communication link quality parameter matrix Q, and calculate the optimal path bypassing the prohibited relay nodes as the demodified communication link redirection path. In the demodified communication link routing relationship, all relay forwarding nodes do not belong to the vulnerable motif instance where the failed node is located. Generate incremental routing table update instructions for the demodified routing relationship according to the format of step S151 and write them into the network topology reconstruction strategy.

[0166] Step S910: Obtain historical archived data of multi-dimensional physical quantity sensing data streams of each sensor node in the sensor network over a long time scale. Perform seasonal periodic slicing processing on the historical archived data according to the natural time period to obtain multiple seasonal periodic slice data. Perform seasonal drift pattern extraction processing on the signal drift characteristics of each sensor node in each seasonal periodic slice data to obtain the seasonal correlation drift pattern curve of each sensor node. The seasonal correlation drift pattern curve includes the seasonal peak phase and seasonal valley phase of the signal drift characteristics in each seasonal periodic slice.

[0167] The seasonal cycle slice segmentation process divides historical archived data into four seasonal cycle slices—spring, summer, autumn, and winter—according to meteorological seasonal division standards. Each seasonal cycle slice contains complete multi-dimensional physical quantity sensing data streams from each sensor node within that seasonal period. Signal drift characteristic extraction is re-executed for the data within each seasonal cycle slice according to steps S120 to S122 to obtain the corresponding signal drift characteristic. The maximum value of the normalized abrupt change slope amplitude in the signal drift characteristic of each seasonal cycle slice is taken as the drift peak for that season, and the occurrence time of this peak on the seasonal time axis is recorded as the seasonal peak phase. The minimum value of the normalized abrupt change slope amplitude is taken as the drift trough, and its occurrence time is recorded as the seasonal trough phase. The peak phases and trough phases of the four seasons are connected in seasonal order to form an annual cycle curve with the season as the horizontal axis. This annual cycle curve is the seasonally associated drift pattern curve.

[0168] Step S920: Construct a mapping table of correspondence between seasonal dimensions and signal drift trends using the seasonal drift pattern curve. Divide the natural time period into multiple drift pattern seasonal intervals through the mapping table. Each drift pattern seasonal interval corresponds to a signal drift dominant trend type.

[0169] The mapping table between seasonal dimensions and signal drift trends is a lookup table structure. The row index is the season type, and the column fields contain the drift peak range, drift trough range, and dominant trend direction of the signal drift trend characteristics corresponding to that season. Adjacent seasons with the same dominant trend direction are merged into a single drift pattern season interval based on the peak and trough change directions of each season. The dominant trend direction is determined by the sign of the mean of the trend direction change sequence in the signal drift trend characteristics within that season: a positive mean indicates an upward dominant trend, a negative mean indicates a downward dominant trend, and a mean approximately zero indicates a stable dominant trend.

[0170] Step S930: When generating the signal drift situation characteristics of each sensor node in real time, the seasonal interval of the drift mode corresponding to the current natural time is used as a seasonal context label and attached to the signal drift situation characteristics to obtain the seasonal perception signal drift situation characteristics carrying the seasonal context label. The seasonal perception signal drift situation characteristics are input into the node health status topology inference logic. The node health status topology inference logic dynamically switches the preset multi-level degradation threshold interval according to the seasonal context label, so that different drift mode seasonal intervals use multi-level degradation threshold intervals with different interval boundaries to evaluate the health status degradation level.

[0171] In step S122, when generating signal drift situation characteristics in real time, the current system time is read to determine the current drift mode seasonal interval, and the interval identifier of this drift mode seasonal interval is attached to the signal drift situation characteristics as a seasonal context label. In step S132, when evaluating the health status degradation level, a set of threshold intervals corresponding to the seasonal context label is retrieved from multiple preset multi-level degradation threshold intervals for evaluation. The threshold interval boundary values ​​of different drift mode seasonal intervals are set independently. The threshold of the rising seasonal interval is appropriately relaxed compared to the threshold of the stable seasonal interval to avoid misjudging seasonal rising drift as abnormal sensor degradation.

[0172] Step S940: Based on the assessed health status degradation level, generate a health status topology snapshot sequence that periodically switches with seasonal intervals of drift mode in the sensor network health status topology map. The health status topology snapshot sequence contains sensor network health status topology maps corresponding to seasonal intervals of different drift modes.

[0173] For each drift mode seasonal interval, an independent sensor network health status topology map is generated. The health status degradation level of each sensor node in the map is determined by the dynamic switching threshold in step S930. The topology maps are arranged in chronological order according to the drift mode seasonal intervals to form a health status topology snapshot sequence.

[0174] Step S950: Perform a difference comparison analysis on the sensor network health status topology map of adjacent drift mode seasonal intervals in the health status topology snapshot sequence, identify sensor nodes that experience a leap in health status degradation level between different drift mode seasonal intervals, and generate a list of seasonally sensitive nodes.

[0175] For the same sensor node in two adjacent sensor network health status topology diagrams, a subtraction operation is performed. If the absolute value of the difference between the degradation level in the later diagram and the degradation level in the earlier diagram is greater than or equal to a preset threshold for a jump (which is 2 levels), then the sensor node has experienced a jump. All sensor nodes that have experienced jumps are added to the seasonally sensitive node list.

[0176] Step S960: During the network topology reconstruction strategy generation process, the seasonal sensitive node list is queried. For sensor nodes in the list, when their health status degradation level does not exceed the preset degradation threshold but is in the seasonal adjacent interval of the preset degradation threshold, a seasonal pre-response network topology reconstruction strategy is generated in advance.

[0177] The seasonal proximity interval is defined as being one level away from the preset degradation threshold level. For sensor nodes in the seasonally sensitive node list whose health status degradation level is in the seasonal proximity interval, the virtual failure level is taken as the degradation level that is expected to jump in the next drift mode seasonal interval. The seasonal pre-response network topology reconstruction strategy is generated in advance according to steps S140 to S149.

[0178] Step S970: Store the seasonal pre-response network topology reconstruction strategy in the form of a strategy template in the aggregation node, and set seasonal time trigger conditions. When the system time enters the corresponding drift mode seasonal interval, the seasonal pre-response network topology reconstruction strategy is automatically loaded and executed.

[0179] The seasonal trigger condition is the start timestamp of the seasonal interval in the drift mode. When the system time reaches this start timestamp, the aggregation node automatically reads the corresponding seasonal pre-response network topology reconstruction strategy from the strategy template library, executes it according to step S150, converts it into a reconstruction execution instruction set, and issues it for execution. After execution, a seasonal pre-response reconstruction completion notification is sent to the ship monitoring and management terminal.

[0180] Figure 2This application illustrates a sensor network health self-assessment and reconstruction system 100 for intelligent ships, comprising a processor 1001 and a memory 1003. The processor 1001 and memory 1003 are connected, for example, via a bus 1002. Optionally, the sensor network health self-assessment and reconstruction system 100 for intelligent ships may further include a transceiver 1004. The transceiver 1004 can be used for data interaction between this sensor network health self-assessment and reconstruction system for intelligent ships and other sensor network health self-assessment and reconstruction systems for intelligent ships, such as sending and / or receiving data. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of this sensor network health self-assessment and reconstruction system 100 for intelligent ships does not constitute a limitation on the embodiments of this application.

[0181] The memory 1003 is used to store program code for executing the embodiments of this application, and its execution is controlled by the processor 1001. The processor 1001 is used to execute the program code stored in the memory 1003 to implement the steps shown in the foregoing method embodiments.

[0182] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application, without departing from the technical concept of this application, also fall within the protection scope of the embodiments of this application.

Claims

1. A method for self-assessment and reconstruction of sensor network health applied to intelligent ships, characterized in that, The method includes: The raw response signal set of each sensor node in the sensor network deployed in the intelligent ship hull structure monitoring area is collected. The raw response signal set contains multi-dimensional physical quantity sensing data streams with acquisition time stamps. The original response signal set is subjected to signal drift situation analysis to obtain the signal drift situation characteristics of each sensor node, and the response dissimilarity characteristics between nodes are generated based on the signal amplitude deviation of different sensor nodes in the original response signal set under the same acquisition time mark. Based on the signal drift characteristics and the inter-node response dissimilarity characteristics, a sensor network health status topology map representing the overall health status of the sensor network is generated through a preset node health status topology deduction logic. The sensor network health status topology map includes the health status degradation level of each sensor node and the availability identifier of functional alternative paths between nodes. Based on the set of failed nodes whose health status degradation level exceeds a preset degradation threshold in the sensor network health status topology diagram and the availability identifier of the functional alternative paths between the nodes, a network topology reconstruction strategy is determined for the set of failed nodes. The network topology reconfiguration strategy is converted into a set of reconfiguration execution instructions that includes routing table update instructions and node working parameter adjustment instructions. The set of reconfiguration execution instructions is then distributed to the aggregation node of the sensor network to trigger a network topology reconfiguration operation.

2. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The step involves performing signal drift state analysis on the original response signal set to obtain the signal drift state characteristics of each sensor node, and generating inter-node response dissimilarity characteristics based on the signal amplitude deviation of different sensor nodes within the original response signal set at the same acquisition time marker, including: The multi-dimensional physical quantity sensing data streams of each sensor node in the original response signal set are arranged in time sequence according to the acquisition time mark to generate a multi-dimensional time sequence signal array for each sensor node. The physical quantity sensing data streams of each dimension in the multi-dimensional time sequence signal array are subjected to signal baseline drift separation processing to extract the signal baseline drift component and signal transient fluctuation component corresponding to the physical quantity of that dimension. Based on the trend direction change sequence and trend inflection point density of the signal baseline drift component within the continuous acquisition time marker interval, a signal drift situation trajectory line describing the long-term evolution trend of the sensor node sensing characteristics is constructed. The trend direction change sequence in the signal drift situation trajectory line is input to a preset trend direction change detector to obtain the change slope amplitude and the trend duration after the change corresponding to the change moment in the trend direction of the signal drift situation trajectory line. The change slope amplitude and the trend duration after the change are encapsulated as the signal drift situation feature. Extract the signal amplitudes of sensor nodes deployed in the same physical monitoring sub-area under the same acquisition time mark from the original response signal set, and calculate the absolute deviation of the signal amplitude between each pair of sensor nodes. Perform statistical distribution analysis on the absolute deviation of the signal amplitude between all pairs of sensor nodes under each acquisition time mark, and extract the median deviation, which characterizes the central tendency of the deviation value distribution, and the interquartile range of the deviation value distribution, which characterizes the dispersion of the deviation value distribution. Based on the median and interquartile range parameters of the deviation under multiple consecutive acquisition time marks, a time series of deviation distribution evolution is constructed. The time series of deviation distribution evolution is then subjected to fluctuation pattern clustering. Acquisition time mark segments with similar deviation distribution evolution trends are divided into the same deviation fluctuation pattern cluster. The mode value of the absolute deviation of signal amplitude between all pairs of sensor nodes under all acquisition time marks in each deviation fluctuation pattern cluster is calculated, and the mode value is used as the reference value of the inter-node response dissimilarity corresponding to the deviation fluctuation pattern cluster. Based on the benchmark value of inter-node response dissimilarity and the slope change direction of the time series of deviation distribution evolution corresponding to the deviation fluctuation mode cluster, an inter-node response dissimilarity feature vector is generated with the time boundary of the deviation fluctuation mode cluster as the index. Each element in the inter-node response dissimilarity feature vector contains a dissimilarity amplitude component and a dissimilarity change trend component.

3. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The process of generating a sensor network health state topology map representing the overall health state of the sensor network based on the signal drift characteristics and the inter-node response dissimilarity characteristics, through a preset node health state topology deduction logic, includes: Obtain the spatial deployment coordinates and initial adjacency topology of each sensor node in the sensor network. The initial adjacency topology includes the communication link connection relationship and communication link quality parameters between adjacent sensor nodes. Based on the abrupt change slope amplitude and the duration of the trend after the abrupt change contained in the signal drift situation characteristics, normalization processing is performed respectively. After mapping the two to the same dimensionless scoring scale, the drift risk score of each sensor node is calculated by weighting. The drift risk score is then compared with the preset multi-level degradation threshold range to determine the initial health status degradation level of each sensor node. Extract the benchmark value of the inter-node response dissimilarity corresponding to the most recent deviation fluctuation mode cluster from the inter-node response dissimilarity feature vector, and construct an inter-node dissimilarity matrix that measures the consistency of the current response between different sensor nodes. The dissimilarity magnitude component of each pair of sensor nodes in the inter-node dissimilarity matrix and the communication link quality parameter of the pair of sensor nodes in the initial adjacency topology are normalized respectively. After mapping the two to the same evaluation scale, they are fused and weighted to generate the inter-node functional coupling strength coefficient that reflects the functional correlation between nodes. Traverse all sensor nodes in the sensor network, take the currently traversed sensor node as the target node, and filter out the adjacent sensor nodes whose inter-node functional coupling strength coefficient with the target node exceeds a preset coupling threshold to form a set of functional replacement candidate nodes for the target node. For each candidate node in the set of candidate nodes for functional replacement of the target node, based on the initial health status degradation level of the candidate node and the functional coupling strength coefficient between the candidate node and the target node, the functional transfer loss parameter when the candidate node carries the sensing function of the target node is evaluated. The functional transfer loss parameters of all candidate nodes in the set of candidate nodes for functional replacement of the target node are integrated, and an ordered list of functional replacement paths for the target node is generated in ascending order of functional transfer loss parameters. Each functional replacement path in the ordered list of functional replacement paths is assigned an availability identifier. The initial health degradation level of each sensor node is combined with the ordered list of alternative functional paths for each sensor node to obtain a set of node-level health status information containing the health degradation levels of all sensor nodes and the availability identifiers of alternative functional paths between nodes. The set of node-level health status information is then mapped to a two-dimensional planar grid using the spatial deployment coordinates of the sensor network, and corresponding node pairs are connected through the availability identifiers of alternative functional paths between nodes to generate the health status topology map of the sensor network.

4. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The step of determining a network topology reconstruction strategy for the set of failed nodes based on the set of failed nodes whose health status degradation level exceeds a preset degradation threshold in the sensor network health status topology map and the availability identifiers of the functional alternative paths between the nodes includes: Traverse all sensor nodes in the sensor network health status topology diagram, mark sensor nodes whose health status degradation level exceeds a preset degradation threshold as failed nodes, and mark sensor nodes whose health status degradation level does not exceed the preset degradation threshold as healthy nodes. Each failed node in the set of failed nodes is taken as a node to be reconstructed. Healthy nodes with availability identifiers that are connected to the functional alternative paths of the node to be reconstructed are extracted from the sensor network health status topology map to form a list of available alternative nodes for the node to be reconstructed. Calculate the number of inter-node functional alternative paths currently carried by each healthy node in the list of available alternative nodes, calculate its theoretical maximum carrying capacity score based on the initial communication link quality parameters of the healthy node, and reduce its theoretical maximum carrying capacity score according to the number of inter-node functional alternative paths already carried to obtain the remaining functional carrying capacity value of the healthy node. Obtain the set of neighboring nodes of the node to be reconstructed in the topology of the sensor network health status, and select the available replacement node with the largest sum of the functional coupling strength coefficients between the node and the healthy node in the set of neighboring nodes from the list of available replacement nodes as the main replacement node; Based on the spatial deployment coordinates of the primary replacement node, the communication link redirection path for the primary replacement node to receive each healthy node in the set of adjacent nodes of the node to be reconstructed is calculated, and the updated communication link routing relationship is generated. The operating parameters of the primary replacement node are adapted and adjusted, including increasing the signal acquisition frequency of the primary replacement node and expanding the signal sensing threshold range of the primary replacement node, to obtain the operating parameter adjustment instruction of the primary replacement node; For the remaining available alternative nodes in the list of available alternative nodes other than the primary alternative node, the set of adjacent nodes of the node to be reconstructed is grouped according to the remaining functional carrying capacity value of the remaining available alternative nodes, and a corresponding backup alternative node is assigned to each group of adjacent nodes to generate an auxiliary routing path. The communication link redirection path, the auxiliary routing path, and the working parameter adjustment instruction are combined to form a single node reconstruction strategy for the node to be reconstructed. The single-node reconstruction strategies of each failed node in the set of all failed nodes are summarized, and the conflict resolution process is performed on the single-node reconstruction strategies that share the same replacement node. The conflicting routing entries are merged and the working parameter adjustment instructions of the shared replacement node are uniformly adjusted to obtain the network topology reconstruction strategy covering the entire set of failed nodes.

5. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The step of converting the network topology reconfiguration strategy into a set of reconfiguration execution instructions containing routing table update instructions and node working parameter adjustment instructions, and distributing the set of reconfiguration execution instructions to the aggregation node of the sensor network to trigger the network topology reconfiguration operation, includes: All descriptive information for communication link redirection paths and auxiliary routing paths is extracted from the network topology reconstruction strategy. The descriptive information is converted into a next-hop routing mapping table record from the perspective of the aggregation node. The next-hop routing mapping table record is compressed and encoded to generate an incremental routing table update instruction for updating the internal routing table of the aggregation node. The incremental routing table update instruction includes an operation type identifier, a destination address field, and a next-hop address field. Extract all working parameter adjustment instructions for the primary replacement node and other specified replacement nodes from the network topology reconstruction strategy. Group the working parameter adjustment instructions according to the target node identifier to obtain a node working parameter adjustment instruction group that corresponds one-to-one with each replacement node. Add an instruction validity period timestamp and instruction version number to each working parameter adjustment instruction in the node working parameter adjustment instruction group to generate a traceable parameter adjustment instruction with timeliness and version control functions. The incremental routing table update instruction and the traceable parameter adjustment instruction are aggregated by instruction type to form the set of reconstruction execution instructions indexed by the instruction target node; Search for currently active aggregation nodes in the sensor network, establish an instruction delivery channel with the aggregation nodes, and split and deliver the set of reconstruction execution instructions to the corresponding aggregation nodes according to the instruction target node identifier; After each aggregation node receives the set of reconstruction execution instructions, it parses the incremental routing table update instructions through the communication protocol stack between the aggregation node and the instruction target node, and writes new routing table entries or overwrites old routing table entries in the routing forwarding table inside the aggregation node. After each aggregation node completes the routing table entry update, the traceable parameter adjustment instruction is transmitted through the aggregation node to the corresponding instruction target node. The instruction target node parses the traceable parameter adjustment instruction and modifies its own signal acquisition frequency parameter and signal sensing threshold range parameter. After the instruction target node completes the modification of the working parameters, the instruction target node reports a parameter effectiveness confirmation message to the aggregation node. After the aggregation node summarizes the parameter effectiveness confirmation messages from all instruction target nodes, it broadcasts a topology update completion notification to the sensor network.

6. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The method further includes: Real-time voltage and current signals of the power supply branches of each sensor node in the sensor network are collected, and an instantaneous power consumption time-series tracking sequence for each sensor node is generated based on the real-time voltage and current signals. The instantaneous power consumption time-series tracking sequence is subjected to power consumption mode decomposition processing, and the transient power consumption pulse component exhibiting a pulse shape and the static power consumption offset component exhibiting a stable offset shape are extracted from the power consumption mode decomposition processing. The power consumption transient pulse component and the power consumption static offset component are combined to obtain the power consumption feature spectrum of the sensor node. The power consumption characteristic spectrum is compared with the preset historical power consumption benchmark spectrum of sensor nodes by performing spectrum deviation analysis to obtain the power supply status change score value of each sensor node. The power supply status change score value is used as a health influence factor as a supplementary dimension alongside the health status degradation level, and the health influence factor is superimposed on the node attributes of the corresponding sensor node in the sensor network health status topology diagram. Obtain the abnormal code sequence returned by the internal hardware self-test status register of each sensor node in the sensor network, and generate a node hardware vulnerability indicator function based on the frequency and timing distribution of the setting of specific error bit masks in the abnormal code sequence. Receive spectral energy distribution data sent by environmental vibration sensors deployed on the hull of the smart ship, and use the cumulative amount of vibration energy in the spectral energy distribution data that matches the inherent resonant frequency band of the sensor node as the long-term physical damage induction parameter of the node. A multi-source anomaly signal fusion model is constructed that couples the power supply status anomaly score value, the node hardware hidden danger indicator function, and the node long-term physical damage induction parameter. Each anomaly signal is analyzed independently through the multi-source anomaly signal fusion model, and the corresponding node health status deterioration correction factor is output. Based on multiple node health status degradation correction factors obtained for each node, a comprehensive health status degradation coefficient is obtained through weighted fusion. According to the comprehensive health status degradation coefficient, the interval boundaries of the multi-level degradation threshold interval used to determine the initial health status degradation level in the node health status topology deduction logic are adjusted. Based on the adjusted multi-level degradation threshold interval, the revised health status degradation level of each sensor node is recalculated, and the sensor network health status topology is updated according to the revised health status degradation level.

7. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The method further includes: Obtain sensor node failure event logs generated by the sensor network during its historical operating cycle. The sensor node failure event logs include node failure timestamps and spatial deployment coordinates of the failed nodes. Based on the node failure timestamps and spatial deployment coordinates of the failure nodes, a scatter plot of the spatiotemporal distribution of failure events is plotted on the time series axis and the three-dimensional model space of the ship structure. Spatiotemporal density clustering analysis is performed on the scatter plot of the spatiotemporal distribution of failure events to extract spatiotemporal hotspot regions where the clustering density of failure events exceeds the clustering threshold. Common damage cause tracing analysis is performed on the failure nodes contained in each spatiotemporal hotspot region. By comparing the temporal clustering of the trend direction change moment in the signal drift characteristics of the failure nodes in the spatiotemporal hotspot region, the spatiotemporal propagation direction vector of the failure propagation path is determined. A failure propagation directed topology graph is constructed with spatiotemporal propagation direction vectors as edges and sensor nodes as nodes. The causal tracing strength value of node failure is calculated along the propagation direction in the failure propagation directed topology graph. Based on the causal tracing strength value, a set of key root cause nodes that trigger cascading failures is extracted. Microenvironmental parameters of the physical installation location of each key root cause node in the set of key root cause nodes are collected. The microenvironmental parameters include the stress value of the installation base, the local environmental corrosion rate, and the impedance fluctuation amplitude of the connecting cable. Each microenvironment parameter is compared with a preset threshold to obtain an independent accelerated failure level for each parameter. By combining the independent accelerated failure levels of each parameter and using a preset failure level-remaining time mapping model, the estimated remaining effective service time of the critical root cause node is obtained. Replace the preset failure time array of sensor nodes belonging to the set of critical root cause nodes in the sensor network health status topology diagram with the estimated remaining effective service time of the critical root cause nodes. Before generating the network topology reconstruction strategy, nodes in the key root cause node set whose estimated remaining effective service time is lower than the preset task cycle are marked as potential failure nodes in advance, and the potential failure nodes are added to the failure node set. The proposed failure node and the original failure node in the set of failure nodes are used together as input objects for the network topology reconstruction strategy to generate a forward-looking network topology reconstruction strategy that simultaneously covers both failure nodes and proposed failure nodes.

8. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The method further includes: After the aggregation node executes the set of reconstruction execution instructions, it receives steady-state response signal samples reported by each sensor node under the new topology of the sensor network. By performing frequency domain energy spectrum analysis on the steady-state response signal samples, it obtains the response energy spectrum distribution of the new topology. Extract the spatial coverage of the physical monitoring sub-region covered by the set of failed nodes from the sensor network health status topology map, and calculate the signal sensing coverage area of ​​the physical monitoring sub-region of the new topology based on the spatial deployment coordinates of the sensor nodes that assume the alternative function in the new topology and their preset effective physical sensing radius. The coverage difference boundary extraction process is performed on the signal sensing coverage area and the historical signal sensing coverage area before the failure node set fails to obtain the spatial boundary contour data of the sensing hole area. Based on the spatial boundary contour data of the sensing hole area, the sensing hole frequency band energy attenuation curve corresponding to the spatial coordinates of the sensing hole area is located in the response energy spectrum distribution, and the attenuation slope and cutoff frequency of the sensing hole frequency band energy attenuation curve are extracted. The preset signal reconstruction compensation algorithm library is invoked, and the corresponding signal reconstruction compensation algorithm is selected according to the attenuation slope and cutoff frequency. The signal reconstruction compensation algorithm is then used to perform inverse filtering compensation processing on the signal in the sensing hole frequency band to generate a signal compensation dataset for the sensing hole area. The signal compensation dataset is spatially aligned and fused with the spatial boundary contour data of the sensing cavity area to generate a complete monitoring area signal field distribution map without gap sensing cavity. The current working condition task sequence of the intelligent ship is obtained, and the key structural safety monitoring requirements are parsed from the working condition task sequence. The key structural safety monitoring requirements include the lower limit requirement of monitoring point density and the range of mandatory monitoring areas for specified structural parts. The signal field distribution map of the complete monitoring area of ​​the seamless sensing cavity is compared with the requirements for safety monitoring of the key structure. The necessary monitoring area of ​​the specified structural parts where the signal sensing capability cannot meet the lower limit requirement of the monitoring point density is identified. When there is a necessary monitoring area of ​​the specified structural parts that does not meet the requirements, a mobile sensing unit scheduling instruction is generated and sent to the inspection equipment equipped with mobile sensors. The inspection equipment is then scheduled to move to the necessary monitoring area of ​​the specified structural parts to enhance the local monitoring capability.

9. The sensor network health self-assessment and reconstruction method for intelligent ships according to claim 1, characterized in that, The method further includes: The communication protocol data stream of the sensor network is monitored in real time, and health status beacon frames periodically uploaded by each sensor node to the aggregation node are captured from the communication protocol data stream. The health status beacon frames carry the sensor node identifier and the current health status degradation level. The health status degradation levels carried by the health status beacon frames received in multiple consecutive cycles are arranged in time sequence to construct a node health degradation rate sequence, and the transition step size of the health status degradation level between adjacent cycles in the node health degradation rate sequence is calculated. The transition step size value is subjected to abnormal transition pattern recognition processing. When the transition step size value exceeds the preset upper limit of the normal transition step size threshold, an alarm message for the sudden deterioration of the health status of the sensor node is generated. Intercept the regular monitoring data messages sent by sensor nodes to the aggregation node within the normal transition step size threshold range, extract the multi-dimensional physical quantity sensing data stream from the application layer payload of the regular monitoring data message, and perform information entropy estimation processing on the multi-dimensional physical quantity sensing data stream to obtain the sensing data information entropy value. The information entropy value of the perceived data is correlated with the signal reliability score calculated based on the signal drift state characteristics of the corresponding sensor node. The information contribution of the sensor node's perceived data is evaluated based on the results of the correlation analysis, and sensor nodes whose information contribution is lower than a preset contribution threshold are marked as inefficient nodes. Before generating the network topology reconstruction strategy, the inefficient node set and the failed node set are compared and deduplicated to construct a comprehensive set of nodes to be optimized, which includes both failed and inefficient nodes. Obtain the node betweenness centrality metric value of each node to be optimized in the sensor network health status topology map of the comprehensive set of nodes to be optimized, and prioritize selecting nodes to be optimized whose node betweenness centrality metric value is lower than a preset centrality threshold and set them as priority replacement nodes. A node sleep flag is added to the priority replacement node in the working parameter adjustment instruction. The node sleep flag is set to enable the priority replacement node to enter a low-power sleep state after receiving the working parameter adjustment instruction. The topology vacancy information after the priority replacement node leaves the network is added to the network topology reconstruction strategy, and the communication link routing relationship in the network topology reconstruction strategy is adjusted so that the communication links of the remaining active nodes completely bypass the priority replacement node that is in a dormant state.

10. A sensor network health self-assessment and reconstruction system for intelligent ships, characterized in that, The device includes a processor and a computer-readable storage medium storing machine-executable instructions that, when executed by the processor, implement the sensor network health self-assessment and reconstruction method for intelligent ships as described in any one of claims 1-9.

Images