Automatic driving circuit board online diagnosis method and system based on digital twinning

Abstract

The application discloses an automatic driving circuit board online diagnosis method and system based on digital twinning, and relates to the technical field of automatic driving. Firstly, high-frequency voltage signals containing background noise and potential fault information on the circuit board are acquired in real time. Then, the signal is deeply time-frequency transformed by introducing the cyclostationary theory, and the characteristic separation of the signal components is realized in the cycle domain through spectral correlation density estimation. Based on this, a cyclostationary purification algorithm is designed to filter out deterministic periodic noise, thereby effectively extracting a non-stationary residual signal. Furthermore, the energy aggregation and quantization of the purified residual spectrum are performed to construct a feature score that can stably represent the severity of weak faults. Finally, an adaptive threshold decision mechanism is established to evaluate the score in real time, and an accurate fault risk level is output. In this way, the strong electromagnetic noise can be effectively penetrated, and high-sensitivity online diagnosis of early weak faults of the circuit board can be realized.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and more specifically, to an online diagnostic method and system for autonomous driving circuit boards based on digital twins. Background Technology

[0002] As the automotive industry deeply integrates with electrification, intelligence, and connectivity, autonomous driving technology has become a strategic high ground in global technological competition. As the core hardware carrier of autonomous driving systems, printed circuit boards (PCBs) integrate a large number of precision electronic components. Their operational stability and reliability directly affect the vehicle's decision-making, control, and execution capabilities, forming the cornerstone of driving safety. However, the complex electromagnetic environment of autonomous vehicles, especially electric vehicles, poses a severe challenge to the reliability of PCBs. High-power power electronic devices such as inverters and drive motors generate strong electromagnetic interference during operation, and this high-intensity electrical noise permeates the entire vehicle's electrical network. This strong noise environment easily masks subtle impedance changes in the circuit board caused by early degradation or microcracks, making it difficult for traditional diagnostic methods to effectively detect potential faults, thus posing a serious safety hazard. Therefore, to ensure the ultimate reliability of autonomous vehicles, there is an urgent need for an advanced diagnostic solution that can penetrate strong noise backgrounds and perform highly sensitive online monitoring and fault warning of the circuit board's health status.

[0003] Against this backdrop, digital twin technology, with its ability to connect the physical and digital worlds and achieve full lifecycle state mapping and predictive maintenance, offers a new approach to addressing the aforementioned challenges. By constructing a high-fidelity digital twin model of an autonomous driving circuit board, it is theoretically possible to simulate and analyze its electrical behavior under complex operating conditions in real time, and compare it with monitoring data from the physical entity to identify potential anomalies. However, existing digital twin-based diagnostic solutions still face bottlenecks in practical applications. Many solutions rely primarily on twin mapping and deviation analysis of macroscopic physical parameters (such as temperature, average voltage, and current). While these methods can capture relatively significant faults, their diagnostic sensitivity and accuracy are severely insufficient for weak signal changes masked by strong electromagnetic noise. These methods fail to delve into the inherent time-frequency structure characteristics of signals; when the energy of weak fault characteristic signals is far lower than the background noise, diagnostic logic based on macroscopic parameter deviations is prone to false alarms or missed alarms.

[0004] Therefore, there is an urgent need for an optimized online diagnostic method and system for autonomous driving circuit boards based on digital twins. Summary of the Invention

[0005] This application is made in order to solve the above-mentioned technical problems.

[0006] According to one aspect of this application, an online diagnostic method for autonomous driving circuit boards based on digital twins is provided, comprising:

[0007] Acquire high-frequency voltage signals collected by the power management chip;

[0008] High-fidelity signal segmentation is performed on the high-frequency voltage signal to obtain a signal frame sequence;

[0009] Spectral correlation density estimation is performed on each signal frame in the signal frame sequence to obtain a signal spectral correlation density map sequence;

[0010] Cyclic stationary purification and non-stationary residual extraction are performed on each signal spectrum correlation density map in the signal spectrum correlation density map sequence to obtain a purified non-stationary residual spectrum sequence.

[0011] The residual energy aggregation and fault feature quantization of the purified non-stationary residual spectrum sequence are performed to obtain the quantized fault score.

[0012] An adaptive threshold decision is made on the quantified fault score to obtain the fault risk level;

[0013] This displays the fault risk level.

[0014] According to another aspect of this application, an online diagnostic system for autonomous driving circuit boards based on digital twins is provided, comprising:

[0015] The high-frequency signal acquisition module is used to acquire the high-frequency voltage signal acquired by the power management chip;

[0016] The high-fidelity signal segmentation module is used to perform high-fidelity signal segmentation on high-frequency voltage signals to obtain signal frame sequences.

[0017] The spectral correlation density estimation module is used to estimate the spectral correlation density of each signal frame in the signal frame sequence to obtain a sequence of signal spectral correlation density maps.

[0018] The cyclic purification residual extraction module is used to perform cyclic stationary purification and non-stationary residual extraction on each signal spectrum correlation density map in the signal spectrum correlation density map sequence to obtain a purified non-stationary residual spectrum sequence.

[0019] The residual energy aggregation and quantization module is used to perform residual energy aggregation and fault feature quantization on the purified non-stationary residual spectrum sequence to obtain the quantized fault score.

[0020] The adaptive threshold decision module is used to make adaptive threshold decisions on the quantified fault scores to obtain the fault risk level.

[0021] The fault risk level display module is used to display the fault risk level.

[0022] Compared with existing technologies, this application provides an online diagnostic method and system for autonomous driving circuit boards based on digital twins. First, it acquires high-frequency voltage signals on the circuit board containing background noise and potential fault information in real time. Then, it introduces cyclostationary theory to perform deep time-frequency transformation on the signal, and achieves feature separation of signal components in the cyclic domain through spectral correlation density estimation. Based on this, a cyclostationary purification algorithm is designed to filter out deterministic periodic noise, thereby effectively extracting non-stationary residual signals. Furthermore, the purified residual spectrum is subjected to energy aggregation and quantization to construct a feature score that stably characterizes the severity of minor faults. Finally, an adaptive threshold decision mechanism is established to evaluate this score in real time and output an accurate fault risk level. This effectively penetrates the masking of strong electromagnetic noise, thereby achieving highly sensitive online diagnosis of early-stage minor faults on the circuit board. Attached Figure Description

[0023] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0024] Figure 1 This is a flowchart of an online diagnostic method for autonomous driving circuit boards based on digital twins, according to an embodiment of this application.

[0025] Figure 2 This is a data flow diagram of an online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application.

[0026] Figure 3 This is a flowchart of sub-step S2 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application.

[0027] Figure 4 This is a flowchart of sub-step S3 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application.

[0028] Figure 5 This is a flowchart of sub-step S4 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application.

[0029] Figure 6 This is a flowchart of sub-step S41 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application.

[0030] Figure 7This is a flowchart of sub-step S5 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application.

[0031] Figure 8 This is a block diagram of an online diagnostic system for autonomous driving circuit boards based on digital twins, according to an embodiment of this application. Detailed Implementation

[0032] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0033] To address the problems mentioned above, this application proposes an online diagnostic method for autonomous driving circuit boards based on digital twins. Figure 1 This is a flowchart of an online diagnostic method for autonomous driving circuit boards based on digital twins, according to an embodiment of this application. Figure 2 This is a data flow diagram of an online diagnostic method for autonomous driving circuit boards based on digital twins, according to an embodiment of this application. Figure 1 and Figure 2 As shown, the online diagnostic method for autonomous driving circuit boards based on digital twins includes the following steps: S1, acquiring a high-frequency voltage signal collected by a power management chip; S2, performing high-fidelity signal segmentation on the high-frequency voltage signal to obtain a signal frame sequence; S3, estimating the spectral correlation density of each signal frame in the signal frame sequence to obtain a signal spectral correlation density map sequence; S4, performing cyclic stationary purification and non-stationary residual extraction on each signal spectral correlation density map in the signal spectral correlation density map sequence to obtain a purified non-stationary residual spectrum sequence; S5, performing residual energy aggregation and fault feature quantization on the purified non-stationary residual spectrum sequence to obtain a quantized fault score; S6, performing adaptive threshold decision on the quantized fault score to obtain a fault risk level; S7, displaying the fault risk level.

[0034] In the aforementioned online diagnostic method for autonomous driving circuit boards based on digital twins, step S1 involves acquiring a high-frequency voltage signal collected by a power management chip. It should be understood that since the core components of an autonomous driving circuit board rely on a stable power supply, faults are manifested through voltage fluctuations. High-frequency voltage signals can capture minute abnormal changes, and conventional acquisition methods are insufficient to meet the accuracy requirements of online diagnostics. Therefore, this application acquires high-frequency voltage signals through a power management chip to provide high-fidelity raw data support for subsequent fault diagnosis. This allows for the accurate capture of subtle fault characteristics in the circuit board's power system, laying a reliable foundation for subsequent diagnostic analysis and ensuring the safety and stability of the autonomous driving system.

[0035] Specifically, in one possible embodiment, step S1 is implemented as follows: First, the acquisition parameters of the power management chip in the autonomous driving circuit board are determined, including the sampling rate, input voltage range, and number of channels, to ensure that the parameters match the operating characteristics of the circuit board. Then, the analog voltage signal is converted into a digital signal through the chip's built-in analog-to-digital converter module. Simultaneously, signal conditioning circuitry is used to amplify, filter, and isolate the signal, eliminating external electromagnetic interference. Finally, the converted high-frequency voltage signal is transmitted in real-time to the data storage unit through a high-speed transmission interface, achieving complete signal preservation and providing high-quality raw data for subsequent processing.

[0036] In the aforementioned online diagnostic method for autonomous driving circuit boards based on digital twins, step S2 involves high-fidelity segmentation of the high-frequency voltage signal to obtain a signal frame sequence. It should be understood that, due to the continuity and redundancy of the high-frequency voltage signal, direct overall processing increases the computational load and can easily lead to fault characteristics being masked by redundant information, affecting diagnostic efficiency and accuracy. Therefore, this application further implements high-fidelity segmentation processing on the high-frequency voltage signal to focus on effective signal segments, preserving fault characteristics while reducing data processing complexity. This allows for precise separation of effective information and redundant parts in the signal, ensuring that fault characteristics are not destroyed, improving the efficiency of subsequent signal analysis, and providing strong support for rapid fault location on the circuit board.

[0037] In particular, in one specific embodiment, Figure 3 This is a flowchart of sub-step S2 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application. Figure 3 As shown, step S2 includes: S21, overlapping and framing the high-frequency voltage signal to obtain the original signal frame sequence; S22, performing intra-frame signal baseline correction on each original signal frame in the original signal frame sequence to obtain the signal frame sequence.

[0038] Specifically, in step S21, the high-frequency voltage signal is subjected to overlapping framing to obtain the original signal frame sequence. It should be understood that since fault features in high-frequency voltage signals may span multiple continuous signal segments, conventional non-overlapping framing easily leads to feature breaks, failing to fully preserve the fault evolution process and affecting the comprehensiveness of the diagnosis. Therefore, this application further employs overlapping framing to process the high-frequency voltage signal, thereby ensuring the continuity of signal features and avoiding the loss of key fault information. This allows adjacent frame signals to maintain effective connection, completely capturing the evolution trajectory of fault features, providing comprehensive raw data for subsequent intra-frame analysis, and improving the accuracy and completeness of fault diagnosis.

[0039] Specifically, in one possible embodiment, step S21 is implemented as follows: First, a fixed frame length and overlap ratio are determined based on the frequency characteristics and fault characteristic period of the high-frequency voltage signal. Then, a sliding window is set, and the signal is segmented step-by-step according to the overlap ratio based on the set frame length, ensuring that there is a preset overlap ratio between adjacent frames. During the framing process, the start and end positions of each window are recorded in real time to avoid signal omission or duplication. Finally, each segmented signal is used as an original signal frame, arranged in chronological order to form an original signal frame sequence, and the relevant parameter information of each frame is stored to prepare for subsequent baseline correction.

[0040] Specifically, step S22 involves performing intra-frame signal baseline correction on each original signal frame in the original signal frame sequence to obtain the signal frame sequence. It should be understood that the original signal frames are susceptible to circuit noise, environmental interference, and equipment drift during acquisition and transmission, causing the baseline to deviate from the normal level, masking the true fault characteristics, and reducing the accuracy of subsequent feature extraction. Therefore, this application further performs intra-frame signal baseline correction on each frame in the original signal frame sequence to eliminate baseline drift and interference effects, restoring the true characteristics of the signal. This ensures that the corrected signal baseline remains stable, highlighting the difference between fault characteristics and normal signals, improving the accuracy of fault feature identification, and providing high-quality signal data for reliable diagnosis of autonomous driving circuit boards.

[0041] Specifically, in one possible embodiment, step S22 is implemented as follows: First, all sample point data within each original signal frame are extracted. The arithmetic mean of the frame signal is calculated by summing the data and dividing by the total number of sample points. This arithmetic mean represents the DC bias component in the current signal frame. Then, a centering operation is performed, that is, the arithmetic mean is subtracted from the value of each sample point in the original signal frame. This mean-reduction process shifts the signal baseline level to zero, effectively eliminating low-frequency DC interference introduced by slow power supply voltage drift or fluctuations in the reference voltage of the acquisition device. The resulting corrected signal frame will no longer contain a DC component, thus preventing high-energy peaks from forming at zero frequency in the subsequent Fourier transform spectrum, preventing them from masking the spectral analysis of weak fault characteristic signals, and ensuring the accuracy of subsequent spectral correlation density estimation.

[0042] In the aforementioned online diagnostic method for autonomous driving circuit boards based on digital twins, step S3 involves estimating the spectral correlation density of each signal frame in the signal frame sequence to obtain a sequence of signal spectral correlation density maps. It should be understood that since fault features in the signal frame sequence are often masked by strong electromagnetic noise, conventional frequency domain analysis can only reflect single-frequency information and cannot distinguish the correlation between periodic noise and non-stationary fault signals, leading to difficulties in fault feature extraction. Therefore, this application further performs spectral correlation density estimation on each signal frame to construct a two-dimensional spectral domain space, revealing the correlation characteristics of different frequency components. This allows for accurate separation of periodic noise and non-stationary fault signals, forming a spectral correlation density map sequence that can intuitively identify fault features, providing a clear analytical basis for subsequent cyclical and stable purification, and improving the accuracy of autonomous driving circuit board fault diagnosis.

[0043] In particular, in one specific embodiment, Figure 4 This is a flowchart of sub-step S3 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application. Figure 4 As shown, step S3 includes: S31, performing time-frequency decomposition on the signal frame to obtain the complex spectrum of the signal frame; S32, performing frequency axis translation and conjugate spectral product calculation on the complex spectrum of the signal frame to obtain a set of spectral product sequences; S33, performing time smoothing and matrix construction on the set of spectral product sequences to obtain a signal spectral correlation density map.

[0044] Specifically, step S31 involves performing time-frequency decomposition on the signal frame to obtain its complex spectrum. It should be understood that since a signal frame can only reflect the change in voltage amplitude over time in the time domain, it cannot reflect the distribution patterns of different frequency components. Fault characteristics are often hidden within specific frequency ranges, making them difficult to capture in time-domain analysis. Therefore, this application further performs time-frequency decomposition on the signal frame to map the time-domain signal to the frequency domain while preserving the signal's amplitude and phase information. This clearly shows the energy distribution of each frequency component in the signal frame, accurately locates the frequency range where fault characteristics may exist, and provides basic frequency domain data for subsequent frequency axis translation and conjugate spectral product calculations, ensuring the accuracy of spectral correlation density estimation.

[0045] Specifically, in one possible embodiment, step S31 is implemented as follows: First, based on the sampling rate of the signal frame and the frequency range of the fault characteristics, the core parameters of time-frequency decomposition are determined, including the number of Fourier transform points, the window function type, and the window length. Then, a Hamming window is used to window the signal frame to suppress spectral leakage; the window length setting must ensure complete coverage of the minimum fault characteristic period. Next, a Fast Fourier Transform (FFT) is performed on the windowed signal frame to convert the discrete-time signal into a complex-frequency signal, where the real part of the complex number corresponds to the amplitude of the frequency components, and the imaginary part corresponds to the phase. Finally, the transform result is calibrated on the frequency axis, negative frequency components are removed, and the frequency scale is adjusted to generate a complex spectrum of the signal frame containing amplitude and phase information, while simultaneously verifying whether the spectral resolution meets the requirements for fault feature capture.

[0046] Specifically, step S32 involves performing frequency axis shifting and conjugate spectral product calculation on the complex spectrum of the signal frame to obtain a set of spectral product sequences. It should be understood that since the complex spectrum of the signal frame only reflects the energy information of a single frequency and cannot reflect the correlation between different frequency components, and the core of spectral correlation density is frequency coupling characteristics, the lack of this correlation information will prevent subsequent estimation. Therefore, this application further performs frequency axis shifting and conjugate spectral product calculation on the complex spectrum to construct the correlation between different frequency points. This allows for the extraction of frequency coupling characteristic information from the complex spectrum, forming a spectral product sequence corresponding to the cyclic frequency, providing crucial data support for subsequent time smoothing, and ensuring that the spectral correlation density estimation accurately reflects the difference between periodic noise and fault signals.

[0047] Specifically, in one possible embodiment, step S32 is implemented as follows: First, a set of cyclic frequencies to be calculated is determined, which must cover the fundamental frequency and harmonics of the main periodic noise in the autonomous driving circuit board. Then, for each cyclic frequency, the complex spectrum of the signal frame is shifted forward and backward along the frequency axis, respectively. During the shifting process, a cyclic shift algorithm is used to ensure the continuity of the frequency axis and avoid boundary truncation errors. Next, the conjugate of the complex spectrum after the backward shift is taken to preserve phase correlation information. Then, the forward-shifted spectrum and the conjugate backward-shifted spectrum are multiplied point-by-point to obtain the spectral product sequence corresponding to that cyclic frequency. Finally, the spectral product sequences of all cyclic frequencies are arranged in cyclic frequency order to form a set of spectral product sequences. Simultaneously, a consistency check is performed to ensure that the frequency axes of each sequence are aligned to avoid calculation deviations.

[0048] Specifically, step S33 involves time smoothing and matrix construction of the spectral product sequence set to obtain a signal spectral correlation density map. It should be understood that due to random noise interference in the spectral product sequence set, direct analysis would lead to unstable spectral correlation features, making it impossible to clearly distinguish between valid signals and noise, thus affecting the reliability of subsequent fault identification. Therefore, this application further performs time smoothing and matrix construction on the spectral product sequence set to reduce the impact of random noise and transform the discrete sequences into structured spectra. This results in a stable and clear signal spectral correlation density map, highlighting the spectral correlation features of periodic noise as ridges and presenting the features of non-stationary fault signals as discrete energy clusters, providing a clear target for subsequent cyclical and stable purification.

[0049] Specifically, in one possible embodiment, step S33 is implemented as follows: First, a first-order infinite impulse response (IIR) filter is selected as the time smoothing tool, and a smoothing factor is determined, with a value ranging from 0.01 to 0.1. The factor size needs to balance noise suppression effect and signal response speed. Then, for each sequence in the spectral product sequence set, smoothing calculation is performed point-by-point along the time dimension, and random fluctuations are eliminated by a weighted average of the current point and historical smoothing results. Next, a two-dimensional matrix framework is constructed based on the cyclic frequency set and the spectral frequency range, where the row index of the matrix corresponds to the cyclic frequency and the column index corresponds to the spectral frequency. Then, the smoothed spectral product sequences are filled into the corresponding rows of the matrix in cyclic frequency order, ensuring that the frequency point of each sequence precisely matches the matrix column index. Finally, the matrix elements are normalized to map the numerical range to the 0-1 interval, generating an intuitive signal spectrum correlation density map. The visibility of features is enhanced through grayscale mapping or pseudo-color rendering, ensuring the distinguishability between periodic noise and fault features.

[0050] In the aforementioned online diagnostic method for autonomous driving circuit boards based on digital twins, step S4 involves performing cyclic stationary purification and non-stationary residual extraction on each signal spectrum correlation density map sequence to obtain a purified non-stationary residual spectrum sequence. It should be understood that since the signal spectrum correlation density maps still contain a large amount of periodic noise generated by normal components of the autonomous driving circuit board (such as inverters and clock modules), this noise, existing in the form of structured ridges, can mask the discrete energy characteristics of non-stationary fault signals, making it impossible to effectively identify fault residuals. Therefore, this application further performs cyclic stationary purification and non-stationary residual extraction on each signal spectrum correlation density map to accurately shield deterministic periodic noise and separate and retain non-stationary fault-related residual signals. This completely eliminates the interference of periodic noise on fault characteristics, obtaining a residual spectrum sequence containing only non-stationary fault signals and a small amount of random noise, providing a high-purity analytical object for subsequent residual energy aggregation and fault quantification, significantly improving the sensitivity of fault diagnosis.

[0051] In particular, in one specific embodiment, Figure 5 This is a flowchart of sub-step S4 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application. Figure 5 As shown, step S4 includes: S41, generating a cleaned mask matrix based on the signal spectrum correlation density map; S42, applying a mask and extracting non-stationary residual spectra from the signal spectrum correlation density map based on the cleaned mask matrix to obtain the cleaned non-stationary residual spectrum.

[0052] Specifically, step S41 generates a cleanup mask matrix based on the signal spectrum correlation density map. It should be understood that since periodic noise (such as inverter switching frequency and SoC clock harmonics) in the signal spectrum correlation density map is distributed in a ridge-like pattern, its position may drift slightly due to fluctuations in the operating conditions of the autonomous driving circuit board (such as Dynamic Voltage Frequency Adjustment (DVFS)). Directly shielding based on a fixed frequency range could easily lead to the accidental deletion of fault signals or the omission of noise. Therefore, this application further generates a cleanup mask matrix based on the actual frequency distribution of the signal spectrum correlation density map to accurately match the noise location under the current operating conditions and clearly mark the areas that need to be shielded. This provides a spatial positioning basis that perfectly matches the actual noise distribution for subsequent mask applications, avoiding damage to non-stationary fault residuals or the omission of periodic noise during the cleanup process, and ensuring the purity and accuracy of non-stationary residual extraction.

[0053] In particular, in one specific embodiment, Figure 6 This is a flowchart of sub-step S41 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application. Figure 6As shown, step S41 includes: S411, obtaining a calibrated healthy cycling frequency set; S412, initializing the mask matrix based on the signal spectrum correlation density map to obtain an initial mask matrix; S413, applying a masking neighborhood and generating a final mask based on the calibrated healthy cycling frequency set to obtain a cleaned mask matrix.

[0054] More specifically, step S411 involves obtaining the calibrated set of healthy cycle frequencies. It should be understood that the preset set of healthy cycle frequencies is the nominal value calibrated at the factory of the circuit board. However, during autonomous driving, the operating frequency of core components of the circuit board (such as PLL phase-locked loops and power modules) dynamically adjusts with operating conditions (such as load changes and energy-saving mode switching), causing the actual healthy noise cycle frequency to deviate from the nominal value. Directly using the nominal set would lead to inaccurate subsequent mask marking. Therefore, this application further calibrates the healthy cycle frequency set in conjunction with real-time operating conditions to obtain healthy frequency data that perfectly matches the current circuit board operating state. This ensures that the subsequently generated purification mask matrix accurately points to the periodic noise under the current operating conditions, avoiding incomplete noise shielding or false signal shielding due to frequency drift, and providing an accurate frequency reference for stable cycle purification.

[0055] Specifically, in one possible embodiment, step S411 is implemented as follows: First, real-time operating parameters of the autonomous driving circuit board are obtained from the vehicle controller local area network (CAN), including the SoC real-time clock frequency, power module output voltage level, and inverter operating mode. Then, a preset set of nominal healthy cycle frequencies is read from non-volatile memory. This set contains the nominal frequencies of each noise source (clock, inverter) and their association rules with the operating parameters, such as clock division by 4 and the mapping relationship between inverter switching frequency and voltage level. Next, based on the association rules and using the real-time operating parameters as a reference, the nominal frequencies are dynamically calculated. For example, if the nominal clock division frequency is "clock / 4", the current clock frequency is divided by 4 to obtain the actual healthy frequency. Then, all calculated actual healthy frequencies are organized into a temporary set, and outliers are removed by comparing them with frequency data under historical healthy operating conditions. Finally, the verified temporary set is determined as the calibrated set of healthy cycle frequencies and stored in a temporary cache unit for subsequent mask generation.

[0056] More specifically, step S412 involves initializing the mask matrix based on the signal spectrum correlation density map to obtain an initial mask matrix. It should be understood that since the cleanup mask matrix must perfectly match the signal spectrum correlation density map in terms of dimension to achieve accurate element-wise calculations, a mismatch between the mask matrix dimension and the spectrum will lead to calculation errors or the inability to clean certain regions. Therefore, this application further initializes the mask matrix based on the dimension of the signal spectrum correlation density map to construct a mask framework that is completely consistent with the spatial structure of the spectrum. This ensures that the subsequent marking of healthy cyclic frequency regions accurately corresponds to their actual positions in the spectrum, avoiding cleanup misalignment due to dimensional deviations. It provides a structurally complete and dimensionally adapted base matrix for subsequent shielding neighborhood applications, ensuring the usability of the final cleanup mask.

[0057] Specifically, in one possible embodiment, step S412 is implemented as follows: First, the metadata of the signal spectrum correlation density map is read through the data interface, and the number of discrete points on the cyclic frequency axis (denoted as p) and the number of discrete points on the spectral frequency axis (denoted as q) are extracted to determine the size of the initial mask matrix as p×q. Then, storage space of the corresponding size is allocated in the memory of the on-board computing unit, and a matrix is ​​created using a two-dimensional array structure. Next, all elements in the matrix are initialized, with each element uniformly set to 1. This value represents that the signal is allowed to pass through in the initial state, and only noise areas that need to be shielded are marked with a 0 value in subsequent steps. Then, the initialized matrix is ​​dimension-verified. By comparing the matrix dimensions with those of the signal spectrum correlation density map, it is confirmed that the number of rows and columns are completely consistent. Finally, the initial mask matrix is ​​stored in a temporary data area, and its creation timestamp and the corresponding signal spectrum correlation density map number are recorded to ensure the traceability of the two.

[0058] More specifically, step S413 involves applying a shielding neighborhood to the initial mask matrix and generating a final mask based on the calibrated healthy cyclic frequency set to obtain a cleaned mask matrix. It should be understood that since the calibrated healthy cyclic frequencies are single-point frequency values, and actual periodic noise is affected by parasitic parameters of the circuit board (such as capacitance and inductance), it forms a certain bandwidth of energy distribution (frequency drift) around the single-point frequency. Simply shielding the single-point frequency cannot completely eliminate the noise. Therefore, this application further applies a shielding neighborhood to the initial mask matrix to expand the shielding range and cover the frequency drift interval of the noise. This completely shields all the energy of the periodic noise, avoiding noise residue caused by frequency drift, while ensuring that signals in non-noise areas pass through completely, providing a noise-free basis for the extraction of non-stationary residuals after subsequent mask application.

[0059] Specifically, in one possible embodiment, step S413 is implemented as follows: First, the actual healthy frequency value corresponding to each noise source is extracted from the calibrated healthy cyclic frequency set. Combined with the cyclic frequency resolution of the signal spectrum correlation density map, the discrete row index corresponding to each frequency value is calculated. Then, based on the circuit board design parameters, such as the frequency drift bandwidth caused by parasitic parameters, the shielding neighborhood range corresponding to each row index is determined, such as three row indices before and after. Next, the initial mask matrix is ​​traversed, and all column elements (covering the full spectrum frequency) within the row index corresponding to each healthy frequency and its neighborhood range are modified from 1 to 0. Then, edge repair is performed on the modified matrix, using a linear interpolation algorithm to process the element values ​​in the overlapping neighborhood areas to avoid energy abrupt changes. Finally, the processed matrix is ​​superimposed on the signal spectrum correlation density map for display. Through visual verification and energy statistical analysis, it is confirmed that the ridge region corresponding to noise has been completely marked with a 0 value, while the non-noise region retains a 1 value, ultimately generating a clean mask matrix.

[0060] Specifically, in step S42, based on the purification mask matrix, a mask is applied to the signal spectrum correlation density map, and non-stationary residual spectrum is extracted to obtain the purified non-stationary residual spectrum. It should be understood that since the purification mask matrix has accurately marked the periodic noise region, noise shielding can be achieved solely through matrix operations. Without mask application, periodic noise would still remain in the spectrum correlation density map, preventing the non-stationary fault residual from being extracted separately. Therefore, this application further performs operations between the purification mask matrix and the signal spectrum correlation density map to physically shield the periodic noise energy, separating and extracting the non-stationary residual signal. This directly eliminates the interference of periodic noise, allowing the non-stationary fault residual to be clearly presented as discrete energy clusters, resulting in a pure residual spectrum. This provides interference-free analytical data for subsequent residual energy aggregation and fault feature quantification, ensuring the accuracy of fault score calculation.

[0061] Specifically, in one possible embodiment, step S42 is implemented as follows: First, a dimensionality consistency check is performed based on the cleaned mask matrix and the signal spectral correlation density map to ensure that the number of rows (cyclic frequencies) and columns (spectral frequencies) are completely matched. Then, element-wise matrix multiplication is performed to zero the energy in regions with a mask value of 0 (periodic noise regions) in the spectral correlation density map, while retaining the energy in regions with a mask value of 1 (potential residual regions). Next, outlier processing is performed on the calculation results, removing isolated high-energy points in the residual regions caused by calculation errors using the 3σ criterion to avoid interfering with subsequent analysis. Then, the processed matrix is ​​normalized to map the numerical range to the 0-1 interval, enhancing the visibility of residual features. Finally, the normalized matrix is ​​trimmed to remove meaningless edge regions at both ends of the frequency axis, generating the final cleaned non-stationary residual spectrum. Simultaneously, the energy distribution statistics of the residual spectrum, such as mean and variance, are recorded to provide a reference benchmark for subsequent fault quantification.

[0062] In the aforementioned online diagnostic method for autonomous driving circuit boards based on digital twins, step S5 involves performing residual energy aggregation and fault feature quantification on the purified non-stationary residual spectrum sequence to obtain a quantified fault score. It should be understood that, because fault features are dispersed as discrete energy clusters across various frames in the purified non-stationary residual spectrum sequence, the energy distribution of a single frame's spectrum cannot intuitively reflect the overall severity of the fault, and lacks a unified measurement standard, making it difficult to directly use for fault judgment. Therefore, this application further performs residual energy aggregation and fault feature quantification on the residual spectrum sequence to integrate multi-frame residual energy information, transforming the dispersed fault features into a single, comparable quantitative indicator. This eliminates the random interference of single-frame data, accurately characterizes the cumulative degree and severity level of the fault, provides a clear and unified judgment basis for subsequent adaptive threshold decisions, and ensures the objectivity and accuracy of fault risk assessment.

[0063] In particular, in one specific embodiment, Figure 7 This is a flowchart of sub-step S5 of the online diagnostic method for autonomous driving circuit boards based on digital twins according to an embodiment of this application. Figure 7 As shown, step S5 includes: S51, calculating the non-stationary residual energy of the purified non-stationary residual spectrum to obtain the total non-stationary residual energy; S52, extracting the total energy of the stationary component from the purified non-stationary residual spectrum; S53, normalizing the total non-stationary residual energy based on the total energy of the stationary component to obtain the quantized fault score.

[0064] Specifically, step S51 involves calculating the non-stationary residual energy of the purified non-stationary residual spectrum to obtain the total non-stationary residual energy. It should be understood that since the fault characteristics in the purified non-stationary residual spectrum exist in the form of scattered energy clusters, the energy value of a single energy cluster cannot reflect the overall scale of the fault and is easily affected by random noise, leading to misjudgment. Only by integrating all residual energies can the true degree of the fault be accurately reflected. Therefore, this application further performs non-stationary residual energy calculation on the purified non-stationary residual spectrum to summarize the energies related to all non-stationary faults in the spectrum, obtaining the total energy value reflecting the overall scale of the fault. This avoids misjudgment of fault severity due to analysis of a single energy cluster, accurately captures the cumulative energy characteristics of the fault, provides core energy data support for subsequent fault score normalization, and ensures that the quantification results truly reflect the fault state.

[0065] Specifically, in one possible embodiment, step S51 is implemented as follows: First, each complex element in the purified non-stationary residual spectrum matrix is ​​traversed, and the square of its modulus (amplitude) is calculated, which is the sum of the squares of the real and imaginary parts. This operation converts the residual spectrum in the complex domain into an energy density spectrum in the non-negative real domain, such that the value at each frequency point represents the instantaneous power or energy density at that point. Subsequently, the values ​​at all effective frequency points in the energy density spectrum are accumulated. During the accumulation process, the cyclic frequency resolution and the spectral frequency resolution (as integral infinitesimals) can be further multiplied to approximate the two-dimensional energy integration. This calculation process covers the entire spectral correlation density plane, thereby aggregating the dispersed fault feature energies into a scalar value. Finally, this scalar value is determined as the total non-stationary residual energy, serving as the molecular input for subsequent fault feature quantization.

[0066] Here, when obtaining the total nonstationary residual energy by performing discrete two-dimensional summation by calculating the sum of squared amplitudes, it is essentially calculating the square of the global L2 norm of the nonstationary residual spectrum, which has the following drawbacks: First, it is too sensitive to diffuse background noise. That is, even after cyclic stationary purification, there are still a large amount of low-energy random background noise in the residual spectrum. These noises are diffusely distributed throughout the two-dimensional spectral plane. Global summation will accumulate these ubiquitous low-energy noises, forming a very high noise floor. As a result, the energy contributed by the real but weak early fault signals (which may manifest as one or a few isolated points with slightly higher energy) is easily submerged by this huge noise floor, resulting in an extremely low signal-to-noise ratio. Secondly, the lack of utilization of fault signal morphology means that physical faults (such as poor contact arcs or microcracks) often manifest as sparse, localized energy spikes or bright spots on the SCD plane. Global summation does not distinguish between these, treating high-energy local spikes and the same amount of energy accumulated from a large number of low-energy points equally. This loses the opportunity to utilize the prior information that fault signals have sparsity and aggregation on the spectral plane.

[0067] Therefore, in another preferred embodiment, this application employs sparse energy aggregation based on statistical significance, that is, improving the calculation of the global total energy to aggregate only the statistically significant anomalous energies, and not treating the residual spectrum as a continuous function that needs to be integrated globally, but rather treating it as an image in which most of it is background (random noise), and what needs to be done is to find the significant bright spots (fault signals) in the image.

[0068] In other words, the background noise level in the residual spectrum needs to be estimated online and adaptively first. Then, based on this noise level, a dynamic threshold is set to represent that, with a 99.9% (or other high confidence) probability, a point is merely the upper limit of the energy that noise can reach. Thus, energy aggregation is performed only on significant points whose energy far exceeds this threshold. Specifically, step S51 includes: calculating the element-wise amplitude square of the purified non-stationary residual spectrum to obtain the residual energy density spectrum; calculating a robust estimate of the standard deviation of the background noise energy using the median absolute deviation based on the residual energy density spectrum, and determining a dynamic significance threshold based on the robust estimate and a preset significance factor; traversing the residual energy density spectrum and aggregating the excess energy of points exceeding the dynamic significance threshold to obtain the total non-stationary residual energy.

[0069] Specifically, firstly, for the non-stationary residual spectrum after cyclical purification, the residual energy density spectrum is obtained by calculating the amplitude square of each element. In order to robustly estimate the background noise level without being contaminated by potential fault bright spots, statistical values ​​such as mean and standard deviation are not used. Instead, the median and median absolute deviation, which are insensitive to outliers, are used to set the significance threshold. That is, when setting a dynamic threshold based on noise statistics, any point with an energy density value exceeding the threshold will be considered statistically significant. The significance threshold is set as the median plus several times the median absolute deviation.

[0070] Here, based on the residual energy density spectrum, a robust estimate of the standard deviation of the background noise energy is calculated using the median absolute deviation, as follows:

[0071]

[0072] in, It is a robust estimate of the standard deviation of background noise energy. It is the median of all elements in the input residual energy density spectral matrix. It is the residual energy density spectrum matrix, and 1.4826 is a correction factor, such that for normally distributed data, the absolute deviation of the median is approximately equal to the standard deviation.

[0073] Then, based on the robust estimate and the preset significance factor, the dynamic significance threshold is determined as follows:

[0074]

[0075] in, This is a significance factor, an adjustable parameter (e.g., equal to 3 or 5), used to determine the strictness of the judgment of abnormality. It is an adaptive energy threshold used to distinguish between background noise and potential anomalous signals.

[0076] Then, the residual energy density spectrum is traversed. For each point, the excess energy of points exceeding the dynamic significance threshold is aggregated to obtain the total nonstationary residual energy; that is, only those exceeding the threshold are accumulated. The energy value is calculated, and to further amplify the anomaly, its excess energy (i.e., the energy value minus the threshold) is accumulated. This makes the contribution of points that just exceed the threshold very small, while the contribution of points that far exceed the threshold is large.

[0077]

[0078] in, This is the final output, improved sparse residual energy. By taking the larger of 0 and the value within parentheses, it ensures that only positive excess energy is accumulated. and It is the resolution of the cycle frequency and the spectral frequency.

[0079] Therefore, by actively ignoring the background noise floor, which accounts for the vast majority of the total energy, even weak fault signals, as long as their energy can penetrate the statistical upper limit of the noise level, will contribute significantly to the final aggregated energy result, resulting in a high signal-to-noise ratio. Furthermore, because it is insensitive to overall fluctuations in background noise levels caused by different vehicle operating conditions (such as temperature and load changes), meaning the threshold is adaptive—as the overall background noise level increases, the threshold also increases—it maintains consistency in the judgment criteria for anomalies, thus enhancing robustness. In addition, by focusing more on finding sparse bright spots that match the physical characteristics of the fault, rather than broad energy increases, it effectively reduces false alarms caused by random energy fluctuations unrelated to the fault, enhancing specificity.

[0080] Specifically, step S52 involves extracting the total energy of the stationary component from the purified non-stationary residual spectrum. It should be understood that since a small amount of stationary random noise unrelated to the fault remains in the purified non-stationary residual spectrum, its energy dynamically changes with the circuit board's operating conditions (such as temperature and voltage fluctuations). If the total non-stationary residual energy is directly used for quantization, the score will fluctuate due to interference from the stationary noise, affecting the stability of fault judgment. Therefore, this application further extracts the total energy of the stationary component from the purified non-stationary residual spectrum to obtain the benchmark energy of the stationary noise under the current operating conditions, providing a reference for residual energy normalization. This eliminates the interference of stationary noise on fault quantization, ensuring that the subsequently obtained fault score is unaffected by operating condition fluctuations, guaranteeing the comparability of fault scores under different operating conditions, and improving the stability and reliability of the quantization results.

[0081] Specifically, in one possible embodiment, step S52 is implemented as follows: First, a stationarity analysis is performed on the purified non-stationary residual spectrum. The stationarity index of the signal at each frequency point in the spectrum is calculated using the autocorrelation function, and frequency regions with stationarity indices higher than a preset threshold are selected. Then, the extraction range of stationary components is determined, and frequency regions meeting the stationarity index are defined as stationary component regions. Next, the trapezoidal integral method, consistent with the calculation of non-stationary residual energy, is used to perform two-dimensional energy integration on the stationary component region to obtain the preliminary total energy of the stationary components. Then, the preliminary total energy is smoothed, and short-term fluctuations are eliminated using a moving average algorithm to obtain a stable total energy of the stationary components. Finally, this energy value is associated with and stored with the current circuit board operating parameters (temperature, voltage), establishing a mapping relationship between stationary energy and operating conditions, providing a reference for energy extraction under similar operating conditions in the future.

[0082] Specifically, step S53 involves normalizing the total non-stationary residual energy based on the total energy of the stationary component to obtain a quantified fault score. It should be understood that because the total energy of the stationary component varies under different operating conditions, the absolute values ​​of the total non-stationary residual energy are not directly comparable. For example, the normal residual energy under high operating conditions may be higher than the minor fault energy under low operating conditions, leading to misjudgment through direct comparison. Therefore, this application further normalizes the total non-stationary residual energy using the total energy of the stationary component to eliminate the influence of operating condition fluctuations on the energy value, obtaining a dimensionless fault score. In a specific example of this application, step S53 includes: normalizing the fault score of the total non-stationary residual energy using the following formula:

[0083]

[0084] in, For the total energy of the steady component, For the total nonstationary residual energy, It is a numerical stability constant (a very small positive number). To terminate the summation index, For the first Spectral correlation density of each signal frame For the cycle frequency parameter, For the first Frequency values ​​at each frequency point Frequency point interval, This is the quantized fault score. That is, first, the sum of the squared amplitudes of the spectral correlation density of the i-th signal frame at all frequency points with a cyclic frequency of 0 (corresponding to stationary signal characteristics) is calculated, and then multiplied by the frequency point interval to obtain the total energy of the stationary random noise under the current operating condition. Compare it with the numerical stability constant The sum of these two values ​​forms the denominator, while the total energy of all fault-related non-stationary signals in the purified residual spectrum forms the numerator. The fault score is obtained by calculating the ratio of these two values. This method eliminates the interference of stationary noise energy fluctuations under different operating conditions on fault judgment and avoids the calculation anomaly of a zero denominator. The resulting dimensionless fault score can stably characterize the severity of faults across operating conditions, providing an accurate and comparable quantitative basis for subsequent adaptive threshold decision-making.

[0085] In the aforementioned online diagnostic method for autonomous driving circuit boards based on digital twins, step S6 involves applying an adaptive threshold decision to the quantified fault scores to obtain the fault risk level. It should be understood that since the normal fault scores of an autonomous driving circuit board dynamically change with operating conditions (such as high temperature and high load), a fixed threshold cannot adapt to these changes, easily leading to false alarms under high operating conditions or missed alarms under low operating conditions. Therefore, this application further implements an adaptive threshold decision on the quantified fault scores to dynamically adjust the judgment threshold, making it match the normal score range under the current operating conditions. This allows for accurate differentiation between normal fluctuations and fault scores, avoiding misjudgments caused by fixed thresholds, ensuring accurate identification of fault risks under any operating condition, and providing timely and reliable fault warnings for the autonomous driving system.

[0086] Specifically, in one possible embodiment, step S6 is implemented as follows: First, during the offline calibration phase, historical operating data of the autonomous driving circuit board under healthy conditions covering the entire operating range (e.g., temperature -40℃ to 125℃, load rate 0% to 100%) is collected, and the corresponding fault scores are calculated. Through statistical analysis, a multi-dimensional preset threshold mapping library is constructed. This library uses operating condition parameters (temperature, load, voltage) as index keys and statistical distribution parameters (mean and standard deviation) of fault scores under healthy conditions as values. Subsequently, the online monitoring phase begins. First, real-time operating condition parameters of the circuit board are obtained from the vehicle CAN bus. Then, based on these real-time operating condition parameters, the corresponding baseline statistical distribution parameters are retrieved from the preset threshold mapping library. To smooth short-term fluctuations, a sliding window algorithm is used to calculate the statistical characteristics of recent real-time fault scores, where the sliding window size is set to 50 to 100 frames. Based on the retrieved baseline parameters and the real-time calculated statistical characteristics, the baseline threshold is dynamically adjusted, for example, setting the threshold to the baseline mean plus 3 times the baseline standard deviation. Next, the current quantified fault score is compared with the adjusted threshold. If the score is lower than the threshold, it is judged as normal; if it is higher than the threshold, the risk level is classified according to the magnitude of the exceedance. Finally, the stability of the decision result is verified by continuously monitoring the scores of multiple frames. After confirming that the risk level continuously meets the conditions, the final fault risk level is output.

[0087] In the aforementioned online diagnostic method for autonomous driving circuit boards based on digital twins, step S7 involves displaying the fault risk level. It should be understood that since the quantified fault score and risk level need to be perceived by the maintenance personnel or control unit of the autonomous driving system to achieve timely fault handling, a lack of intuitive display prevents the diagnostic results from being translated into actual maintenance actions, leading to the continued existence of potential fault hazards. Therefore, this application further displays the fault risk level to transform the abstract risk level into intuitive visual information, ensuring that relevant personnel or the system can quickly obtain the fault status. This achieves transparency of fault risk, enabling maintenance personnel to promptly grasp the health status of the circuit board, or allowing the autonomous driving system to trigger emergency plans based on the risk level, ensuring the safe operation of the autonomous driving circuit board and reducing the probability of safety accidents caused by faults.

[0088] Specifically, in one possible embodiment, step S7 is implemented as follows: First, the display carriers for the fault risk level are determined, including the vehicle-mounted central display screen, the remote operation and maintenance platform, and the local status indicator lights on the circuit boards. Then, based on the characteristics of the display carriers, corresponding display formats are designed. For example, the vehicle-mounted screen uses a combination of text and icons, where green icons represent normal, yellow represents a warning, and red represents a fault. The indicator lights use different colors to remain constantly lit or flashing, with green constantly lit for normal, yellow flashing for a warning, and red constantly lit for a fault. Next, the fault risk level information and associated data (including fault score, occurrence time, and circuit board location) are integrated into a display data packet and transmitted to each display carrier via the vehicle-mounted Ethernet. The display content is then updated in real time to ensure that the displayed information is synchronized with the latest diagnostic results, and the update frequency is consistent with the diagnostic cycle. Finally, a display alarm mechanism is set up. When the risk level is a warning or a fault, a vehicle-mounted voice prompt or a remote platform pop-up notification is triggered to ensure that relevant personnel are aware of the fault risk immediately, and a display log is recorded for subsequent operation and maintenance traceability.

[0089] In summary, the online diagnostic method for autonomous driving circuit boards based on digital twins, as described in this application, is explained. First, it acquires high-frequency voltage signals on the circuit board containing background noise and potential fault information in real time. Then, it introduces cyclostationary theory to perform a deep time-frequency transformation on the signal, and achieves feature separation of signal components in the cyclic domain through spectral correlation density estimation. Based on this, a cyclostationary purification algorithm is designed to filter out deterministic periodic noise, thereby effectively extracting non-stationary residual signals. Furthermore, the purified residual spectrum is subjected to energy aggregation and quantization to construct a feature score that stably characterizes the severity of minor faults. Finally, an adaptive threshold decision mechanism is established to evaluate this score in real time and output an accurate fault risk level. This effectively penetrates the masking of strong electromagnetic noise, thereby achieving highly sensitive online diagnosis of early-stage minor faults on the circuit board.

[0090] Figure 8This is a block diagram of an online diagnostic system for autonomous driving circuit boards based on digital twins, according to an embodiment of this application. Figure 8 As shown, the online diagnostic system 100 for autonomous driving circuit boards based on digital twins according to an embodiment of this application includes: a high-frequency signal acquisition module 110 for acquiring high-frequency voltage signals acquired by a power management chip; a high-fidelity signal segmentation module 120 for performing high-fidelity signal segmentation on the high-frequency voltage signals to obtain a signal frame sequence; a spectral correlation density estimation module 130 for estimating the spectral correlation density of each signal frame in the signal frame sequence to obtain a signal spectral correlation density map sequence; a cyclic purification residual extraction module 140 for performing cyclic stationary purification and non-stationary residual extraction on each signal spectral correlation density map in the signal spectral correlation density map sequence to obtain a purified non-stationary residual spectrum sequence; a residual energy aggregation and quantization module 150 for performing residual energy aggregation and fault feature quantization on the purified non-stationary residual spectrum sequence to obtain a quantized fault score; an adaptive threshold decision module 160 for performing adaptive threshold decision on the quantized fault score to obtain a fault risk level; and a fault risk level display module 170 for displaying the fault risk level.

[0091] As described above, the online diagnostic system 100 for autonomous driving circuit boards based on digital twins according to embodiments of this application can be implemented in various wireless terminals, such as servers with online diagnostic algorithms for autonomous driving circuit boards based on digital twins. In one possible implementation, the online diagnostic system 100 for autonomous driving circuit boards based on digital twins according to embodiments of this application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the online diagnostic system 100 for autonomous driving circuit boards based on digital twins can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the online diagnostic system 100 for autonomous driving circuit boards based on digital twins can also be one of many hardware modules of the wireless terminal.

[0092] Alternatively, in another example, the digital twin-based online diagnostic system 100 for autonomous driving circuit boards and the wireless terminal can also be separate devices, and the digital twin-based online diagnostic system 100 for autonomous driving circuit boards can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0093] Here, those skilled in the art will understand that the specific operations of each step in the above-described online diagnostic system for autonomous driving circuit boards based on digital twins have been referenced above. Figures 1 to 7 The online diagnostic method for autonomous driving circuit boards based on digital twins has been described in detail, and therefore, its repeated description will be omitted.

Claims

1. A method for online diagnostics of autonomous driving circuit boards based on digital twins, characterized in that, include: Acquire high-frequency voltage signals collected by the power management chip; High-fidelity signal segmentation is performed on the high-frequency voltage signal to obtain a signal frame sequence; Spectral correlation density estimation is performed on each signal frame in the signal frame sequence to obtain a signal spectral correlation density map sequence; Cyclic stationary purification and non-stationary residual extraction are performed on each signal spectrum correlation density map in the signal spectrum correlation density map sequence to obtain a purified non-stationary residual spectrum sequence. The residual energy aggregation and fault feature quantization of the purified non-stationary residual spectrum sequence are performed to obtain the quantized fault score. An adaptive threshold decision is made on the quantified fault score to obtain the fault risk level; This displays the fault risk level.

2. The online diagnostic method for autonomous driving circuit boards based on digital twins according to claim 1, characterized in that, High-fidelity signal segmentation is performed on the high-frequency voltage signal to obtain a signal frame sequence, including: The high-frequency voltage signal is overlapped and framed to obtain the original signal frame sequence; Intra-frame signal baseline correction is performed on each original signal frame in the original signal frame sequence to obtain the signal frame sequence.

3. The online diagnostic method for autonomous driving circuit boards based on digital twins according to claim 1, characterized in that, Spectral correlation density estimation is performed on each signal frame in the signal frame sequence to obtain a signal spectral correlation density map sequence, including: Time-frequency decomposition of the signal frame is performed to obtain the complex spectrum of the signal frame; The complex spectrum of the signal frame is shifted along the frequency axis and the conjugate spectral product is calculated to obtain a set of spectral product sequences. Time smoothing and matrix construction are performed on the spectral product sequence set to obtain the signal spectral correlation density map.

4. The online diagnostic method for autonomous driving circuit boards based on digital twins according to claim 1, characterized in that, Cyclic stationary cleansing and non-stationary residual extraction are performed on each signal spectral correlation density map in the signal spectral correlation density map sequence to obtain a cleaned non-stationary residual spectral sequence, including: Based on the signal spectrum correlation density map, generate a cleanup mask matrix; Based on the purification mask matrix, a mask application and non-stationary residual spectrum extraction are performed on the signal spectrum correlation density map to obtain the purified non-stationary residual spectrum.

5. The online diagnostic method for autonomous driving circuit boards based on digital twins according to claim 4, characterized in that, Based on the signal spectral correlation density map, a cleanup mask matrix is ​​generated, including: Obtain the calibrated set of healthy cycle frequencies; The mask matrix is ​​initialized based on the signal spectrum correlation density map to obtain the initial mask matrix; Based on the calibrated set of healthy cycle frequencies, a masking neighborhood is applied to the initial mask matrix and a final mask is generated to obtain a cleaned mask matrix.

6. The online diagnostic method for autonomous driving circuit boards based on digital twins according to claim 1, characterized in that, The purified non-stationary residual spectrum sequence is subjected to residual energy aggregation and fault feature quantization to obtain the quantized fault score, including: The nonstationary residual energy of the purified nonstationary residual spectrum is calculated to obtain the total nonstationary residual energy. Extract the total energy of the stationary component from the purified non-stationary residual spectrum; The total non-stationary residual energy is normalized based on the total energy of the stationary component to obtain the quantified fault score.

7. The online diagnostic method for autonomous driving circuit boards based on digital twins according to claim 6, characterized in that, The nonstationary residual energy of the purified nonstationary residual spectrum is calculated to obtain the total nonstationary residual energy, including: The residual energy density spectrum is obtained by calculating the square of the amplitude of each element on the purified non-stationary residual spectrum. Based on the residual energy density spectrum, a robust estimate of the standard deviation of background noise energy is calculated using the median absolute deviation, and a dynamic significance threshold is determined based on the robust estimate and a preset significance factor. The excess energy of points exceeding the dynamic significance threshold is aggregated by traversing the residual energy density spectrum to obtain the total nonstationary residual energy.

8. The online diagnostic method for autonomous driving circuit boards based on digital twins according to claim 6, characterized in that, The fault score is normalized based on the total energy of the stationary component to obtain the quantified fault score. This includes normalizing the fault score of the total non-stationary residual energy using the following formula: in, For the total energy of the steady component, The total nonstationary residual energy, It is the numerical stability constant. To terminate the summation index, For the first Spectral correlation density of each signal frame For the cycle frequency parameter, For the first Frequency values ​​at each frequency point Frequency point interval, This represents the quantified fault score.

9. An online diagnostic system for autonomous driving circuit boards based on digital twins, characterized in that, include: The high-frequency signal acquisition module is used to acquire the high-frequency voltage signal acquired by the power management chip; The high-fidelity signal segmentation module is used to perform high-fidelity signal segmentation on high-frequency voltage signals to obtain signal frame sequences. The spectral correlation density estimation module is used to estimate the spectral correlation density of each signal frame in the signal frame sequence to obtain a sequence of signal spectral correlation density maps. The cyclic purification residual extraction module is used to perform cyclic stationary purification and non-stationary residual extraction on each signal spectrum correlation density map in the signal spectrum correlation density map sequence to obtain a purified non-stationary residual spectrum sequence. The residual energy aggregation and quantization module is used to perform residual energy aggregation and fault feature quantization on the purified non-stationary residual spectrum sequence to obtain the quantized fault score. The adaptive threshold decision module is used to make adaptive threshold decisions on the quantified fault scores to obtain the fault risk level. The fault risk level display module is used to display the fault risk level.

Images