Electronic component operation state monitoring method and system based on artificial intelligence

Abstract

The invention relates to the technical field of operation state monitoring, and discloses an electronic component operation state monitoring method and system based on artificial intelligence, and the method comprises the steps: carrying out the evaluation of the operation life of an electronic component through an operation life evaluation model, obtaining the operation life evaluation result of the electronic component and the uncertainty of the operation life evaluation result; and carrying out multi-source fusion on the multi-source sensing detection data in combination with the operation life evaluation result of the electronic component, carrying out deep feature extraction on the multi-source fusion result in combination with the uncertainty of the operation life evaluation result, obtaining operation state features, and carrying out feature mapping to obtain a real-time operation state. According to the method, multi-source data such as spectrums, sound waves and vibration are fused, through uncertainty modeling and multi-damage feature extraction, accurate evaluation of the service life of the electronic component and real-time monitoring and correction of the state are achieved, the recognition capacity of faults such as unbalance, resonance and bearing damage is improved, and the monitoring accuracy of the operation state is improved.

Description

Technical Field

[0001] The present invention relates to the field of operating state monitoring, and particularly to a method and system for monitoring the operating state of electronic components based on artificial intelligence. Background Art

[0002] Electronic components are widely used in key fields such as aerospace, automotive electronics, communication equipment, medical instruments, and industrial control. They are the core foundation for the operation of modern electronic systems, and their reliability directly affects the safety and stability of the entire system. With the continuous improvement of the integration level of electronic systems, components operate under complex working conditions such as high temperature, high pressure, strong vibration, and electromagnetic interference for a long time, and are extremely prone to failures such as performance degradation, contact aging, and material fatigue. Therefore, carrying out the monitoring of the operating state and life assessment of electronic components has become a key technical path to ensure system safety, extend the service life, and reduce maintenance costs. At present, in the field of electronic component state monitoring, there are already some research methods and patented technologies at home and abroad, mainly including monitoring methods based on statistical process control and monitoring methods based on electrical parameter characteristics. For example, in the patent application CN102361014A, statistical techniques such as principal component analysis (PCA) and mean square prediction error (SPE) are used to monitor the state and diagnose faults in the semiconductor manufacturing process. However, this method has limitations in processing high-dimensional, non-linear, and multi-modal data and is difficult to effectively capture the changes in the operating state of complex systems. For example, in the patent application CN209690455U, measurable signals such as voltage and current are used, combined with the switching characteristics and junction temperature changes of the device, to achieve online state monitoring and fault determination of power semiconductor devices. Although this method improves the accuracy of monitoring, its adaptability in dealing with multi-source heterogeneous data and complex operating conditions still needs to be improved. The above methods have achieved the evaluation of the operating state of components and fault early warning to a certain extent, but there are still the following main technical bottlenecks: most methods only rely on electrical parameters or time series signals and cannot comprehensively reflect the multi-dimensional operating state of the device; some methods rely on empirical rules or fixed thresholds and are difficult to adapt to complex and dynamically changing working conditions; there is still a lack of a unified data fusion framework to integrate multi-modal signals such as spectroscopy, heat, sound, and vibration. Summary of the Invention

[0003] The present invention provides a method for monitoring the operating state of electronic components based on artificial intelligence. By fusing multi-source detection data and uncertainty modeling, key damage features are extracted to achieve accurate assessment of the operating life of electronic components and correction of state monitoring, and improve the fault early warning ability and the reliability of the operating state monitoring results.

[0004] To achieve the above object, a method for monitoring the operating state of electronic components based on artificial intelligence provided by the present invention includes the following steps: S1: Collect multi-source detection data of electronic components using multiple sensors and perform standardization processing to obtain multi-source sensing detection data of the electronic components; S2: Use the operating life evaluation model to evaluate the operating life of the electronic components to obtain the operating life evaluation result of the electronic components and the uncertainty of the operating life evaluation result; S3: Combine the operating life evaluation result of the electronic components to perform multi-source fusion on the multi-source sensing detection data to obtain a multi-source fusion result, and perform deep feature extraction on the multi-source fusion result in combination with the uncertainty of the operating life evaluation result to obtain the operating state characteristics of the electronic components; S4: Perform feature mapping on the operating state characteristics of the electronic components to obtain the real-time operating state of the electronic components, obtain the historical operating state sequence of the electronic components, and perform monitoring and correction on the real-time operating state of the electronic components to obtain the real-time operating state monitoring result of the electronic components.

[0005] As a further improvement method of the present invention: Optionally, collecting multi-source detection data of electronic components using multiple sensors includes; The sensors include current / voltage sensors, infrared thermal imagers, accelerometers, ultrasonic sensors, and spectrometers, and the multi-source detection data includes the current value, voltage value, heat value, vibration data, acoustic wave data, and spectral analysis data of the electronic components; Use the current / voltage sensor to collect the current value and voltage value of the electronic component respectively, use the infrared thermal imager to collect the heat value of the electronic component, use the ultrasonic sensor to collect the acoustic wave data of the electronic component, use the accelerometer to collect the vibration data of the electronic component, and use the spectrometer to collect the spectral analysis data of the electronic component; The current value, voltage value, and heat value are sequence data, the acoustic wave data is signal data, the vibration data is acceleration sequence data of the electronic component in a fixed direction, and the spectral analysis data is the intensity value of the electronic component at different wavelengths; specifically, the fixed direction is the direction perpendicular to the surface of the electronic component; Perform standardization processing on the multi-source detection data. The standardization processing methods for the current value, voltage value, acoustic wave data, and heat value are normalization processing. The standardization processing results of the current value, voltage value, and heat value are the sequence data after normalization processing. The standardization processing result of the acoustic wave data is the signal data after normalization processing. Extract the standardization features of the vibration data and spectral analysis data as the standardization processing results of the vibration data and spectral analysis data.

[0006] Optionally, extracting the standardization features of the vibration data and spectral analysis data includes: Perform a fast Fourier transform on the vibration data to obtain the spectral characteristics of the vibration data. The spectral characteristics are in the form of a spectrogram, where the horizontal axis of the spectral characteristics is frequency and the vertical axis is amplitude; Successively extract the unbalance feature, bearing damage feature, and resonance feature from the spectral characteristics as the standardized features of the vibration data. The unbalance feature is the amplitude neighborhood information of the first natural frequency in the vibration data. The bearing damage feature is the amplitude neighborhood information of multiple damage feature frequencies in the vibration data. The resonance feature is the amplitude neighborhood information of the second natural frequency in the vibration data; Specifically, the amplitude neighborhood information of frequency f is an amplitude sequence composed of the amplitudes of the frequencies E before and after frequency f as the center in the spectral characteristics; The first natural frequency is the rotational speed frequency of the electronic component. The second natural frequency is twice the rotational speed frequency of the electronic component. The damage feature frequency is the structural frequency of the electronic component. The damage feature frequency is calculated based on the structural information of the electronic component. The damage feature frequency includes the outer race fault frequency, inner race fault frequency, and rolling fault frequency. The structure of the electronic component includes rolling elements and bearings. The structural information includes the number of rolling elements, the diameter of the rolling elements, the diameter of the bearings, and the contact angle between the rolling elements and the bearing surface; Calculate the peak area and the maximum curvature of the spectral analysis data as the standardized features of the spectral analysis data. Specifically, the form of the spectral analysis data is: ; Where: represents the s-th wavelength, represents the intensity value of the electronic component at wavelength s represents the total number of wavelengths, represents s wavelengths, and the values of the wavelengths increase in sequence; The peak area of the spectral analysis data is : ; The maximum curvature of the spectral analysis data is : ; ; Where: represents the mean value of the wavelength differences; represents selecting the wavelength that makes reach the maximum as the maximum curvature, represents the intensity value of the electronic component at wavelength The intensity value under represents the th wavelength, , s represents the total number of wavelengths, represents the intensity value of the electronic component at the wavelength under, represents the intensity value of the electronic component at the wavelength under, successively represent the th wavelength and the th wavelength respectively.

[0007] Optionally, the operating life evaluation model includes an input layer, a multi-damage attention branch module, a resonance perception module, a time-frequency convolution module, a multi-modal gating fusion module, and an output layer; The input layer is used to receive the physical state data of the electronic component. The multi-damage attention branch module is used to receive the standardized processing result of the vibration data, extract the unbalance feature and the bearing damage feature, construct an MLP + attention channel for the unbalance feature and the amplitude domain information of different damage feature frequencies respectively, and weight different channels using the attention weight to obtain a multi-channel weighted damage feature. The resonance perception module performs an exponential mapping on the resonance feature based on the second-order frequency amplification factor to obtain a resonance amplification feature. The time-frequency convolution module performs a fast Fourier transform on the standardized processing result of the acoustic wave data to obtain an acoustic wave spectrum feature, and performs a convolution process with a partial frequency mask on the acoustic wave spectrum feature to obtain a time-frequency convolution feature representing the local amplitude change of the spectrum. The multi-modal gating fusion module is used to generate a gating weight for the output features of the multi-damage attention branch module, the resonance perception module, and the time-frequency convolution module, and performs a gating weighted fusion on the multi-channel weighted damage feature, the resonance amplification feature, and the time-frequency convolution feature using the gating weight to obtain a gating fusion feature. The output layer is composed of a residual dense connection network and an uncertainty perception head. The output layer performs a residual dense connection and an uncertainty perception optimization process on the gating fusion feature to generate an operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result.

[0008] Specifically, the loss function of the operating life evaluation model is: ; Where: represents the loss function of the operating life evaluation model, represents the model parameters of the operating life evaluation model, represents the uncertainty of the nth training sample, represents the operating life evaluation result of the nth training sample, , where N represents the total number of training samples, represents the true remaining operating life of the nth training sample.

[0009] Optionally, the operating life of the electronic component is evaluated using the operating life evaluation model, including: The exponential mapping formula of the resonance perception module for the resonance feature based on the second-order frequency amplification factor is: ; Where: represents the second-order frequency amplification factor, represents the resonance feature, represents the resonance amplification feature; specifically: ; The generation process of the gating weight is: Calculate the fusion guiding factors of the multi-channel weighted damage feature , resonance amplification feature and time-frequency convolution feature , the fusion guiding factors of the multi-channel weighted damage feature , resonance amplification feature and time-frequency convolution feature are successively and , and the calculation formula of the fusion guiding factor is: ; Where: represents the L2 norm; Use the gating activation function to perform convolution processing on the multi-channel weighted damage feature , resonance amplification feature and time-frequency convolution feature , and use the fusion guiding factor as the bias to generate the gating weights of the multi-channel weighted damage feature , resonance amplification feature and time-frequency convolution feature ; the generation formula of the gating weights of the multi-channel weighted damage feature , resonance amplification feature and time-frequency convolution feature is: ; ; ; Where: represents the gated convolution parameter, and b represents the gated bias. represents the gated activation function, and the selected gated activation function is the Sigmoid function; The multi-channel weighted damage feature , the resonance amplification feature and the time-frequency convolution feature have gated weights in sequence as and ; The output layer performs residual dense connection and uncertainty-aware optimization processing on the gated fusion feature z to generate the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result. The process is as follows: The residual dense connection network includes L residual blocks and a dense connection block. The gated fusion feature z is subjected to residual convolution processing using L residual blocks to obtain the L-layer residual features of the gated fusion feature z. The L-layer residual feature of the gated fusion feature z is: ; Where: represents the L-layer residual feature of the gated fusion feature z, represents the convolution parameter of the L-th residual block, represents the bias of the L-th residual block, represents the ReLU function; The dense connection block is used to perform dense connection on the L-layer residual features and generate the operating life evaluation result of the electronic component: ; Where: represents the operating life evaluation result of the electronic component, represents the dense connection convolution parameter, represents the dense connection bias, represents the operating life evaluation convolution parameter; Generate the uncertainty of the operating life evaluation result : ; Where: represents the uncertainty evaluation convolution parameter.

[0010] Optionally, obtain the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result , and perform multi-source fusion on the multi-source sensing detection data in combination with the operating life evaluation result of the electronic component, including: Obtain the multi-source sensing detection data, and use the operating life evaluation model to receive the physical state data of the electronic component and generate a physical state feature vector, where the physical state feature vector includes the multi-channel weighted damage feature, resonance amplification feature, and time-frequency convolution feature calculated in step S2; Perform convolution processing on the electrothermal state data in the multi-source sensing detection data to obtain an electrothermal state feature vector; Take the operating life evaluation result of the electronic component as the Query vector, calculate the attention of the physical state feature vector and the electrothermal state feature vector respectively, and perform attention weighting on the physical state feature vector and the electrothermal state feature vector to obtain the multi-source fusion result F.

[0011] Optionally, combine the uncertainty of the operating life evaluation result Perform deep feature extraction on the multi-source fusion result F to obtain the operating state features of the electronic component, including: Perform standard deep feature extraction on the multi-source fusion result F to obtain the standard deep features of the multi-source fusion result F : ; Where: BN represents the batch normalization layer, represents the convolution weight of the batch normalization layer, represents the bias of the batch normalization layer, represents the ReLU function; Based on the uncertainty of the operating life evaluation result , calculate the suppression degree g of the standard deep feature : ; Where: represents the suppression coefficient, represents the suppression control parameter; represents the Sigmoid function; Based on the suppression degree g, perform feature modulation on the standard deep feature to obtain the operating state features of the electronic component : ; Where: represents element-wise multiplication. Specifically, introducing the operating life evaluation result to suppress the standard deep feature, when the uncertainty is large, it will dynamically weaken or re-weight the response of some channels to avoid the interference of low-confidence features on the final evaluation result.

[0012] Optionally, a fully connected layer is constructed to perform feature mapping on the operating state characteristics of the electronic component to obtain the real-time operating state A of the electronic component. The fully connected layer includes fully connected convolution parameters and fully connected bias amounts. Obtaining the historical operating state sequence of the electronic component and monitoring and correcting the real-time operating state of the electronic component includes: Specifically, the real-time operating state is within the range of 0-1. The higher the value of the real-time operating state, the better the working performance of the electronic component; The historical operating state sequence of the electronic component is the operating state after monitoring and correction of the electronic component at M historical moments. The monitoring and correction formula for the real-time operating state A is: ; Where: represents the monitoring and correction result of the real-time operating state A and serves as the real-time operating state monitoring result of the electronic component, represents the L2 norm, k represents the correction control coefficient, represents the uncertainty of the operating life evaluation result, represents the mean value of the historical operating state sequence.

[0013] To solve the above problems, the present invention provides a monitoring system for the operating state of an electronic component, characterized in that the monitoring system for the operating state of the electronic component includes a server and a data acquisition device. The server includes an operating life evaluation module and an operating state monitoring module: The operating life evaluation module is used to evaluate the operating life of the electronic component by using an operating life evaluation model to obtain the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result; The operating state monitoring module is used to perform multi-source fusion on multi-source sensing detection data by combining the operating life evaluation result of the electronic component to obtain a multi-source fusion result, and perform deep feature extraction on the multi-source fusion result by combining the uncertainty of the operating life evaluation result to obtain the operating state characteristics of the electronic component, perform feature mapping on the operating state characteristics of the electronic component to obtain the real-time operating state of the electronic component, and obtain the historical operating state sequence of the electronic component, and monitor and correct the real-time operating state of the electronic component to obtain the real-time operating state monitoring result of the electronic component; The data acquisition device is used to collect multi-source detection data of the electronic component by using a variety of sensors and perform standardization processing to obtain multi-source sensing detection data of the electronic component.

[0014] To solve the above problems, the present invention also provides an electronic device, which includes: A memory that stores at least one instruction; A communication interface that enables the electronic device to communicate; and a processor that executes the instructions stored in the memory to implement the above-mentioned method for monitoring the operating state of electronic components based on artificial intelligence.

[0015] To solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned method for monitoring the operating state of electronic components based on artificial intelligence.

[0016] Compared with the prior art, the present invention proposes a method for monitoring the operating state of electronic components based on artificial intelligence, and this technology has the following advantages: First, this solution proposes a method for evaluating the operating life of electronic components. It collects multi-source detection data, standardizes the detection data representing the physical state of electronic components, extracts features characterizing the imbalance degree, resonance phenomenon, and bearing damage of electronic components, and uses an operating life evaluation model to evaluate the operating life of electronic components. The operating life evaluation model includes modules of various forms. The multi-damage attention branch module is specifically constructed for vibration signals, and amplitude-domain information of the imbalance feature and bearing damage feature frequencies is extracted respectively. Through the MLP + channel attention mechanism, weighted modeling is performed on different damage features, which can finely depict the performance characteristics of various damages. The resonance perception module enhances the sensitivity of the second-order resonance region through the exponential mapping method based on the second-order frequency amplification factor, making it easier for the model to capture tiny abnormal signals under resonance phenomena, especially suitable for fault warning of high-speed electronic components. The time-frequency convolution module first obtains the acoustic wave spectrum features through FFT, and then introduces local frequency masking + convolution to extract local fluctuation features, improving the model's recognition ability for high-frequency tiny fluctuations in the acoustic field, which is beneficial for the detection of early defects such as material micro-cracks and structural looseness. The multi-modal gating fusion module designs a gating weight generation mechanism for features from different signal sources (vibration, acoustic wave, resonance), adaptively learning the importance of each modality in different states. The residual dense connection network enhances the model's feature transfer and non-linear modeling capabilities, avoiding gradient disappearance and being suitable for processing deep structures. The uncertainty perception head not only outputs the life prediction value but also provides the prediction uncertainty, making the output result more credible and facilitating subsequent risk assessment and maintenance decision-making. And the learning rate update strategy is adjusted, adopting a learning rate warming strategy for early learning rate changes, making the model not oscillate in the initial stage of training, and introducing the gradient of model parameters. Only when the gradient fluctuation of the model is "stable" enough, the learning rate is rapidly increased, making it easier to find the gradient stable interval in the early iteration and helping the model to converge more efficiently. Moreover, in the initial stage of network training, the parameters have not converged, and the gradient usually fluctuates greatly or even does not converge. If a large learning rate is directly used at this time, it will cause the loss function to oscillate or even diverge. By adopting the updated learning rate update strategy, gradient explosion or oscillation is avoided.

[0017] Meanwhile, this solution proposes a method for monitoring the operating state, which conducts attention guidance based on the results of the operating life assessment, enabling the multi-source fusion process to extract effective features related to the degradation trend and operating life in a targeted manner instead of blindly extracting features. During the process of deep feature extraction, the uncertainty of the operating life assessment results is introduced to suppress the standard deep features. When the uncertainty is large, the responses of certain channels will be dynamically weakened or reweighted to avoid the interference of low-confidence features on the final assessment results. The introduction of the historical operating state sequence for operating state monitoring correction helps to identify trend changes, provides a more meaningful discriminant basis, and introduces uncertainty during the correction process. When the uncertainty is high, more reliance is placed on the historical sequence, thereby obtaining high-quality, task-related operating features and interpretable, real-time state monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 FIG. is a schematic flow chart of a method for monitoring the operating state of electronic components based on artificial intelligence provided by an embodiment of the present invention; Figure 2 FIG. is a functional module diagram of an operating state monitoring system for electronic components provided by an embodiment of the present invention; Figure 2 In the figure: 100 is an operating state monitoring system for electronic components, 101 is an operating life assessment module, 102 is an operating state monitoring module, and 103 is a data acquisition device; The implementation, functional features, and advantages of the objectives of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0020] An embodiment of the present application provides a method for monitoring the operating state of electronic components based on artificial intelligence. The execution subject of the method for monitoring the operating state of electronic components based on artificial intelligence includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for monitoring the operating state of electronic components based on artificial intelligence can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0021] Referring to Figure 1 , Embodiment 1 of the present invention is as follows: S1: Use a variety of sensors to collect multi-source detection data of electronic components and perform standardization processing to obtain multi-source sensing detection data of electronic components.

[0022] Collect multi-source detection data of electronic components using multiple sensors, including; The sensors include a current / voltage sensor, an infrared thermal imager, an accelerometer, an ultrasonic sensor, and a spectral analyzer. The multi-source detection data includes the current value, voltage value, heat value, vibration data, acoustic wave data, and spectral analysis data of the electronic components; Use the current / voltage sensor to collect the current value and voltage value of the electronic component respectively, use the infrared imager to collect the heat value of the electronic component, use the ultrasonic sensor to collect the acoustic wave data of the electronic component, use the accelerometer to collect the vibration data of the electronic component, and use the spectral analyzer to collect the spectral analysis data of the electronic component; The current value, voltage value, and heat value are sequence data, the acoustic wave data is signal data, the vibration data is the acceleration sequence data of the electronic component in a fixed direction, and the spectral analysis data is the intensity value of the electronic component at different wavelengths; specifically, the fixed direction is the direction perpendicular to the surface of the electronic component; Perform standardization processing on the multi-source detection data. The standardization processing methods for the current value, voltage value, acoustic wave data, and heat value are normalization processing. The standardization processing results of the current value, voltage value, and heat value are the sequence data after normalization processing. The standardization processing result of the acoustic wave data is the signal data after normalization processing. Extract the standardized features of the vibration data and spectral analysis data as the standardization processing results of the vibration data and spectral analysis data.

[0023] Extracting the standardized features of the vibration data and spectral analysis data includes: Perform a fast Fourier transform on the vibration data to obtain the frequency spectrum features of the vibration data. The frequency spectrum features are in the form of a spectrum. The horizontal axis of the frequency spectrum features is the frequency, and the vertical axis is the amplitude; Successively extract the imbalance feature, bearing damage feature, and resonance feature from the frequency spectrum features as the standardized features of the vibration data. The imbalance feature is the amplitude neighborhood information of the first natural frequency in the vibration data. The bearing damage feature is the amplitude neighborhood information of multiple damage characteristic frequencies in the vibration data. The resonance feature is the amplitude neighborhood information of the second natural frequency in the vibration data; Specifically, the amplitude neighborhood information of frequency f is the amplitude sequence composed of the amplitudes of the frequencies before and after frequency f centered on frequency f in the frequency spectrum features; The first natural frequency is the rotational frequency of the electronic component, the second natural frequency is twice the rotational frequency of the electronic component, the damage characteristic frequency is the structural frequency of the electronic component, the damage characteristic frequency is calculated based on the structural information of the electronic component, the damage characteristic frequency includes the outer race fault frequency, the inner race fault frequency, and the rolling fault frequency, the structure of the electronic component includes rolling elements and bearings, and the structural information includes the number of rolling elements, the diameter of the rolling elements, the diameter of the bearing, and the contact angle between the rolling elements and the bearing surface; As an embodiment of the present invention, the first natural frequency of the electronic component is , and the outer race fault frequency is , the inner race fault frequency is , and the rolling fault frequency is : ; Where: represents the number of rolling elements, represents the diameter of the rolling elements, represents the diameter of the bearing, represents the contact angle between the rolling elements and the bearing surface; The bearing is a mechanical element used to support the rotating part and enable it to rotate freely on the shaft. The rolling elements are located between the inner and outer rings of the bearing to achieve rolling motion and reduce friction. The rolling elements and the bearing form a rolling bearing. The rolling bearing includes the rotor, fan blades, etc. in the electronic component; Calculate the peak area and the maximum curvature of the spectral analysis data as the standardized features of the spectral analysis data. Specifically, the form of the spectral analysis data is: ; Where: represents the s-th wavelength, represents the intensity value of the electronic component at the wavelength , s represents the total number of wavelengths, represents s wavelengths, and the values of the wavelengths increase in sequence; The peak area of the spectral analysis data is : ; The maximum curvature of the spectral analysis data is : ; ; Where: represents the mean value of the wavelength difference; represents the selection such that Reach the maximum wavelength , as the maximum curvature, represents the intensity value of the electronic component at wavelength . represents the th wavelength, , s represents the total number of wavelengths, represents the intensity value of the electronic component at wavelength . represents the intensity value of the electronic component at wavelength . successively represent the th wavelength and the th wavelength respectively.

[0024] S2: Use the operating life evaluation model to evaluate the operating life of the electronic component, and obtain the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result.

[0025] The operating life evaluation model includes an input layer, a multi-damage attention branch module, a resonance perception module, a time-frequency convolution module, a multi-modal gating fusion module, and an output layer; The input layer is used to receive the physical state data of the electronic component. The multi-damage attention branch module is used to receive the standardized processing result of the vibration data, extract the imbalance feature and the bearing damage feature, construct an MLP + attention channel for the imbalance feature and the amplitude domain information of different damage feature frequencies respectively, and use the attention weight to weight different channels to obtain a multi-channel weighted damage feature. The resonance perception module performs an exponential mapping on the resonance feature based on the second-order frequency amplification factor to obtain a resonance amplification feature. The time-frequency convolution module performs a fast Fourier transform on the standardized processing result of the acoustic wave data to obtain an acoustic wave spectrum feature, and performs a convolutional processing with a partial frequency mask on the acoustic wave spectrum feature to obtain a time-frequency convolution feature representing the local amplitude change of the spectrum. The multi-modal gating fusion module is used to generate a gating weight for the output features of the multi-damage attention branch module, the resonance perception module, and the time-frequency convolution module, and use the gating weight to perform gating weighted fusion on the multi-channel weighted damage feature, the resonance amplification feature, and the time-frequency convolution feature to obtain a gating fusion feature. The output layer is composed of a residual dense connection network and an uncertainty perception head. The output layer performs residual dense connection and uncertainty perception optimization processing on the gating fusion feature to generate the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result.

[0026] Specifically, the loss function of the operating life evaluation model is: ; Among them: represents the loss function of the operating life evaluation model, represents the model parameters of the operating life evaluation model, represents the uncertainty of the nth training sample, represents the operating life evaluation result of the nth training sample, , where N represents the total number of training samples, represents the true remaining operating life of the nth training sample.

[0027] In the embodiments of the present invention, the AdamW optimizer is used to iteratively optimize the model parameters in the loss function, and the learning rate update strategy in the AdamW optimizer is optimized. The optimized learning rate update strategy is: ; where: represents the learning rate during the t-th iterative optimization process of the model parameters, represents the preset maximum learning rate, represents the preset minimum learning rate, represents the gradient of the model parameters during the t-th iterative optimization process, represents the preset safety gradient, represents the number of steps for learning rate warming up, represents the iterative optimization step size of the model parameters, represents the slope adjustment factor. The learning rate warming-up strategy is adopted for early learning rate changes, so that the model does not oscillate at the initial stage of training, and the gradient of the model parameters is introduced. Only when the gradient fluctuation of the model is "stable" enough, the learning rate is increased rapidly, which is easier to find the gradient stable interval in the early stage of iteration, helps the model to converge more efficiently. And since the parameters have not converged at the initial stage of network training, the gradient usually fluctuates greatly or even does not converge. If a large learning rate is directly used at this time, it will cause the loss function to oscillate or even diverge. By adopting the updated learning rate update strategy, gradient explosion or oscillation is avoided.

[0028] Evaluating the operating life of electronic components using the operating life evaluation model includes: The exponential mapping formula of the resonance sensing module for resonance characteristics based on the second-order frequency amplification factor is: ; where: represents the second-order frequency amplification factor, represents the resonance characteristic, represents the resonance amplification characteristic; specifically:

[0029] The generation process of the gated weights is as follows: Calculate the multi-channel weighted damage features , resonance amplification features and time-frequency convolution features of the fusion guiding factors. The fusion guiding factors of the multi-channel weighted damage features , resonance amplification features and time-frequency convolution features are successively and . The calculation formula of the fusion guiding factor is: ; Where: represents the L2 norm; Use the gated activation function to perform convolution processing on the multi-channel weighted damage features , resonance amplification features and time-frequency convolution features , and use the fusion guiding factor as the bias to generate the gated weights of the multi-channel weighted damage features , resonance amplification features and time-frequency convolution features . The generation formula of the gated weights of the multi-channel weighted damage features , resonance amplification features and time-frequency convolution features is: ; Where: represents the gated convolution parameter, b represents the gated bias, represents the gated activation function, and the selected gated activation function is the Sigmoid function; The gated weights of the multi-channel weighted damage features , resonance amplification features and time-frequency convolution features are successively and ; The process of the output layer performing residual dense connection and uncertainty perception optimization processing on the gated fusion feature z to generate the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result is as follows: The residual dense connection network includes L residual blocks and a dense connection block. Use the L residual blocks to perform residual convolution processing on the gated fusion feature z to obtain the L-layer residual features of the gated fusion feature z. The L-layer residual feature of the gated fusion feature z is: ; Where: represents the L-th layer residual feature of the gated fusion feature z, represents the convolution parameter of the L-th residual block, represents the bias of the L-th residual block, represents the ReLU function; Use the dense connection block to densely connect the L-layer residual features and generate the operation life evaluation result of the electronic component: ; Where: represents the operation life evaluation result of the electronic component, represents the dense connection convolution parameter, represents the dense connection bias, represents the operation life evaluation convolution parameter; The uncertainty perception head generates the uncertainty of the operation life evaluation result : ; Where: represents the uncertainty evaluation convolution parameter.

[0030] S3: Combine the operation life evaluation result of the electronic component to perform multi-source fusion on the multi-source sensing detection data, obtain the multi-source fusion result, and combine the uncertainty of the operation life evaluation result to perform deep feature extraction on the multi-source fusion result to obtain the operation state feature of the electronic component.

[0031] Obtain the operation life evaluation result of the electronic component and the uncertainty of the operation life evaluation result , and combine the operation life evaluation result of the electronic component to perform multi-source fusion on the multi-source sensing detection data, including: Obtain the multi-source sensing detection data, use the operation life evaluation model to receive the physical state data of the electronic component, and generate a physical state feature vector, where the physical state feature vector includes the multi-channel weighted damage feature, resonance amplification feature, and time-frequency convolution feature calculated in step S2; Perform convolution processing on the electrothermal state data in the multi-source sensing detection data to obtain an electrothermal state feature vector; Take the operation life evaluation result of the electronic component as the Query vector, calculate the attention of the physical state feature vector and the electrothermal state feature vector respectively, perform attention weighting on the physical state feature vector and the electrothermal state feature vector, and obtain the multi-source fusion result F.

[0032] Combine the uncertainty of the operation life evaluation result Perform deep feature extraction on the multi-source fusion result F to obtain the operation state features of the electronic component, including: Perform standard deep feature extraction on the multi-source fusion result F to obtain the standard deep features of the multi-source fusion result F : ; Where: BN represents the batch normalization layer, represents the convolution weight of the batch normalization layer, represents the bias of the batch normalization layer, represents the ReLU function; Based on the uncertainty of the operation life evaluation result , calculate the suppression degree g of the standard deep features : ; Where: represents the suppression coefficient, represents the suppression control parameter; represents the Sigmoid function; Based on the suppression degree g, perform feature modulation on the standard deep features to obtain the operation state features of the electronic component : ; Where: represents element-wise multiplication. Specifically, introducing the uncertainty of the operation life evaluation result to suppress the standard deep features. When the uncertainty is large, the response of some channels will be dynamically weakened or re-weighted to avoid the interference of low-confidence features on the final evaluation result.

[0033] S4: Perform feature mapping on the operation state features of the electronic component to obtain the real-time operation state of the electronic component, obtain the historical operation state sequence of the electronic component, and perform monitoring and correction on the real-time operation state of the electronic component to obtain the real-time operation state monitoring result of the electronic component.

[0034] Construct a fully connected layer for the operation state features of the electronic component Perform feature mapping to obtain the real-time operating state A of the electronic component. The fully connected layer includes fully connected convolution parameters and fully connected bias amounts. Obtain the historical operating state sequence of the electronic component, and monitor and correct the real-time operating state of the electronic component, including: Specifically, the real-time operating state is within the range of 0-1. The higher the value of the real-time operating state, the better the working performance of the electronic component. As an embodiment of the present invention, if the real-time operating state , it indicates that the real-time operating state is a normal operating state, that is, the electronic component has good working performance and no obvious performance decline; if the real-time operating state , it indicates that the real-time operating state is a sub-healthy state, that is, there are slight abnormalities but no impact on the function, and there are potential fault hazards; if the real-time operating state , it indicates that the real-time operating state is a sub-healthy state, that is, there are slight abnormalities but no impact on the function, and there are potential fault hazards; if the real-time operating state , it indicates that the real-time operating state is a fault warning state, that is, obvious fault precursors are shown and the service life decreases significantly; if the real-time operating state , it indicates that the real-time operating state is a fault state, that is, structural and functional damage occurs, and the performance decreases or is lost; if the real-time operating state , it indicates that the real-time operating state is a failure state, that is, the electronic component cannot work completely; the , is a preset state threshold; The historical operating state sequence of the electronic component is the operating state after monitoring and correction of the electronic component at M historical moments. The monitoring and correction formula for the real-time operating state A is: ; Where: represents the monitoring and correction result of the real-time operating state A, and serves as the real-time operating state monitoring result of the electronic component, represents the L2 norm, k represents the correction control coefficient, represents the uncertainty of the operating life evaluation result, represents the mean value of the historical operating state sequence.

[0035] Embodiment 2: As Figure 2 shown, it is a functional module diagram of the electronic component operating state monitoring system 100 provided by an embodiment of the present invention, which can implement a method for monitoring the operating state of an electronic component based on artificial intelligence in Embodiment 1.

[0036] According to the implemented functions, the electronic component operation status monitoring system 100 may include an operation life evaluation module 101, an operation status monitoring module 102, and a data acquisition device 103. The modules in the present invention may also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions.

[0037] The operation life evaluation module 101 is used to evaluate the operation life of the electronic component by using an operation life evaluation model, and obtain the operation life evaluation result of the electronic component and the uncertainty of the operation life evaluation result. The operation status monitoring module 102 is used to perform multi-source fusion on multi-source sensing detection data in combination with the operation life evaluation result of the electronic component to obtain a multi-source fusion result, and perform deep feature extraction on the multi-source fusion result in combination with the uncertainty of the operation life evaluation result to obtain the operation status feature of the electronic component, perform feature mapping on the operation status feature of the electronic component to obtain the real-time operation status of the electronic component, and obtain the historical operation status sequence of the electronic component, and perform monitoring and correction on the real-time operation status of the electronic component to obtain the real-time operation status monitoring result of the electronic component. The data acquisition device 103 is used to collect multi-source detection data of the electronic component by using a variety of sensors and perform standardization processing to obtain multi-source sensing detection data of the electronic component.

[0038] Specifically, each module in the electronic component operation status monitoring system 100 in the embodiment of the present invention adopts the same technical means as those described in the Figure 1 a method for monitoring the operation status of electronic components based on artificial intelligence described above, and can produce the same technical effects, which will not be elaborated here.

[0039] It should be understood that the described embodiments are only for illustration purposes and are not limited by this structure in the scope of the patent application.

[0040] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the term "including" or "comprising" or any other variation thereof in this article is intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements not only includes those elements but also includes other elements not expressly listed, or also includes elements inherent to such process, device, article or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, device, article or method including the element.

[0041] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described example methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium as described above (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0042] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. An operating state monitoring method for electronic components based on artificial intelligence, characterized in that, The method includes: S1: Collect multi-source detection data of electronic components using multiple sensors and perform standardization processing to obtain multi-source sensing detection data of the electronic components. The multi-source sensing detection data includes physical state data and electrothermal state data; The physical state data includes the standardized processing results of spectral analysis data, acoustic wave data, and vibration data of the electronic components. The electrothermal state data includes the standardized processing results of current value, voltage value, and heat value during the operation of the electronic components; S2: Use the operation life evaluation model to evaluate the operation life of the electronic components. The operation life evaluation model takes the physical state data of the electronic components as input, calculates the multi-channel weighted damage characteristics, resonance amplification characteristics, and time-frequency convolution characteristics corresponding to the physical state data, and outputs the operation life evaluation result of the electronic components and the uncertainty of the operation life evaluation result; S3: Perform multi-source fusion on the multi-source sensing detection data in combination with the operation life evaluation result of the electronic components to obtain a multi-source fusion result, and perform deep feature extraction on the multi-source fusion result in combination with the uncertainty of the operation life evaluation result to obtain the operation state characteristics of the electronic components; S4: Perform feature mapping on the operation state characteristics of the electronic components to obtain the real-time operation state of the electronic components, obtain the historical operation state sequence of the electronic components, and perform monitoring and correction on the real-time operation state of the electronic components to obtain the real-time operation state monitoring result of the electronic components.

2. The method for monitoring the operating state of electronic components based on artificial intelligence according to claim 1, wherein, Collect multi-source detection data of electronic components using multiple sensors, including; The sensors include current / voltage sensors, infrared thermal imagers, accelerometers, ultrasonic sensors, and spectrometers. The multi-source detection data includes the current value, voltage value, heat value, vibration data, acoustic wave data, and spectral analysis data of the electronic components; Use the current / voltage sensor to collect the current value and voltage value of the electronic components respectively, use the infrared thermal imager to collect the heat value of the electronic components, use the ultrasonic sensor to collect the acoustic wave data of the electronic components, use the accelerometer to collect the vibration data of the electronic components, and use the spectrometer to collect the spectral analysis data of the electronic components; The current value, voltage value, and heat value are sequence data, the acoustic wave data is signal data, the vibration data is the acceleration sequence data of the electronic components in a fixed direction, and the spectral analysis data is the intensity value of the electronic components at different wavelengths; Perform standardization processing on the multi-source detection data. The standardization processing methods for the current value, voltage value, acoustic wave data, and heat value are normalization processing. The standardized processing results of the current value, voltage value, and heat value are the sequence data after normalization processing. The standardized processing result of the acoustic wave data is the signal data after normalization processing. Extract the standardized features of the vibration data and spectral analysis data as the standardized processing results of the vibration data and spectral analysis data.

3. The method for monitoring the operating state of electronic components based on artificial intelligence according to claim 2, wherein, Extracting the standardized features of the vibration data and spectral analysis data includes: Perform a fast Fourier transform on the vibration data to obtain the spectral characteristics of the vibration data. The spectral characteristics are in the form of a spectrogram, with the horizontal axis of the spectral characteristics being frequency and the vertical axis being amplitude; Successively extract the unbalance feature, bearing damage feature, and resonance feature from the spectral characteristics as the standardized features of the vibration data. The unbalance feature is the amplitude neighborhood information of the first natural frequency in the vibration data. The bearing damage feature is the amplitude neighborhood information of multiple damage characteristic frequencies in the vibration data. The resonance feature is the amplitude neighborhood information of the second natural frequency in the vibration data; The first natural frequency is the rotational frequency of the electronic component. The second natural frequency is twice the rotational frequency of the electronic component. The damage characteristic frequency is the structural frequency of the electronic component. The damage characteristic frequency is calculated based on the structural information of the electronic component. The damage characteristic frequencies include the outer race fault frequency, inner race fault frequency, and rolling element fault frequency. The structure of the electronic component includes rolling elements and bearings, and the structural information includes the number of rolling elements, the diameter of the rolling elements, the diameter of the bearing, and the contact angle between the rolling elements and the bearing surface; Calculate the peak area and the maximum curvature of the spectral analysis data as the standardized features of the spectral analysis data.

4. The method for monitoring the operating state of electronic components based on artificial intelligence according to claim 1, wherein The operating life evaluation model includes an input layer, a multi-damage attention branch module, a resonance perception module, a time-frequency convolution module, a multi-modal gating fusion module, and an output layer; The input layer is used to receive the physical state data of the electronic component. The multi-damage attention branch module is used to receive the standardized processing results of the vibration data, extract the unbalance feature and the bearing damage feature, construct an MLP + attention channel for the unbalance feature and the amplitude neighborhood information of different damage characteristic frequencies respectively, and weight different channels using attention weights to obtain multi-channel weighted damage features. The resonance perception module performs an exponential mapping on the resonance feature based on the second-order frequency amplification factor to obtain a resonance amplification feature. The time-frequency convolution module performs a fast Fourier transform on the standardized processing results of the acoustic wave data to obtain the acoustic wave spectral characteristics, and performs a convolution process with a partial frequency mask on the acoustic wave spectral characteristics to obtain time-frequency convolution features representing local amplitude changes in the spectrum. The multi-modal gating fusion module is used to generate gating weights for the output features of the multi-damage attention branch module, the resonance perception module, and the time-frequency convolution module, and perform gating weighted fusion on the multi-channel weighted damage features, the resonance amplification features, and the time-frequency convolution features using the gating weights to obtain gating fusion features. The output layer consists of a residual dense connection network and an uncertainty perception head. The output layer performs residual dense connection and uncertainty perception optimization processing on the gating fusion features to generate the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result.

5. The method for monitoring the operating state of electronic components based on artificial intelligence according to claim 4, wherein Use the operating life evaluation model to evaluate the operating life of the electronic component, including: The exponential mapping formula for the resonance perception module to perform on the resonance feature based on the second-order frequency amplification factor is: ； Where: represents the second-order frequency amplification factor, represents the resonance characteristic, represents the resonance amplification characteristic; The generation process of the gating weights is: Calculate the multi-channel weighted damage feature , resonance amplification feature and time-frequency convolution feature of the fusion guiding factor, the multi-channel weighted damage feature , resonance amplification feature and time-frequency convolution feature of the fusion guiding factor are successively and ; Using a gated activation function for the multi-channel weighted damage features , resonance amplification features , and time-frequency convolution features Perform convolution processing, and use the fusion guiding factor as a bias to generate the multi-channel weighted damage features , resonance amplification features , and time-frequency convolution features The gated weights of; the multi-channel weighted damage features , resonance amplification features , and time-frequency convolution features The gated weight generation formula is: ； ； ； Where: represents the gated convolution parameter, and b represents the gated bias, represents the gated activation function; The multi-channel weighted damage feature , the resonance amplification feature , and the time-frequency convolution feature have gating weights of and ; The process of the output layer performing residual dense connection and uncertainty-aware optimization processing on the gated fusion feature z to generate the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result is as follows: The residual dense connection network includes L residual blocks and a dense connection block. The L residual blocks are used to perform residual convolution processing on the gated fusion feature z to obtain L-layer residual features of the gated fusion feature z. The L-layer residual feature of the gated fusion feature z is: ； Where: Denote the residual feature of the L-th layer of the gated fusion feature z, Denote the convolutional parameters of the L-th residual block, Denote the bias of the L-th residual block, Denote the ReLU function; The dense connection block is used to perform dense connection on the L-layer residual features and generate the operating life evaluation result of the electronic component: ； Where: Indicates the operation life evaluation result of the said electronic component, Indicates the densely connected convolution parameter, Indicates the densely connected bias, Indicates the operation life evaluation convolution parameter; Indicates the L-layer residual feature of the gated fusion feature z; The uncertainty of the uncertainty perception head generates the operating life evaluation result : ； Where: Indicates the convolution parameter for uncertainty evaluation.

6. The method for monitoring the operating state of electronic components based on artificial intelligence according to claim 1, characterized in that Obtain the operation life evaluation result of the electronic component and the uncertainty of the operation life evaluation result , and perform multi-source fusion on the multi-source sensing detection data in combination with the operation life evaluation result of the electronic component, including: Obtain the multi-source sensing detection data. Use the operating life evaluation model to receive the physical state data of the electronic component and generate a physical state feature vector. The physical state feature vector includes the multi-channel weighted damage feature, resonance amplification feature, and time-frequency convolution feature calculated in step S2; Perform convolution processing on the electrothermal state data in the multi-source sensing detection data to obtain an electrothermal state feature vector; Use the operating life evaluation result of the electronic component as the Query vector, calculate the attention of the physical state feature vector and the electrothermal state feature vector respectively, and perform attention weighting on the physical state feature vector and the electrothermal state feature vector to obtain the multi-source fusion result F.

7. The method for monitoring the operating state of electronic components based on artificial intelligence according to claim 6, wherein, Combined with the uncertainty of the operating life evaluation result Perform deep feature extraction on the multi-source fusion result F to obtain the operating state characteristics of the electronic components, including: Perform standard depth feature extraction on the multi-source fusion result F to obtain the standard depth features of the multi-source fusion result F : ； Where: BN represents the batch normalization layer, represents the convolution weights of the batch normalization layer, represents the bias of the batch normalization layer, represents the ReLU function; Uncertainty based on the operating life evaluation result , calculate the suppression degree g for the standard depth feature : ； Where: represents the suppression coefficient, represents the suppression control parameter; denotes the Sigmoid function; Perform feature modulation on the standard depth features based on the suppression degree g to obtain the operating state features of the electronic components : ； Where: Denotes element-wise multiplication.

8. The method for monitoring the operating state of electronic components based on artificial intelligence according to claim 7, characterized in that, Construct a fully connected layer for the operating state characteristics of the electronic components Perform feature mapping to obtain the real-time operating state A of the electronic components. The fully connected layer includes fully connected convolution parameters and fully connected bias amounts. Obtain the historical operating state sequence of the electronic components and monitor and correct the real-time operating state of the electronic components, including: The historical operating state sequence of the electronic component is the operating state after monitoring and correction of the electronic component at M historical moments. The monitoring and correction formula for the real-time operating state A is: ； Where: Indicates the monitoring and correction result of the real-time operation state A, which is used as the real-time operation state monitoring result of electronic components. Indicates the L2 norm, and k represents the correction control coefficient. Indicates the uncertainty of the operation life evaluation result. Indicates the mean value of the historical operation state sequence.

9. An operating state monitoring system for electronic components, characterized in that, The electronic component operating state monitoring system includes a server and a data acquisition device. The server includes an operating life evaluation module and an operating state monitoring module: The operating life evaluation module is used to evaluate the operating life of the electronic component using the operating life evaluation model to obtain the operating life evaluation result of the electronic component and the uncertainty of the operating life evaluation result; The operating state monitoring module is used to perform multi-source fusion on the multi-source sensing detection data in combination with the operating life evaluation result of the electronic component to obtain a multi-source fusion result, and perform deep feature extraction on the multi-source fusion result in combination with the uncertainty of the operating life evaluation result to obtain the operating state feature of the electronic component, perform feature mapping on the operating state feature of the electronic component to obtain the real-time operating state of the electronic component, and obtain the historical operating state sequence of the electronic component, and perform monitoring and correction on the real-time operating state of the electronic component to obtain the real-time operating state monitoring result of the electronic component; The data acquisition device is used to collect multi-source detection data of the electronic component using multiple sensors and perform standardization processing to obtain the multi-source sensing detection data of the electronic component; To implement an artificial intelligence-based electronic component operating state monitoring method as described in any one of claims 1-8.

Images