An electric control panel fault prediction and self-recovery method and device

Abstract

The application discloses an electric control board fault prediction and self-healing method and device, and belongs to the field of electronic equipment health management. The method comprises the following steps: generating a probe excitation signal; injecting the probe excitation signal into a measured circuit of the electric control board, and synchronously capturing the global field response data of the measured circuit under the joint action of the probe excitation signal and a working signal; reconstructing a real-time physical parameter atlas of the measured circuit by solving a physical inverse problem according to the global field response data and the probe excitation signal; generating corresponding parameter harmonization instructions according to the deviation between the parameter values of each node in the real-time physical parameter atlas and the corresponding nominal values; and executing the parameter harmonization instructions to dynamically adjust the adjustable element parameters of the corresponding nodes in the measured circuit, so that the electrical performance of the nodes approaches the nominal values, early and accurate prediction and unconscious self-healing of the electric control board fault can be realized, and the stability and reliability of the control system are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of health management technology for electronic devices, and in particular to a method and apparatus for predicting and self-healing faults in electronic control boards. Background Technology

[0002] As the brain and nerve center of various modern industrial control, communication, aerospace, and automation equipment, the functional integrity and operational reliability of the electronic control board directly determine the stability and safety of the entire system. These electronic control boards integrate a large number of electronic components, and when operating for a long time in complex electromagnetic and temperature and humidity environments, they will inevitably experience performance degradation or even sudden failures due to factors such as component aging and environmental stress.

[0003] In existing technologies, to ensure the reliable operation of the electronic control board, the commonly used technical means include passive monitoring and periodic maintenance. Passive monitoring mainly involves installing sensors on the electronic control board to monitor macroscopic operating parameters such as voltage, current, and temperature of key nodes in real time. When these parameters exceed preset fixed thresholds, an alarm is triggered. In addition, some solutions introduce statistical analysis or machine learning algorithms based on historical operating data to predict the probability of failure and issue early warnings. For recovery after a failure, it often relies on system restart via watchdog circuits or primary / backup channel switching through hardware redundancy design.

[0004] Chinese Patent Publication No. CN201810680112.4 discloses a control device and a machine learning device for predicting faults in printed circuit boards and components constituting a numerical control device. The machine learning device included in the fault prediction device comprises: a state observation unit that observes operating status data representing the operating condition of the managed equipment and equipment structure data representing the equipment structure as state variables representing the current state of the environment; a tag data acquisition unit that acquires maintenance history data representing the maintenance history of the managed equipment as tag data; and a learning unit that uses the state variables and tag data to learn how to correlate the fault periods of the printed circuit boards constituting the managed equipment with the operating status data and equipment structure data.

[0005] However, the aforementioned existing technologies have significant technical shortcomings. Passive monitoring methods based on fixed thresholds are slow to respond, often only detecting problems when the control board is already clearly malfunctioning or about to fail, missing the opportunity for early intervention. While statistical prediction methods have some foresight, their predictions are usually probabilistic, making it difficult to accurately pinpoint specific faulty components and failure modes, and they are poor at identifying novel or complex failure modes not seen in the training data. Furthermore, existing fault recovery measures, such as system restarts or primary / backup switching, are drastic, often causing service interruptions, and cannot address the slow performance degradation caused by gradual drift in component parameters. The prediction and repair processes are disconnected, lacking an effective coordination mechanism. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a method and apparatus for predicting and self-healing electronic control board faults. By actively injecting detection excitation signals, reconstructing real-time physical parameter maps, and dynamically harmonizing parameters, it can achieve early and accurate prediction and imperceptible self-healing of electronic control board faults, significantly improving the stability and reliability of the control system.

[0007] The above objectives can be achieved through the following approach:

[0008] A method for predicting and self-healing faults in an electronic control board, the method comprising:

[0009] Generate exploration stimulus signal;

[0010] The probe excitation signal is injected into the circuit under test on the electronic control board, and the global field response data of the circuit under test under the combined action of the probe excitation signal and the working signal is captured simultaneously.

[0011] Based on the global field response data and the probe excitation signal, the real-time physical parameter spectrum of the circuit under test is reconstructed by solving the physical inverse problem;

[0012] Based on the deviation between the parameter values ​​of each node in the real-time physical parameter map and the corresponding nominal values, generate corresponding parameter harmonization instructions;

[0013] The parameter adjustment command is executed to dynamically adjust the adjustable component parameters of the corresponding node in the circuit under test, so that the electrical performance of the node approaches the nominal value.

[0014] Optionally, generating the probe excitation signal includes: determining at least one key signal injection node based on the topology of the circuit under test; generating a scanning sequence covering a frequency range as a basic excitation signal; and modulating the basic excitation signal according to encoding rules to generate the probe excitation signal with specific spectral and temporal characteristics.

[0015] Optionally, the synchronous acquisition of the global field response data of the circuit under test includes: synchronously acquiring electrical signals, electromagnetic field distribution signals, and thermal radiation signals through multiple miniature sensing units arranged at key locations on the circuit board under test; performing synchronization and time-frequency alignment processing on the acquired multimodal raw signals, and fusing them to generate the global field response data.

[0016] Optionally, the step of reconstructing the real-time physical parameter map of the circuit under test by solving a physical inverse problem based on the global field response data and the probe excitation signal includes: establishing an idealized multi-physics coupling model of the circuit under test as an initial forward model; inputting the probe excitation signal into the initial forward model to calculate the theoretical field response data; constructing an inversion optimization function with the objective of minimizing the difference between the theoretical field response data and the global field response data; solving the inversion optimization function, iteratively updating the component parameter set in the initial forward model until the difference satisfies a convergence condition, and expressing the final component parameter set as the real-time physical parameter map.

[0017] Optionally, the method further includes: identifying abnormal nodes whose parameter drift exceeds a first threshold from the real-time physical parameter map; extracting the parameter drift pattern of the abnormal nodes, the parameter drift pattern including drift direction, drift rate and correlation degree; mapping potential performance degradation paths according to the parameter drift pattern, and estimating the remaining effective lifetime.

[0018] Optionally, the step of generating corresponding parameter harmonization instructions based on the deviation between the parameter values ​​of each node in the real-time physical parameter map and the corresponding nominal values ​​includes: determining whether the deviation exceeds a second threshold; if it does not exceed, calculating the compensation amount based on the negative feedback control law, and generating the parameter harmonization instructions for fine-tuning the parameters of the adjustable element; if it exceeds, combining the potential performance degradation path and the current operating load information, selecting a degraded operation strategy or a redundancy switching strategy from the strategy library, and generating a composite parameter harmonization instruction containing a sequence of actions.

[0019] Optionally, the execution parameter harmonization instruction includes: if the parameter harmonization instruction is a fine-tuning instruction, driving the digital potentiometer or programmable gain amplifier in the adjustable element parameters to make continuous or step-by-step adjustments according to the compensation amount; if the parameter harmonization instruction is a composite instruction, first controlling the analog switch to switch to the backup function circuit, then updating the idealized multiphysics coupling model based on the switched circuit state, and regenerating the fine-tuning instruction for the new circuit.

[0020] Optionally, the method further includes: recording the real-time physical parameter spectrum before and after each execution of the parameter harmonization command to form a harmonization effect sample; correcting the solution process of the physical inverse problem based on the harmonization effect sample to update the parameter mapping relationship in the initial forward model; and training a self-healing efficacy evaluation model using historical harmonization effect samples to optimize the parameters of the second threshold and the negative feedback control law.

[0021] Optionally, injecting the probe excitation signal into the circuit under test of the electronic control board includes: injecting the probe excitation signal during the idle time window of the main business logic operation of the electronic control board; or inserting and injecting the probe excitation signal when the transient characteristics of the working signal are detected to be stable.

[0022] The present invention also provides an apparatus for implementing the above method, the apparatus comprising:

[0023] An active stimulus source is used to generate exploration stimulus signals;

[0024] A global field response sensor network, distributed on the electronic control board, is used to synchronously capture the global field response data of the circuit under test under the combined action of the detection excitation signal and the working signal;

[0025] A real-time physical feature reconstructor, connected to the global field response sensor network, is used to reconstruct the real-time physical parameter spectrum of the circuit under test based on the global field response data and the detection excitation signal.

[0026] A dynamic parameter tuner, connected to the real-time physical feature reconstructor, is used to generate and execute parameter tuning instructions based on the real-time physical parameter spectrum to dynamically adjust the parameters of adjustable components in the circuit under test.

[0027] Compared with the prior art, the present invention has the following advantages:

[0028] This invention actively injects a probing excitation signal into the circuit under test and captures its multi-physics response data, enabling reverse reconstruction of the real-time physical parameter spectrum of the circuit's internal components. This allows for accurate identification and location of faults at the nascent stage of slight drift in their physical parameters, before significant degradation or failure of component functionality occurs, thus improving the earlyness and accuracy of fault prediction.

[0029] This invention precisely maps the fault diagnosis output to the parameter deviation values ​​of specific components, directly mapping it to the input of self-healing execution, i.e., the compensation amount for adjustable components, thus constructing a closed loop from diagnosis to repair. This integrated diagnosis and repair mechanism eliminates the delays and information distortions in traditional solutions from early warning, analysis to decision-making and execution, achieving near-instantaneous adaptive compensation and repair for early faults.

[0030] This invention, through continuous and fine-tuning parameter harmonization, enables imperceptible self-healing against component aging or parameter drift without interrupting the main business functions of the electronic control board. This mechanism of actively maintaining the internal steady state of the system ensures that the overall output performance of the electronic control board remains stable over a long period, avoiding sudden shutdowns caused by the accumulation of performance degradation, and significantly improving the overall availability and operational reliability of the control system.

[0031] The self-learning closed loop designed in this invention enables the system to accumulate experience from the success or failure of each self-healing action, continuously calibrating its internal physical model and optimizing its decision-making logic using harmonic effect samples. This allows the health management capability of the control board to self-evolve over time, exhibiting stronger adaptability to different operating conditions and aging paths, ultimately extending the entire lifespan of the equipment. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0034] Figure 2 This is a schematic diagram illustrating the generation principle and modulation characteristics of the detection excitation signal of the present invention;

[0035] Figure 3 This is a three-dimensional scatter view and noise annotation diagram collected by the sensor array of the present invention.

[0036] Figure 4 This is a schematic diagram of the parameter drift pattern recognition and lifetime prediction logic of the present invention;

[0037] Figure 5 This is a schematic diagram of the hierarchical closed-loop self-healing control logic of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] A method for predicting and self-healing faults in an electronic control board, the method comprising:

[0040] Generate exploration stimulus signal;

[0041] The probe excitation signal is injected into the circuit under test on the electronic control board, and the global field response data of the circuit under test under the combined action of the probe excitation signal and the working signal is captured simultaneously.

[0042] Based on the global field response data and the probe excitation signal, the real-time physical parameter spectrum of the circuit under test is reconstructed by solving the physical inverse problem;

[0043] Based on the deviation between the parameter values ​​of each node in the real-time physical parameter map and the corresponding nominal values, generate corresponding parameter harmonization instructions;

[0044] The parameter adjustment command is executed to dynamically adjust the adjustable component parameters of the corresponding node in the circuit under test, so that the electrical performance of the node approaches the nominal value.

[0045] Specifically, refer to Figure 1 As shown, by actively injecting a specially coded probing excitation signal into the circuit, and simultaneously capturing multi-dimensional, global physical field response data generated by the combined action of this excitation signal and the normal operating signal through a sensor network distributed throughout the circuit board, the circuit utilizes a pre-established mathematical model describing the ideal physical characteristics of the circuit. By solving the physical inverse problem (e.g., using a regularized iterative algorithm or gradient descent method), the actual captured field response data is compared and iteratively optimized with the theoretical response data predicted by the model. This allows for the reverse calculation of the current real physical parameter values ​​of each key component within the circuit, thereby constructing an accurate real-time physical parameter map. By comparing this map with the design nominal values, nodes where parameters drift can be directly identified. Based on the degree of deviation, precise parameter harmonization commands are generated and executed. Through adjustable components integrated in the drive circuit, these parameter deviations are dynamically compensated in real time, ensuring that the overall performance of the circuit remains close to the ideal state.

[0046] This enables early prediction of control board failures, accurately locating problems before components completely fail and only when physical parameters experience minor drifts. It elevates traditional fault warning to the level of fault prevention. This method deeply integrates diagnosis and self-healing processes, directly mapping diagnostic results to precise compensation actions. It completely eliminates the time delay and information loss between problem discovery and solution development and execution. Through continuous and gradual parameter adjustment, self-healing can be achieved without interrupting main business functions. Even with the degradation of some components, the overall functionality and performance of the control board remain stable over a long period, significantly improving equipment availability and reliability. Furthermore, it forms a complete self-learning closed loop, where each prediction and self-healing action becomes valuable experience for optimizing its internal diagnostic model and decision-making logic, allowing the level of intelligence to continuously evolve over time and extending the effective service life of the control board and the entire equipment system.

[0047] Optionally, generating a probe excitation signal includes: determining at least one key signal injection node based on the topology of the circuit under test; generating a scanning sequence covering a frequency range as a basic excitation signal; and modulating the basic excitation signal according to encoding rules to generate a probe excitation signal with specific spectral and temporal characteristics.

[0048] Specifically, refer to Figure 2As shown, a diagnostic signal with low energy but high information capacity, easily extracted from complex background noise, is generated without interfering with the normal operation of the control board. The signal generation is multi-stage, ensuring both a comprehensive evaluation of the circuit's frequency response characteristics and high-precision time-domain localization. Based on a pre-loaded topology design file of the circuit under test, such as a Gerber file or circuit netlist, critical paths with the greatest impact on overall circuit performance or historically high failure rates are identified. These paths typically include the main path of the power distribution network, the high-frequency clock distribution network, and critical input / output interface circuits. One or more physical locations on these paths are designated as critical signal injection nodes, ensuring the injected signal flows through the maximum number of critical components. Subsequently, an active excitation source generates a scanning sequence covering the operating frequency bands and potential failure frequency bands of the critical components of the circuit under test as the base excitation signal. This base excitation signal is typically a linearly modulated (LFM) signal, whose frequency is linearly scanned from a lower starting frequency, such as 10 kHz, to a higher cutoff frequency, such as 10 MHz, within a preset time period, such as 1 to 10 milliseconds. This scanning sequence design ensures that subsequent diagnostics can comprehensively reflect the health status of frequency-sensitive components such as capacitors and inductors at different frequencies, forming a broad-spectrum characteristic of the basic excitation signal. To enhance the signal's anti-interference capability and give it precise timestamp characteristics, a preset encoding rule in memory is invoked to modulate the basic excitation signal. This encoding rule typically uses a pseudo-random noise sequence, such as an M-sequence or Gold code, due to its sharp autocorrelation peak characteristics. The modulation process generates the final probe excitation signal by multiplying or XORing the basic excitation signal with this pseudo-random noise sequence. This process can be represented by the following formula:

[0049] ,

[0050] in, The final generated exploration excitation signal possesses specific spectral and temporal characteristics. It is the basic excitation signal generated in the previous step, namely the linear frequency modulation signal, which is dynamically generated by the signal generator according to the set scanning frequency range (which covers the characteristic impedance sensitive frequency band of key capacitors and inductors in the circuit) and duration. The pre-defined encoding rules are represented by a digital pseudo-random code sequence with excellent autocorrelation characteristics read from the onboard non-volatile memory. By modulating the excitation signal in this way, not only is wideband test information carried, but its energy is also extended to a wider spectrum, making its power spectral density lower than the background noise of the circuit, thus achieving low detectability. At the same time, its unique encoding gives it a unique identity, providing a technical basis for accurately capturing its response signal and calculating the propagation delay in the full-domain field response data through correlation operations.

[0051] Optionally, the global field response data of the circuit under test can be captured synchronously, including: simultaneously acquiring electrical signals, electromagnetic field distribution signals and thermal radiation signals through multiple miniature sensing units arranged at key locations on the circuit board under test; synchronizing and time-frequency aligning the acquired multimodal raw signals and fusing them to generate global field response data.

[0052] Specifically, by constructing an ultra-high-dimensional dataset that reflects the coupled responses of the electronic control board under transient excitation across multiple physical fields (electrical, electromagnetic, and thermal), this dataset must be highly aligned in both time and space to provide accurate and distortion-free input for subsequent inverse physical problems. This process relies on an array of multiple miniature sensing units deployed at key locations on the circuit board under test, referring to... Figure 3 As shown, these key locations are high-stress areas pre-determined through thermal and electromagnetic compatibility simulations of the circuit, such as near the pins of power switching devices, corners of high-speed digital signal lines, and the bottom of large-capacity electrolytic capacitors. The micro-sensing units are heterogeneously integrated, specifically including a high-impedance differential probe and an integrated Hall current sensor for acquiring electrical signals; a micro-loop antenna array for acquiring electromagnetic field distribution signals, which can be integrated in a patch form; and a non-contact infrared thermopile sensor array based on microelectromechanical systems (MEMS) technology for acquiring thermal radiation signals. The analog outputs of all micro-sensing units are connected to an analog-to-digital converter driven by a unified high-precision clock source to ensure that all channels synchronously begin high-speed sampling at the same trigger signal, i.e., the start time of the probe excitation signal injection. The sampling rate is typically set in the range of 200 MHz to 1 GHz to fully capture the response details of the probe excitation signal, thereby obtaining multi-mode raw signals. The processor performs synchronization and time-frequency alignment processing on these parallel and discrete multi-mode raw signal data streams; the processing is not a simple data merging. First, digital clock calibration is performed. A specific pseudo-random code contained in the probe excitation signal is used as a global synchronization reference. By detecting relevant peaks in the data from each channel, the transmission delay differences of different sensing paths are accurately calculated and compensated, achieving nanosecond-level time alignment. Next, data reconstruction and fusion are performed. All time-aligned discrete data points are mapped onto a unified three-dimensional spatiotemporal grid based on the physical coordinates of each micro-sensor unit on the circuit board. The output of this process is the global field response data. Structurally, this data is a four-dimensional tensor, with its dimensions representing time, two-dimensional spatial coordinates, and the type of physical field. This structured global field response data comprehensively describes snapshots of the electrical, electromagnetic, and thermal field distributions of the entire circuit board's critical areas at every time slice from the injection of the probe excitation signal. This provides a complete and high-fidelity data foundation for subsequently inferring the health status of local components from the overall physical performance.

[0053] Optionally, based on the global field response data and the probe excitation signal, the real-time physical parameter map of the circuit under test is reconstructed by solving the physical inverse problem. This includes: establishing an idealized multi-physics coupling model of the circuit under test as the initial forward model; inputting the probe excitation signal into the initial forward model to calculate the theoretical field response data; constructing an inversion optimization function with the objective of minimizing the difference between the theoretical field response data and the global field response data; solving the inversion optimization function, iteratively updating the component parameter set in the initial forward model until the difference satisfies a convergence condition, and expressing the final component parameter set as the real-time physical parameter map.

[0054] Specifically, by solving the inverse physical problem to reconstruct the real-time physical parameter spectrum of the circuit under test, the collected field data characterizing the external physical performance of the circuit is transformed into a set of quantitative parameters describing the health status of the internal components of the circuit. This set of parameters is then imported into the electronic design automation (EDA) file of the control board, such as an ODB++ or IPC-2581 format file, and constructed offline using finite element analysis software. It is a digital twin containing information such as the geometry of all traces on the circuit board, the dielectric constant of the materials, and the thermal resistance of the components. To adapt to the computing power resources of the onboard processor, this digital twin undergoes model order reduction (MOR) processing and is transformed into a high-precision surrogate model, which serves as the initial forward model. It should be noted that, considering the computing power limitations of the embedded processor on the control board, and to meet real-time requirements, the initial forward model is preferably a 'Reduced Order Model (ROM)' or 'Surrogate Model' generated based on the above finite element analysis results during actual deployment. This reduced-order model retains the physical field mapping relationships of key nodes, but the computational complexity is reduced exponentially, enabling millisecond-level iterative solutions within the microprocessor of this device. Its internal component parameter set is set to its design nominal values. Then, the real-time physical feature reconstructor takes the digital waveform of the probe excitation signal as input and loads it into the initial forward model for a multi-physics joint simulation. The simulation calculates the voltage, current, electromagnetic field, and thermal field distribution generated when the excitation signal propagates in an ideal circuit. The simulation results are the theoretical field response data. Then, an inversion optimization function is constructed with the objective of minimizing the difference between the theoretical field response data and the actually acquired global field response data. This function solves a nonlinear least squares problem. To address the inherent ill-posedness of the electromagnetic inverse problem, this embodiment introduces a Tikhonov regularization constraint, mathematically expressed as:

[0055] ,

[0056] in, This is the inversion optimization function. The first term is the data fitting error, and the second term is the regularization constraint term. It is the set of component parameters to be solved in the initial forward model, including the values ​​of key resistors, capacitors, inductors, etc. The probe stimulus signal represents the known injection. This represents the forward model operator, which runs a simulation once and outputs the theoretical field response data. It is the global field response data that was previously obtained as known input. It is a regularization parameter, dynamically selected using the L-curve method or generalized cross-validation (GCV), used to balance the degree of data fit with the smoothness of the solution. It is the design nominal value vector of the component, used as prior information to constrain the solution process. As a weighting matrix, it should be noted that although the spatial distribution of micro-sensing units may be less than the number of component parameters to be solved, the detection excitation signal injected in this invention... It is a scanning sequence covering a wide frequency band (e.g., 100kHz-20MHz), and the full-domain field response data $D_m$ has extremely high degrees of freedom in the frequency domain. By utilizing the differences in the impedance characteristics of components at different frequencies, and combining a small number of spatial measurement points with a large amount of frequency data, the equations are mathematically transformed into an overdetermined system. This allows for the unique determination of the parameter values ​​of each key component in the circuit. Subsequently, to ensure the convergence of the iterative solution, an iterative optimization algorithm, such as the Levenberg-Marquardt algorithm or the conjugate gradient method, is used to solve the aforementioned inversion optimization function. Starting from the initial nominal parameter set, the algorithm calculates the error function in each iteration. For parameter set The gradient is calculated, and the parameter set P is finely adjusted along the direction that results in the fastest reduction of error. This iterative process continues, constantly updating the component parameter set until the error... The value is lower than the preset convergence threshold, for example, less than 10 to the power of -6, or the parameter set The update rate tends to stabilize. At this point, the finally converged set of component parameters P is the real-time physical parameter map, which, in the form of a matrix or vector, accurately reflects the actual physical parameter values ​​of each key component in the circuit under test at the current moment, rather than just whether its function is normal.

[0057] Optionally, the method further includes: identifying abnormal nodes whose parameter drift exceeds a first threshold from the real-time physical parameter map; extracting the parameter drift pattern of the abnormal node, the parameter drift pattern including drift direction, drift rate and correlation; mapping potential performance degradation paths according to the parameter drift pattern, and estimating the remaining effective lifetime.

[0058] Specifically, by extracting dynamic trend information from quantified parameter drift, the system extends its capabilities from assessing current health status to predicting future failure risks. The latest real-time physical parameter map is compared element-by-element with the stored nominal design parameter map of the circuit. By calculating the relative deviation rate of each parameter, abnormal nodes where parameter drift exceeds a first preset threshold are identified from the real-time physical parameter map. This first preset threshold is not a single fixed value but is set based on the reliability manuals (such as MIL-HDBK-217F) and historical maintenance data for different types of components. For example, the threshold for precision resistors might be set to 1%, while the capacitance reduction threshold for electrolytic capacitors might be set to 20%. Simultaneously, for each identified abnormal node, its historical real-time physical parameter map sequence is retrieved from the onboard memory. This sequence records parameter snapshots since the device was commissioned. The parameter drift pattern of the abnormal node can be extracted by performing time series analysis on the parameter change sequence over time, for example, by applying the least squares method for linear or nonlinear fitting. The result of this analysis is the parameter drift pattern, which includes the following key information: drift direction, i.e., whether the parameter value increases or decreases; drift rate, i.e., the amount of parameter change per unit time, which is the first derivative of the fitted curve; and correlation, i.e., the temporal correlation coefficient between the parameter drift of this node and the parameter drift of other abnormal nodes, used to identify common cause failure modes. Based on the extracted parameter drift pattern, a pre-defined physical failure model library is called to map potential performance degradation paths and estimate the remaining effective lifetime. For example, if the on-resistance Rds(on) of a power MOSFET exhibits a continuously increasing parameter drift pattern, it will be mapped to a chip overheating aging failure model. The estimation of the remaining effective lifetime is achieved by extrapolating the fitted parameter drift curve, i.e., calculating the time required for the parameter to reach its failure threshold (e.g., Rds(on) increases by 50%). This process can be expressed as:

[0059] ,

[0060] in, This represents the estimated remaining effective lifespan. It is a parameter threshold defined as complete failure, obtained from the component specification sheet. It is the parameter value of the node obtained from the real-time physical parameter map at the current moment. It is the average drift rate calculated based on the parameter drift mode. (Refer to...) Figure 4 As shown, this step not only identifies where the problem is, but also quantifies the severity of the problem and how long it can be used, providing precise time window guidance for preventative maintenance.

[0061] Optionally, based on the deviation between the parameter values ​​of each node in the real-time physical parameter map and the corresponding nominal values, a corresponding parameter harmonization instruction is generated, including: determining whether the deviation exceeds a second threshold; if it does not exceed, calculating the compensation amount based on the negative feedback control law, and generating a parameter harmonization instruction for fine-tuning the parameters of the adjustable components; if it exceeds, combining potential performance degradation paths and current operating load information, selecting a degraded operation strategy or a redundancy switching strategy from the strategy library, and generating a composite parameter harmonization instruction containing a sequence of actions.

[0062] Specifically, by diagnosing quantified parameter deviations and translating them into specific, executable, graded repair or compensation actions, seamless integration of diagnosis and decision-making is achieved. The parameter values ​​of each node in the real-time physical parameter map are compared with preset nominal values ​​to calculate the absolute deviation or relative deviation rate of each parameter. Then, it is determined whether the deviation exceeds a preset second threshold. This second preset threshold is a tiered threshold system that defines the sensitivity and strength of the self-healing response. For example, for a filter capacitor, a 10% decrease in capacitance may not exceed the second preset threshold, while a decrease exceeding 30% is considered exceeding it. This judgment is the trigger condition for generating parameter harmonization commands at different levels. If the judgment result is that the deviation does not exceed the second preset threshold, it indicates that the parameter drift is still in an early, slight stage, and a fine-tuning compensation mechanism is initiated. Based on classic negative feedback control laws, such as proportional-integral-derivative (PID) control algorithms, the parameter deviation value is used as the error input to calculate the required compensation amount. The input to this calculation process is the parameter deviation value, and the output is a precise compensation amount used to generate fine-tuning parameter harmonization commands. For example, if the resistance value of a certain feedback network increases by ΔR, the compensation will drive the digital potentiometer connected in parallel to it, causing its resistance value to decrease accordingly, thereby restoring the total resistance after parallel connection to the nominal value. The data format of this parameter harmonization command is a digital message containing the target component address and the target set value. Conversely, if the judgment result is that the deviation has exceeded the second preset threshold, it indicates that the component performance has been severely degraded, and simple parameter compensation can no longer guarantee performance or is about to fail. At this time, the optimal repair strategy will be selected from the preset strategy library by combining the estimated potential performance degradation path and the current operating load information obtained in real time from the bus. This strategy library is a rule-based decision table that stores response plans for different severe fault scenarios. For example, if the potential performance degradation path points to the rapid failure of the capacitor, and the current operating load is low, the redundancy switching strategy will be selected first; if the operating load is in a high-priority task stage, a degraded operation strategy may be selected, that is, by adjusting the operating frequency or processing accuracy, temporarily sacrificing some performance to ensure that the core function is not interrupted. After selecting a strategy, a composite parameter harmonization command is generated. The instruction is not a single adjustment value, but a sequence of multi-step actions, such as "1. Disconnect the main channel switch; 2. Close the backup channel switch; 3. Read the initial parameters of the new channel; 4. Restart the fine-tuning compensation loop", and is encapsulated into a transaction sequence and sent to the execution module.

[0063] Optionally, the parameter tuning instruction is executed, including: if the parameter tuning instruction is a fine-tuning instruction, driving the digital potentiometer or programmable gain amplifier in the adjustable component parameters to make continuous or step adjustments according to the compensation amount; if the parameter tuning instruction is a composite instruction, first controlling the analog switch to switch to the backup function circuit, then updating the idealized multiphysics coupling model based on the circuit state after switching, and regenerating the fine-tuning instruction for the new circuit.

[0064] Specifically, refer to Figure 5 As shown, a closed loop from decision-making to physical repair is completed through precise and controlled modification of circuit parameters. The control board employs a reconfigurable hardware architecture during the design phase, meaning adjustable components are pre-installed in critical signal paths or feedback networks. The execution process branches according to the type of parameter tuning instruction received. If the parameter tuning instruction is a fine-tuning instruction, the dynamic parameter tuner parses the instruction content, identifies the address code of the target adjustable component, and the compensation amount to be set. These adjustable components are pre-integrated on the circuit board, such as digital potentiometers controlled by I2C or SPI interfaces, programmable gain amplifiers, or tunable capacitor arrays. The dynamic parameter tuner, acting as the master device, sends write instructions to the target component via the corresponding communication bus, writing the compensation amount into its internal register. For example, for an 8-bit digital potentiometer, the compensation amount is converted into a digital value between 0 and 255, which determines its tap position, thereby enabling continuous or step-by-step adjustment of its resistance value, with an adjustment accuracy of up to 1 / 256 of its total resistance. If the parameter harmonization command is a composite command, the dynamic parameter harmonizer acts as a sequence controller, executing actions one by one according to the preset order in the command. For example, when executing a redundancy switching strategy, the harmonizer first sends control signals to a set of high-side or low-side driver chips to drive analog switches or MOSFET switch arrays, physically disconnecting the power or signal path of the main functional channel. After confirming that the main channel has been safely disconnected, the harmonizer sends signals to another set of driver chips to close the corresponding switch of the backup functional circuit, switching the circuit to the hardware-redundant backup channel. This switching process must follow the principle of disconnecting before connecting to avoid bus conflicts or short circuits. After the switching is completed, the dynamic parameter harmonizer immediately triggers a rapid probe and reconstruction process for the newly enabled channel, reading the initial parameter state of the new channel and confirming that it is in a healthy state. In the fourth step, based on the state of the new channel, the idealized multiphysics coupling model is automatically updated, the simulation object is switched to the topology of the backup channel, and then fine-tuning instructions for this new circuit are regenerated and executed to make it work at the optimal state point. This series of actions ensures a smooth and safe transition to backup resources in the event of a serious failure, and allows for immediate and meticulous management of these resources.

[0065] Optionally, the method further includes: recording real-time physical parameter maps before and after each execution of the parameter harmonization command to form harmonization effect samples; correcting the solution process of the physical inverse problem based on the harmonization effect samples to update the parameter mapping relationship in the initial forward model; and training a self-healing efficacy evaluation model using historical harmonization effect samples to optimize the parameters of the second threshold and the negative feedback control law.

[0066] Specifically, by enabling the health management of the control board to learn from its own experience, its diagnostic model and decision-making logic can self-improve over time, achieving continuous performance enhancement and a higher level of intelligence. After each parameter harmonization command is executed, the exploration and reconstruction process is immediately run again to obtain the real-time physical parameter map after harmonization. The command content of this harmonization action, the real-time physical parameter map before harmonization, and the real-time physical parameter map after harmonization are used as structured recording units, namely harmonization effect samples, and stored in the onboard non-volatile memory to form a continuously growing experience database. For example, during equipment maintenance or low-load periods, the solution process of the physical inverse problem is corrected. This means optimizing the initial forward model. The harmonization effect samples are analyzed, especially those cases where the compensation effect is not ideal or unexpected side effects occur. The actual field response data in these "erroneous" cases are compared with the model simulation data to identify the error sources in the model. Subsequently, by adjusting secondary parameters in the model, such as the nonlinear coefficients of materials or the parasitic parameter coupling model between components, i.e., updating the parameter mapping relationship in the initial forward model, the simulation results of the model can better fit all known "input and output" sample pairs. This process is equivalent to allowing the digital twin model to correct its understanding of the physical world through practical experience, making its predictions more accurate. Using all historical harmonization effect samples, a self-healing efficacy evaluation model is trained offline or online. This model can be a small neural network or a reinforcement learning agent. Specifically, a multilayer perceptron (MLP) neural network or a random forest regression model can be used. In this embodiment, a three-layer BP neural network is preferred. Its input layer receives a feature vector containing 'the parameter drift of the current component, the current operating load rate, and the ambient temperature'; the output layer outputs a normalized prediction value, representing the 'expected performance recovery' after applying a specific harmonization strategy. The training process adopts a supervised learning method, using 'fault-strategy' pairs in the historical database as input and the 'harmonization effect score' recorded after actual execution as label data, and iteratively updating the network weights through an error backpropagation algorithm. Through this specific predictive model, the system can dynamically optimize the second preset threshold. For example, if the model predicts that for a certain deviation, even if it does not exceed the current threshold, early, small compensation can bring significant long-term benefits, it will adaptively lower the trigger threshold for that type of deviation. Simultaneously, the model can optimize the parameters of the negative feedback control law, such as the proportional, integral, and derivative coefficients in the PID controller, enabling it to respond faster, more stably, and with less overshoot to different types and rates of parameter drift, thus achieving both prediction and continuous evolution.

[0067] Optionally, the probe excitation signal is injected into the circuit under test of the electronic control board, including: injecting the probe excitation signal during the idle time window of the main business logic operation of the electronic control board; or inserting and injecting the probe excitation signal when the transient characteristics of the working signal are detected to be stable.

[0068] Specifically, by injecting the probe stimulus signal into the circuit under test on the control board, the diagnostic operation and the main business function are decoupled in the time domain, ensuring that the diagnostic data with the highest signal-to-noise ratio is obtained without interfering with the normal operation of the circuit and without introducing additional noise. This objective is achieved through two intelligent injection strategies with optimal execution. The first strategy injects the signal within the idle time window of the main business logic operation of the control board, which relies on deep collaboration between the health management system and the main processor or operating system of the control board. The health management system monitors the task scheduler status of the main processor in real time through a preset interface. When the task scheduler enters an idle loop or issues a low-priority system heartbeat signal, indicating that there are no high-priority computation or I / O tasks in the next defined time period, such as 20 to 50 milliseconds, this state is identified as an idle time window. Once the health management system captures this window, it immediately triggers an active stimulus source to fully inject the probe stimulus signal into the circuit and complete the response acquisition, ensuring that the entire diagnostic process is completed within the "breathing gap" of the business logic. The second strategy inserts and injects the signal when the transient characteristics of the working signal tend to stabilize, which is a timing-based method based on the electrical state of the circuit. The health management system continuously utilizes high-speed sampling channels in its sensor network to monitor critical operating signals of the circuit under test in real time, such as the current waveform of the main power bus or the differential signal of the high-speed communication bus. Within a very short sliding time window, such as 100 microseconds, the short-term variance or peak value of the operating signal is continuously calculated. When the calculated variance value remains below a preset stability threshold for a continuous period, such as more than 500 microseconds (e.g., below 0.5% of its average value), the electromagnetic environment and load state of the circuit are considered to have entered a stable state. At this moment, a probe excitation signal is immediately inserted and injected. This method is particularly suitable for control boards whose main business logic runs continuously with few significant idle cycles. It can accurately capture the microsecond-level calm window during load switching intervals, avoiding measurements during moments of rapid current changes or high-speed data flipping, thereby improving the purity and repeatability of diagnostic data.

[0069] The present invention also provides an apparatus for implementing the above method, the apparatus comprising:

[0070] An active stimulus source is used to generate exploration stimulus signals;

[0071] A global field response sensor network, distributed on the electronic control board, is used to synchronously capture the global field response data of the circuit under test under the combined action of the detection excitation signal and the working signal.

[0072] A real-time physical feature reconstructor, connected to a global field response sensor network, is used to reconstruct the real-time physical parameter spectrum of the circuit under test based on global field response data and detection excitation signals.

[0073] The dynamic parameter tuner, connected to the real-time physical characteristic reconstructor, is used to generate and execute parameter tuning instructions based on the real-time physical parameter spectrum to dynamically adjust the parameters of adjustable components in the circuit under test.

[0074] In this embodiment, to verify the feasibility and technical effectiveness of the present invention in practical applications, the method and apparatus are applied to the I / O processing board of a certain type of UAV flight control computer. This I / O processing board is responsible for acquiring and processing analog signals from multiple attitude sensors and a barometric altimeter; its operational stability and signal integrity are crucial for flight safety. Due to severe temperature variations and vibrations in the working environment, passive components on the board (especially capacitors in the filter circuit) are prone to parameter drift, leading to signal distortion or interruption.

[0075] This embodiment aims to monitor and reconstruct the physical parameters of an MLCC (multilayer ceramic capacitor) C1 in the key filter circuit of the I / O processing board in real time, predict its performance degradation trend, and ensure the continuous and stable operation of the flight control system through self-healing function when its performance deteriorates severely.

[0076] The object under test is the main filter capacitor C1 in the π-type RC filter circuit connected to the output of a gyroscope sensor on the flight controller I / O board. Its design nominal value is... Based on the ODB++ design file of the I / O board, an idealized multiphysics coupling model was established using finite element analysis software as the initial forward model. The parameters of capacitor C1 in the model are... Set its nominal value to 100.0 nF.

[0077] The current of the main power rail of the I / O board is monitored in real time. When the variance of the current is less than 0.5% of its average value within a 500-microsecond sliding window, it is determined that the transient characteristics of the working signal are stable, and the injection process is started immediately.

[0078] An active excitation source generates a linear frequency modulated signal as the fundamental excitation signal. Its frequency linearly scans from 100kHz to 20MHz within 5 milliseconds. Subsequently, a 1023-bit Gold code pseudo-random sequence is retrieved from memory. The final detection excitation signal is generated through multiplication modulation. The signal is injected into the input of the filter circuit.

[0079] A miniature sensor array was deployed around the physical location C1, including a high-impedance probe for acquiring voltage signals, a miniature loop antenna for acquiring electromagnetic field distribution, and an infrared thermopile sensor for acquiring thermal radiation. At the instant the probe excitation signal was injected, all sensors were triggered by a unified clock source, synchronously acquiring data at a sampling rate of 500 MS / s. The acquired multimodal raw signals were time-aligned to nanosecond levels by performing correlation peak detection on the Gold code, and then fused and reconstructed into global field response data based on the sensor physical coordinates. .

[0080] The physical feature real-time reconstructor constructs the inversion optimization function. ,in Includes the capacitance value to be determined The Levenberg-Marquardt algorithm is used to analyze data from... The iterative solution process begins, which can also be executed via the FPGA hardware acceleration unit on the control board or offloaded to the connected edge computing gateway to ensure convergence within a 250-millisecond control cycle. The value is less than the convergence threshold of 10⁻ 6 When the iteration stops, it eventually converges. The value is the real-time physical parameter of C1.

[0081] Following this, fault prediction and self-healing process verification were conducted. During an accelerated aging test lasting 2000 flight hours, the above process was periodically executed, and capacitance values ​​were recorded. Real-time parameter graphs.

[0082] In the early stages of testing, The parameters fluctuated slightly around the nominal values. As the test progressed, the capacitance value showed a clear decreasing trend. This process was recorded and is shown in the table below.

[0083] Table 1. Performance degradation monitoring and lifespan prediction data for capacitor C1

[0084]

[0085] After 1500 hours of testing, the reconstructed capacitance value of C1 was 81.9 nF, and its relative deviation of -18.1% exceeded the first preset threshold (-15%) set for this type of component for the first time. This triggered a lifetime estimation procedure. By performing least-squares fitting on the historical data in Table 1, the parameter drift pattern was extracted, and the average drift rate was calculated. / hour. According to the component datasheet, its failure threshold... The capacitance has decreased by 30% (i.e., 70.0 nF). Based on this, the remaining effective lifespan is estimated as follows: The timeframe was calculated. Based on this, a preventative maintenance alert was generated, indicating that C1 was expected to enter a high-risk state approximately 661 flight hours later.

[0086] During testing, slight drift in other components was also detected. For example, at the 1850th hour of testing, the reconstructed value of a resistor R5 (nominal value 10.0kΩ) used to set the amplifier gain was 10.18kΩ, a deviation of +1.8%. This deviation did not exceed the second preset threshold (±2.0%) set for this resistor. Fine-tuning instructions were generated based on the PID control law.

[0087] Table 2 Data on the fine-tuning compensation process of resistor R5

[0088]

[0089] As shown in the table, the dynamic parameter tuner executes the instruction and controls an 8-bit digital potentiometer DP1 (total resistance 50kΩ) connected in parallel with R5 via the I2C bus, adjusting its setting value from the initial 255 (open circuit) to 204 (corresponding to a resistance value of approximately 40kΩ), so that the equivalent resistance after R5 and DP1 are connected in parallel is restored to 10.00kΩ, thus achieving precise compensation for the gain.

[0090] During the 2168th hour of testing, a routine check reconstructed the capacitance value of C1 to 68.5nF, with a deviation of -31.5%, exceeding its second preset threshold (-30%). This indicated severe performance degradation of the component. At this point, the flight control computer was consulted and determined that the current operating load was level flight cruise (a non-critical maneuver). The "redundancy switching strategy" was selected from the strategy library, and a composite parameter harmonization command was generated.

[0091] Table 3 Record of the self-healing process of capacitor C1 in case of severe fault

[0092]

[0093] As shown in Table 3, the dynamic parameter harmonizer completed the entire process from fault detection to switching to the hardware redundancy backup channel and restoring normal monitoring within 250 milliseconds. The entire self-healing operation did not cause any perceptible impact on the UAV's flight attitude, successfully avoiding a potential flight control risk caused by sensor signal filtering failure, and verifying the technical effect of the present invention in achieving "diagnosis and healing integration" without interrupting the main business functions.

[0094] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0095] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for predicting and self-healing faults in an electronic control board, characterized in that, The methods include: Generate exploration stimulus signal; The probe excitation signal is injected into the circuit under test on the electronic control board, and the global field response data of the circuit under test under the combined action of the probe excitation signal and the working signal is captured simultaneously. Based on the global field response data and the probe excitation signal, the real-time physical parameter spectrum of the circuit under test is reconstructed by solving the physical inverse problem; Based on the deviation between the parameter values ​​of each node in the real-time physical parameter map and the corresponding nominal values, generate corresponding parameter harmonization instructions; Execute parameter tuning instructions to dynamically adjust the adjustable component parameters of the corresponding node in the circuit under test, so that the electrical performance of the node approaches the nominal value; The process of reconstructing the real-time physical parameter map of the circuit under test by solving a physical inverse problem based on global field response data and probe excitation signal includes: establishing an idealized multi-physics coupling model of the circuit under test as an initial forward model; inputting the probe excitation signal into the initial forward model to calculate theoretical field response data; constructing an inversion optimization function with the objective of minimizing the difference between the theoretical field response data and the global field response data; solving the inversion optimization function, iteratively updating the component parameter set in the initial forward model until the difference satisfies a convergence condition, and expressing the final component parameter set as the real-time physical parameter map.

2. The method for predicting and self-healing faults in an electronic control board according to claim 1, characterized in that, The generation of the probe excitation signal includes: determining at least one key signal injection node based on the topology of the circuit under test; generating a scanning sequence covering a frequency range as a basic excitation signal; and modulating the basic excitation signal according to encoding rules to generate the probe excitation signal with specific spectral and temporal characteristics.

3. A method for predicting and self-healing faults in an electronic control board according to claim 2, characterized in that, The method of synchronously capturing the global field response data of the circuit under test includes: synchronously acquiring electrical signals, electromagnetic field distribution signals, and thermal radiation signals through multiple miniature sensing units arranged at key locations on the circuit board under test; performing synchronization and time-frequency alignment processing on the acquired multimodal raw signals, and fusing them to generate the global field response data.

4. A method for predicting and self-healing faults in an electronic control board according to claim 1, characterized in that, The method further includes: identifying abnormal nodes whose parameter drift exceeds a first threshold from the real-time physical parameter map; extracting the parameter drift pattern of the abnormal nodes, the parameter drift pattern including drift direction, drift rate and correlation degree; mapping potential performance degradation paths according to the parameter drift pattern, and estimating the remaining effective lifetime.

5. A method for predicting and self-healing faults in an electronic control board according to claim 4, characterized in that, The step of generating corresponding parameter harmonization instructions based on the deviation between the parameter values ​​of each node in the real-time physical parameter map and the corresponding nominal values ​​includes: determining whether the deviation exceeds a second threshold. If the error is not exceeded, the compensation amount is calculated based on the negative feedback control law, and the parameter harmonization instruction for fine-tuning the parameters of the adjustable element is generated; if the error is exceeded, the potential performance degradation path and the current operating load information are combined, and a degraded operation strategy or a redundancy switching strategy is selected from the strategy library to generate a composite parameter harmonization instruction containing a sequence of actions.

6. A method for predicting and self-healing faults in an electronic control board according to claim 5, characterized in that, The execution parameter harmonization instruction includes: if the parameter harmonization instruction is a fine-tuning instruction, then driving the digital potentiometer or programmable gain amplifier in the adjustable element parameters to make continuous or step-by-step adjustments according to the compensation amount; if the parameter harmonization instruction is a composite instruction, then first controlling the analog switch to switch to the backup function circuit, then updating the idealized multiphysics coupling model based on the circuit state after switching, and regenerating the fine-tuning instruction for the new circuit.

7. A method for predicting and self-healing faults in an electronic control board according to claim 5, characterized in that, The method further includes: recording the real-time physical parameter spectrum before and after each execution of the parameter harmonization command to form a harmonization effect sample; correcting the solution process of the physical inverse problem based on the harmonization effect sample to update the parameter mapping relationship in the initial forward model; and training a self-healing efficacy evaluation model using historical harmonization effect samples to optimize the parameters of the second threshold and the negative feedback control law.

8. A method for predicting and self-healing faults in an electronic control board according to claim 1, characterized in that, The step of injecting the probe excitation signal into the circuit under test of the electronic control board includes: injecting the probe excitation signal during the idle time window of the main business logic operation of the electronic control board; or inserting and injecting the probe excitation signal when the transient characteristics of the working signal are detected to be stable.

9. A fault prediction and self-healing device for an electronic control board, characterized in that, The apparatus for implementing the method of any one of claims 1 to 8 comprises: An active stimulus source is used to generate exploration stimulus signals; A global field response sensor network, distributed on the electronic control board, is used to synchronously capture the global field response data of the circuit under test under the combined action of the detection excitation signal and the working signal; A real-time physical feature reconstructor, connected to the global field response sensor network, is used to reconstruct the real-time physical parameter spectrum of the circuit under test based on the global field response data and the detection excitation signal. A dynamic parameter tuner, connected to the real-time physical feature reconstructor, is used to generate and execute parameter tuning instructions based on the real-time physical parameter spectrum to dynamically adjust the parameters of adjustable components in the circuit under test.

Images