Performance detection and analysis method and system for industrial control mainboard

Abstract

The invention relates to the field of industrial control systems, in particular to a performance detection analysis method and system for an industrial control mainboard, and the method achieves the comprehensive optimization of performance stability, energy efficiency improvement and failure rate reduction through multi-modal data collection, causal relationship model construction and multi-objective optimization. Integrating multi-device data by adopting federated learning, and dynamically updating a causal path and generating an optimization result through closed-loop feedback in combination with a deep reinforcement learning optimization path adjustment strategy; according to the method, through dynamic adjustment of the path and global adaptability improvement of the model, the performance detection accuracy and the optimization effect of the industrial control mainboard in diversified application scenes are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of industrial control systems, and in particular to a performance detection and analysis method and system for an industrial control mainboard. Background Art

[0002] The industrial control motherboard is the core component of the industrial control system, and its performance directly affects the reliability and stability of the entire system. With the continuous development of industrial automation and intelligence, the performance detection and optimization of the industrial control motherboard has become an important part of ensuring the safe and efficient operation of the system. However, traditional detection methods often use static parameter collection, single environment testing and other means, which are difficult to adapt to complex and changeable industrial application scenarios. Therefore, a more accurate and dynamic performance detection and optimization solution is urgently needed.

[0003] The existing technology (Chinese invention patent, publication number: CN118688617B, name: A performance detection and analysis method and system for integrated circuit motherboards) adopts a performance detection method based on application data collection and scenario establishment, performs adaptive clustering analysis on the test results, and generates detection results through a performance evaluation network. Although it can improve the detection accuracy to a certain extent, this method has the following defects: the test scenario lacks a dynamic adjustment mechanism and cannot fully reflect the performance of the motherboard in a diverse environment; the existing method mainly relies on a fixed test sequence and static sample analysis, and fails to achieve dynamic adjustment of the optimization path, resulting in limited optimization effect; the test data of different devices is not fully utilized, the optimization results lack global adaptability, and it is difficult to provide effective optimization strategies for new scenarios. Summary of the invention

[0004] In view of the many problems existing in the above-mentioned prior art, the present invention provides a performance detection and analysis method and system for an industrial control motherboard. The present invention collects multimodal operation data of the industrial control motherboard, constructs a causal relationship model, combines federated learning with deep reinforcement learning, and optimizes the performance detection path. The present invention uses dynamic weight adjustment and closed-loop feedback to ensure that the model can generate accurate optimization paths in different scenarios, thereby improving the accuracy of performance detection and the optimization effect.

[0005] A performance detection and analysis method for an industrial control motherboard comprises the following steps:

[0006] Collect multi-modal data of the industrial control motherboard during operation, pre-process the data including electrical signals, acoustic signals, thermal imaging and optical signals, and generate performance characteristic data;

[0007] Based on the performance characteristic data, a causal relationship model is constructed using a structural equation modeling method to extract the causal relationship between the characteristics, and combined with the weight adjustment of multimodal features, causal reasoning network data is generated, wherein the causal reasoning network data includes effective causal relationships and abnormal impact paths;

[0008] According to the causal reasoning network data, a multi-objective optimization model is constructed, and an optimization path is generated by quantitatively defining the performance stability, energy efficiency improvement and failure rate reduction goals, and dynamically adjusting the weights between the goals to balance the priorities; the performance of the optimization path is verified, and the optimization path is adjusted according to the verification results, and optimization feedback data is output;

[0009] Based on the optimization feedback data, the causal relationship model is updated, and the dynamic adaptability of the optimization path is optimized through a collaborative learning mechanism. The collaborative learning mechanism adopts a distributed learning framework, integrates multiple rounds of test data through federated learning, and combines a reinforcement learning algorithm to enhance the dynamic adjustment capability of the optimization path, and generates closed-loop optimization feedback data to further improve the effect of performance detection and optimization.

[0010] Preferably, the preprocessing of the multimodal data comprises the following steps:

[0011] The electrical signal data is corrected for baseline drift, the signal baseline is calculated by the sliding window method and differential adjustment is performed;

[0012] Band-pass filtering is performed on the acoustic signal data, and a finite impulse response filter is used to filter the signal in the range of 20 Hz to 20 kHz;

[0013] Perform median filtering on the thermal imaging data, select a 3x3 window to calculate the median point by point to eliminate random noise in the image;

[0014] The optical signal data is normalized and the signal amplitude is adjusted to the [0,1] interval using the maximum and minimum value normalization method.

[0015] Preferably, the bandpass filtering of the acoustic signal data comprises the following steps:

[0016] Convert the time domain signal into the frequency domain signal through fast Fourier transform;

[0017] In the frequency domain, keep the frequency band from 20Hz to 20kHz and set other frequency bands to zero;

[0018] The signal is reconstructed from the frequency domain to the time domain using an inverse Fourier transform to generate filtered acoustic signal data.

[0019] Preferably, the construction of the causal relationship model comprises the following steps:

[0020] Select key parameters in the performance characteristic data, including voltage fluctuation amplitude, temperature rise rate, vibration acceleration and optical interference intensity;

[0021] Calculate conditional probabilities based on performance characteristic parameters and generate causal graph structures using Bayesian network methods;

[0022] The parameter weights in the causal graph are optimized by maximum likelihood estimation, and the edges with causal strength lower than a preset threshold are removed to determine the causal path.

[0023] Preferably, the optimization of the causal path comprises the following steps:

[0024] Calculate the probability gain of each causal path and set the minimum threshold of causal strength;

[0025] Eliminate low-gain paths and retain high-gain paths to reduce the size of the causal network;

[0026] The conditional probabilities are recalculated for the remaining paths, and the path strengths are adjusted dynamically.

[0027] Preferably, the multi-objective optimization model includes the following objective functions:

[0028] Performance stability objective function, which calculates the stability of performance indicators through variance;

[0029] Energy efficiency improvement objective function, which calculates energy efficiency by the ratio of input power to output power;

[0030] The failure rate reduction objective function calculates the failure rate optimization target by the inverse ratio of the failure frequency;

[0031] The linear weighted method is used to comprehensively calculate each objective function to generate the optimized path.

[0032] Preferably, the linear weighting method comprises the following steps:

[0033] Dynamically adjust weights based on real-time calculation results of each objective function;

[0034] The weight updates are optimized using gradient descent to ensure that the optimization path is responsive to key objectives.

[0035] Preferably, the collaborative learning mechanism is implemented by the following steps:

[0036] Collect optimized path data from different test devices, encrypt it and transmit it to the central server;

[0037] Aggregate data in a central server to build a federated learning model;

[0038] Distribute the trained model parameters to each test device to achieve distributed optimization path adjustment.

[0039] Preferably, the federated learning model is combined with deep reinforcement learning, iteratively updates the path adjustment strategy through a policy gradient optimization algorithm, and automatically selects a suitable optimization path in a new test scenario.

[0040] A system for implementing the performance detection and analysis method of the industrial control mainboard, comprising:

[0041] The data acquisition module is used to collect multi-modal data during the operation of the industrial control mainboard, including electrical signal data, acoustic signal data, thermal imaging data and optical signal data;

[0042] A data preprocessing module, used for performing noise filtering, time alignment and normalization processing on the multimodal data to generate performance characteristic data;

[0043] A causal relationship modeling module, used to build a causal relationship model based on the performance feature data, extract the causal relationship between features, and generate causal reasoning network data according to the weight adjustment of multimodal features, wherein the causal reasoning network data includes effective causal relationships and abnormal impact paths;

[0044] An optimization model building module is used to build a multi-objective optimization model based on the causal reasoning network data, generate an optimization path by quantitatively defining the performance stability, energy efficiency improvement and failure rate reduction goals, and dynamically adjusting the weights between the goals;

[0045] A performance verification module is used to perform performance verification on the optimization path, adjust the optimization path according to the verification result, and generate optimization feedback data;

[0046] The collaborative learning module is used to update the causal relationship model based on the optimization feedback data, integrate multiple rounds of test data through a distributed learning framework and a federated learning mechanism, and combine the dynamic adaptability of the optimization path of the reinforcement learning algorithm to generate closed-loop optimization feedback data.

[0047] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0048] The present invention realizes dynamic adjustment and global adaptability of the optimization path by combining federated learning with deep reinforcement learning.

[0049] The present invention realizes real-time correction of causal path optimization by dynamically updating the causal relationship model technology;

[0050] The present invention realizes efficient adaptation of the optimization model in new scenarios through collaborative learning mechanism and closed-loop optimization feedback technology.

[0051] The present invention achieves the comprehensive optimization effects of performance stability, energy efficiency improvement and failure rate reduction through the construction of a multi-objective optimization model and dynamic weight adjustment technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the process of the present invention;

[0053] Figure 2 A schematic diagram of the causal relationship model update process in the present invention;

[0054] Figure 3 It is a schematic diagram of the multi-objective optimization model process in the present invention;

[0055] Figure 4 Schematic diagram of collaborative learning mechanism in the present invention;

[0056] Figure 5 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0057] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0058] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0059] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0060] like Figure 1 As shown, a performance detection and analysis method for an industrial control motherboard includes the following steps:

[0061] Collect multi-modal data of the industrial control motherboard during operation, pre-process the data including electrical signals, acoustic signals, thermal imaging and optical signals, and generate performance characteristic data;

[0062] Preferably, the preprocessing of the multimodal data comprises the following steps:

[0063] The electrical signal data is corrected for baseline drift, the signal baseline is calculated by the sliding window method and differential adjustment is performed;

[0064] Band-pass filtering is performed on the acoustic signal data, and a finite impulse response filter is used to filter the signal in the range of 20 Hz to 20 kHz;

[0065] Perform median filtering on the thermal imaging data, select a 3x3 window to calculate the median point by point to eliminate random noise in the image;

[0066] The optical signal data is normalized and the signal amplitude is adjusted to the [0,1] interval using the maximum and minimum value normalization method.

[0067] Through electrical sensors, acoustic sensors, thermal imaging equipment and optical detection equipment, multimodal data of the industrial control motherboard during operation is collected, including electrical signals, acoustic signals, thermal imaging images and optical interference waveforms.

[0068] These data represent different physical and environmental properties of the operating status of the industrial control motherboard. For example, electrical signal data can reflect the current and voltage fluctuations on the motherboard; acoustic signal data captures possible mechanical failures through the vibration spectrum; thermal imaging data reveals the temperature distribution in key areas of the motherboard; and optical signal data analyzes micro-vibrations on the motherboard surface through changes in interference fringes.

[0069] During the acquisition process, the electrical signal may experience baseline drift due to sensor drift or ambient temperature changes, which will affect the accuracy of the data. The drift can be eliminated by calculating the dynamic baseline of the signal through the sliding window method and performing differential adjustment. Define the sliding window size as W, and calculate the average value B of the electrical signal in each window. i , as the baseline value of the window; perform differential processing on the original signal S(t) to obtain the drift-corrected signal S′(t):

[0070] S′(t)=S(t)-B i

[0071] Among them, B i is the mean value of the signal in the window, t is the time, and i is the window index. It can effectively correct signal drift and ensure the accuracy of subsequent analysis.

[0072] Vibroacoustic signals may contain a lot of environmental noise. By bandpass filtering, the signal frequency components in the range of 20Hz to 20kHz are retained, which is the key frequency band for vibration analysis. Finite impulse response (FIR) filters are used to set the bandpass range f 1 =20Hz, f 2 =20kHz, the signal after filtering is A'(t). The filtering expression is:

[0073]

[0074] Where h[n] is the impulse response of the filter, N is the filter order, and A(t) is the original signal. It eliminates interference noise in irrelevant frequency bands and highlights the key frequency components related to mechanical vibration.

[0075] Thermal imaging data may contain random noise introduced by sensor noise or environmental interference. Median filtering can effectively remove noise while retaining edge features by replacing pixel values ​​with the median value in a selected local window. Set the window size to 3×3 and replace each pixel in the thermal imaging data T(x,y) with the median value of its neighboring pixels:

[0076] T′(x,y)=Median({T(i,j)|(i,j)∈Neighborhood (x,y)})

[0077] Among them, T′(x,y) is the pixel value after filtering. It reduces the image noise level, improves the clarity of temperature distribution, and provides accurate data for subsequent temperature rise analysis.

[0078] The amplitude of an optical signal may fluctuate due to changes in sensor sensitivity or light source intensity. Normalization improves the comparability of data from different acquisition environments by mapping the signal amplitude to a fixed range.

[0079] For the optical signal data O(t), the maximum and minimum value normalization method is used:

[0080]

[0081] Among them, O min and O max are the minimum and maximum values ​​of the signal, respectively, and O′(t) is the normalized signal. This ensures the consistency of signals collected under different optical devices and enhances the reliability of subsequent analysis.

[0082] The pre-processed multi-modal data is input into the feature extraction module to extract key performance parameters (such as voltage fluctuation, vibration spectrum peak, abnormal heat distribution area, etc.) and generate performance characteristic data. The performance characteristic data is used for causal relationship modeling and optimization path generation, providing a basis for performance detection and optimization.

[0083] Preferably, the bandpass filtering of the acoustic signal data comprises the following steps:

[0084] Convert the time domain signal into the frequency domain signal through fast Fourier transform;

[0085] In the frequency domain, keep the frequency band from 20Hz to 20kHz and set other frequency bands to zero;

[0086] The signal is reconstructed from the frequency domain to the time domain using an inverse Fourier transform to generate filtered acoustic signal data.

[0087] The bandpass filtering of acoustic signal data in the present invention is intended to extract the effective frequency band (20Hz to 20kHz) related to vibration characteristics from the acoustic signal collected when the industrial control motherboard is running. Bandpass filtering retains the signal components within the target frequency band through frequency domain analysis and eliminates irrelevant noise or interference in other frequency bands, thereby enhancing the clarity of the signal and the accuracy of analysis.

[0088] In the specific implementation process, the time domain acoustic signal is first converted into a frequency domain signal through Fast Fourier Transform (FFT) to facilitate the identification of different frequency components. Then, only the frequencies between 20Hz and 20kHz are retained in the frequency domain, and other frequency components are set to zero. Finally, the processed frequency domain signal is reconstructed into a time domain signal through Inverse Fast Fourier Transform (IFFT) to generate filtered acoustic signal data.

[0089] Through the bandpass filtering process, the key frequency signals related to the vibration of the industrial control motherboard are effectively retained, and low-frequency environmental noise (such as mechanical resonance) and high-frequency electromagnetic interference are eliminated. This signal processing method ensures the analysis quality of acoustic data and can more accurately reflect the mechanical performance status of the motherboard, such as loose bolts, abnormal fan vibration and other problems.

[0090] Embodiment: During the operation of a certain type of industrial control motherboard, the original signal collected by the acoustic sensor is mixed with a large amount of low-frequency noise. Through the above-mentioned bandpass filtering method, the mechanical resonance below 20Hz and the electromagnetic noise above 20kHz in the signal are completely removed, and only the signal in the key frequency band is retained. The filtered data clearly shows the abnormal vibration of the fan inside the motherboard, and its main vibration frequency is concentrated in the range of 3.2kHz to 3.8kHz. Further analysis confirmed that the fan bearing was worn, so it was replaced and maintained in time.

[0091] In another embodiment, in a test of an industrial control motherboard under high load, a significant 5.5kHz spike signal was successfully captured after processing the acoustic signal by a bandpass filtering method. The signal originated from micro-vibration caused by loosening of the fixing bolts on the motherboard. The problem was solved by re-reinforcing the bolts, significantly improving the stability of the equipment.

[0092] Based on the performance characteristic data, a causal relationship model is constructed using a structural equation modeling method to extract the causal relationship between the characteristics, and combined with the weight adjustment of multimodal features, causal reasoning network data is generated, wherein the causal reasoning network data includes effective causal relationships and abnormal impact paths;

[0093] The present invention constructs a causal relationship model through performance characteristic data to reveal the direct causal relationship between the multimodal characteristics of the industrial control motherboard. The Structural Equation Modeling (SEM) method is used to extract implicit variable associations from multimodal data and form a causal network. Structural equation modeling has the ability to process both observed variables and latent variables, and is very suitable for processing the complex relationship of multimodal data during the operation of the industrial control motherboard.

[0094] In the present invention, the performance characteristic data includes voltage fluctuation amplitude, temperature rise rate, vibration acceleration and optical interference intensity, etc. These characteristics are closely related to the operating status of the industrial control motherboard. The conditional probability distribution between the characteristics is calculated by the SEM method to generate a causal graph structure, in which the nodes represent the performance characteristics and the edges represent the causal relationship between the characteristics. Further combined with the weight adjustment of multimodal features, the accuracy and adaptability of the causal reasoning network are enhanced.

[0095] Extract the multi-modal performance characteristic data of the industrial control motherboard during operation and set the key performance indicator P 1 ,P 2 ,…,P n As the observed variable of the model. For example: P 1 Indicates the voltage fluctuation amplitude; P 2 Indicates the rate of temperature rise; P 3 Indicates vibration acceleration; P 4 Represents the optical interference intensity.

[0096] The causal diagram structure was constructed using SEM, and the model parameters were optimized by the Maximum Likelihood Estimation (MLE) method to obtain the path coefficient β between the characteristics. ij , indicating that from the feature P i To P j The causal strength of:

[0097]

[0098] Among them, ∈ j is the error term, which represents random fluctuations that cannot be explained by other characteristics.

[0099] In the causal inference network, the feature weight w is dynamically adjusted according to the importance of different modal data. k :

[0100]

[0101] Among them, Var(P k ) represents the feature P k The weight adjustment enhances the influence of high-variability features in causal inference.

[0102] According to the optimized causal graph structure, causal reasoning network data is generated, which contains effective causal relationships and abnormal impact paths. Effective causal relationships are used to describe feature associations under normal operating conditions, and abnormal impact paths are used to mark key causal chains that may cause performance problems.

[0103] Preferably, Figure 2 As shown, the construction of the causal relationship model includes the following steps:

[0104] Select key parameters in the performance characteristic data, including voltage fluctuation amplitude, temperature rise rate, vibration acceleration and optical interference intensity;

[0105] Calculate conditional probabilities based on performance characteristic parameters and generate causal graph structures using Bayesian network methods;

[0106] The parameter weights in the causal graph are optimized by maximum likelihood estimation, and the edges with causal strength lower than a preset threshold are removed to determine the causal path.

[0107] In this invention, in order to reveal the causal relationship between the multimodal performance characteristics of the industrial control motherboard during operation, a causal relationship model based on Bayesian network is constructed. The model generates a causal graph structure through conditional probability calculation of performance characteristic parameters, and further optimizes the parameter weights of the model through maximum likelihood estimation, eliminating edges with low causal strength to clarify the main causal path.

[0108] The performance characteristic data of industrial control motherboards include voltage fluctuation amplitude, temperature rise rate, vibration acceleration, and optical interference intensity. There may be complex correlations between these parameters. For example, voltage fluctuations may cause temperature rise, which in turn may affect vibration characteristics. Through Bayesian network modeling, the conditional dependency between these parameters can be captured, and the causal strength can be quantified from a statistical perspective, providing a basis for performance optimization and fault diagnosis.

[0109] Before modeling, we first select key parameters closely related to the operating status of the industrial control motherboard from the performance characteristic data: voltage fluctuation amplitude (P 1 ) indicates power supply stability; temperature rise rate (P 2 ) describes the motherboard's heat dissipation performance; vibration acceleration (P 3 ) indicates that it is used to monitor the mechanical vibration state; optical interference intensity (P 4 ) indicates the presence of minute vibrations or mechanical deformations.

[0110] Based on the above parameters, a Bayesian network is constructed to quantify the relationship between the parameters through conditional probability distribution. The core of the Bayesian network is the Conditional Probability Table (CPT), whose formula is:

[0111]

[0112] Among them, P(P i |P j ) means that when P is known j In the case of P i The probability of occurrence, P(P i |P j ) is the joint probability, P(P j ) is the marginal probability. A causal graph structure is generated based on the calculation results, where nodes represent performance parameters and edges represent conditional dependencies.

[0113] The parameter weights in the causal graph are optimized using Maximum Likelihood Estimation (MLE). The optimized path weight β ij Represents the parameter P i To P j If the path weight is lower than the preset threshold τ, the corresponding edge is removed, and finally the path with high causal strength is retained, thus clarifying the causal path. The optimization formula of path weight is:

[0114]

[0115] Among them, the parent node (P k ) indicates that k Directly related predecessor nodes.

[0116] By constructing the above causal relationship model, the causal relationship between the multi-modal characteristic parameters of the industrial control motherboard during operation can be clearly described, helping to locate performance bottlenecks or causes of failures. For example, if the model shows that voltage fluctuations have a significant impact on vibration acceleration through temperature rise, there may be problems with the power module or heat dissipation design. The clarification of the causal path makes the optimization direction more precise and avoids ineffective extensive testing.

[0117] Example 1: When an industrial control motherboard is running at high frequency, the voltage fluctuation (P 1 ) and the temperature rise rate (P 2 ) showed abnormal characteristic parameters. The causal graph generated by Bayesian network modeling showed that P 1 P 2 The conditional probability of 2 |P 1 )=0.92, and the path weight β 12 =0.87. Further optimization eliminated other low-intensity paths and finally determined that the main cause of temperature rise was voltage fluctuation. Based on this result, the power module parameters were adjusted and the temperature rise rate was reduced by 15%.

[0118] Example 2: Vibration acceleration (P 3 ) increased abnormally. Through the analysis of the causal relationship model, it was found that the optical interference intensity (P 4 ) 3 The conditional probability of 3 |P 4 )=0.81, and the path weight β 34 =0.79. Combined with the abnormal impact path, it was located that the mechanical structure of the mainboard was loose. To address this problem, the bolts were re-reinforced and the vibration amplitude was successfully reduced.

[0119] Preferably, the optimization of the causal path comprises the following steps:

[0120] Calculate the probability gain of each causal path and set the minimum threshold of causal strength;

[0121] Eliminate low-gain paths and retain high-gain paths to reduce the size of the causal network;

[0122] The conditional probabilities are recalculated for the remaining paths, and the path strengths are adjusted dynamically.

[0123] The present invention models the causal relationship of the performance characteristic data of the industrial control motherboard, generates a causal graph structure, and then optimizes the causal path in order to improve the accuracy and computational efficiency of the model. The core of causal path optimization is to quantify the importance of each path through probability gain calculation, eliminate low-gain paths according to the minimum threshold of causal strength, and only retain high-gain paths that have a significant impact on key performance parameters, thereby reducing the complexity of the causal network. For the retained high-gain paths, the conditional probability is further recalculated and the path strength is dynamically adjusted to ensure the accuracy and adaptability of the model.

[0124] For each path P in the causal network i →P j Calculate its probability gain ΔG ij , which is used to quantify the contribution of the path. The calculation formula of probability gain is:

[0125] ΔG ij =P(P j |P i )-P(P j )

[0126] Among them, P(P j |P i ) is the conditional probability, indicating that in P i In the presence of P j The probability of occurrence; P(P j ) is the marginal probability, which means that P does not depend on any predecessor node.j The probability of occurrence. The larger the probability gain value, the higher the probability of occurrence. i →P j For the target node P j The more significant the impact.

[0127] The calculated probability gain of each path is compared with the preset minimum threshold τ of causal strength:

[0128] If ΔG ij <τ, then eliminate path P i →P j , then eliminate the path P i →P j After removing the low-gain paths, only the high-gain paths are retained. This method can effectively reduce the redundant paths in the causal network, reduce the computational complexity, and highlight the important causal relationships.

[0129] For the remaining paths after removing the low-gain paths, the conditional probability is recalculated and the path strength is dynamically adjusted. The dynamic adjustment of path strength is achieved by updating the path weight β ij To achieve this, the formula is:

[0130] β ij = argmaxP(P j |Parent Node(P j ))

[0131] Among them, the parent node (P j ) indicates that the causal network is directly connected to P j Dynamically adjusting the path strength can further improve the adaptability of the causal model to different operating conditions and ensure that the model can accurately reflect the relationship between performance characteristics in a variety of scenarios.

[0132] Through causal path optimization, the structure of the causal network can be significantly simplified, computing efficiency can be improved, and the most critical causal relationship can be retained, which helps to quickly locate the root cause of the abnormal performance of the industrial control motherboard. After eliminating the low-gain path, the accuracy and reliability of the causal model are further enhanced, providing efficient and streamlined data support for the subsequent optimization model construction. Dynamic adjustment of path strength can also adapt to the characteristic changes of the motherboard under different operating conditions and improve the robustness of the model.

[0133] Example 1: When an industrial control motherboard is operating in a high temperature environment, the performance characteristic data collected includes the voltage fluctuation amplitude (P 1 ), temperature rise rate (P 2 ), vibration acceleration (P 3 ). The causal network analysis results show that path P 1 →P 2 The probability gain ΔG 12=0.78, path P 2 →P 3 The probability gain ΔG 23 =0.81, and path P 1 →P 3 The probability gain is only 0.05. According to the threshold τ = 0.10, the low gain path P is eliminated. 1 →P 3 Finally, two high-gain paths are retained. By further adjusting the path P 1 →P 2 The weight of the error was determined, confirming that poor heat dissipation of the power module was the root cause of the abnormal vibration.

[0134] Example 2: In the performance test of a certain model of industrial control motherboard, 8 low-gain paths were eliminated through causal path optimization, and only 4 high-gain paths were retained, among which the optical interference intensity (P 4 ) to vibration acceleration (P 3 ) of the conditional probability P(P 3 |P 4 ) was dynamically adjusted to 0.89. Further analysis revealed that the main contribution of this path came from the loosening of a bolt on the motherboard. The vibration problem was solved by reinforcing the bolt and successfully improved the system stability.

[0135] According to the causal reasoning network data, a multi-objective optimization model is constructed, and an optimization path is generated by quantitatively defining the performance stability, energy efficiency improvement and failure rate reduction goals, and dynamically adjusting the weights between the goals to balance the priorities; the performance of the optimization path is verified, and the optimization path is adjusted according to the verification results, and optimization feedback data is output;

[0136] The present invention constructs a multi-objective optimization model based on causal reasoning network data to achieve systematic improvement of industrial control motherboard performance detection and optimization. The core of the multi-objective optimization model is to quantify and balance multiple optimization goals, including performance stability, energy efficiency improvement and failure rate reduction. By defining each objective function and dynamically adjusting its weight, an optimal path that can simultaneously meet each optimization goal, i.e., an optimized path, is generated. The optimized path is further tested and adjusted in combination with performance verification to generate optimization feedback data to support subsequent optimization iterations.

[0137] The performance stability objective function measures the performance fluctuation of the industrial control motherboard under different loads and is specifically defined by the variance of the voltage fluctuation amplitude:

[0138] f 1 =Var(V)

[0139] Among them, V represents the voltage fluctuation data, Var(V) is its variance, and the smaller the value, the more stable the system performance.

[0140] The energy efficiency improvement objective function is defined as the ratio of the motherboard power output to the power input:

[0141]

[0142] Among them, P out and P in They are output power and input power respectively. The larger the value, the higher the energy efficiency.

[0143] The failure rate reduction objective function is defined as the inverse of the failure frequency:

[0144]

[0145] Among them, F failure Indicates the number of faults that occur within a certain period of time. The larger the value, the lower the fault rate.

[0146] In order to balance the priorities between different goals, a dynamic weight adjustment mechanism is introduced to update the weight of each goal in real time according to the current operating status. i :

[0147]

[0148] The objective functions are combined into a comprehensive optimization objective through linear weighted summation:

[0149]

[0150] Among them, F is the comprehensive optimization goal, w i is the weight of the ith target.

[0151] The optimized path is generated using the Multi-Objective Deep Deterministic Policy Gradient (MO-DDPG) algorithm, which searches for the optimal path points in the continuous action space through reinforcement learning to ensure that the path can effectively balance the needs of different objectives.

[0152] The generated optimized path is validated for performance, including thermal field simulation, electric field simulation, and vibration analysis. The feasibility and optimization effect of the path points are verified by simulating the real operating environment. If the optimization effect of some path points does not meet expectations, the path is adjusted according to the simulation results to ensure that the final path can achieve comprehensive optimization.

[0153] According to the verification results, optimization feedback data is output, including optimization path points, performance improvement indicators and path adjustment records. The optimization feedback data is used to update the causal reasoning network and the input of the next round of optimization model, forming a closed loop of optimization.

[0154] Preferably, Figure 3 As shown, the multi-objective optimization model includes the following objective functions:

[0155] Performance stability objective function, which calculates the stability of performance indicators through variance;

[0156] Energy efficiency improvement objective function, which calculates energy efficiency by the ratio of input power to output power;

[0157] The failure rate reduction objective function calculates the failure rate optimization target by the inverse ratio of the failure frequency;

[0158] The linear weighted method is used to comprehensively calculate each objective function to generate the optimized path.

[0159] The multi-objective optimization model proposed in this invention is used to comprehensively balance the three optimization goals of performance stability, energy efficiency improvement and failure rate reduction during the operation of the industrial control motherboard. The optimal optimization path is generated by quantifying the function form of each goal and performing linear weighted comprehensive calculation. The multi-objective optimization model can accurately reflect the performance requirements of the industrial control motherboard under different working conditions and provide data support to achieve dynamic adjustment.

[0160] The performance stability objective function is used to quantify the performance fluctuation of the chemical control motherboard under different load conditions, and the variance of the voltage fluctuation amplitude is used to measure the stability.

[0161] f 1 =Var(V)

[0162] Among them, V represents voltage fluctuation data, Var(V) is its variance, and the smaller the value, the more stable the system performance. By collecting voltage signal data in real time, the local variance is calculated using the sliding window method to determine the dynamic stability of the performance.

[0163] The energy efficiency improvement objective function reflects the ratio of the input power to the output power of the motherboard, and aims to improve the efficiency of energy use.

[0164]

[0165] Among them, P out and P in The output power and input power are respectively, and the larger the value, the higher the energy efficiency. The input power and output power are collected in real time through the mainboard power sensor, the energy efficiency target function is calculated, and the power consumption-related nodes in the optimization path are dynamically adjusted.

[0166] The failure rate reduction objective function is used to quantify the reliability of the motherboard operation and is expressed as the inverse of the failure frequency. The lower the frequency, the lower the failure rate.

[0167]

[0168] Among them, Ffailure Indicates the number of failures within a certain period of time. The larger the value, the lower the failure rate. The failure frequency of the mainboard is counted through the fault log and online monitoring system, the failure rate objective function is calculated, and the nodes in the path that reduce the risk of failure are optimized.

[0169] The above three objective functions are linearly weighted summed through dynamic weights wiw_iwi to comprehensively calculate the optimization path.

[0170]

[0171] Among them, F is the comprehensive optimization goal, w i is the weight of the i-th target, which is dynamically adjusted through real-time monitoring data.

[0172] The multi-objective optimization model achieves comprehensive optimization of the operating status of the industrial control motherboard by balancing the three key goals of performance stability, energy efficiency and failure rate. The model can dynamically respond to different operating conditions, ensure that the system operates in an efficient and safe state, and effectively improve the reliability of the system by optimizing the path.

[0173] Example 1: In a high-load operation test of a certain industrial control motherboard, the variance of the collected voltage fluctuation amplitude data is 0.15, the input power is 120W, the output power is 90W, and the failure frequency is 1 time every 5 hours. The performance stability objective function calculated by the multi-objective optimization model is 0.15, the energy efficiency improvement objective function is 0.75, and the failure rate reduction objective function is 5. According to the optimization path, after adjusting the heat dissipation module settings and power load distribution, the performance stability objective function is reduced to 0.10, the energy efficiency is improved to 0.85, and the failure rate objective function is increased to 10. After optimization, the system stability and energy efficiency are significantly improved.

[0174] Example 2: Under low load conditions, the power conversion efficiency of an industrial control motherboard is low. Through optimization path analysis, it is found that the cooling system has a significant impact on power consumption. After optimization, the power conversion efficiency is increased from the original 70% to 85%, and the vibration frequency is reduced by reducing the cooling system load, further optimizing performance stability.

[0175] Preferably, the linear weighting method comprises the following steps:

[0176] Dynamically adjust weights based on real-time calculation results of each objective function;

[0177] The weight updates are optimized using gradient descent to ensure that the optimization path is responsive to key objectives.

[0178] The linear weighted method in the present invention is used to integrate the performance stability, energy efficiency improvement and failure rate reduction objective functions of the multi-objective optimization model. By calculating the results of each objective function in real time, the weights are dynamically adjusted to adapt to the changes in the operating status of the industrial control motherboard. At the same time, the gradient descent method is used to optimize the weight update to ensure that the optimization path can quickly respond to the changing needs of key objectives.

[0179] The core idea of ​​dynamic weight adjustment is to dynamically allocate weights according to the relative importance of each objective function. The weight optimization process is iterated through the gradient descent method, gradually approaching the optimal weight configuration, so that the result of the comprehensive optimization function can reflect the best optimization path under the current operating conditions.

[0180] For the objective functions of performance stability, energy efficiency improvement and failure rate reduction, the weights w are defined respectively 1 、w 2 and w 3 , the comprehensive optimization objective function is:

[0181] F=w 1 ·f 1 +w 2 ·f 2 +w 3 ·f 3

[0182] Among them, f 1 、f 2 、f 3 They are the objective function values ​​of performance stability, energy efficiency improvement and failure rate reduction respectively.

[0183] The weight adjustment is updated based on the objective function value calculated in real time:

[0184]

[0185] Through this dynamic adjustment mechanism, the larger the weight, the higher the importance of the corresponding objective function, and the optimization path will prioritize responding to objectives with higher weights.

[0186] In order to further optimize the weights, the gradient descent method is used to iteratively update the weights. The update formula of gradient descent is:

[0187]

[0188] in, represents the weight at the tth iteration, η is the learning rate, Represents the comprehensive optimization function for weight w i Through multiple iterations, the weights are gradually optimized to the optimal value.

[0189] In the specific implementation, a dynamic adjustment threshold ∈ is set to control the update step size to ensure the stability and convergence of weight optimization.

[0190] The optimized path is generated based on the optimized weights, and each node in the path reflects the optimal configuration under different operating conditions. The generated path is further adjusted after performance verification to ensure that the optimization effect meets expectations.

[0191] Through the dynamic weight adjustment of the linear weighted method and the optimization of the gradient descent method, the key optimization targets of the industrial control motherboard under different operating conditions can be accurately reflected, and the flexibility and adaptability of the comprehensive optimization model can be improved. This method ensures that the generation of the optimization path can quickly respond to real-time needs, thereby improving the stability and energy efficiency of the industrial control motherboard performance, while effectively reducing the failure rate.

[0192] Example 1: Performance stability objective function f of an industrial control motherboard when running under high load 1 is 0.2, and the energy efficiency improvement objective function f 2 is 0.8, the failure rate reduction objective function f 3 is 0.1. The initial weight w 1 =5,w 2 =1.25, w 3 = 10. The weight is optimized by gradient descent method. After 10 iterations, the final weight is adjusted to w 1 =4.8,w 2 =1.5,w 3 =9.6. The optimized comprehensive objective function value is 15% lower than the initial value. The generated optimization path reduces the voltage fluctuation by 12% and improves the energy efficiency by 8%.

[0193] Example 2: Under low load conditions, the weight of the objective function for reducing the failure rate of a motherboard increased significantly. Through dynamic weight adjustment and optimization, it was found that the energy efficiency improvement had little impact on the system operation status. The optimized path focused on optimizing the heat dissipation configuration and load distribution, improving the stability of the system and reducing the failure rate by 50%.

[0194] Based on the optimization feedback data, the causal relationship model is updated, and the dynamic adaptability of the optimization path is optimized through a collaborative learning mechanism. The collaborative learning mechanism adopts a distributed learning framework, integrates multiple rounds of test data through federated learning, and combines a reinforcement learning algorithm to enhance the dynamic adjustment capability of the optimization path, and generates closed-loop optimization feedback data to further improve the effect of performance detection and optimization.

[0195] In the present invention, the optimization feedback data is used to dynamically update the causal relationship model, and by continuously correcting the causal path weights in the model, it is ensured that the causal network can adapt to the actual operating state of the industrial control motherboard. At the same time, by introducing a collaborative learning mechanism, the dynamic adaptive adjustment of the optimization path is achieved, ensuring that the optimization model has the ability to operate efficiently in multiple scenarios. The collaborative learning mechanism adopts a distributed learning framework, uses federated learning to integrate multiple rounds of data from different test equipment, combines the reinforcement learning algorithm to optimize the path adjustment strategy, and finally generates closed-loop optimization feedback data.

[0196] Federated learning ensures the privacy and security of data at each node through distributed data aggregation. At the same time, it optimizes path adjustment in the ever-changing operating environment of industrial control motherboards through reinforcement learning algorithms, improving the model's adaptability to actual scenarios. Closed-loop optimization feedback data provides input for the next round of optimization cycles, further improving performance detection and optimization effects.

[0197] Optimization feedback data includes adjustment records of optimization path points, performance improvement indicators, characteristic parameter changes, etc. Through feedback data, the impact of nodes in the optimization path on the overall performance is analyzed.

[0198] Update the causal relationship model and adjust the causal path weight β according to the optimization feedback data ij :

[0199] β ij =β ij +Δβ ij

[0200] Among them, Δβ ij Determined by the gain change of the path in the feedback data, the weight of the path with high gain is increased, and the weight of the path with low gain is appropriately reduced. The updated causal model can more accurately reflect the causal relationship between multimodal features under the current operating state.

[0201] The distributed learning framework runs the optimization model in parallel on multiple test devices, and each device generates and updates the optimization path locally. Through federated learning, the optimization results of each device are encrypted and transmitted to the central server for data aggregation and model update. The federated learning aggregation formula is:

[0202]

[0203] Among them, θ is the global model parameter, θ k is the local model parameter of the kth device, and N is the total number of devices.

[0204] Reinforcement learning optimizes path adjustment. Based on federated learning, the reinforcement learning algorithm is combined to optimize the path adjustment strategy. Through the Deep Deterministic Policy Gradient (DDPG) algorithm, the optimization path is dynamically adjusted in the continuous action space. The reward function of reinforcement learning is defined as:

[0205] R=α·Δstability+β·Δenergy efficiency-γ·Δfailure rate

[0206] Among them, α, β, γ are weight coefficients, and Δ represents the performance change before and after optimization.

[0207] Through the collaborative learning mechanism, the optimization path adjusted by reinforcement learning is integrated with the performance verification results to generate closed-loop optimization feedback data. The closed-loop optimization feedback data includes path adjustment records, causal model update parameters, performance improvement indicators, etc., providing data support for the next round of optimization and realizing the cyclic iteration of the optimization process.

[0208] Preferably, Figure 4 As shown, the collaborative learning mechanism is implemented by the following steps:

[0209] Collect optimized path data from different test devices, encrypt it and transmit it to the central server;

[0210] Aggregate data in a central server to build a federated learning model;

[0211] Distribute the trained model parameters to each test device to achieve distributed optimization path adjustment.

[0212] The present invention realizes the distributed dynamic adjustment of the optimization path of the industrial control motherboard through a collaborative learning mechanism. Collaborative learning adopts a federated learning architecture, which aims to integrate the optimization path data generated by multiple test devices while ensuring data privacy and security. Federated learning improves the generalization ability and adaptability of the model by encrypting data transmission, aggregating and distributing optimization model parameters on a central server. By integrating the local optimization path of each test device, the impact of different operating environments on the performance of the industrial control motherboard can be more effectively captured, thereby providing a global optimization basis for path adjustment.

[0213] Each test device runs the optimization model locally to generate optimized path data, including adjustment records of path points and corresponding performance indicators. The path data is encrypted using a data encryption method based on homomorphic encryption to ensure that it cannot be decrypted or tampered with during transmission. The encrypted data is transmitted to the central server via a secure communication protocol (such as the TLS protocol).

[0214] The central server receives the encrypted path data from each test device and aggregates the local model parameters using the Federated Averaging (FedAvg) algorithm. The formula is as follows:

[0215]

[0216] Among them, θ is the global model parameter, θ k is the local model parameter of the kth device, w k is the data weight of the corresponding device, and N is the total number of devices. The aggregated federated learning model can integrate the optimized path data of multiple devices and adapt to the performance requirements of different operating conditions.

[0217] The aggregated global model parameters are distributed to each test device by the central server, and each device updates the local optimization path according to the global model. When updating the path, the reinforcement learning strategy is combined to dynamically optimize the path adjustment to ensure that each device can quickly respond to optimization needs under a specific operating environment. The policy gradient update formula of reinforcement learning is:

[0218]

[0219] Among them, η is the learning rate and R is the cumulative reward function of the optimization path.

[0220] Through the collaborative learning mechanism, the optimization path data of different test devices can be effectively integrated, and the generated federated learning model has strong generalization ability and can adapt to various operating conditions. Distributed optimization path adjustment ensures that each device dynamically optimizes performance according to its own environment, while the global aggregation of the central server improves the overall performance detection and optimization effect of the model. This distributed learning architecture fully guarantees the system in terms of data privacy and security.

[0221] Example 1: The performance of an industrial control motherboard is optimized under three operating environments: environment 1 is high temperature and high humidity, environment 2 is low temperature and low humidity, and environment 3 is normal temperature and normal humidity. The optimization paths generated locally by each test device have significant differences in performance stability and energy efficiency improvement. Through the collaborative learning mechanism, the optimization path data under the three environments are transmitted to the central server for encrypted aggregation to generate a global optimization model. After the model parameters are distributed to each device, the performance improvement of each device in its specific environment is more than 15%, and the convergence speed of the optimization path is increased by 30%.

[0222] Example 2: In the parallel operation test of multiple industrial control motherboards, the distributed optimization path adjustment dynamically optimizes the path of each device through the reinforcement learning strategy. Based on the central server aggregation, the optimization result of the failure rate reduction objective function is improved by 20% compared with the initial state, and the variance of the performance stability objective function is reduced by 25%. The optimized system shows higher stability and adaptability under different load conditions.

[0223] Preferably, the federated learning model is combined with deep reinforcement learning, iteratively updates the path adjustment strategy through a policy gradient optimization algorithm, and automatically selects a suitable optimization path in a new test scenario.

[0224] The present invention optimizes the performance path adjustment strategy of the industrial control motherboard by combining the federated learning model with deep reinforcement learning. Federated learning is responsible for integrating the optimization data from different test devices to generate a model with global adaptability, while deep reinforcement learning iteratively adjusts the optimization path through the policy gradient optimization algorithm to dynamically adapt to new test scenarios. This method not only improves the efficiency and adaptability of path optimization, but also ensures the performance stability of the model in diversified scenarios.

[0225] The policy gradient optimization algorithm is a reinforcement learning method that aims to optimize the strategy by maximizing the cumulative reward function. During the path adjustment process, deep reinforcement learning selects the optimal path point by balancing exploration and utilization, so that the performance goals of the industrial control motherboard can be achieved in different operating environments.

[0226] Run the optimization model on multiple test devices and collect optimization path data under different operating conditions, including the selection of path points and corresponding performance improvement indicators. Use the federated average algorithm to aggregate local model parameters and generate a federated learning model. The formula is:

[0227] Based on the federated learning model, deep reinforcement learning is introduced, and the policy gradient optimization algorithm is used to adjust the path. The optimization goal is to maximize the cumulative reward function R, and the optimization strategy is updated by the following formula:

[0228] The design of the reward function comprehensively considers performance stability, energy efficiency improvement and failure rate reduction, and is defined as: R = α·Δstability + β·Δenergy efficiency - γ·Δfailure rate.

[0229] In the new test scenario, the deep reinforcement learning model selects the optimal path point based on the current state and reward feedback. After each iteration, the path point selection results are compared with the actual performance improvement to further update the model parameters to ensure the dynamic adaptability of path optimization.

[0230] After the combination of federated learning and deep reinforcement learning, the adjustment results of the optimization path are fed back to the central server to update the causal relationship model and form a closed-loop optimization cycle. The closed-loop feedback data is used to guide the next round of optimization and improve the accuracy and applicability of the model.

[0231] Through the combination of federated learning and deep reinforcement learning, the present invention realizes dynamic adjustment of the optimization path and enhanced global adaptability. In the new test scenario, the optimization path can automatically adapt to environmental changes and achieve comprehensive optimization of performance stability, energy efficiency and failure rate. Federated learning ensures the privacy and security of data, while deep reinforcement learning improves the real-time response capability and iteration efficiency of path optimization.

[0232] Example 1: In the test of industrial control motherboards under three different operating conditions, federated learning integrated the optimization data of high temperature, high humidity and low load equipment to generate a global optimization model. The path adjustment strategy was optimized through deep reinforcement learning, and the path points that prioritized reducing the temperature rise rate were selected in the high temperature scenario, which ultimately reduced the temperature rise by 20% and improved the performance stability by 15%.

[0233] Example 2: When an industrial control motherboard runs in a high-load environment, the energy efficiency improvement objective function value of the initial optimization path is 0.75, and the failure rate objective function value is 5. Through deep reinforcement learning to optimize the path adjustment strategy, the model selects a path that balances stability and energy efficiency in the new scenario. The results show that the optimized path improves energy efficiency to 0.85, the failure rate target to 8, and the variance of the performance stability target is reduced by 10%.

[0234] like Figure 5 As shown, a system for implementing the performance detection and analysis method of the industrial control motherboard includes:

[0235] The data acquisition module is used to collect multi-modal data during the operation of the industrial control mainboard, including electrical signal data, acoustic signal data, thermal imaging data and optical signal data;

[0236] A data preprocessing module, used for performing noise filtering, time alignment and normalization processing on the multimodal data to generate performance characteristic data;

[0237] A causal relationship modeling module, used to build a causal relationship model based on the performance feature data, extract the causal relationship between features, and generate causal reasoning network data according to the weight adjustment of multimodal features, wherein the causal reasoning network data includes effective causal relationships and abnormal impact paths;

[0238] An optimization model building module is used to build a multi-objective optimization model based on the causal reasoning network data, generate an optimization path by quantitatively defining the performance stability, energy efficiency improvement and failure rate reduction goals, and dynamically adjusting the weights between the goals;

[0239] A performance verification module is used to perform performance verification on the optimization path, adjust the optimization path according to the verification result, and generate optimization feedback data;

[0240] The collaborative learning module is used to update the causal relationship model based on the optimization feedback data, integrate multiple rounds of test data through a distributed learning framework and a federated learning mechanism, and combine the dynamic adaptability of the optimization path of the reinforcement learning algorithm to generate closed-loop optimization feedback data.

[0241] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0242] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A performance detection and analysis method for an industrial control motherboard, characterized in that: The following steps are involved: Collect multi-modal data of the industrial control motherboard during operation, pre-process the data including electrical signals, acoustic signals, thermal imaging and optical signals, and generate performance characteristic data; Based on the performance characteristic data, a causal relationship model is constructed using a structural equation modeling method to extract the causal relationship between the characteristics, and combined with the weight adjustment of multimodal features, causal reasoning network data is generated, wherein the causal reasoning network data includes effective causal relationships and abnormal impact paths; According to the causal reasoning network data, a multi-objective optimization model is constructed, and an optimization path is generated by quantitatively defining the performance stability, energy efficiency improvement and failure rate reduction goals, and dynamically adjusting the weights between the goals to balance the priorities; the performance of the optimization path is verified, and the optimization path is adjusted according to the verification results, and optimization feedback data is output; Based on the optimization feedback data, the causal relationship model is updated, and the dynamic adaptability of the optimization path is optimized through a collaborative learning mechanism. The collaborative learning mechanism adopts a distributed learning framework, integrates multiple rounds of test data through federated learning, and combines a reinforcement learning algorithm to enhance the dynamic adjustment capability of the optimization path, and generates closed-loop optimization feedback data to further improve the effect of performance detection and optimization.

2. The method according to claim 1, characterized in that The preprocessing of the multimodal data comprises the following steps: The electrical signal data is corrected for baseline drift, the signal baseline is calculated by the sliding window method and differential adjustment is performed; Band-pass filtering is performed on the acoustic signal data, and a finite impulse response filter is used to filter the signal in the range of 20 Hz to 20 kHz; Perform median filtering on the thermal imaging data, select a 3x3 window to calculate the median point by point to eliminate random noise in the image; The optical signal data is normalized and the signal amplitude is adjusted to the [0,1] interval using the maximum and minimum value normalization method.

3. The method according to claim 2, characterized in that The bandpass filtering of the acoustic signal data comprises the following steps: Convert the time domain signal into the frequency domain signal through fast Fourier transform; In the frequency domain, keep the frequency band from 20Hz to 20kHz and set other frequency bands to zero; The signal is reconstructed from the frequency domain to the time domain using an inverse Fourier transform to generate filtered acoustic signal data.

4. The method according to claim 1, characterized in that: The construction of the causal relationship model includes the following steps: Select key parameters in the performance characteristic data, including voltage fluctuation amplitude, temperature rise rate, vibration acceleration and optical interference intensity; Calculate conditional probabilities based on performance characteristic parameters and generate causal graph structures using Bayesian network methods; The parameter weights in the causal graph are optimized by maximum likelihood estimation, and the edges with causal strength lower than a preset threshold are removed to determine the causal path.

5. The method according to claim 4, characterized in that The optimization of the causal path comprises the following steps: Calculate the probability gain of each causal path and set the minimum threshold of causal strength; Eliminate low-gain paths and retain high-gain paths to reduce the size of the causal network; The conditional probabilities are recalculated for the remaining paths, and the path strengths are adjusted dynamically.

6. The method according to claim 1, characterized in that The multi-objective optimization model includes the following objective functions: Performance stability objective function, which calculates the stability of performance indicators through variance; Energy efficiency improvement objective function, which calculates energy efficiency by the ratio of input power to output power; The failure rate reduction objective function calculates the failure rate optimization target by the inverse ratio of the failure frequency; The linear weighted method is used to comprehensively calculate each objective function to generate the optimized path.

7. The method according to claim 6, characterized in that The linear weighting method comprises the following steps: Dynamically adjust weights based on real-time calculation results of each objective function; The weight updates are optimized using gradient descent to ensure that the optimization path is responsive to key objectives.

8. The method according to claim 1, characterized in that: The collaborative learning mechanism is implemented by the following steps: Collect optimized path data from different test devices, encrypt it and transmit it to the central server; Aggregate data in a central server to build a federated learning model; Distribute the trained model parameters to each test device to achieve distributed optimization path adjustment.

9. The method according to claim 8, characterized in that The federated learning model combines deep reinforcement learning, iteratively updates the path adjustment strategy through the policy gradient optimization algorithm, and automatically selects a suitable optimization path in a new test scenario.

10. A system for implementing the performance detection and analysis method of the industrial control motherboard according to any one of claims 1 to 9, characterized in that: include: The data acquisition module is used to collect multi-modal data during the operation of the industrial control mainboard, including electrical signal data, acoustic signal data, thermal imaging data and optical signal data; A data preprocessing module, used for performing noise filtering, time alignment and normalization processing on the multimodal data to generate performance characteristic data; A causal relationship modeling module, used to build a causal relationship model based on the performance feature data, extract the causal relationship between features, and generate causal reasoning network data according to the weight adjustment of multimodal features, wherein the causal reasoning network data includes effective causal relationships and abnormal impact paths; An optimization model building module is used to build a multi-objective optimization model based on the causal reasoning network data, generate an optimization path by quantitatively defining the performance stability, energy efficiency improvement and failure rate reduction goals, and dynamically adjusting the weights between the goals; A performance verification module is used to perform performance verification on the optimization path, adjust the optimization path according to the verification result, and generate optimization feedback data; The collaborative learning module is used to update the causal relationship model based on the optimization feedback data, integrate multiple rounds of test data through a distributed learning framework and a federated learning mechanism, and combine the dynamic adaptability of the optimization path of the reinforcement learning algorithm to generate closed-loop optimization feedback data.

Images