HDI communication board fault prediction method and system based on neural network model

Abstract

This invention provides a method and system for HDI communication board fault prediction based on a neural network model, belonging to the field of artificial intelligence technology. First, it collects various fault signal manifestations of the HDI communication board to generate a fault signal demand set, constructs an initial neural network model, and obtains an adaptation relationship framework. Based on this, the model is adjusted to obtain a neural network model adapted to the fault signal demand. Next, it obtains a set of real-time operating signals and inputs them into the neural network model. Through hierarchical extraction and association mining, a fault feature association set is obtained. Then, combined with the corresponding relationships, the fault type and trend are determined, generating a fault prediction result set. Finally, it binds the results with the operating signal acquisition time to generate a fault prediction report set containing time-related information. This invention improves the accuracy and timeliness of HDI communication board fault prediction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method and system for predicting HDI communication board faults based on a neural network model. Background Technology

[0002] In the field of electronic communications, HDI (High-Density Interconnect) communication boards have become a key component of many high-end electronic devices due to their high integration and miniaturization. However, due to their complex structure and sophisticated manufacturing process, HDI communication boards are prone to various malfunctions during operation. These malfunctions can not only affect the performance of the communication board but may also paralyze the entire electronic device system, causing serious economic losses.

[0003] Currently, fault prediction for HDI communication boards mainly relies on traditional detection methods and experience-based judgment. Traditional detection methods typically involve testing and repairing after a fault occurs, a reactive approach that cannot prevent faults from happening in advance and is insufficient to meet the high reliability and stability requirements of modern electronic equipment. Experience-based judgment, on the other hand, is limited by the professional level and experience of the testing personnel, resulting in strong subjectivity and low accuracy, making it difficult to accurately predict some highly concealed faults. Furthermore, most existing fault prediction technologies only consider single fault signals, lacking comprehensive analysis and correlation mining of multiple fault signals, and thus failing to fully and accurately reflect the operating status and fault trends of the HDI communication board. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for predicting HDI communication board faults based on a neural network model, the method comprising: The fault signal manifestations corresponding to various faults of the HDI communication board are collected, and a fault signal demand set is generated in combination with the operating mechanism of the HDI communication board. At the same time, an initial neural network model with basic signal processing functions is constructed. Based on the signal dimensions in the fault signal demand set and the hierarchical structure of the initial neural network model, a matching framework for the adaptation relationship between the fault signal demand set and the initial neural network model is obtained. Based on the adaptation framework between the fault signal demand set and the initial neural network model, the number of hierarchical nodes, hierarchical connection method and signal processing channel of the initial neural network model are adapted and adjusted to obtain a neural network model adapted to the fault signal demand. The real-time operating signal set of the HDI communication board is obtained. The real-time operating signal set of the HDI communication board is input into a neural network model that adapts to the fault signal requirements. Through the fault signal guidance mechanism preset in the model, the fault-related signals in the real-time operating signal set are extracted and associated in a hierarchical manner to obtain the fault feature association set. Based on the fault feature association set and combined with the correspondence between fault signals and fault types in the fault signal demand set, the fault type and fault occurrence trend of the HDI communication board are judged, and a fault prediction result set of the HDI communication board is generated. The set of HDI communication board fault prediction results is bound to the HDI communication board's operating signal acquisition time to generate a set of HDI communication board fault prediction reports containing time-related information.

[0005] In another aspect, embodiments of the present invention also provide an HDI communication board fault prediction system based on a neural network model, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to run the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0006] Based on the above, this embodiment of the invention comprehensively collects the fault signal manifestations corresponding to various faults of the HDI communication board and generates a fault signal demand set in combination with its operating mechanism, effectively avoiding prediction deviations caused by missing or inaccurate information. Based on the adaptation framework between the fault signal demand set and the initial neural network model, the initial neural network model is precisely adapted and adjusted to obtain a neural network model adapted to the fault signal demand, enabling the model to better handle the fault signals of the HDI communication board and improving the model's relevance and effectiveness. After obtaining the real-time operating signal set of the HDI communication board and inputting it into the adaptation model, the fault-related signals in the real-time operating signals are extracted and correlated hierarchically through the fault signal guidance mechanism preset within the model. This allows for in-depth exploration of the intrinsic connections between fault signals, discovering potential fault features that are difficult to detect using traditional methods, thereby generating an accurate fault feature correlation set. Based on the fault feature association set and the correspondence between fault signals and fault types, the fault type and fault occurrence trend of HDI communication board are judged, and a fault prediction report set containing time correlation information is generated. This enables early prediction and accurate location of HDI communication board faults, improves the accuracy and timeliness of HDI communication board fault prediction, effectively reduces the probability of fault occurrence, and enhances the reliability and stability of electronic equipment. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the HDI communication board fault prediction method based on a neural network model provided in an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of the hardware architecture of the HDI communication board fault prediction system based on a neural network model provided in an embodiment of the present invention. Detailed Implementation

[0009] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a fault prediction method for HDI communication boards based on a neural network model, provided in one embodiment of the present invention. The following is a detailed description of this fault prediction method for HDI communication boards based on a neural network model.

[0010] Step S110: Collect the fault signal manifestations corresponding to various faults of the HDI communication board, generate a fault signal demand set in combination with the operating mechanism of the HDI communication board, and at the same time construct an initial neural network model with basic signal processing functions. Based on the signal dimensions in the fault signal demand set and the hierarchical structure of the initial neural network model, match them to obtain the adaptation relationship framework between the fault signal demand set and the initial neural network model.

[0011] In this embodiment, a certain type of HDI communication board is used as the application object. This HDI communication board mainly consists of a signal processing module, a power management module, a data transmission module, and an interface module. During actual operation, various fault types may occur, such as signal transmission interruption, power fluctuation, and poor interface contact. First, signal acquisition and model building are required to address these fault types. During the signal acquisition phase, corresponding sensors are installed at the test points of each key module of the HDI communication board. For example, voltage and current sensors are installed at the output of the power management module, signal oscilloscope probes are installed on the differential signal lines of the data transmission module, and contact resistance testers are installed at the pins of the interface module to obtain operational signal data when various faults occur. Simultaneously, when building the initial neural network model, considering the complexity of signal processing on the HDI communication board, a basic architecture combining convolutional neural networks and recurrent neural networks is chosen. The signal input layer is responsible for receiving raw operational signals from multiple channels. The multi-level signal processing layer includes convolutional layers for extracting spatial features, recurrent layers for capturing temporal features, an attention mechanism for weighted fusion of features at different levels in the signal fusion layer, and the signal output layer outputs the preliminary feature processing results. When matching signal dimensions with model hierarchical structure, it is necessary to map the voltage signal dimension and current signal dimension of the power supply module, and the signal amplitude dimension, frequency dimension, and phase dimension of the data transmission module to different levels of the model. For example, basic signal dimensions such as voltage and current are assigned to the first-level signal processing layer of the initial neural network model, while complex dimensions such as the spectral characteristics of the signal are assigned to higher levels, thus forming an adaptation framework.

[0012] Step S111: Collect operating signal data of the HDI communication board when various faults occur under different operating conditions, classify and organize the collected operating signal data, and generate a list of fault signal manifestations of the HDI communication board.

[0013] In this embodiment, different operating conditions include the HDI communication board's operation under full load, half load, no load, and different ambient temperatures (e.g., T1, T2, T3 degrees Celsius). For signal transmission interruption faults, simulation tests were conducted under different operating conditions. When the HDI communication board experiences a signal transmission interruption, the signal acquired by the signal oscilloscope through the data transmission module shows that the differential signal amplitude suddenly drops to zero, and the duration exceeds T units of time. Simultaneously, the contact resistance value of the interface module rises sharply to above R1 units of resistance. For power fluctuation faults, under full load and high temperature conditions, the voltage signal at the output of the power management module fluctuates around the standard output voltage, with a fluctuation amplitude reaching P percentage of the standard voltage and a fluctuation frequency of F Hz. The current signal exhibits periodic changes synchronized with the voltage fluctuations. The collected signal data of different faults under different operating conditions are classified according to the fault type, such as signal transmission interruption, power fluctuation, poor interface contact, etc. For each type of fault, the signal data is recorded with information such as signal type (voltage, current, resistance, signal amplitude, etc.), signal change range, duration, etc. Finally, a list of fault signal manifestations of HDI communication board is compiled.

[0014] Step S112: Analyze the operating mechanism of the HDI communication board, extract the signal transmission paths and signal interaction methods of each component of the HDI communication board, and generate an HDI communication board operating mechanism description document.

[0015] In this embodiment, the operation mechanism of the HDI communication board involves the collaborative workflow between its various components. The signal processing module, as the core, receives raw data signals from the external interface module. After filtering and decoding by its internal digital signal processing unit, the processed signal is transmitted to the data transmission module. The data transmission module interacts with external devices at high speed via differential signal lines. Simultaneously, the power management module provides a stable DC power supply to the signal processing module, data transmission module, and interface module, and includes voltage regulation and overcurrent protection circuits. Regarding the signal transmission path, external data signals enter through the pins of the interface module, pass through the electrostatic discharge protection circuit, and are transmitted to the input terminal of the signal processing module. The output signal of the signal processing module is transmitted to the transmitting terminal of the data transmission module via an internal bus. The output voltage of the power management module is connected to the power input terminals of each module via copper foil traces. In terms of signal interaction, the signal processing module and the data transmission module use a serial communication protocol for data exchange. Each transmitted data frame includes a start bit, data bits, parity bit, and stop bit. The power management module interacts with other modules via a power enable signal. When the enable signal is high, the power module starts outputting voltage. The signal transmission paths and signal interaction methods of each of the above components are recorded in detail to generate an HDI communication board operation mechanism description document.

[0016] Step S113: Based on the HDI communication board fault signal performance list and the HDI communication board operation mechanism description document, extract the signal dimensions corresponding to various faults, describe the signal characteristics and signal acquisition requirements of each signal dimension, and generate a fault signal dimension description set.

[0017] In this embodiment, taking a signal transmission interruption fault as an example, based on the signal data of the fault in the fault signal manifestation list and the signal transmission path of the data transmission module in the operation mechanism description document, the corresponding signal dimensions are extracted, including differential signal amplitude, signal transmission frequency, signal rise time, and interface contact resistance. For the differential signal amplitude dimension, its signal characteristic is that when the fault occurs, the amplitude suddenly drops from the normal operating voltage unit V1 to zero, and the time of the drop is less than time unit T1. The signal acquisition requirement is to use an oscilloscope probe with a bandwidth of not less than frequency unit B1 to acquire the signal at the differential signal output of the data transmission module, with the sampling frequency set to frequency unit F1, and the acquisition duration not less than time unit T2. For the interface contact resistance dimension, the signal characteristic is that when the fault occurs, the resistance value rapidly rises from the normal resistance unit less than R2 to above R1. The signal acquisition requirement is to use a high-precision milliohm meter to measure between the pins of the interface module and the corresponding connector, with a measurement accuracy of R3 resistance units and a measurement interval of time unit T3. In the same way, output voltage, output current, voltage ripple and other signal dimensions are extracted from power supply fluctuation faults, and their signal characteristics and acquisition requirements are described respectively. All this information is then integrated to generate a fault signal dimension description set.

[0018] Step S1131: Select a single type of fault signal from the list of fault signal manifestations of the HDI communication board, and in conjunction with the HDI communication board operation mechanism documentation, define the HDI communication board operation phase corresponding to this type of fault signal manifestation.

[0019] In this embodiment, a power fluctuation fault signal is selected from the HDI communication board fault signal list. According to the HDI communication board operation mechanism documentation, the power management module is responsible for converting the input AC voltage into the DC voltage required by each module, and it internally includes rectifier circuits, filter circuits, voltage regulator circuits, etc. When the feedback resistor in the voltage regulator circuit of the power management module ages, it will cause the output voltage to fluctuate. Therefore, the HDI communication board operation corresponding to this power fluctuation fault signal is the operation of the voltage regulator circuit of the power management module.

[0020] Step S1132: Analyze the signal transmission paths and signal interaction nodes involved in this operation process, and define the signal types related to the performance of this type of fault signal.

[0021] In this embodiment, the signal transmission path involved in the operation of the voltage regulator circuit of the power management module includes the transmission of the input voltage signal from the output of the rectifier circuit to the input of the filter circuit, the output of the filter circuit being connected to the input of the voltage regulator chip, and the output of the voltage regulator chip being connected to the power input of each module. Signal interaction nodes include the connection nodes between the rectifier circuit and the filter circuit, the connection nodes between the filter circuit and the voltage regulator chip, and the connection nodes between the voltage regulator chip and the power input of each module. Analysis shows that the signal types related to power fluctuation fault signals include the input voltage signal, output voltage signal, feedback voltage signal of the voltage regulator circuit, and the current signal passing through the voltage regulator chip.

[0022] Step S1133: Extract signal dimensions that can characterize the fault signal performance from the defined signal type. Each signal dimension corresponds to a specific attribute of the signal.

[0023] In this embodiment, for power fluctuation faults, signal dimensions are extracted from the signal types defined above. For output voltage signals, the extracted signal dimensions include voltage amplitude, voltage fluctuation amplitude, and fluctuation frequency; for feedback voltage signals, the extracted signal dimensions include feedback voltage value and feedback voltage change rate; for current signals passing through the voltage regulator chip, the extracted signal dimensions include current amplitude and current change rate. These signal dimensions correspond to different specific attributes of the signal, such as voltage amplitude representing the magnitude of the output voltage, voltage fluctuation amplitude representing the degree to which the voltage deviates from the standard value, and fluctuation frequency representing the speed of voltage fluctuation, etc.

[0024] Step S1134: Describe the features of each extracted signal dimension, define the specific manifestation and change pattern of the signal dimension when the fault occurs, and generate a signal dimension feature description.

[0025] In this embodiment, taking the voltage fluctuation amplitude of the output voltage signal as an example, its specific manifestation when a power fluctuation fault occurs is that the voltage value changes periodically around the standard output voltage V0, exhibiting a sinusoidal pattern. The fluctuation amplitude gradually increases from ΔV1 during normal operation to ΔV2, where ΔV2 is greater than ΔV1. For the feedback voltage change rate dimension of the feedback voltage signal, when a fault occurs, its absolute value increases from the normal K1 to K2, and the direction of the change rate is opposite to the direction of the output voltage change. That is, when the output voltage increases, the feedback voltage change rate is negative, and when the output voltage decreases, the feedback voltage change rate is positive. Each extracted signal dimension is characterized in the above manner to generate a signal dimension feature description.

[0026] Step S1135: Based on the HDI communication board operation mechanism documentation and actual operation requirements, define the signal acquisition method, acquisition frequency and acquisition range for each signal dimension, and generate signal dimension acquisition requirements.

[0027] In this embodiment, for the voltage amplitude dimension of the output voltage signal, based on the output voltage range of the power management module in the operating mechanism documentation and the actual operating requirements, the acquisition method uses a differential probe connected to the output pin of the voltage regulator chip. The acquisition frequency is set to F2 frequency units to ensure that rapid voltage changes can be captured. The acquisition range is set to Vmin to Vmax voltage units, where Vmin is P1 percentage of the standard output voltage and Vmax is P2 percentage of the standard output voltage. For the current amplitude dimension of the current signal passing through the voltage regulator chip, the acquisition method uses a series current probe, with the acquisition frequency also set to F2 frequency units. The acquisition range is Imax to Imin current units, where Imax is P3 percentage of the maximum output current of the power management module and Imin is zero current units. Following the above method, the acquisition method, acquisition frequency, and acquisition range are defined for each signal dimension, generating signal dimension acquisition requirements.

[0028] Step S1136: Bind the signal dimension, signal dimension feature description and signal dimension acquisition requirements corresponding to the single type of fault signal to generate a signal dimension description subset for the single type of fault.

[0029] In this embodiment, the signal dimensions corresponding to power fluctuation faults, such as the voltage amplitude, voltage fluctuation range, and fluctuation frequency of the output voltage, the feedback voltage value and feedback voltage change rate of the feedback voltage, and the current amplitude and current change rate of the current signal, are bound to their corresponding signal dimension feature descriptions and signal dimension acquisition requirements. For example, the output voltage amplitude dimension is bound to its feature description "During the fault, the voltage fluctuates around V0 according to a sine wave pattern, and the fluctuation amplitude increases from ΔV1 to ΔV2" and the acquisition requirement "Differential probe acquisition, acquisition frequency F2, range Vmin to Vmax" to form a complete record. All the above records are combined to generate a subset of signal dimension descriptions for power fluctuation faults.

[0030] Step S1137: Perform the above-mentioned selection, definition, extraction, feature description, definition of acquisition requirements and binding processing steps on all fault signal manifestations in the HDI communication board fault signal manifestation list to generate a signal dimension description subset for each type of fault.

[0031] In this embodiment, in addition to power fluctuation faults, steps S1131 to S1136 are also performed on all fault signal manifestations in the HDI communication board fault signal manifestation list, such as signal transmission interruption faults and interface contact failure faults. For signal transmission interruption faults, after selecting its signal manifestation, the corresponding operating link is defined as the differential signal transmission link of the data transmission module. Signal dimensions such as differential signal amplitude, signal transmission frequency, and signal rise time are extracted to describe the characteristics of each dimension. For example, the differential signal amplitude drops rapidly from the normal V3 to zero during a fault. The acquisition method is defined as oscilloscope probe acquisition, and the acquisition frequency is F3, etc. Then, the above information is bound to generate a signal dimension description subset for signal transmission interruption faults. Other faults are processed in the same way to obtain signal dimension description subsets for various faults.

[0032] Step S1138: Summarize the signal dimension description subsets of all types of faults and remove duplicate signal dimension descriptions.

[0033] In this embodiment, the signal dimension description subsets of various faults, such as power fluctuation faults, signal transmission interruption faults, and interface contact failure faults, are summarized. During the summarization process, it was found that different faults may involve the same signal dimension. For example, both interface contact failures and signal transmission interruption faults involve the signal dimension of interface contact resistance, and their signal dimension feature descriptions and acquisition requirements are the same. In this case, duplicate descriptions need to be removed, and only one complete description of the signal dimension needs to be retained.

[0034] Step S1139: Classify and sort the summarized signal dimension descriptions, and divide the signal dimension descriptions into different categories according to the signal type.

[0035] In this embodiment, the summarized signal dimension descriptions include voltage-related signal dimensions, current-related signal dimensions, resistance-related signal dimensions, signal amplitude-related signal dimensions, and frequency-related signal dimensions. These descriptions are categorized according to signal type: all voltage-related signal dimensions, such as output voltage amplitude and feedback voltage value, are classified as voltage; current amplitude and current change rate are classified as current; interface contact resistance is classified as resistance, and so on, forming different categories.

[0036] Step S11310: Integrate the classified and sorted signal dimension descriptions to generate a fault signal dimension description set.

[0037] In this embodiment, the classified and sorted descriptions of various signal dimensions such as voltage, current, and resistance are integrated in a certain order, such as first voltage, then current, then resistance, etc., to form a fault signal dimension description set, which contains detailed information on all signal dimensions corresponding to various faults of the HDI communication board.

[0038] Step S114: Perform correlation analysis on the signal dimensions in the fault signal dimension description set, define the signal dimension combinations corresponding to different fault types, and generate a fault signal requirement subset.

[0039] In this embodiment, a correlation analysis is performed on the signal dimensions in the fault signal dimension description set to analyze which signal dimensions will simultaneously exhibit anomalies when different fault types occur. For example, when a signal transmission interruption fault occurs, anomalies in signal dimensions such as differential signal amplitude dropping to zero, interface contact resistance value increasing significantly, and signal transmission frequency becoming zero are typically observed. Therefore, these signal dimensions are combined together as the signal dimension combination corresponding to the signal transmission interruption fault. For power fluctuation faults, the corresponding signal dimension combination includes increased output voltage fluctuation amplitude, abnormal feedback voltage change rate, and current amplitude fluctuation. The signal dimension combinations corresponding to each fault type are organized to generate a fault signal requirement subset, with each subset corresponding to one fault type.

[0040] Step S115: Integrate all fault signal requirement subsets, supplement the priority information and signal transmission timing information of various fault signals, and generate a fault signal requirement set.

[0041] In this embodiment, the various fault signal requirement subsets are integrated together, and priority information and signal transmission timing information of the fault signals are supplemented. The priority information is set according to the severity of the fault and the importance of the signal; for example, in a power fluctuation fault, the output voltage signal has a higher priority than the current signal. The signal transmission timing information describes the order in which different signal dimensions appear during the fault occurrence process; for example, in a signal transmission interruption fault, the interface contact resistance value rises before the differential signal amplitude decreases. These supplementary information are added to the integrated fault signal requirement subsets to form a complete fault signal requirement set.

[0042] Step S116: Construct the basic architecture of the initial neural network model. This basic architecture includes a signal input layer, a multi-level signal processing layer, a signal fusion layer, and a signal output layer. The signal input layer is configured to receive and process signal data. The multi-level signal processing layer is configured to process signal features layer by layer. The signal fusion layer is configured to fuse signal features from different levels. The signal output layer is configured to output the preliminary signal processing results.

[0043] In this embodiment, the basic architecture for constructing the initial neural network model is defined. The signal input layer employs a multi-channel input structure, capable of receiving various types of operational signal data from different sensors, such as voltage, current, and resistance signals. Each channel corresponds to a signal type, and the number of neurons in the input layer is determined based on the number of signal channels and the number of sampling points per channel. The multi-level signal processing layer includes a first-level processing layer, a second-level processing layer, and a third-level processing layer. The first-level processing layer uses convolutional layers to extract local features of the signal; the second-level processing layer uses recurrent neural network layers, such as LSTM layers, to capture the temporal features of the signal; and the third-level processing layer uses fully connected layers to further abstract and process the features. The signal fusion layer uses an attention mechanism to weight and fuse the features output from the multi-level signal processing layers, giving higher weights to important features. The signal output layer uses fully connected layers to output preliminary signal processing results, such as feature vectors.

[0044] Step S117: Extract the signal processing capability information of each layer of the initial neural network model, covering the range of signal dimensions that each layer can process, the signal processing speed, and the signal transmission format between layers, and generate a list of the layer capabilities of the initial neural network model.

[0045] In this embodiment, for the signal input layer, the processed signal dimensions include various signal types such as voltage, current, and resistance. The number of sampling points for each signal type ranges from N1 to N2, the signal processing speed is S1 samples per second, and the signal transmission format between layers is tensor format with a shape of [B, C, L], where B is the batch size, C is the number of channels, and L is the signal length. For the first-level convolutional processing layer, the processed signal dimensions are the same as the feature dimensions output by the input layer, the signal processing speed is S2 samples per second, and the transmission format is [B, C1, L1], where C1 is the number of channels after convolution, and L1 is the signal length after convolution. Similarly, the signal processing capability information of the multi-level signal processing layer, signal fusion layer, and signal output layer is extracted and summarized to generate an initial neural network model layer capability list.

[0046] Step S118: Compare the signal dimensions and signal processing requirements in the fault signal requirement set with the initial neural network model level capability list one by one, define the initial neural network model level and processing node corresponding to each signal dimension, and generate a correspondence table between signal dimensions and model levels.

[0047] In this embodiment, the signal dimensions in the fault signal requirement set, such as differential signal amplitude and output voltage amplitude, each have their own processing requirements, such as feature extraction and time series analysis. These signal dimensions and their processing requirements are compared one by one with the range of signal dimensions and processing functions that each layer in the initial neural network model's layer capability list can handle. For example, the differential signal amplitude signal dimension requires local feature extraction, which matches the function of the first-level convolutional processing layer; therefore, it is mapped to the M1-th processing node of the first-level convolutional processing layer. The time series feature analysis requirement of the output voltage amplitude matches the function of the second-level LSTM processing layer, corresponding to the M2-th processing node of the second-level LSTM processing layer. The correspondence between all signal dimensions and model layers and processing nodes is recorded to generate a correspondence table between signal dimensions and model layers.

[0048] Step S119: Based on the correspondence table between signal dimensions and model levels, construct the connection mapping relationship between signal nodes in the fault signal demand set and processing nodes at each level of the initial neural network model, forming the initial adaptation relationship framework between the fault signal demand set and the initial neural network model.

[0049] In this embodiment, based on the correspondence table between signal dimensions and model levels, a connection mapping relationship is established between the signal nodes corresponding to each signal dimension in the fault signal requirement set and the processing nodes of the corresponding level in the initial neural network model. For example, the differential signal amplitude signal node is connected to the M1th processing node of the first-level convolutional processing layer, and the output voltage amplitude signal node is connected to the M2th processing node of the second-level LSTM processing layer. Simultaneously, the signal transmission paths and transmission methods between nodes are determined, such as using a fully connected method or a skip connection method, forming an initial adaptation relationship framework. This initial adaptation relationship framework describes the preliminary correspondence between fault signal requirements and model structure.

[0050] Step S1110: Based on the signal dimensions in the fault signal requirement set, check whether there are any uncovered signal dimensions in the initial adaptation relationship framework. For uncovered signal dimensions, add the corresponding model level and processing node to the initial adaptation relationship framework to generate the adaptation relationship framework between the fault signal requirement set and the initial neural network model.

[0051] In this embodiment, all signal dimensions in the fault signal requirement set are traversed, and it is checked whether these signal dimensions have corresponding model levels and processing nodes in the initial adaptation relationship framework. Suppose that the interface contact resistance signal dimension is found to be not covered in the initial adaptation relationship framework. In this case, a new processing level needs to be added to the initial neural network model, or a processing node needs to be added to an existing level to process this signal dimension. For example, a fourth processing level can be added to the multi-level signal processing layer, specifically for processing resistance-type signal dimensions. Corresponding processing nodes are configured for this level, and then a connection mapping relationship is established between the interface contact resistance signal node and the processing node of the newly added processing layer, ultimately generating a complete adaptation relationship framework.

[0052] Step S120: Based on the adaptation relationship framework between the fault signal demand set and the initial neural network model, the number of hierarchical nodes, hierarchical connection method and signal processing channel of the initial neural network model are adapted and adjusted to obtain a neural network model adapted to the fault signal demand.

[0053] In this embodiment, the initial neural network model is adjusted according to the adaptation framework. Regarding the number of nodes at each level, for the signal input layer, the number of neurons is increased or decreased based on the number of signal channels in the fault signal demand set to match the number of signal inputs. Regarding the hierarchical connection method, the connection structure between levels is adjusted according to the connection mapping relationship between signal nodes and processing nodes in the adaptation framework, such as adding skip connections to achieve cross-level signal transmission. Regarding signal processing channels, independent processing channels are configured for different types of signal dimensions, such as voltage signal channels and current signal channels, ensuring that various signals can be processed specifically. Through these adjustments, the neural network model can better adapt to the fault signal demand.

[0054] Step S121: Analyze the adaptation relationship framework between the fault signal requirement set and the initial neural network model, extract the number of signal dimensions, signal processing accuracy requirements and signal transmission timing requirements that need to be adapted for each level, and generate a hierarchical adaptation requirement list.

[0055] In this embodiment, the adaptation framework is analyzed to determine the number of signal dimensions that each layer needs to process. For example, the signal input layer needs to process D1 signal dimensions, the first-level processing layer needs to process D2 signal dimensions, and so on. Signal processing accuracy requirements include the measurement accuracy of each signal dimension and the accuracy of feature extraction, such as the processing accuracy requirement for voltage signals to reach V4 voltage units and the processing accuracy requirement for current signals to reach I1 current units. Signal transmission timing requirements include the transmission delay of signals between layers and the synchronization method, such as requiring that the delay of the signal transmission from the first-level processing layer to the second-level processing layer does not exceed T4 time units and that a synchronous transmission method is used. After organizing the above information, a layer adaptation requirement list is generated.

[0056] Step S122: For the signal input layer of the initial neural network model, adjust the number of signal receiving nodes and the signal receiving format of the signal input layer according to the number of signal dimensions and signal acquisition requirements corresponding to the signal input layer in the hierarchical adaptation requirement list, so that the signal input layer can receive the signal type of the HDI communication board operation signal and receive the signal at the signal transmission rate of the HDI communication board operation signal, thus obtaining the adapted signal input layer structure.

[0057] In this embodiment, the number of signal dimensions corresponding to the signal input layer in the hierarchical adaptation requirement list is D1. The signal acquisition requirements include signal type (voltage, current, resistance, etc.) and signal transmission rate R1 data units per second. Initially, the number of signal receiving nodes in the signal input layer is N3, which cannot meet the reception requirements of D1 signal dimensions. Therefore, the number of signal receiving nodes needs to be increased to D1. Simultaneously, the signal reception format is adjusted to support different signal types, such as analog signals being converted to digital signals via an A / D conversion module before input, and digital signals being directly input. Furthermore, the reception rate of the signal input layer is configured to R1 data units per second to match the transmission rate of the HDI communication board's operating signals. These adjustments result in the adapted signal input layer structure.

[0058] Step S123: For the multi-level signal processing layer of the initial neural network model, adjust the number of nodes in each level of the signal processing layer according to the signal dimension and processing accuracy requirements in the hierarchical adaptation requirement list, so that the number of nodes meets the processing requirements of the corresponding signal dimension. At the same time, modify the signal processing logic of each node to increase the ability to identify fault signals, and obtain the adapted signal processing layer structure of each level.

[0059] In this embodiment, the multi-level signal processing layer includes a first-level, a second-level, and a third-level processing layer. For the first-level processing layer, the signal dimension count corresponding to the layer adaptation requirement list is D2, and the processing accuracy requirement is P4 percentage. Initially, the number of nodes in the first-level processing layer is N4. Based on the signal dimension count D2 and the processing accuracy requirement P4, the required number of nodes is calculated to be N5 (N5 is greater than N4), therefore the number of nodes is increased to N5. The signal processing logic of each node is modified. Based on the original feature extraction algorithm, a fault signal feature template matching function is added. By comparing the input signal features with the preset fault signal feature template, the ability to identify fault signals is improved. In the same way, the number of nodes and the processing logic of the second-level and third-level processing layers are adjusted, resulting in the adapted signal processing layer structure for each level.

[0060] Step S1231: Extract the number of signal dimensions, processing accuracy requirements, and fault signal perception requirements for multi-level signal processing layers from the hierarchical adaptation requirement list, and generate an adaptation parameter table for each level of signal processing layer.

[0061] In this embodiment, information related to multi-level signal processing layers is extracted from the hierarchical adaptation requirement list. For the first-level processing layer, the signal dimension is D2, the processing accuracy requirement is P4%, and the fault signal perception requirement is the ability to identify basic fault characteristics such as signal amplitude abrupt changes and frequency anomalies. For the second-level processing layer, the signal dimension is D3, the processing accuracy requirement is P5%, and the fault signal perception requirement is the ability to capture short-term temporal variation characteristics of the signal. For the third-level processing layer, the signal dimension is D4, the processing accuracy requirement is P6%, and the fault signal perception requirement is the ability to analyze the long-term trend and complex correlation characteristics of the signal. The above information is organized into a table to generate an adaptation parameter table for each level of signal processing layer.

[0062] Step S1232: Select the first-level signal processing layer of the initial neural network model, analyze the current number of nodes, node processing logic and signal processing range of the first-level signal processing layer, and generate the current state information of the first-level signal processing layer.

[0063] In this embodiment, the first-level signal processing layer of the initial neural network model is selected. By checking the model structure configuration file, the current number of nodes is found to be N4. The node processing logic adopts standard convolution operation, that is, feature extraction is performed by convolution operation between the convolution kernel and the input feature map. The signal processing range is the feature map output by the input layer, with a size of [B, C, L]. The above information is recorded to generate the current state information of the first-level signal processing layer.

[0064] Step S1233: Based on the number of signal dimensions corresponding to the first-level signal processing layer in the adaptation parameter table of each level of signal processing layer, calculate the number of nodes required for the first-level signal processing layer, and adjust the number of nodes of the first-level signal processing layer so that the number of nodes can handle all the corresponding signal dimensions.

[0065] In this embodiment, the number of signal dimensions corresponding to the first-level signal processing layer in the adaptation parameter table of each level of signal processing layer is D2. According to signal processing theory, the number of nodes N5 required to process D2 signal dimensions can be expressed by the formula N5=D2. K is calculated, where K is the number of nodes required for each signal dimension (determined empirically and experimentally). Assuming K is K1, then N5 = D2. K1. The initial number of nodes in the first-level signal processing layer is N4. If N4 is less than N5, the number of nodes is increased to N5; if N4 is greater than N5, the number of nodes is reduced to N5, to ensure that the number of nodes can handle all corresponding signal dimensions.

[0066] Step S1234: Based on the processing accuracy requirements of the first-level signal processing layer in the adaptation parameter table of each level of signal processing layer, modify the signal processing logic of each node of the first-level signal processing layer, and adjust the signal filtering rules and signal enhancement logic inside the node.

[0067] In this embodiment, the processing accuracy requirement for the first-level signal processing layer is P4 percentage. To achieve this accuracy requirement, the signal filtering rules of the nodes are modified, and an adaptive median filtering algorithm is adopted. By dynamically adjusting the size of the filtering window, noise interference in the signal is removed, improving the purity of the signal. At the same time, the signal enhancement logic is adjusted to weight and enhance the effective features in the signal, such as assigning higher weights to the edge features of the signal, making these features more prominent in subsequent processing, thereby improving the processing accuracy.

[0068] Step S1235: Based on the fault signal perception requirements corresponding to the first-level signal processing layer in the adaptation parameter table of each level of signal processing layer, add fault signal feature recognition rules to the processing logic of each node to improve the node's ability to perceive fault signals.

[0069] In this embodiment, the fault signal perception requirement of the first-level signal processing layer is to identify basic fault characteristics such as sudden changes in signal amplitude and frequency anomalies. Fault signal characteristic identification rules are added to the node processing logic. For example, an amplitude change threshold T5 is set; when the signal change exceeds T5 within a time unit T6, it is determined to be an amplitude change fault characteristic. A frequency range [F4, F5] is set; when the signal frequency exceeds this range, it is determined to be a frequency anomaly fault characteristic. By adding these rules, the node can promptly perceive basic fault characteristics during signal processing.

[0070] Step S1236: Adjust the signal transmission connection relationship between nodes in the first-level signal processing layer so that the signal transmission between nodes meets the preset transmission conditions, and generate the first-level adapted signal processing layer structure.

[0071] In this embodiment, the preset transmission conditions include a signal transmission delay not exceeding T7 time units and a signal transmission bandwidth not less than B2 frequency units. Initially, the nodes in the first-level signal processing layer use a fully connected approach, which may result in excessive transmission delay. Therefore, the connection relationship is adjusted to a partial connection approach, retaining connections only with adjacent nodes to reduce the number of connections and thus reduce transmission delay. Simultaneously, signal routing between nodes is optimized to ensure that the signal transmission bandwidth meets the requirements. These adjustments generate the adapted first-level signal processing layer structure.

[0072] Step S1237: Select the second-level signal processing layer of the initial neural network model, repeat the processing steps of parsing, adjusting the number of nodes, optimizing the processing logic, enhancing the fault signal perception, and adjusting the connection relationship for the first-level signal processing layer, and generate the second-level adapted signal processing layer structure according to the adaptation parameters corresponding to the second-level signal processing layer in the adaptation parameter table of each level of signal processing layer.

[0073] In this embodiment, the second-level signal processing layer is selected, and its current node count is N6, the node processing logic is an LSTM basic unit, and the signal processing range is the feature sequence output by the first-level processing layer. Based on the second-level signal dimension count D3 in the adaptation parameter table, the required number of nodes N7 = D3 is calculated. K2 (where K2 is the number of nodes required for each signal dimension in the second level) is adjusted to N7. The processing accuracy requirement is P5 percentage; the processing logic is modified to use a bidirectional LSTM structure to improve the accuracy of temporal feature extraction, and a dropout layer is added to prevent overfitting. The fault signal perception requirement is to capture short-term temporal change features; therefore, a sliding window analysis is added to the processing logic to judge the signal change trend within the window. The node connection relationship is adjusted to a cyclic connection to adapt to temporal signal processing, generating the adapted signal processing layer structure for the second level.

[0074] Step S1238: Select the third-level signal processing layer of the initial neural network model. Following the same processing steps, based on the adaptation parameters corresponding to the third-level signal processing layer in the adaptation parameter table of each level of signal processing layer, complete the adjustment of the number of nodes, optimization of processing logic, and adjustment of connection relationships to generate the adapted signal processing layer structure of the third level.

[0075] In this embodiment, the initial number of nodes in the third-level signal processing layer is N8, the number of signal dimensions in the adaptation parameter table is D4, and the required number of nodes is calculated as N9 = D4. K3 (K3 is the number of nodes required for each signal dimension in the third level), adjust the number of nodes to N9. The processing accuracy requirement is P6 percentage points. The processing logic adopts a fully connected layer and introduces batch normalization technology to stabilize the training process. The fault signal perception requirement is to analyze long-term trends and complex correlation features. A feature correlation analysis module is added to the processing logic to mine correlation relationships by calculating the correlation coefficient between features. The connection relationship adopts cross-level connection, receiving feature inputs from the first and second level processing layers to generate the adapted signal processing layer structure for the third level.

[0076] Step S1239: Perform correlation verification on the first, second and third level adapted signal processing layer structures according to the original hierarchical order to ensure that the signal processing flow between each level meets the preset coherence conditions. Integrate the adapted signal processing layer structures of each level after correlation verification to generate the final adapted signal processing layer structure of each level.

[0077] In this embodiment, the adapted signal processing layer structures of the first, second, and third levels are connected sequentially. The continuity conditions are checked during the signal transmission process from the first-level output to the second-level input and from the second-level output to the third-level input, including feature dimension matching, data format consistency, and transmission delay within acceptable limits. If the feature dimension of the second-level processing layer output does not match the dimension required by the third-level input, a fully connected layer is added for dimension transformation. If the transmission delay exceeds a preset value, the signal transmission path is optimized. After correlation verification and adjustment, the structures at each level are integrated to form the final adapted signal processing layer structure.

[0078] Step S124: Based on the signal transmission timing requirements in the fault signal demand set, adjust the connection method between the adapted signal processing layers at each level, establish a cross-level signal transmission path, so as to realize the interactive transmission of fault signals processed at different levels according to the timing requirements, and obtain the adjusted hierarchical connection structure.

[0079] In this embodiment, the signal transmission timing requirements in the fault signal demand set stipulate that certain fault signals need to be directly transmitted from the first-level processing layer to the third-level processing layer for comprehensive analysis, without passing through the second-level processing layer. The initial hierarchical connection method is a sequential connection, i.e., first level → second level → third level. To meet the timing requirements, cross-level signal transmission paths are established between the adapted signal processing layers, such as adding a direct connection channel between the first-level and third-level processing layers. Simultaneously, timing control logic for signal transmission is configured to ensure that cross-level transmitted signals are synchronized with sequentially transmitted signals in time, avoiding timing chaos, resulting in the adjusted hierarchical connection structure.

[0080] Step S125: For the signal fusion layer of the initial neural network model, adjust the node configuration and fusion logic of signal fusion according to the signal fusion requirements in the hierarchical adaptation requirement list, so as to increase the transmission weight of signal features strongly related to fault prediction in the fusion process, and obtain the adapted signal fusion layer structure.

[0081] In this embodiment, the signal fusion requirements in the hierarchical adaptation requirement list include fusing features from multiple signal processing layers and highlighting the weights of features strongly correlated with fault prediction. The initial signal fusion layer is configured with a fixed number of fusion nodes, and the fusion logic uses a simple weighted average. The node configuration is adjusted to increase the number of fusion nodes, so that each node is specifically responsible for fusing features related to a certain type of fault. Regarding the fusion logic, an attention mechanism is used to calculate the correlation between each input feature and the fault prediction target. Features with high correlation are assigned higher weights, and features with low correlation are assigned lower weights. Then, the features are weighted and summed according to their weights to obtain the fused features. Through these adjustments, the adapted signal fusion layer structure is obtained.

[0082] Step S126: For the signal output layer of the initial neural network model, adjust the number of output nodes and the output signal format of the signal output layer according to the number of fault types and the fault prediction accuracy requirements in the fault signal demand set, so that the signal output layer can output prediction-related signals of various faults, and obtain the adapted signal output layer structure.

[0083] In this embodiment, the number of fault types in the fault signal requirement set is T8, and the fault prediction accuracy requirement is a prediction accuracy of P7%. The initial number of output nodes in the signal output layer is N10, which is less than T8. Therefore, the number of output nodes is increased to T8, with each node corresponding to one fault type. Regarding the output signal format, initially it is a single probability value output. This is adjusted to an output format that includes multi-dimensional information such as fault probability, fault confidence, and fault feature importance. For example, each element of the output vector corresponds to the probability of a fault type, and a scalar representing the confidence level and a vector representing the importance of each feature are added. Through these adjustments, the signal output layer can meet the output requirements of fault prediction, resulting in the adapted signal output layer structure.

[0084] Step S127: Integrate the adapted signal input layer structure, the adapted signal processing layer structures at each level, the adjusted hierarchical connection structure, the adapted signal fusion layer structure, and the adapted signal output layer structure to generate a preliminary adapted neural network model.

[0085] In this embodiment, following the sequence of signal processing, the adapted signal input layer structure is used as the input of the model, and its output is connected to the input of each adapted signal processing layer structure. Each signal processing layer transmits signals through the adjusted hierarchical connection structure. The output of the signal processing layer is connected to the adapted signal fusion layer structure, and the output of the signal fusion layer is connected to the adapted signal output layer structure. During the integration process, the interface matching between each layer and the correct data flow are ensured to generate a preliminary adapted neural network model.

[0086] Step S128: Extract the signal processing results of each level of the initially adapted neural network model, compare them with the signal processing requirements in the fault signal requirement set, and define the differences between the signal processing results of each level and the required signals.

[0087] In this embodiment, fault signal data for testing is input into the preliminarily adapted neural network model, and the processing results of the signal input layer, each level of signal processing layer, signal fusion layer, and signal output layer are extracted. These processing results are compared with the signal processing requirements in the fault signal requirement set, such as whether the output of the signal input layer completely receives all signal dimensions, whether the features extracted by the signal processing layer include the key features of the fault signal, whether the fusion result of the signal fusion layer highlights important features, and whether the output format and accuracy of the signal output layer meet the requirements. Discrepancies are identified, such as a feature extracted by a certain level of signal processing layer omitting the fault feature of frequency anomaly, or the prediction accuracy of the signal output layer not reaching P7 percentage.

[0088] Step S129: Based on the differences, make secondary adjustments to the signal processing layers and signal fusion layers of the initially adapted neural network model, modify the node configuration and processing logic to reduce the difference between the signal processing results and the required signals, generate a secondary adapted neural network model, fix the overall structure of the secondary adapted neural network model, and generate a neural network model adapted to the fault signal requirements.

[0089] In this embodiment, secondary adjustments are made to the corresponding layers to address the differences. For the issue of missing frequency anomaly features in a certain signal processing layer, the node processing logic of that layer is modified, a frequency analysis module is added, and the signal is converted to the frequency domain using Fourier transform for analysis to extract frequency features. For the issue of insufficient prediction accuracy in the signal output layer, the attention weight calculation method of the signal fusion layer is adjusted, employing a more complex correlation calculation method, such as introducing mutual information to measure the correlation between features and fault types. Simultaneously, the number of nodes in the signal output layer is increased to improve the model's expressive power. After these secondary adjustments, a secondary-adapted neural network model is generated. The overall structure of this model is fixed, including the number of nodes in each layer, connection methods, and processing logic, ultimately resulting in a neural network model adapted to the fault signal requirements.

[0090] Step S130: Obtain the real-time operating signal set of the HDI communication board, input the real-time operating signal set of the HDI communication board into the neural network model adapted to the fault signal requirements, and through the fault signal guidance mechanism preset inside the model, perform hierarchical extraction and correlation mining of fault-related signals in the real-time operating signal set to obtain the fault feature association set.

[0091] In this embodiment, sensors installed on various modules of the HDI communication board collect operational signals in real time. For example, voltage sensors collect the output voltage signal of the power management module, and current sensors collect the operating current signal of the data transmission module. These signals are categorized and organized according to signal type to form a real-time operational signal set. This set is preprocessed according to the input format required by the model, such as normalization, and then input into a neural network model adapted to fault signal requirements. The fault signal guidance mechanism within the model guides each processing layer to extract features from the input signal based on preset fault signal feature parameters, progressing from basic features to detailed features and then to deep features, gradually deepening the process. Correlation analysis is performed on the extracted features to uncover the relationships between different features, ultimately forming a fault feature association set.

[0092] Step S131: Obtain various real-time operating signals of the HDI communication board during operation, classify and organize the obtained real-time operating signals according to signal type, and generate a set of real-time operating signals of the HDI communication board.

[0093] In this embodiment, during operation, the HDI communication board continuously collects real-time operating signals through various sensors, including the output voltage and current signals of the power management module, the differential signal amplitude and signal transmission frequency of the data transmission module, and the contact resistance signals of the interface module. These signals are categorized according to signal type, such as voltage signals, current signals, resistance signals, signal amplitude signals, and frequency signals. For each type of signal, the acquisition time, acquisition value, and signal unit are recorded. All categorized signals are then integrated to generate a real-time operating signal set for the HDI communication board.

[0094] Step S132: Input the real-time operation signal set of the HDI communication board into the signal input layer of the neural network model that adapts to the fault signal requirements one by one according to the preset input order. The real-time operation signal is converted into a signal format that the model can process through the signal conversion node of the signal input layer, and the real-time operation signal set after format conversion is obtained.

[0095] In this embodiment, the preset input order is arranged from highest to lowest importance for fault prediction, such as first inputting power-related signals, then data transmission-related signals, and finally interface-related signals. Signals from the real-time operating signal set are input to the signal input layer one by one in this order. The signal conversion nodes in the signal input layer perform A / D conversion on the input analog signals, converting them into digital signals with a sampling bit depth of B3 bits; the digital signals are then format-converted into a tensor format that the model can process, with a shape of [B, C, L], where B is the batch size, C is the number of signal channels, and L is the signal length. After conversion, the format-converted real-time operating signal set is obtained.

[0096] Step S133: Extract signal feature descriptions corresponding to various faults from the fault signal demand set, convert the signal feature descriptions into guidance parameters that the model can recognize, generate a fault signal guidance set, and input the fault signal guidance set into the signal processing layers of the neural network model that adapts to the fault signal demand.

[0097] In this embodiment, signal feature descriptions for each fault type are extracted from the fault signal requirement set. For example, the feature descriptions for signal transmission interruption faults include "differential signal amplitude suddenly drops to zero" and "interface contact resistance value increases significantly." These textual descriptions are converted into numerical guidance parameters that the model can recognize. For instance, "amplitude drops to zero" is converted into an amplitude threshold V5, and "resistance value increases significantly" is converted into a resistance threshold R4. These guidance parameters are organized according to fault type and level to generate a fault signal guidance set. This set is then input to each level of signal processing layer through the model's parameter input interface to guide the model in fault feature extraction.

[0098] Step S1331: Analyze the fault signal demand set, extract the various fault types contained therein, classify the various fault types according to the location where the fault occurs, and generate a fault type classification table.

[0099] In this embodiment, the fault signal requirement set is analyzed, and the extracted fault types include power fluctuation faults, signal transmission interruption faults, interface contact failures, and signal processing module calculation errors. Based on the location of the fault, power fluctuation faults and signal processing module calculation errors occur in the power management module and signal processing module, respectively, and are considered core module faults; signal transmission interruption faults and interface contact failures occur in the data transmission module and interface module, respectively, and are considered connection module faults. The above classification results are then compiled into a table to generate a fault type classification table.

[0100] Step S1332: For each fault type in the fault type classification table, extract the corresponding signal feature description, covering the signal fluctuation pattern, signal duration, signal change trend and signal interaction relationship, and generate a subset of signal feature descriptions for each fault type.

[0101] In this embodiment, a power fluctuation fault in the core module is taken as an example to extract its signal feature description. The signal fluctuation pattern is a sinusoidal wave; the duration is T9 time units; the trend is that the fluctuation amplitude gradually increases; the interaction relationship of the signals is that the output voltage fluctuation and the feedback voltage change rate are negatively correlated. These descriptions are then combined to generate a signal feature description subset for the power fluctuation fault. Similarly, signal feature descriptions are extracted for other fault types, generating their respective signal feature description subsets.

[0102] Step S1333: Perform semantic normalization on the signal feature descriptions in the signal feature description subset of each type of fault, unify the expression of signal feature descriptions, and convert the normalized signal feature description subset of each type of fault into numerical guidance parameters that can be recognized by the neural network model that adapts to the fault signal requirements according to the preset parameter conversion rules, thereby generating a guidance parameter subset for each type of fault.

[0103] In this embodiment, the signal feature descriptions are semantically normalized. For example, "fluctuation amplitude gradually increases" is uniformly expressed as "fluctuation amplitude monotonically increases with time." The preset parameter conversion rules map textual descriptions to specific numerical values ​​or ranges. For instance, "sine wave fluctuation" is converted to the fluctuation type parameter value 1, "duration T9 time units" is converted to the duration parameter T9, "fluctuation amplitude monotonically increases with time" is converted to the amplitude change rate parameter K4 (positive value), and "output voltage fluctuation is negatively correlated with feedback voltage change rate" is converted to the correlation coefficient parameter -0.8. These numerical guidance parameters are integrated to generate a subset of guidance parameters for each type of fault.

[0104] Step S1334: Based on the processing range of each level of signal processing layer of the neural network model that adapts to the fault signal requirements, divide the subset of guidance parameters for each type of fault into basic guidance parameters, detailed guidance parameters, and deep guidance parameters, which correspond to the first level signal processing layer, the second level signal processing layer, and the third level signal processing layer of the neural network model, respectively.

[0105] In this embodiment, the first-level signal processing layer of the neural network model adapted to fault signal requirements processes basic features, such as signal amplitude and frequency; the second level processes detailed features, such as signal fluctuation patterns and rate of change; and the third level processes deep features, such as the interaction relationships between signals and long-term trends. The parameters in the subset of guidance parameters for each type of fault are divided according to their processing range. For example, amplitude threshold V5 and frequency range [F4, F5] are basic guidance parameters and are assigned to the first-level processing layer; fluctuation type parameters and amplitude rate of change parameter K4 are detailed guidance parameters and are assigned to the second-level processing layer; and correlation coefficient parameters are deep guidance parameters and are assigned to the third-level processing layer.

[0106] Step S1335: Integrate the basic guidance parameters of all fault types to generate the guidance parameter group corresponding to the first-level signal processing layer; integrate the detailed guidance parameters of all fault types to generate the guidance parameter group corresponding to the second-level signal processing layer; integrate the deep guidance parameters of all fault types to generate the guidance parameter group corresponding to the third-level signal processing layer.

[0107] In this embodiment, the basic guidance parameters for all fault types, such as amplitude thresholds and frequency ranges for various faults, are integrated into a guidance parameter group containing multiple sub-parameters, which serves as the guidance parameters for the first-level signal processing layer. Similarly, all detailed guidance parameters are integrated to generate the second-level guidance parameter group, and all deep guidance parameters are integrated to generate the third-level guidance parameter group. The parameters in each guidance parameter group are organized according to the fault type so that the model can correspond to the specific fault type during processing.

[0108] Step S1336: Encapsulate the guidance parameter groups corresponding to the first, second and third level signal processing layers according to the model hierarchy to generate a fault signal guidance set.

[0109] In this embodiment, the first-level, second-level, and third-level guidance parameter groups are arranged in hierarchical order and encapsulated using a dictionary data structure. The dictionary keys are hierarchical identifiers (such as "first-level", "second-level", and "third-level"), and the values ​​are the corresponding guidance parameter groups. In this way, the model can quickly obtain the corresponding guidance parameters based on the hierarchical identifiers and generate a fault signal guidance set when reading the data.

[0110] Step S1337: Determine the guidance parameter receiving interface of each level of the signal processing layer in the neural network model that meets the fault signal requirements, and send the guidance parameter group of the corresponding level in the fault signal guidance set to the receiving interface of each level of the signal processing layer.

[0111] In this embodiment, each signal processing layer of the neural network model adapted to fault signal requirements has a dedicated guidance parameter receiving interface. These interfaces have specific communication protocols and data formats. By consulting the model's interface definition document, the receiving interfaces for the first-level signal processing layer are determined to be I1, the second-level to I2, and the third-level to I3. The first-level guidance parameter set from the fault signal guidance set is sent to the first-level processing layer through interface I1, the second-level guidance parameter set is sent to the second-level processing layer through interface I2, and the third-level guidance parameter set is sent to the third-level processing layer through interface I3.

[0112] Step S134: In the first-level signal processing layer, based on the basic fault signal feature guidance parameters in the fault signal guidance set, the basic fault signal features of the format-converted real-time running signal set are extracted to obtain the first-level fault signal feature set.

[0113] In this embodiment, the first-level signal processing layer receives basic fault signal feature guidance parameters, such as amplitude threshold V5 and frequency range [F4, F5]. Each signal in the real-time running signal set after format conversion is processed. For example, for differential amplitude signals, it is compared with the amplitude threshold V5; if the signal value is lower than V5, amplitude anomaly features are extracted. For frequency signals, it is determined whether they are within the range [F4, F5]; if not, frequency anomaly features are extracted. All extracted basic fault signal features are integrated to form a first-level fault signal feature set. Each feature in this first-level fault signal feature set contains information such as feature type, feature value, and corresponding signal source.

[0114] Step S135: Input the first-level fault signal feature set into the second-level signal processing layer, and extract the detailed fault signal features from the first-level fault signal feature set according to the detailed fault signal feature guidance parameters in the fault signal guidance set to obtain the second-level fault signal feature set.

[0115] In this embodiment, the second-level signal processing layer receives detailed fault signal feature guidance parameters, such as fluctuation type parameters and amplitude change rate parameter K4. It further analyzes the features in the first-level fault signal feature set. For example, for amplitude anomaly features, it combines the fluctuation type parameter to determine whether the fluctuation pattern is a sine wave, square wave, etc.; it calculates the amplitude change rate and compares it with K4 to determine whether it conforms to a monotonically increasing trend. These detailed features are extracted to generate a second-level fault signal feature set, which contains more detailed fault feature information than the first-level set.

[0116] Step S136: Input the second-level fault signal feature set into the third-level signal processing layer, and extract the deep fault signal features from the second-level fault signal feature set according to the deep fault signal feature guidance parameters in the fault signal guidance set to obtain the third-level fault signal feature set.

[0117] In this embodiment, the third-level signal processing layer uses deep fault signal feature guidance parameters, such as correlation coefficient parameters, to analyze the second-level fault signal feature set. For example, it calculates the correlation coefficients between different fault features to determine their interaction relationships, such as whether the output voltage fluctuation feature and the feedback voltage change rate feature are negatively correlated; it analyzes the long-term trend of fault features over time to predict their development direction. These deep features are then extracted to form the third-level fault signal feature set.

[0118] Step S137: Perform feature normalization processing on the first-level fault signal feature set, the second-level fault signal feature set, and the third-level fault signal feature set respectively to generate normalized first-level, second-level, and third-level fault signal feature sets; input the normalized first-level, second-level, and third-level fault signal feature sets into the signal fusion layer of the neural network model adapted to fault signal requirements according to their corresponding hierarchical information; the signal fusion layer performs correlation and fusion of normalized fault signal feature sets of different levels according to the signal correlation requirements in the fault signal requirement set to obtain a fused fault feature set.

[0119] In this embodiment, the feature normalization process employs the Z-score normalization method, processing the feature values ​​in each feature set to ensure that their mean is 0 and their standard deviation is 1. For example, for the amplitude anomaly feature values ​​in the first-level fault signal feature set, the transformation is performed using the formula (feature value - feature mean) / feature standard deviation. The normalized feature sets at each level are then labeled according to their hierarchical information (e.g., "first level", "second level", "third level") and input into the signal fusion layer. Based on signal correlation requirements, such as "power-related features and data transmission-related features need to be correlated," the signal fusion layer uses an attention mechanism to weight and fuse features from different levels, calculating the weight of each feature, and then combining the features according to their weights to obtain the fused fault feature set.

[0120] Step S138: Perform correlation analysis on the fault features in the fused fault feature set, define the dependency and co-occurrence relationships between different fault features, and generate a fault feature correlation table.

[0121] In this embodiment, all fault features in the fused fault feature set are paired to analyze their temporal order of occurrence, spatial signal source, and correlation in numerical changes. For example, the feature of increased interface contact resistance precedes the feature of decreased differential signal amplitude. Spatially, both originate from the interface module and the data transmission module, and their numerical changes are negatively correlated. Therefore, a dependency relationship (increased interface contact resistance leads to decreased differential signal amplitude) and a co-occurrence relationship (the two often occur simultaneously) are defined between them. The dependency and co-occurrence relationships of all feature pairs are recorded to generate a fault feature association table.

[0122] Step S1381: Extract all fault features from the fused fault feature set, assign a unique feature identifier to each fault feature, and generate a fault feature identifier list.

[0123] In this embodiment, traverse the fused fault feature set, and extract all different fault features, such as amplitude anomaly features, frequency anomaly features, increased contact resistance features, etc. Assign a unique feature identifier to each fault feature, such as F1, F2, F3, etc., and associate the feature name with the feature identifier to generate a fault feature identifier list.

[0124] Step S1382: Combine all the fault features in the fault feature identifier list in pairs to generate all possible pairs of fault feature combinations.

[0125] In this embodiment, there are M fault features in the fault feature identifier list. Use a combination algorithm to generate all possible pairs of combinations, such as (F1,F2), (F1,F3), (F2,F3), etc., and a total of M (M - 1) / 2 combination pairs are generated.

[0126] Step S1383: For each pair of fault feature combinations, extract the occurrence time, signal source, and signal intensity change information of the two fault features in the pair from the fused fault feature set.

[0127] In this embodiment, taking the combination pair (F1,F2) as an example, F1 is the feature of increased interface contact resistance, and F2 is the feature of decreased differential signal amplitude. Extract the occurrence time T10 of F1 from the fused fault feature set, the signal source is the interface module, and the signal intensity change information is from the R2 resistance unit to the R1 resistance unit; the occurrence time T11 of F2, the signal source is the data transmission module, and the signal intensity change information is from the V3 voltage unit to the zero voltage unit.

[0128] Step S1384: Analyze whether there is an overlap in the occurrence times of the two fault features, whether the signal sources belong to the same operation link of the HDI communication board, and whether there is synchronization in the signal intensity changes.

[0129] In this embodiment, for the combination pair (F1,F2), there is an overlapping part between the occurrence time T10 of F1 and the occurrence time T11 of F2 (T10 < T11 < T10 + ΔT, where ΔT is the duration of F1); the signal sources, the interface module and the data transmission module, both belong to the connection module operation link; the increase in the signal intensity of F1 is synchronized with the decrease in the signal intensity of F2 in time, that is, F2 decreases while F1 increases.

[0130] Step S1385: Define the dependency relationship between the two fault features according to the overlap situation of the occurrence times, the relevance of the signal sources, and the synchronization of the signal intensity changes, that is, whether the occurrence of one fault feature depends on the occurrence of the other fault feature.

[0131] In this embodiment, since F1 appears earlier than F2, and their appearance times overlap, their signal sources are related, and their intensity changes are synchronized, the occurrence of F2 depends on the occurrence of F1, that is, the increase in interface contact resistance leads to a decrease in the amplitude of the differential signal.

[0132] Step S1386: Simultaneously define the co-occurrence relationship between the two fault features, that is, whether the two fault features frequently occur at the same time.

[0133] In this embodiment, the co-occurrence relationship between F1 and F2 is determined by statistically analyzing the proportion of times F1 and F2 occur simultaneously in the total number of faults. If this proportion exceeds P8, then the co-occurrence relationship between the two is defined. Statistical analysis shows that the proportion of F1 and F2 occurring simultaneously is P9 (P9 > P8), therefore, a co-occurrence relationship between them is confirmed.

[0134] Step S1387: Describe the dependency and co-occurrence relationships of each fault feature combination pair and generate a single fault feature association record.

[0135] In this embodiment, the association relationship of the combination pair (F1, F2) is described as follows: the dependency relationship is that F2 depends on F1, and the co-occurrence relationship is that they often appear simultaneously. The above description, along with the feature identifier, is recorded to generate a single fault feature association record.

[0136] Step S1388: Summarize all individual fault feature association records, sort the summarized fault feature association records according to the fault feature identifier, and generate a fault feature association table.

[0137] In this embodiment, the associated records of all fault feature pairs are summarized together and sorted according to the alphabetical order of the feature identifiers, such as sorting by the first feature identifier first and then by the second feature identifier, to generate a fault feature association table, which shows the association between all fault features.

[0138] Step S139: Based on the fault feature association table, group the fault features in the fused fault feature set according to the association relationship to obtain multiple fault feature association groups.

[0139] In this embodiment, based on the fault feature correlation table, fault features that have a dependency relationship or co-occurrence relationship are grouped together. For example, F1 (increased interface contact resistance), F2 (decreased differential signal amplitude), and F3 (abnormal signal transmission frequency) are correlated and grouped into one group; F4 (output voltage fluctuation), F5 (abnormal feedback voltage change rate), and F6 (current amplitude fluctuation) are correlated and grouped into another group. This process continues to obtain multiple fault feature correlation groups.

[0140] Step S1310: Integrate all fault feature association groups, supplement the signal source information and timing information corresponding to each fault feature association group, and generate a fault feature association set.

[0141] In this embodiment, for each fault feature association group, signal source information for all fault features in the group is supplemented, such as from the interface module, data transmission module, etc., as well as timing information, such as the order in which each feature appears and its duration. This supplementary information is added to the association group, and then all association groups are integrated to generate a fault feature association set.

[0142] Step S140: Based on the fault feature association set and the correspondence between fault signals and fault types in the fault signal demand set, determine the fault type and fault occurrence trend of the HDI communication board, and generate a set of HDI communication board fault prediction results.

[0143] In this embodiment, the fault feature association set contains multiple fault feature association groups, each corresponding to a set of related fault features. The fault signal requirement set stores the correspondence between fault signals (i.e., fault features) and fault types. For example, the association group containing features F1, F2, and F3 corresponds to a signal transmission interruption fault. By comparing each fault feature association group with the fault signal type correspondence table, its corresponding fault type is determined. Simultaneously, the temporal changes of fault features in the association groups are analyzed, such as the trend of feature intensity changes and frequency of occurrence, to determine the development trend of the fault, such as whether the fault is worsening and the possible time range. The judgment results of fault type and fault occurrence trend are integrated to generate the HDI communication board fault prediction result set.

[0144] Step S141: Extract all fault feature association groups, signal source information and time sequence information from the fault feature association set, and generate fault feature association details.

[0145] In this embodiment, the fault feature association set is traversed, and each fault feature association group, the signal source information of the group (such as interface module, data transmission module, etc.), and the timing information (such as feature appearance time, duration, and order) are extracted and organized into a structured table to generate a fault feature association detail, which records the specific information of each association group in detail.

[0146] Step S142: Extract the correspondence between fault signals and fault types from the fault signal demand set, and generate a fault signal type correspondence table.

[0147] In this embodiment, the fault signal requirement set includes the fault signal feature combination corresponding to each fault type. For example, signal transmission interruption faults correspond to feature combinations F1, F2, and F3, and power fluctuation faults correspond to feature combinations F4, F5, and F6. The above correspondence is extracted and presented in tabular form. Each row of the table represents a fault type, and the corresponding column lists the fault signal feature combination for that fault type, generating a fault signal type correspondence table.

[0148] Step S143: Compare each fault feature association group in the fault feature association details with the fault signals in the fault signal type correspondence table one by one, define the fault type corresponding to each fault feature association group, and generate preliminary fault type matching results.

[0149] In this embodiment, for a fault feature association detail, the fault feature identifiers contained therein, such as F1, F2, and F3, are extracted. The fault type containing these feature identifier combinations is searched in the fault signal type correspondence table. It is found that the feature combination for a signal transmission interruption fault is exactly F1, F2, and F3. Therefore, this association group is initially matched as a signal transmission interruption fault. Following the same method, all association groups are compared and matched to generate preliminary fault type matching results.

[0150] Step S144: Extract the time series information corresponding to each fault feature association group, analyze the signal change trend of the fault feature association group at different time stages, and generate the fault feature time series change curve.

[0151] In this embodiment, taking a certain associated group as an example, the signal strength of fault feature F1 increases from R2 to R1 during the time period T12 to T13, while F2 decreases from V3 to zero during the time period T11 to T14. With time as the horizontal axis and signal strength as the vertical axis, the change curves of F1 and F2 are plotted respectively. These curves are then integrated to form the time-series change curve of the fault feature for this associated group, visually demonstrating the change of the fault feature over time.

[0152] Step S145: Based on the fault characteristic time-series change curve, determine the development trend of the fault characteristics, define the possible time range and development speed of the fault, and generate the fault occurrence trend judgment result.

[0153] In this embodiment, the time-series change curve of fault characteristics is analyzed. If the curve shows that the signal strength of the fault characteristics is continuously increasing and the rate of increase is gradually accelerating, it indicates that the fault is worsening and developing rapidly. By fitting the curve, the time point when the signal strength reaches the fault threshold is predicted, thereby defining the possible time range of the fault occurrence. For example, if the current signal strength of F1 is R5, and it continues to increase at the current rate of increase K5, it is expected to reach the fault threshold R1 at time point T15. Therefore, the possible time range of the fault occurrence is around T15, and the development rate is K5 units per unit of resistance per unit of time. The above judgment results are integrated to generate the fault occurrence trend judgment result.

[0154] For example, step S1451: Extract all data points in the fault characteristic time-series change curve, define the time information and signal strength information corresponding to each data point, and generate time-series curve data details.

[0155] In this embodiment, the fault characteristic timing change curve consists of multiple discrete data points. Each data point contains time information (such as T16, T17, T18, etc.) and corresponding signal strength information (such as R6, R7, R8, etc.). The above data points are arranged in chronological order to generate a detailed timing curve data.

[0156] Step S1452: Analyze the signal strength change pattern in the time series curve data details to determine whether the signal strength shows an upward trend, a downward trend, or a stable trend.

[0157] In this embodiment, by calculating the signal strength difference between adjacent data points in the time-series curve data details, if most of the differences are positive, the signal strength is judged to be on an upward trend; if most of the differences are negative, the signal strength is judged to be on a downward trend; if the differences fluctuate within a small range, the signal strength is judged to be on a stable trend. For example, if the data point differences for a certain feature are +ΔR1, +ΔR2, and +ΔR3, which are all positive, then the signal strength is judged to be on an upward trend.

[0158] Step S1453: If the signal strength is increasing, calculate the amplitude of the signal strength change between adjacent time points and define the rate of increase of the signal strength.

[0159] In this embodiment, for an upward trend in signal strength, the amplitude of signal strength change between adjacent time points is calculated, i.e., the signal strength of the later data point minus the signal strength of the previous data point, resulting in ΔR1, ΔR2, etc. Then, the average value of these amplitude changes is calculated as the average growth rate of the signal strength, K6 resistance unit per time unit, defining the growth rate of the signal strength.

[0160] Step S1454: Based on the growth rate of the signal strength and the fault occurrence threshold corresponding to the fault feature in the fault signal demand set, predict the time required for the fault feature to reach the fault occurrence threshold and define the possible time range of the fault occurrence.

[0161] In this embodiment, the fault occurrence threshold for this fault feature in the fault signal demand set is R1. The current signal strength is R5, the growth rate is K6, and the predicted time required to reach the threshold is ΔT = (R1 - R5) / K6. The current time is T19, so the possible occurrence time range of the fault is [T19 + ΔT - ΔT1, T19 + ΔT + ΔT1], where ΔT1 is the prediction error time range.

[0162] Step S1455: If the signal strength shows a decreasing trend, calculate the decrease in signal strength between adjacent time points, define the rate of decrease in signal strength, and determine whether the fault characteristic will gradually disappear. If it will not disappear, predict the signal strength after it stabilizes and the corresponding time range.

[0163] In this embodiment, for the signal strength exhibiting a downward trend, the decrease in amplitude between adjacent time points is calculated, i.e., the signal strength of the previous data point minus the signal strength of the next data point, yielding ΔR4, ΔR5, etc. The average decrease rate K7 per unit of resistance per unit of time is calculated. It is determined whether the downward trend will continue until the signal strength reaches zero. If it does not disappear, the signal strength will stabilize at a certain value as the decrease rate gradually approaches zero. By fitting the decrease curve, the stable signal strength R9 and the time range [T20, T21] are predicted.

[0164] Step S1456: If the signal strength shows a stable trend, analyze whether the signal strength is within the normal signal strength range. If it exceeds the normal range, determine that a fault has occurred and define the duration of the fault. If it is within the normal range, determine that the fault characteristics will not cause a fault in the short term.

[0165] In this embodiment, the signal strength of a stable trend has a normal range [R10, R11]. If the current signal strength R12 is within this range, it is determined that the fault feature will not cause a fault in the short term; if R12 exceeds [R10, R11], it is determined that a fault has occurred. The duration of the fault is defined as T23-T22, from the start time T22 when the feature appears to the current time T23.

[0166] Step S1457: Integrate the signal strength change trend, change rate, possible fault occurrence time range or fault duration to generate a fault development trend description.

[0167] In this embodiment, for fault characteristics with an upward trend, the fault development trend is described as "the signal strength is increasing at a rate of K6 resistance units per time unit, and is expected to reach the fault threshold within the time range of [T19+ΔT-ΔT1, T19+ΔT+ΔT1]"; for characteristics with a downward trend that do not disappear, it is described as "the signal strength is decreasing at a rate of K7 resistance units per time unit, and is expected to stabilize at R9 resistance units within the time range of [T20, T21]"; for characteristics that are stable and exceed the normal range, it is described as "the signal strength is stable, the current value R12 exceeds the normal range [R10, R11], and the fault has lasted for time units T23-T22".

[0168] Step S1458: Based on the description of the fault development trend, define the urgency of the fault occurrence and generate a fault urgency indicator.

[0169] In this embodiment, the urgency of a fault is divided into three levels: high, medium, and low. Faults expected to occur within a time unit T24 are classified as high urgency; those between T24 and T25 are classified as medium; and those exceeding T25 are classified as low. Based on the time range described in the fault development pattern, such as [T19+ΔT-ΔT1, T19+ΔT+ΔT1], if it is less than T24, the urgency is marked as "high".

[0170] Step S1459: Bind the description of the fault development trend with the fault urgency level indicator to generate the fault occurrence trend judgment result.

[0171] In this embodiment, the description of the fault development trend and the corresponding urgency level are combined together, such as "the signal strength is on the rise, the growth rate is K6 resistance units per time unit, and it is expected to reach the fault threshold within the time range of [T19+ΔT-ΔT1, T19+ΔT+ΔT1], urgency level: high", to generate a fault occurrence trend judgment result.

[0172] Step S146: Bind the preliminary fault type matching result with the fault occurrence trend judgment result to generate a single fault prediction record. Each fault prediction record includes fault type, fault feature association group, signal source information, time sequence information and fault occurrence trend.

[0173] In this embodiment, the preliminary matching result of the fault type (such as signal transmission interruption fault) is bound with the corresponding fault occurrence trend judgment result (such as development trend description and urgency level indicator), and the fault feature association group, signal source information (interface module and data transmission module) and time sequence information (occurrence time of each feature, etc.) corresponding to the fault are attached to generate a complete fault prediction record.

[0174] Step S147: Classify and organize all individual fault prediction records, and divide the fault prediction records into different part fault prediction sets according to the components of the HDI communication board.

[0175] In this embodiment, the HDI communication board comprises a power management module, a signal processing module, a data transmission module, and an interface module. Based on the signal source information in each fault prediction record, it is assigned to the corresponding component. For example, records originating from the interface module and data transmission module are assigned to the connection module fault prediction set, and records originating from the power management module are assigned to the power module fault prediction set, resulting in different component fault prediction sets.

[0176] Step S148: Extract the fault occurrence trend judgment results from the fault prediction set of each part, analyze the correlation between different fault types in the same part, and generate a description of the correlation between part faults.

[0177] In this embodiment, taking the fault prediction set of the connection module as an example, this set may include signal transmission interruption faults and interface contact failures. Analyzing the fault occurrence trends of both, if the urgency of the interface contact failure is high and its development may lead to the aggravation of the signal transmission interruption fault, it indicates that there is a correlation between the two, that is, interface contact failure will exacerbate the signal transmission interruption fault. Describing the above correlation effects generates a fault correlation effect description for the connection module.

[0178] Step S149: Add the description of the associated impact of the part failure to the corresponding part failure prediction set to generate the part failure prediction set.

[0179] In this embodiment, the description of the fault association effect of the connection module part is added to the fault prediction set of the connection module part, so that the fault prediction set of the connection module part not only contains a single fault prediction record, but also contains the association effect information between faults, thus forming a part fault prediction set.

[0180] Step S1410: Integrate all fault prediction sets to generate a set of HDI communication board fault prediction results that includes various fault types, fault occurrence trends, and location-related impacts.

[0181] In this embodiment, the fault prediction sets of all parts such as the power module, signal processing module, and connection module are integrated together to form the HDI communication board fault prediction result set. This HDI communication board fault prediction result set comprehensively reflects the various fault conditions, development trends, and correlation effects between faults in different parts of the HDI communication board.

[0182] Step S150: Bind the HDI communication board fault prediction result set with the HDI communication board's operating signal acquisition time to generate an HDI communication board fault prediction report set containing time-related information.

[0183] In this embodiment, the operating signal acquisition time of the HDI communication board refers to the specific time point or time period during which the sensor acquires the real-time operating signal. Each fault prediction record in the fault prediction result set is bound to its corresponding signal acquisition time. For example, if a fault prediction record is generated based on signals acquired during the time period from T26 to T27, then that time period information is added to the record. Then, all fault prediction records bound with time-related information are organized according to time order and location classification to generate an HDI communication board fault prediction report set. This HDI communication board fault prediction report set facilitates users' understanding of the fault situation and development process at different time points.

[0184] Figure 2 The diagram illustrates the hardware structure of an HDI communication board fault prediction system 100 based on a neural network model, as provided in an embodiment of the present invention. Figure 2 As shown, the HDI communication board fault prediction system 100 based on a neural network model may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0185] Machine-readable storage medium 120 can store data and / or instructions. In some embodiments, machine-readable storage medium 120 can store data acquired from an external terminal. In some embodiments, machine-readable storage medium 120 can store data and / or instructions used by the HDI communication board fault prediction system 100 based on a neural network model to execute or use in order to complete the exemplary methods described in this invention. In a specific implementation, one or more processors 110 execute the computer-executable instructions stored in machine-readable storage medium 120, enabling processor 110 to execute the HDI communication board fault prediction method based on a neural network model as described in the above method embodiments. Processor 110, machine-readable storage medium 120, and communication unit 140 are connected via bus 130, and processor 110 can be used to control the transmission and reception operations of communication unit 140. The specific implementation process of processor 110 can be found in the various method embodiments executed by the HDI communication board fault prediction system 100 based on a neural network model described above, and their implementation principles and technical effects are similar, so they will not be repeated here.

[0186] Furthermore, embodiments of the present invention also provide a readable storage medium containing computer-executable instructions. When the processor executes the computer-executable instructions, the above-described HDI communication board fault prediction method based on a neural network model is implemented.

[0187] It should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof. Similarly, it should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof.

Claims

1. A fault prediction method for HDI communication boards based on a neural network model, characterized in that, The method includes: The fault signal manifestations corresponding to various faults of the HDI communication board are collected, and a fault signal demand set is generated in combination with the operating mechanism of the HDI communication board. At the same time, an initial neural network model with basic signal processing functions is constructed. Based on the signal dimensions in the fault signal demand set and the hierarchical structure of the initial neural network model, a matching framework for the adaptation relationship between the fault signal demand set and the initial neural network model is obtained. Based on the adaptation framework between the fault signal demand set and the initial neural network model, the number of hierarchical nodes, hierarchical connection method and signal processing channel of the initial neural network model are adapted and adjusted to obtain a neural network model adapted to the fault signal demand. The real-time operating signal set of the HDI communication board is obtained. The real-time operating signal set of the HDI communication board is input into a neural network model that adapts to the fault signal requirements. Through the fault signal guidance mechanism preset in the model, the fault-related signals in the real-time operating signal set are extracted and associated in a hierarchical manner to obtain the fault feature association set. Based on the fault feature association set and combined with the correspondence between fault signals and fault types in the fault signal demand set, the fault type and fault occurrence trend of the HDI communication board are judged, and a fault prediction result set of the HDI communication board is generated. The set of HDI communication board fault prediction results is bound to the HDI communication board's operating signal acquisition time to generate a set of HDI communication board fault prediction reports containing time-related information.

2. The HDI communication board fault prediction method based on a neural network model according to claim 1, characterized in that, The method involves collecting fault signal manifestations corresponding to various faults of the HDI communication board, generating a fault signal demand set based on the operating mechanism of the HDI communication board, and simultaneously constructing an initial neural network model with basic signal processing functions. A matching framework between the signal dimensions in the fault signal demand set and the hierarchical structure of the initial neural network model is obtained, resulting in an adaptation relationship framework between the fault signal demand set and the initial neural network model, including: Collect operating signal data of HDI communication board when various faults occur under different operating conditions, classify and organize the collected operating signal data, and generate a list of fault signal manifestations of HDI communication board; Analyze the operating mechanism of the HDI communication board, extract the signal transmission paths and signal interaction methods of each component of the HDI communication board, and generate an HDI communication board operating mechanism description document; Based on the HDI communication board fault signal performance list and HDI communication board operation mechanism description document, extract the signal dimensions corresponding to various faults, describe the signal characteristics and signal acquisition requirements of each signal dimension, and generate a fault signal dimension description set. Perform correlation analysis on the signal dimensions in the fault signal dimension description set, define the signal dimension combinations corresponding to different fault types, and generate a fault signal requirement subset. Integrate all fault signal requirement subsets, supplement the priority information and signal transmission timing information of various fault signals, and generate a fault signal requirement set. The basic architecture for building the initial neural network model includes a signal input layer, a multi-level signal processing layer, a signal fusion layer, and a signal output layer. The signal input layer is configured to receive and process signal data, the multi-level signal processing layer is configured to process signal features layer by layer, the signal fusion layer is configured to fuse signal features from different levels, and the signal output layer is configured to output the initial signal processing results. Extract signal processing capability information of each layer of the initial neural network model, covering the signal dimension range that each layer can process, signal processing speed and signal transmission format between layers, and generate a list of the layer capabilities of the initial neural network model. The signal dimensions and signal processing requirements in the fault signal requirement set are compared one by one with the initial neural network model level capability list. The initial neural network model level and processing node corresponding to each signal dimension are defined, and a correspondence table between signal dimensions and model levels is generated. Based on the correspondence table between signal dimensions and model levels, a connection mapping relationship is constructed between signal nodes in the fault signal demand set and processing nodes at each level of the initial neural network model, forming an initial adaptation relationship framework between the fault signal demand set and the initial neural network model. Based on the signal dimensions in the fault signal requirement set, check whether there are any uncovered signal dimensions in the initial adaptation relationship framework. For uncovered signal dimensions, add corresponding model levels and processing nodes to the initial adaptation relationship framework to generate an adaptation relationship framework between the fault signal requirement set and the initial neural network model.

3. The HDI communication board fault prediction method based on a neural network model according to claim 1, characterized in that, Based on the adaptation framework between the fault signal demand set and the initial neural network model, the number of hierarchical nodes, hierarchical connection methods, and signal processing channels of the initial neural network model are adapted and adjusted to obtain a neural network model adapted to the fault signal demand, including: The framework for analyzing the adaptation relationship between the fault signal requirement set and the initial neural network model is used to extract the number of signal dimensions, signal processing accuracy requirements, and signal transmission timing requirements that each level needs to adapt to, and generate a list of hierarchical adaptation requirements. For the signal input layer of the initial neural network model, the number of signal dimensions and signal acquisition requirements corresponding to the signal input layer in the hierarchical adaptation requirement list are adjusted to adjust the number of signal receiving nodes and the signal receiving format of the signal input layer so that the signal input layer can receive the signal type of the HDI communication board operation signal and receive the signal at the signal transmission rate of the HDI communication board operation signal, thus obtaining the adapted signal input layer structure. For the multi-level signal processing layer of the initial neural network model, the number of nodes in each level of the signal processing layer is adjusted according to the signal dimension and processing accuracy requirements in the hierarchical adaptation requirement list so that the number of nodes meets the processing requirements of the corresponding signal dimension. At the same time, the signal processing logic of each node is modified to increase the ability to identify fault signals, thus obtaining the adapted signal processing layer structure. Based on the signal transmission timing requirements in the fault signal demand set, the connection method between the adapted signal processing layers at each level is adjusted, and a cross-level signal transmission path is established to enable the fault signals processed at different levels to be transmitted interactively according to the timing requirements, resulting in the adjusted hierarchical connection structure. For the signal fusion layer of the initial neural network model, the node configuration and fusion logic of signal fusion are adjusted according to the signal fusion requirements in the hierarchical adaptation requirement list, so as to increase the transmission weight of signal features strongly related to fault prediction in the fusion process, and obtain the adapted signal fusion layer structure. For the signal output layer of the initial neural network model, the number of output nodes and the output signal format of the signal output layer are adjusted according to the number of fault types and the fault prediction accuracy requirements in the fault signal demand set, so that the signal output layer can output prediction-related signals of various faults, thus obtaining the adapted signal output layer structure. The adapted signal input layer structure, the adapted signal processing layer structures at each level, the adjusted hierarchical connection structure, the adapted signal fusion layer structure, and the adapted signal output layer structure are integrated to generate a preliminary adapted neural network model. Extract the signal processing results of each level of the initially adapted neural network model, compare them with the signal processing requirements in the fault signal requirement set, and define the differences between the signal processing results of each level and the required signals. Based on the differences, the signal processing layers and signal fusion layers of the initially adapted neural network model are adjusted a second time, and the node configuration and processing logic are modified to reduce the difference between the signal processing results and the required signals, thereby generating a second-adapted neural network model. The overall structure of the second-adapted neural network model is then fixed to generate a neural network model that adapts to the fault signal requirements.

4. The HDI communication board fault prediction method based on a neural network model according to claim 1, characterized in that, The process involves acquiring the real-time operating signal set of the HDI communication board, inputting this set into a neural network model adapted to fault signal requirements, and using a pre-set fault signal guidance mechanism within the model to perform hierarchical extraction and correlation mining of fault-related signals in the real-time operating signal set, thereby obtaining a fault feature association set, including: Acquire various real-time operating signals of the HDI communication board during operation, classify and organize the acquired real-time operating signals according to signal type, and generate a set of real-time operating signals of the HDI communication board. The real-time operating signal set of the HDI communication board is input into the signal input layer of the neural network model that adapts to the fault signal requirements one by one according to the preset input order. The real-time operating signal is converted into a signal format that the model can process through the signal conversion node of the signal input layer, and the real-time operating signal set after format conversion is obtained. Extract signal feature descriptions corresponding to various faults from the fault signal demand set, convert the signal feature descriptions into guidance parameters that the model can recognize, generate a fault signal guidance set, and input the fault signal guidance set into the signal processing layers of the neural network model that adapts to the fault signal demand. In the first-level signal processing layer, based on the basic fault signal feature guidance parameters in the fault signal guidance set, the basic fault signal features are extracted from the format-converted real-time running signal set to obtain the first-level fault signal feature set. The first-level fault signal feature set is input into the second-level signal processing layer. Based on the detailed fault signal feature guidance parameters in the fault signal guidance set, the detailed fault signal features of the first-level fault signal feature set are extracted to obtain the second-level fault signal feature set. The second-level fault signal feature set is input into the third-level signal processing layer. Based on the deep fault signal feature guidance parameters in the fault signal guidance set, the deep fault signal feature is extracted from the second-level fault signal feature set to obtain the third-level fault signal feature set. The first-level fault signal feature set, the second-level fault signal feature set, and the third-level fault signal feature set are respectively processed for feature normalization to generate normalized first-level, second-level, and third-level fault signal feature sets. The normalized first-level, second-level, and third-level fault signal feature sets are then input into the signal fusion layer of a neural network model adapted to fault signal requirements according to their corresponding hierarchical information. The signal fusion layer performs correlation and fusion of the normalized fault signal feature sets of different levels according to the signal correlation requirements in the fault signal requirement set to obtain a fused fault feature set. A correlation analysis is performed on the fault features in the fused fault feature set to define the dependency and co-occurrence relationships between different fault features and generate a fault feature correlation table. Based on the fault feature association table, the fault features in the fused fault feature set are grouped according to their association relationships to obtain multiple fault feature association groups; All fault feature association groups are integrated, and the signal source information and timing information corresponding to each fault feature association group are supplemented to generate a fault feature association set.

5. The HDI communication board fault prediction method based on a neural network model according to claim 4, characterized in that, The process involves extracting signal feature descriptions corresponding to various faults from the fault signal demand set, converting these descriptions into model-recognizable guidance parameters, generating a fault signal guidance set, and inputting this set into the signal processing layers of the neural network model adapted to the fault signal demand. This includes: Analyze the fault signal demand set, extract the various fault types contained therein, classify the various fault types according to the location where the fault occurs, and generate a fault type classification table; For each fault type in the fault type classification table, extract the corresponding signal feature description, covering the signal fluctuation pattern, signal duration, signal change trend and signal interaction relationship, and generate a subset of signal feature descriptions for each fault type. Semantic normalization is performed on the signal feature descriptions in the subset of signal feature descriptions for each type of fault to unify the expression of signal feature descriptions. The normalized subset of signal feature descriptions for each type of fault is then converted into numerical guidance parameters that can be recognized by the neural network model that is adapted to the fault signal requirements according to the preset parameter conversion rules, thereby generating a subset of guidance parameters for each type of fault. Based on the processing range of each signal processing layer of the neural network model that adapts to the fault signal requirements, the subset of guidance parameters for each type of fault is divided into basic guidance parameters, detailed guidance parameters, and deep guidance parameters, which correspond to the first-level signal processing layer, the second-level signal processing layer, and the third-level signal processing layer of the neural network model, respectively. The basic guidance parameters for all fault types are integrated to generate the guidance parameter set corresponding to the first-level signal processing layer; the detailed guidance parameters for all fault types are integrated to generate the guidance parameter set corresponding to the second-level signal processing layer; and the deep guidance parameters for all fault types are integrated to generate the guidance parameter set corresponding to the third-level signal processing layer. The guidance parameter groups corresponding to the first, second, and third level signal processing layers are encapsulated in the order of model hierarchy to generate a fault signal guidance set; Determine the guidance parameter receiving interface of each level of the signal processing layer in the neural network model that is adapted to the fault signal requirements, and send the guidance parameter group of the corresponding level in the fault signal guidance set to the receiving interface of each level of the signal processing layer.

6. The HDI communication board fault prediction method based on a neural network model according to claim 1, characterized in that, The method, based on the fault feature association set and combined with the correspondence between fault signals and fault types in the fault signal demand set, determines the fault type and fault occurrence trend of the HDI communication board, and generates a set of HDI communication board fault prediction results, including: Extract all fault feature association groups, signal source information, and time sequence information from the fault feature association set to generate fault feature association details; Extract the correspondence between fault signals and fault types from the fault signal demand set, and generate a fault signal type correspondence table; Each fault feature association group in the fault feature association details is compared with the fault signals in the fault signal type correspondence table one by one to define the fault type corresponding to each fault feature association group and generate a preliminary fault type matching result. Extract the time series information corresponding to each fault feature association group, analyze the signal change trend of the fault feature association group at different time stages, and generate fault feature time series change curves. Based on the time-series change curve of fault characteristics, the development trend of fault characteristics is judged, the possible time range and development speed of fault occurrence are defined, and the fault occurrence trend judgment result is generated. The preliminary fault type matching results are bound with the fault occurrence trend judgment results to generate a single fault prediction record. Each fault prediction record includes fault type, fault feature association group, signal source information, time sequence information and fault occurrence trend. All individual fault prediction records are classified and organized, and the fault prediction records are divided into different part fault prediction sets according to the components of the HDI communication board. Extract the fault occurrence trend judgment results from the fault prediction set of each part, analyze the correlation between different fault types in the same part, and generate a description of the correlation between part faults. Add the description of the associated impact of the component failures to the corresponding component failure prediction set to generate the component failure prediction set; The fault prediction sets of all parts are integrated to generate a set of HDI communication board fault prediction results that includes various fault types, fault occurrence trends and the correlation between parts.

7. The HDI communication board fault prediction method based on a neural network model according to claim 2, characterized in that, Based on the HDI communication board fault signal manifestation list and HDI communication board operation mechanism description document, the signal dimensions corresponding to various faults are extracted, the signal characteristics and signal acquisition requirements of each signal dimension are described, and a fault signal dimension description set is generated, including: Select a single type of fault signal from the list of fault signals of HDI communication board, and in conjunction with the HDI communication board operation mechanism documentation, define the corresponding HDI communication board operation stage for this type of fault signal. Analyze the signal transmission paths and signal interaction nodes involved in this operation to identify the signal types related to the performance of this type of fault signal. Extract signal dimensions that can characterize the fault signal performance from the defined signal types, with each signal dimension corresponding to a specific attribute of the signal; Each extracted signal dimension is described by features, defining the specific manifestation and change pattern of the signal dimension when the fault occurs, and generating a signal dimension feature description. Based on the HDI communication board's operating mechanism documentation and actual operating requirements, the signal acquisition method, acquisition frequency, and acquisition range for each signal dimension are defined, and signal dimension acquisition requirements are generated. The signal dimension, signal dimension feature description and signal dimension acquisition requirements corresponding to the single type of fault signal are bound together to generate a subset of signal dimension descriptions for single type of fault. All fault signal manifestations in the HDI communication board fault signal manifestation list are processed using the above steps of selection, definition, extraction, feature description, definition of acquisition requirements and binding, generating a signal dimension description subset for each type of fault. Summarize all signal dimension description subsets for various types of faults and remove duplicate signal dimension descriptions. The summarized signal dimension descriptions are categorized and sorted, and the signal dimension descriptions are divided into different categories according to signal type. The classified and sorted signal dimension descriptions are integrated to generate a fault signal dimension description set.

8. The HDI communication board fault prediction method based on a neural network model according to claim 3, characterized in that, The multi-level signal processing layer for the initial neural network model adjusts the number of nodes in each level of the signal processing layer according to the signal dimension and processing accuracy requirements in the level adaptation requirement list, so that the number of nodes meets the processing requirements of the corresponding signal dimension. At the same time, the signal processing logic of each node is modified to increase the ability to identify fault signals, resulting in the adapted signal processing layer structure for each level, including: Extract the number of signal dimensions, processing accuracy requirements, and fault signal perception requirements for multi-level signal processing layers from the hierarchical adaptation requirement list, and generate an adaptation parameter table for each level of signal processing layer. Select the first-level signal processing layer of the initial neural network model, analyze the current number of nodes, node processing logic and signal processing range of the first-level signal processing layer, and generate the current state information of the first-level signal processing layer. Based on the number of signal dimensions corresponding to the first-level signal processing layer in the adaptation parameter table of each level of signal processing layer, calculate the number of nodes required for the first-level signal processing layer, and adjust the number of nodes of the first-level signal processing layer so that the number of nodes can handle all the corresponding signal dimensions. Based on the processing accuracy requirements of the first-level signal processing layer in the adaptation parameter table of each level of signal processing layer, modify the signal processing logic of each node of the first-level signal processing layer, and adjust the signal filtering rules and signal enhancement logic inside the node. Based on the fault signal perception requirements of the first-level signal processing layer in the adaptation parameter table of each level of signal processing layer, fault signal feature recognition rules are added to the processing logic of each node to improve the node's ability to perceive fault signals. Adjust the signal transmission connection relationship between nodes in the first-level signal processing layer so that the signal transmission between nodes meets the preset transmission conditions, and generate the first-level adapted signal processing layer structure. The second-level signal processing layer of the initial neural network model is selected, and the processing steps of parsing, adjusting the number of nodes, optimizing the processing logic, enhancing the fault signal perception, and adjusting the connection relationship for the first-level signal processing layer are repeated. Based on the adaptation parameters corresponding to the second-level signal processing layer in the adaptation parameter table of each level of signal processing layer, the adapted signal processing layer structure of the second level is generated. Select the third-level signal processing layer of the initial neural network model, and follow the same processing steps. Based on the adaptation parameters of the third-level signal processing layer in the adaptation parameter table of each level of signal processing layer, complete the adjustment of the number of nodes, optimization of processing logic, and adjustment of connection relationship to generate the adapted signal processing layer structure of the third level. The adapted signal processing layer structures of the first, second, and third levels are correlated and verified according to the original hierarchical order to ensure that the signal processing flow between each level meets the preset coherence conditions. The adapted signal processing layer structures of each level after correlation verification are integrated as a whole to generate the final adapted signal processing layer structure.

9. The HDI communication board fault prediction method based on a neural network model according to claim 4, characterized in that, The process involves performing correlation analysis on fault features in the fused fault feature set, defining the dependencies and co-occurrence relationships between different fault features, and generating a fault feature correlation table, including: Extract all fault features from the fused fault feature set, assign a unique feature identifier to each fault feature, and generate a fault feature identifier list. All fault features in the fault feature identifier list are paired up to generate all possible fault feature combination pairs. For each fault feature pair, extract the occurrence time, signal source, and signal strength change information of the two fault features in the fused fault feature set; Analyze whether the occurrence times of the two fault characteristics overlap, whether the signal source belongs to the same operating stage of the HDI communication board, and whether the signal strength changes are synchronous; Based on the overlap of occurrence time, correlation of signal sources, and synchronicity of signal strength changes, the dependency relationship between two fault characteristics is defined, that is, whether the occurrence of one fault characteristic depends on the occurrence of another fault characteristic. Simultaneously define the co-occurrence relationship between two fault characteristics, that is, whether the two fault characteristics frequently occur at the same time; The dependency and co-occurrence relationships of each fault feature combination pair are described, and a single fault feature association record is generated. Summarize all individual fault feature association records, sort the summarized fault feature association records according to fault feature identifiers, and generate a fault feature association table.

10. A fault prediction system for HDI communication boards based on a neural network model, characterized in that, The device includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to run the programs, instructions, or code in the memory to implement the HDI communication board fault prediction method based on a neural network model as described in any one of claims 1-9.

Images