Embedded ship-borne recording equipment fault log analysis system

Abstract

The invention discloses an embedded ship-borne recording equipment fault log analysis system, and relates to the technical field of ship-borne equipment fault analysis. The embedded ship-borne recording equipment fault log analysis system comprises an operation state calling unit, an operation feature extraction unit, an operation feature analysis unit and an operation state judgment unit, and performs cleaning, feature extraction, state index calculation and judgment analysis on computing power and electrical operation state data of ship-borne equipment in a set period; according to the invention, through the design of the operation feature extraction unit and the combination of a bidirectional gating circulation unit network and a time sequence convolution network structure, operation state data of temperature, voltage, current and the like of the ship-borne recording equipment can be accurately converted into computing power and electrical state features; therefore, the recognition precision and timeliness of the abnormal state are enhanced, and the overall fault judgment capability of the system is improved.

Description

Technical Field

[0001] This invention relates to the field of shipborne equipment fault analysis technology, specifically to an embedded shipborne recording equipment fault log analysis system. Background Technology

[0002] With the rapid development of modern shipborne electronic equipment, the critical embedded systems relied upon by ships in combat, patrol, and support missions are becoming increasingly complex. These embedded shipborne recording devices are widely used in multiple key areas such as information acquisition, data processing, control command issuance, and environmental perception, bearing the important responsibility of ensuring the stable operation of shipborne mission systems. They typically feature high integration, low power consumption, and strong real-time performance, making them suitable for complex marine environments such as high temperature, high humidity, and strong interference.

[0003] During long-term operation, embedded shipborne recording equipment faces a variety of potential risk factors, such as power supply fluctuations, electrical interference, and chip thermal overload, which may lead to equipment performance degradation or even failure. To ensure the stability and availability of the equipment, shipborne systems are typically equipped with fault logging functions to collect and record various operational data and abnormal information during equipment operation for maintenance or fault tracing purposes.

[0004] Current fault log analysis methods for embedded shipborne recording devices mainly rely on single-point anomaly detection and post-event log screening. They lack systematic data retrieval, feature extraction, and status determination processes to support the device's operating status. In particular, it is difficult to achieve collaborative perception and real-time assessment of computing power and electrical status within a short period, making it difficult to expose hidden risks in critical operating phases in a timely manner.

[0005] For example, although the equipment may have shown early signs such as unstable power supply voltage and chip overheating during operation, traditional methods, due to the lack of a unified data acquisition cycle and the lack of clear multi-source feature division and analysis path, ultimately result in problems such as redundant log content, low value density, and delayed risk level identification, which affect the equipment's early warning capabilities and maintenance response efficiency. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an embedded shipborne recording device fault log analysis system, which solves the problems of unsystematic retrieval of shipborne recording device operating status data, unclear status feature extraction, and delayed fault risk assessment in existing technologies.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an embedded shipborne recording device fault log analysis system, comprising: a running status retrieval unit, used to retrieve running status data of the shipborne recording device for a set time period and perform data cleaning processing, the running status data including computing power and electrical running status data; a running feature extraction unit, used to perform dimensional division and behavioral deconstruction analysis on the cleaned running status data based on a preset feature extraction strategy to extract and form a running status feature set of the shipborne recording device, including computing power and electrical status features; a running feature analysis unit, used to analyze the running status index of the shipborne recording device based on the running status feature set, including computing power and electrical status indices; and a running status determination unit, used to match and analyze the running status index of the shipborne recording device with multiple preset running status determination interval sets, each running status determination interval set consisting of a computing power status interval and an electrical status interval, and each running status determination interval set corresponding to a fault risk level.

[0008] Furthermore, the computing power operation status data includes the chip core temperature and chip core power supply voltage at several time points, and the electrical operation status data includes the device input voltage, device input current, and device ground potential noise at several time points.

[0009] Furthermore, the preset feature extraction strategy specifically involves inputting the cleaned computing power, electrical, and structural operating status data into a pre-trained bidirectional gated recurrent unit network and a temporal convolutional network for dimensionality partitioning and behavioral deconstruction analysis.

[0010] Furthermore, the computing power status characteristics include the maximum temperature rise rate of the chip, the duration of chip overheating, the minimum power supply voltage of the chip, the number of voltage drops in the chip power supply, and the standard deviation of the chip power supply voltage fluctuation. The electrical status characteristics include the maximum amplitude of the device input voltage fluctuation, the number of sudden changes in the device input voltage, the peak value of the device input current, the number of sudden changes in the device input current, the maximum amplitude of the device ground potential noise, and the number of sudden jumps in the device ground potential.

[0011] Furthermore, the bidirectional gated recurrent unit network includes an input pre-solution layer, a sequence dependency construction layer, a trend aggregation analysis layer, a fluctuation feature deconstruction layer, and a feature encapsulation output layer.

[0012] Further, the specific steps for obtaining the computing power status characteristics of the shipborne recording equipment are as follows: In the input pre-decomposition layer of the bidirectional gated cyclic unit network, the computing power operation status data of the shipborne recording equipment is time-axis aligned to obtain a dual-channel time series, including temperature and power supply voltage channel sequences; in the sequence dependency construction layer of the bidirectional gated cyclic unit network, bidirectional gated cyclic operation is performed on the dual-channel time series to extract the context state dependency features at each time point; in the trend aggregation analysis layer of the bidirectional gated cyclic unit network, trend evolution analysis is performed on the context state feature sequence to obtain the maximum temperature rise rate of the chip, the minimum power supply voltage value of the chip, and the duration of chip overheating; in the fluctuation feature deconstruction layer of the bidirectional gated cyclic unit network, fluctuation statistics and mutation detection are performed on the power supply voltage channel sequence to obtain the number of chip power supply voltage drops and the standard deviation of chip power supply voltage fluctuation; in the feature encapsulation output layer of the bidirectional gated cyclic unit network, the output results of the trend aggregation analysis layer and the fluctuation feature deconstruction layer are encoded and integrated to form the computing power status feature set of the shipborne recording equipment.

[0013] Furthermore, the temporal convolutional network includes a channel sorting layer, a multi-scale convolutional extraction layer, a feature response aggregation layer, a mutation event recognition layer, and a feature encapsulation output layer.

[0014] Further, the specific steps for obtaining the electrical state characteristics of the shipborne recording equipment are as follows: In the channel processing layer of the temporal convolutional network, the electrical operating state data of the shipborne recording equipment is processed by channel format processing to obtain a three-channel time series; in the multi-scale convolution extraction layer of the temporal convolutional network, multi-scale convolution processing is performed on the three-channel time series to obtain the local variation characteristics of voltage, current and ground potential; in the feature response aggregation layer of the temporal convolutional network, response intensity aggregation processing is performed on the local variation characteristics to obtain the maximum amplitude of equipment input voltage fluctuation, the peak value of equipment input current and the maximum amplitude of equipment ground potential noise; in the abrupt event identification layer of the temporal convolutional network, abrupt point detection processing is performed on the local variation characteristics to obtain the number of abrupt changes in equipment input voltage, the number of abrupt changes in equipment input current and the number of abrupt jumps with equipment ground potential; in the feature encapsulation output layer of the temporal convolutional network, the output results of the feature response aggregation layer and the abrupt event identification layer are encoded and integrated, and output to form the electrical state characteristic set of the shipborne recording equipment.

[0015] Further, the specific steps for analyzing the computing power status index of the shipborne recording equipment are as follows: read the maximum temperature rise rate of the chip, the duration of chip overheating, the minimum power supply voltage of the chip, and the standard deviation of chip power supply voltage fluctuation of the shipborne recording equipment, and perform standardization processing; then, combine the standardized maximum temperature rise rate of the chip, the duration of chip overheating, the minimum power supply voltage of the chip, and the standard deviation of chip power supply voltage fluctuation with the number of chip power supply voltage drops to obtain the computing power status index of the shipborne recording equipment.

[0016] Further, the specific steps for analyzing the electrical status index of the shipborne recording equipment are as follows: read the maximum amplitude of the equipment input voltage fluctuation, the peak value of the equipment input current, and the maximum amplitude of the equipment ground potential noise of the shipborne recording equipment, and perform standardization processing; combine the standardized maximum amplitude of the equipment input voltage fluctuation, the peak value of the equipment input current, and the maximum amplitude of the equipment ground potential noise with the number of sudden changes in equipment input voltage, the number of sudden changes in equipment input current, and the number of sudden jumps in equipment ground potential to obtain the computing power status index of the shipborne recording equipment.

[0017] The present invention has the following beneficial effects:

[0018] (1) The embedded shipborne recording equipment fault log analysis system, through the design of the running feature extraction unit, adopts a structure combining bidirectional gated recurrent unit network and temporal convolutional network to transform the time series data such as chip core temperature, voltage, and current into computing power status features and electrical status features. The bidirectional gated recurrent unit network is good at extracting time-dependent features, such as trend indicators such as temperature rise rate and overheating duration. The temporal convolutional network can efficiently identify local fluctuations and sudden changes, such as the number of input voltage changes and the number of ground potential jumps. This model structure not only enhances the depth and accuracy of feature deconstruction, but also enables the model to have cross-time point pattern recognition capabilities, which can effectively uncover hidden abnormal states that are difficult to capture by traditional statistical methods. Compared with the traditional processing method that only uses statistical quantities such as maximum value and mean, this scheme significantly improves feature granularity and judgment accuracy, and is especially suitable for real-time identification of hidden risks such as sudden and electrical anomalies in shipborne environment.

[0019] (2) The embedded shipborne recording equipment fault log analysis system, through the design of operation feature analysis unit and operation status judgment unit, constructs two major index systems: computing power status index and electrical status index. The system not only performs fusion modeling based on multiple feature values, such as temperature rise rate, voltage fluctuation amplitude, and number of mutations, but also introduces preset weighting coefficients in the database, such as temperature rise rate coefficient and voltage fluctuation coefficient, to enhance the model's sensitivity and adjustment capability to different risk sources. Finally, by setting multiple status judgment interval sets, it maps different levels of computing power and electrical status indices and outputs specific fault risk levels. This mechanism supports the application of analysis results to the decision triggering of the operation and maintenance system, and has functions such as real-time early warning, level classification, and strategy response. It truly upgrades fault logs from post-event tracing to pre-event intervention, greatly improving the reliability and safety management level of shipborne equipment operation.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Figure 1 This is a block diagram of an embedded shipborne recording device fault log analysis system according to the present invention.

[0022] Figure 2 This is a flowchart illustrating the specific steps involved in obtaining the computing power status characteristics of a shipborne recording device in an embedded shipborne recording device fault log analysis system according to the present invention.

[0023] Figure 3 This is a flowchart illustrating the specific steps involved in obtaining the electrical status characteristics of a shipborne recording device in an embedded shipborne recording device fault log analysis system according to the present invention. Detailed Implementation

[0024] Please see Figure 1This invention provides a technical solution: an embedded shipborne recording device fault log analysis system, comprising: a running status retrieval unit, used to retrieve running status data of the shipborne recording device within a set time period (e.g., five minutes) and perform data cleaning processing; the running status data includes computing power and electrical running status data; a running feature extraction unit, used to perform dimensional division and behavioral deconstruction analysis on the cleaned running status data based on a preset feature extraction strategy to extract a running status feature set of the shipborne recording device, including computing power and electrical status features; a running feature analysis unit, used to analyze the running status index of the shipborne recording device based on the running status feature set, including computing power and electrical status indices; and a running status determination unit, used to match and analyze the running status index of the shipborne recording device with multiple preset running status determination interval sets, each running status determination interval set consisting of a computing power status interval and an electrical status interval, and each running status determination interval set corresponding to a fault risk level, including but not limited to the following examples:

[0025] Operating status determination interval set 1 (low temperature rise + low electrical disturbance):

[0026] Computing power range: 0.00~0.35;

[0027] Electrical condition range: 0.00~0.25;

[0028] Risk level: No risk of failure;

[0029] Note: The chip temperature rises slowly, the voltage fluctuation is stable, the electrical input is stable, there are no sudden changes or jumps, and the system is in extremely healthy operating condition.

[0030] Operating status determination interval set 2 (minor thermal disturbance + acceptable electrical disturbance):

[0031] Computing power range: 0.36~0.55;

[0032] Electrical condition range: 0.26~0.45;

[0033] Risk level: Low risk warning;

[0034] Note: The chip temperature rises slightly faster, there are occasional power supply fluctuations or slight drops, and the peak voltage and noise amplitude are slightly increased. Regular inspection is recommended.

[0035] Operating status determination interval set 3 (moderate thermal stress + multi-point electrical fluctuation):

[0036] Computing power range: 0.56~0.85;

[0037] Electrical condition range: 0.46~0.70;

[0038] Risk level: Medium failure risk;

[0039] Note: If the system shows a significant increase in the rate of temperature rise or an extended period of overheating, a low minimum power supply voltage with significant fluctuations, or an increased number of abrupt changes at the input, it is recommended to interrupt operation and perform preventative testing.

[0040] Operating status determination interval set 4 (high thermal load + frequent electrical disturbances):

[0041] Computing power range: 0.86~1.20;

[0042] Electrical condition range: 0.71~1.00;

[0043] Risk level: High-risk fault warning;

[0044] Note: The equipment is operating abnormally violently. The chip is under high temperature stress, the power supply system is unstable, and there are obvious and drastic changes in voltage and current. It is very likely in a critical fault state and needs to be investigated immediately.

[0045] Operating status determination interval set 5 (critical thermal runaway + electrical breakdown precursor):

[0046] Computing power range: 1.21 to 1.50;

[0047] Electrical condition range: 1.01~1.20;

[0048] Risk level: Severe fault condition;

[0049] Note: The chip is on the verge of thermal collapse, with frequent power supply drops, and abnormally severe input-side spikes and peaks, indicating a potential risk of chip damage or power system breakdown. It is recommended to immediately disconnect the power and send the chip for repair.

[0050] The operating status retrieval unit, feature extraction unit, feature analysis unit, and judgment unit are all integrated and deployed within the embedded processing platform of the shipborne recording device, running in an embedded operating system environment, to achieve local acquisition, edge computing, and risk identification of the device's own operating status.

[0051] The computing power operation status data includes the chip core temperature and chip core power supply voltage at several time points, and the electrical operation status data includes the device input voltage, device input current and device ground potential noise at several time points.

[0052] Among them, the core temperature value of the chip refers to the actual operating temperature of the core area of ​​the main control chip in the shipborne recording equipment at each time point. It is used to reflect the chip's heat load and heat dissipation efficiency. It can be measured and obtained by onboard temperature sensors (such as thermistors or thermocouples), and the sampling interval is usually within 100ms.

[0053] The core power supply voltage value of the chip refers to the actual power supply voltage received by the core circuit of the main control chip in the shipborne recording equipment at each time point. It is used to reflect the stability and load response capability of the chip power supply system. It can be acquired in real time by a high-precision ADC chip in conjunction with a voltage divider sampling circuit.

[0054] The device input voltage value refers to the DC or AC voltage value provided by the external power supply input terminal at each time point during the operation of the shipborne recording equipment. It is used to reflect the power supply quality of the power system and can be obtained by measuring through a voltage sampling module (such as an operational amplifier isolation circuit combined with ADC sampling).

[0055] The input current value of the equipment refers to the amount of current absorbed by the shipborne recording equipment from the power supply at each point in time. It is used to reflect the changes in equipment power and load behavior. It can be obtained in real time by using a Hall current sensor or a current sampling resistor in conjunction with a differential amplifier.

[0056] Equipment ground potential noise refers to the non-zero small voltage fluctuation that appears at the ground (GND) reference point of the shipborne recording equipment at each time point. It is usually caused by poor grounding or system interference and is used to reflect the electrical integrity of the system. It can be obtained by measuring between the ground wire and the regulated reference point through a high impedance differential measurement circuit.

[0057] Specifically, the preset feature extraction strategy involves inputting the cleaned computing power, electrical, and structural operating status data into a pre-trained bidirectional gated recurrent unit network and a temporal convolutional network for dimensionality partitioning and behavioral deconstruction analysis.

[0058] The pre-training steps for the bidirectional gated recurrent unit network are as follows:

[0059] Raw data of chip core temperature and power supply voltage of shipborne recording equipment under different operating conditions were collected and constructed into a dual-channel sequence sample group with a time step length of N. Corresponding anomaly labels were then marked based on the equipment maintenance logs, including various states such as "rapid temperature rise" and "voltage drop." Specifically, each segment of raw data was synchronized and aligned to construct time series samples of <T1,V1>, <T2,V2>, ..., <Tn,Vn>. Segments with abnormal events were marked as 1, and those without anomalies were marked as 0, using manual review or event logs. This process was used to construct the training and test sets required for supervised learning.

[0060] The constructed training samples are input into a bidirectional gated recurrent unit network (Bi-GRU), which includes forward and backward GRU state units, capable of simultaneously capturing the bidirectional dependence of temperature and voltage over time. The labeled samples are trained using a cross-entropy loss function, and weights are updated using the Adam optimizer until the validation set accuracy converges. Specifically, at each time step, the network calculates the context state vector, merges them to form a complete state tensor, and predicts whether it is an anomalous state through the classification output layer. The optimization objective is to minimize the cross-entropy loss between the predicted label and the actual label. Dropout is used during training to suppress overfitting.

[0061] The anomaly detection accuracy, recall, and F1 score of the trained model were evaluated on the test set, and the generalization ability of the model under different decision thresholds was verified using ROC-AUC curves. Once the set criteria (e.g., F1 > 0.92) were met, the trained weight parameters were frozen and embedded into the system for subsequent state feature extraction. Specifically, the trained network was tested on data not used in training. If it could accurately detect more than 90% of known thermal runaway / voltage anomaly segments and maintain a false positive rate below 5%, the model was considered to have met the deployment requirements, and the structure and parameters were fixed.

[0062] The pre-training steps for temporal convolutional networks are as follows:

[0063] The system collects three-channel signals (voltage, current, and ground potential noise) from the shipborne recording equipment and marks corresponding abrupt events (such as voltage drops, current surges, and ground potential jumps) based on manual analysis or equipment anomaly logs. Samples are divided into fixed time windows (e.g., 20 seconds) to form input sequences and event label sets. Specifically, each three-channel signal sequence is uniformly sampled to construct a sample <U(t), I(t), N(t)>, labeled with whether a sudden change or fluctuation exceeding a set threshold occurs within the window. Anomalies are marked as 1, and normalities as 0, forming a supervised sample group.

[0064] The samples are input into a Temporal Convolutional Network (TCN), and three convolutional kernels with different receptive fields are used to extract local variation features at multiple scales. The network structure includes a channel integration layer, a multi-scale convolutional layer, a channel attention fusion layer, and a classification output layer. FocalLoss is chosen as the loss function to enhance the model's ability to learn low-probability mutation events. Specifically, the convolutional kernels slide across the three channels of data at 3, 5, and 7 time steps to extract change maps, and after fusing the feature maps, predict whether mutation events exist. The training objective is to maximize the detection capability of small-sample abnormal events and avoid the mainstream normal data from masking weak variations.

[0065] The model's accuracy and false positive rate were evaluated using a validation set. Further robustness under non-standard conditions was verified using multiple measured interference signals. Once the training accuracy threshold was reached (e.g., accuracy ≥ 95%, false positive rate ≤ 3%), the model structure and parameters were frozen and imported into the actual fault log analysis module. Specifically, non-training data, such as power fluctuation anomaly test groups and electromagnetic interference test groups, were used to perform stress testing on the model. After confirming that it still possesses the ability to identify sudden changes during periods of strong interference or electrical anomalies, it was finally integrated into the embedded processing framework.

[0066] In this implementation scheme, by pre-training and solidifying the weight parameters of a bidirectional gated recurrent unit network and a temporal convolutional network, the system can achieve high-precision, low-latency automatic identification of the computing power and electrical status of shipboard recording equipment in actual deployment. The bidirectional gated recurrent unit network significantly improves the modeling ability for abnormal temperature rise and power supply drop trends through temporal coupling learning of chip temperature and voltage. Meanwhile, the temporal convolutional network, based on multi-scale receptive fields and mutation mode training, effectively enhances the robustness of identifying weak electrical disturbances and sudden anomalies. Both types of networks construct training sets based on actual historical fault logs and manual labels, making the model more consistent with the real operating characteristics of the shipboard environment. The pre-training strategy not only improves the performance stability and generalization ability of the model after it goes online, but also avoids relying on a large amount of computing resources for dynamic learning during the deployment stage, providing a reliable guarantee for lightweight fault identification in embedded scenarios.

[0067] Specifically, the computing power status characteristics include the chip's maximum temperature rise rate, the duration of chip overheating, the chip's minimum power supply voltage, the number of times the chip's power supply voltage drops, and the standard deviation of the chip's power supply voltage fluctuation.

[0068] The bidirectional gated recurrent unit network includes an input pre-solution layer, a sequence dependency construction layer, a trend aggregation analysis layer, a fluctuation feature deconstruction layer, and a feature encapsulation output layer.

[0069] like Figure 2 As shown, the specific steps to obtain the computing power status characteristics of the shipborne recording equipment are as follows: In the input pre-decomposition layer of the bidirectional gated cyclic unit network, the computing power operation status data of the shipborne recording equipment is processed by time axis alignment to obtain a dual-channel time series, including temperature and power supply voltage channel sequences. Specifically, the temperature channel and voltage channel sequences are constructed, and a set of <temperature, voltage> sample pairs are formed at each time point with consistent time steps, forming a two-dimensional sequence input tensor.

[0070] In the sequence dependency construction layer of the bidirectional gated recurrent unit network, a bidirectional gated recurrent operation is performed on the dual-channel time series to extract the context state dependency features of each time point. Specifically, for each time point, temperature-voltage state dependency vectors are constructed in the forward and backward sequences respectively, and fused into a time context expression sequence tensor for subsequent trend analysis.

[0071] In the trend aggregation analysis layer of the bidirectional gated recurrent unit network, trend evolution analysis is performed on the context state feature sequence to obtain the chip's maximum temperature rise rate, the chip's minimum power supply voltage value, and the chip's overheating duration. Specifically, the temperature rise rate is calculated by the maximum value of the temperature difference / time difference between adjacent time points, the overheating duration is the total continuous time for which the temperature exceeds a preset threshold, and the minimum voltage value is the minimum sampled value in the entire sequence.

[0072] In the fluctuation feature deconstruction layer of the bidirectional gated recurrent unit network, fluctuation statistics and mutation detection are performed on the power supply voltage channel sequence to obtain the number of chip power supply voltage drops and the standard deviation of chip power supply voltage fluctuation. Specifically, the number of drops is counted by detecting events that continuously decrease beyond a threshold, and the standard deviation of fluctuation is the standard deviation of the sampled voltage of the entire sequence, which is used to measure the overall stability of the power supply voltage.

[0073] In the feature encapsulation output layer of the bidirectional gated cyclic unit network, the output results of the trend aggregation analysis layer and the fluctuation feature deconstruction layer are encoded and integrated to form the computing power status feature set of the shipborne recording device.

[0074] The specific steps for analyzing the computing power status index of shipborne recording equipment are as follows: Read the maximum temperature rise rate of the chip, the duration of chip overheating, the minimum power supply voltage of the chip, and the standard deviation of chip power supply voltage fluctuation of the shipborne recording equipment, and perform standardization processing (i.e., unit removal); Combine the standardized maximum temperature rise rate of the chip, the duration of chip overheating, the minimum power supply voltage of the chip, and the standard deviation of chip power supply voltage fluctuation with the number of chip power supply voltage drops to obtain the computing power status index of the shipborne recording equipment.

[0075] The specific formula for calculating the computing power status index of shipborne recording equipment is as follows: ;in, , , , , , The following are, in order: computing power status index of the shipborne recording equipment, maximum chip temperature rise rate, chip overheating duration, minimum chip power supply voltage, number of chip power supply voltage drops, and standard deviation of chip power supply voltage fluctuation. , , The values ​​are, in order, the temperature rise rate coefficient, the overheating duration coefficient, and the voltage regulation factor (used to prevent the denominator from being 0) stored in the database, and in this embodiment, they are 0.72, 0.65, and 0.01, respectively.

[0076] In this implementation scheme, the computing power status feature extraction and exponential calculation mechanism of the bidirectional gated cyclic unit network possesses strong engineering practicality and technological advancement. On the one hand, the system can simultaneously mine the dynamic coupling characteristics of chip temperature and power supply voltage through the construction of dual-channel time series; by leveraging the deep modeling of the time context through the bidirectional cyclic structure, it can effectively capture potential thermal runaway trends and power supply anomaly signals. On the other hand, through two layered processing modules, trend aggregation analysis and fluctuation deconstruction, the system can extract core features with high sensitivity and representativeness, such as the maximum temperature rise rate, the minimum power supply voltage, and the number of drops, laying a solid foundation for subsequent risk index assessment. At the same time, the index calculation formula integrates standardized feature values ​​and weighted adjustment coefficients, which not only avoids the deviation problem caused by the direct superposition of different physical quantities, but also allows for flexible adaptation to the actual operating characteristics of different shipborne equipment by adjusting parameters. Ultimately, the computing power status index formed by the system has high comprehensiveness and identification capabilities, and can serve as a key reference indicator in embedded fault log analysis, effectively supporting the judgment of operational health status and the optimization of maintenance strategies.

[0077] Specifically, electrical condition characteristics include the maximum amplitude of equipment input voltage fluctuation, the number of abrupt changes in equipment input voltage, the peak value of equipment input current, the number of abrupt changes in equipment input current, the maximum amplitude of equipment ground potential noise, and the number of abrupt jumps in equipment ground potential.

[0078] Temporal convolutional networks include channel sorting layers, multi-scale convolutional extraction layers, feature response aggregation layers, mutation event recognition layers, and feature encapsulation output layers.

[0079] like Figure 3 As shown, the specific steps to obtain the electrical state characteristics of the shipborne recording equipment are as follows: In the channel processing layer of the temporal convolutional network, the electrical operating state data of the shipborne recording equipment is processed by channel format to obtain a three-channel time series. Specifically, the three signals are constructed into a time series of [voltage(t), current(t), and ground noise(t)] according to the time steps to form a unified 3×T channel tensor structure.

[0080] In the multi-scale convolution extraction layer of the temporal convolutional network, multi-scale convolution processing is performed on the three-channel time series to obtain the local variation features of voltage, current and ground potential. Specifically, feature maps are extracted by scanning with convolution kernels of different widths to obtain local feature responses such as voltage peaks, abnormal current rises and small-period ground noise jumps.

[0081] In the feature response aggregation layer of the temporal convolutional network, the response intensity aggregation processing is performed on the local change features to obtain the maximum amplitude of device input voltage fluctuation, device input current peak value and device ground potential noise. Specifically, the maximum amplitude of voltage fluctuation is the maximum amplitude difference in the local convolutional response, the current peak value is the maximum current response point, and the maximum amplitude of ground noise is the maximum difference amplitude of the ground potential channel.

[0082] In the abrupt event recognition layer of the temporal convolutional network, abrupt point detection processing is performed on local change features to obtain the number of abrupt changes in device input voltage, the number of abrupt changes in device input current, and the number of abrupt jumps in device ground potential. Specifically, the number of abrupt voltage changes is the number of times the voltage difference between adjacent time points exceeds the threshold, the number of abrupt current changes is identified by the current slope exceeding the limit, and the number of abrupt jumps in ground potential is the detection result of instantaneous jump events.

[0083] In the feature encapsulation output layer of the temporal convolutional network, the output results of the feature response aggregation layer and the mutation event recognition layer are encoded and integrated, and the output forms the electrical state feature set of the shipborne recording device.

[0084] The specific steps for analyzing the electrical status index of shipborne recording equipment are as follows: Read the maximum amplitude of the equipment input voltage fluctuation, the peak value of the equipment input current, and the maximum amplitude of the equipment ground potential noise from the shipborne recording equipment, and perform standardization processing (i.e., unit removal). Then, combine the standardized maximum amplitude of the equipment input voltage fluctuation, the peak value of the equipment input current, and the maximum amplitude of the equipment ground potential noise with the number of abrupt changes in equipment input voltage, the number of abrupt changes in equipment input current, and the number of abrupt jumps in equipment ground potential to obtain the computing power status index of the shipborne recording equipment.

[0085] The specific formula for calculating the electrical status index of shipborne recording equipment is as follows: ;in, , , , , , , The following are, in order: electrical condition index of the shipborne recording equipment, maximum amplitude of equipment input voltage fluctuation, peak value of equipment input current, maximum amplitude of equipment ground potential noise, number of abrupt changes in equipment input voltage, number of abrupt changes in equipment input current, and number of abrupt jumps in equipment ground potential. , , , The values ​​are, in order, the voltage fluctuation coefficient, peak current coefficient, noise amplitude coefficient, and joint mutation risk coefficient stored in the database, and in this embodiment, they are 0.673, 0.521, 0.438, and 0.137, respectively.

[0086] In this implementation scheme, the sensitivity and intelligence level of electrical anomaly identification of shipborne recording equipment are significantly improved by using an electrical state feature extraction and index calculation method based on temporal convolutional networks. First, through collaborative modeling of the channel sorting layer and the multi-scale convolutional extraction layer, local temporal features can be extracted from the three channels of input voltage, current and ground potential noise, effectively capturing complex phenomena such as drastic voltage fluctuations, current surges or ground disturbances. Second, the system not only extracts intensity-type indicators such as peak value and maximum fluctuation amplitude, but also accurately counts electrical sudden anomalies in conjunction with the abrupt event identification layer, ensuring coverage and identification of boundary problems and discontinuous anomalies. Finally, through the fusion and integration of the feature encapsulation output layer, and using a weighted index formula designed with parameters such as voltage fluctuation coefficient and peak current coefficient, the multi-dimensional electrical anomaly characterization indicators are condensed into a single electrical state index, realizing a quantitative expression of the electrical operation stability and risk of the equipment.

[0087] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0088] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An embedded shipborne recording device fault log analysis system, characterized in that, include: The operation status retrieval unit is used to retrieve the operation status data of the shipborne recording equipment for a set time period and perform data cleaning processing. The operation status data includes computing power and electrical operation status data. The feature extraction unit is used to perform dimensional division and behavioral deconstruction analysis on the cleaned and processed operational status data based on a preset feature extraction strategy, and extract the operational status feature set of the shipborne recording equipment, including computing power and electrical status features. The operational feature analysis unit is used to analyze the operational status index of the shipborne recording equipment based on the operational status feature set, including computing power and electrical status index. The operation status determination unit is used to match and analyze the operation status index of the shipborne recording equipment with multiple preset operation status determination interval sets. Each operation status determination interval set consists of a computing power status interval and an electrical status interval, and each operation status determination interval set corresponds to a fault risk level.

2. The embedded shipborne recording device fault log analysis system according to claim 1, characterized in that, The computing power operation status data includes the chip core temperature and chip core power supply voltage at several time points, and the electrical operation status data includes the device input voltage, device input current and device ground potential noise at several time points.

3. The embedded shipborne recording device fault log analysis system according to claim 1, characterized in that, The preset feature extraction strategy specifically involves inputting the cleaned computing power, electrical, and structural operation status data into a pre-trained bidirectional gated recurrent unit network and a temporal convolutional network for dimensionality partitioning and behavioral deconstruction analysis.

4. The embedded shipborne recording device fault log analysis system according to claim 1, characterized in that, The computing power status characteristics include the maximum temperature rise rate of the chip, the duration of chip overheating, the minimum power supply voltage of the chip, the number of voltage drops in the chip power supply, and the standard deviation of the chip power supply voltage fluctuation. The electrical status characteristics include the maximum amplitude of the device input voltage fluctuation, the number of sudden changes in the device input voltage, the peak value of the device input current, the number of sudden changes in the device input current, the maximum amplitude of the device ground potential noise, and the number of sudden jumps in the device ground potential.

5. The embedded shipborne recording device fault log analysis system according to claim 3, characterized in that, The bidirectional gated recurrent unit network includes an input pre-solution layer, a sequence dependency construction layer, a trend aggregation analysis layer, a fluctuation feature deconstruction layer, and a feature encapsulation output layer.

6. The embedded shipborne recording device fault log analysis system according to claim 5, characterized in that, The specific steps to obtain the computing power status characteristics of the shipborne recording equipment are as follows: In the input pre-decomposition layer of the bidirectional gated cyclic unit network, the computing power operation status data of the shipborne recording equipment is time-axis aligned to obtain a dual-channel time series, including temperature and power supply voltage channel sequences. In the sequence dependency construction layer of the bidirectional gated recurrent unit network, a bidirectional gated recurrent operation is performed on the dual-channel time series to extract the context state dependency features at each time point. In the trend aggregation analysis layer of the bidirectional gated cyclic unit network, trend evolution analysis is performed on the context state feature sequence to obtain the chip's maximum temperature rise rate, the chip's minimum power supply voltage, and the chip's overheating duration. In the fluctuation feature deconstruction layer of the bidirectional gated cyclic unit network, fluctuation statistics and mutation detection are performed on the power supply voltage channel sequence to obtain the number of chip power supply voltage drops and the standard deviation of chip power supply voltage fluctuation. In the feature encapsulation output layer of the bidirectional gated cyclic unit network, the output results of the trend aggregation analysis layer and the fluctuation feature deconstruction layer are encoded and integrated to form the computing power status feature set of the shipborne recording device.

7. The embedded shipborne recording device fault log analysis system according to claim 3, characterized in that, Temporal convolutional networks include channel sorting layers, multi-scale convolutional extraction layers, feature response aggregation layers, mutation event recognition layers, and feature encapsulation output layers.

8. The embedded shipborne recording device fault log analysis system according to claim 7, characterized in that, The specific steps for obtaining the electrical status characteristics of shipborne recording equipment are as follows: In the channel formatting layer of the temporal convolutional network, the electrical operating status data of the shipborne recording equipment is processed by channel formatting to obtain a three-channel time series. In the multi-scale convolution extraction layer of the temporal convolutional network, multi-scale convolution processing is performed on the three-channel time series to obtain the local variation features of voltage, current and ground potential; In the feature response aggregation layer of the temporal convolutional network, the response intensity of local change features is aggregated to obtain the maximum amplitude of device input voltage fluctuation, the peak value of device input current, and the maximum amplitude of device ground potential noise. In the abrupt event recognition layer of the temporal convolutional network, abrupt point detection processing is performed on the local change features to obtain the number of abrupt changes in device input voltage, the number of abrupt changes in device input current, and the number of abrupt jumps with device ground potential. In the feature encapsulation output layer of the temporal convolutional network, the output results of the feature response aggregation layer and the mutation event recognition layer are encoded and integrated, and the output forms the electrical state feature set of the shipborne recording device.

9. The embedded shipborne recording device fault log analysis system according to claim 4, characterized in that, The specific steps for analyzing the computing power status index of shipborne recording equipment are as follows: Read the maximum temperature rise rate of the chip, the duration of chip overheating, the minimum power supply voltage of the chip, and the standard deviation of chip power supply voltage fluctuation of the shipborne recording equipment, and perform standardization processing. The computing power status index of the shipborne recording equipment will be obtained by comprehensively analyzing the standardized chip maximum temperature rise rate, chip overheating duration, chip minimum power supply voltage, and chip power supply voltage fluctuation standard deviation, combined with the number of chip power supply voltage drops.

10. The embedded shipborne recording device fault log analysis system according to claim 4, characterized in that, The specific steps for analyzing the electrical status index of shipborne recording equipment are as follows: The maximum amplitude of the device input voltage fluctuation, the peak value of the device input current, and the maximum amplitude of the device ground potential noise of the shipborne recording equipment are read and standardized. The maximum amplitude of the standardized equipment input voltage fluctuation, the peak value of the equipment input current, and the maximum amplitude of the equipment ground potential noise are combined with the number of sudden changes in equipment input voltage, the number of sudden changes in equipment input current, and the number of sudden jumps in equipment ground potential to obtain the computing power status index of the shipborne recording equipment.

Images