A method and system for remote monitoring of the operating state of an optoelectronic imaging assembly

Abstract

The application relates to the technical field of data processing, and discloses a working state remote monitoring method and system for an optoelectronic imaging assembly. The method comprises the following steps: performing structured processing on multi-source data of the optoelectronic imaging assembly to obtain a standardized data set, extracting a working state feature vector by using a variational autoencoder, identifying a real-time working state by using a multi-modal fusion algorithm, performing predictive fault diagnosis by using a time series analysis network, realizing remote transmission and decision support based on a distributed architecture, and finally obtaining a remote monitoring result of the optoelectronic imaging assembly. The application solves the technical problems of insufficient multi-modal data fusion, lack of predictive fault diagnosis capability and low distributed monitoring processing efficiency in the existing optoelectronic imaging assembly monitoring technology, and improves the accuracy and intelligent level of the working state monitoring of the optoelectronic imaging assembly.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a working state remote monitoring method and system for optoelectronic imaging assemblies. BACKGROUND

[0002] The existing optoelectronic imaging assembly monitoring technology mainly adopts a simple monitoring method based on threshold judgment. Temperature sensors, vibration sensors, and electrical parameter monitoring devices are deployed on key components such as CCD sensors, CMOS sensors, optical lenses, and optical filters to collect basic physical parameter data during device operation. The traditional monitoring system compares the collected sensor data with the preset safety threshold, triggers an alarm mechanism when the monitoring parameters exceed the normal range, and uses simple data recording and remote transmission functions to send the monitoring data to the monitoring center through a fixed communication protocol, thereby achieving basic state monitoring of optoelectronic imaging devices.

[0003] However, the existing technology has significant technical defects and application limitations. First, the traditional threshold judgment method can only perform passive fault detection and cannot provide early warnings before faults occur, resulting in high equipment maintenance costs and a high risk of sudden failure downtime. Second, the existing monitoring system lacks an effective data fusion mechanism, and the correlation between different parameters cannot be established, making it difficult to accurately reflect the overall health status of optoelectronic imaging assemblies. In addition, the existing technology relies mainly on manual experience for fault diagnosis and lacks intelligent analysis algorithms, making it difficult to handle complex multi-modal data and time series evolution patterns, resulting in low fault recognition accuracy.

[0004] Based on the deep analysis of the above technical defects, the optoelectronic imaging assembly monitoring field faces the fundamental technical challenge of transitioning from passive monitoring to active prediction. Since the existing technology cannot effectively fuse multi-source heterogeneous monitoring data, it is difficult to construct an accurate device state feature representation, which in turn affects the accuracy of fault pattern recognition and prediction ability. At the same time, the lack of intelligent time series analysis capabilities makes it difficult for existing systems to capture the evolution of device states and potential fault development trends, and this limitation is more pronounced in complex distributed environments, as traditional centralized processing architectures cannot meet the real-time monitoring needs of large-scale device groups and cannot provide flexible decision support and visualization analysis functions. SUMMARY

[0005] The present application provides a working state remote monitoring method and system for optoelectronic imaging assemblies, which solves the technical problems of insufficient multi-modal data fusion, lack of predictive fault diagnosis capability, and low efficiency of distributed monitoring in existing optoelectronic imaging assembly monitoring technology, and improves the accuracy and intelligence level of optoelectronic imaging assembly working state monitoring.

[0006] In a first aspect, the application provides a working state remote monitoring method for an optoelectronic imaging assembly, the working state remote monitoring method for the optoelectronic imaging assembly comprising: structurally processing temperature data, vibration data, electrical parameter data and image quality data of the optoelectronic imaging assembly to obtain a standardized monitoring data set; performing feature extraction and encoding processing on the standardized monitoring data set according to a variational autoencoder algorithm to obtain an optoelectronic imaging assembly working state feature vector; performing state recognition processing on the optoelectronic imaging assembly working state feature vector through a multi-modal fusion algorithm to obtain an optoelectronic imaging assembly real-time working state identifier; performing predictive fault diagnosis processing on the optoelectronic imaging assembly real-time working state identifier according to a time series analysis network to obtain optoelectronic imaging assembly fault early warning information; performing remote transmission and decision support processing on the optoelectronic imaging assembly fault early warning information through a distributed data processing architecture to obtain an optoelectronic imaging assembly remote monitoring result.

[0007] In a second aspect, the application provides an optoelectronic imaging assembly working state remote monitoring system, the optoelectronic imaging assembly working state remote monitoring system comprising:

[0008] a processing module configured to structurally process temperature data, vibration data, electrical parameter data and image quality data of the optoelectronic imaging assembly to obtain a standardized monitoring data set;

[0009] an encoding module configured to perform feature extraction and encoding processing on the standardized monitoring data set according to a variational autoencoder algorithm to obtain an optoelectronic imaging assembly working state feature vector;

[0010] a recognition module configured to perform state recognition processing on the optoelectronic imaging assembly working state feature vector through a multi-modal fusion algorithm to obtain an optoelectronic imaging assembly real-time working state identifier;

[0011] a diagnosis module configured to perform predictive fault diagnosis processing on the optoelectronic imaging assembly real-time working state identifier according to a time series analysis network to obtain optoelectronic imaging assembly fault early warning information;

[0012] a transmission module configured to perform remote transmission and decision support processing on the optoelectronic imaging assembly fault early warning information through a distributed data processing architecture to obtain an optoelectronic imaging assembly remote monitoring result.

[0013] In a third aspect, an optoelectronic imaging assembly working state remote monitoring device is provided, comprising: a memory and at least one processor, the memory having instructions stored therein; the at least one processor invoking the instructions in the memory to cause the optoelectronic imaging assembly working state remote monitoring device to perform the above-mentioned working state remote monitoring method for the optoelectronic imaging assembly.

[0014] In a fourth aspect, a computer readable storage medium is provided, in which instructions are stored, when running on a computer, cause the computer to execute the working state remote monitoring method for photoelectric imaging assembly.

[0015] In the technical solutions provided in the present application, the temperature data, vibration data, electrical parameter data and image quality data of the photoelectric imaging assembly are structured to obtain a standardized monitoring data set, effectively solving the technical problems of non-uniformity of multi-source heterogeneous data formats and uneven data quality in the prior art. Through unified data standardization processing, the accuracy and reliability of subsequent algorithm analysis are ensured. The variational autoencoder algorithm is used to extract and encode the standardized monitoring data set to obtain a photoelectric imaging assembly working state feature vector, breaking through the limitations of traditional monitoring technology relying on manual feature selection. The probabilistic distribution modeling capability of the variational autoencoder enables the feature extraction process to have stronger generalization performance and robustness, and can automatically discover potential patterns and key features in the data, significantly improving the accuracy and integrity of state feature representation. The photoelectric imaging assembly working state feature vector is processed by a multi-modal fusion algorithm to obtain a real-time working state identifier, innovatively establishing the correlation between different data modalities. The multi-modal fusion algorithm effectively integrates temperature, vibration, electrical and image information through attention mechanism and cross correlation calculation, avoiding the one-sidedness and limitations of single modal data, making the state recognition result more comprehensive and accurate, and significantly improving the recognition accuracy compared with traditional independent parameter monitoring methods. According to the predictive fault diagnosis processing of the real-time working state identifier by the time series analysis network, the fault warning information is obtained, realizing the technical leap from passive fault detection to active fault prediction. The long short-term dependency relationship modeling capability of the time series analysis network enables the system to capture the time evolution law and potential fault development trend of the device state, and the 2-4 weeks early warning capability greatly reduces the risk of sudden equipment failure, effectively reducing maintenance cost and downtime loss.

[0016] The remote monitoring result is obtained by remote transmission and decision support processing of the fault early warning information through the distributed data processing architecture, effectively solving the performance bottleneck and scalability problem of the traditional centralized processing architecture in large-scale equipment monitoring, and the cloud collaborative computing capability of the distributed architecture enables the system to flexibly cope with different scale monitoring requirements, while providing high availability and fault tolerance capability. In the specific application field of optoelectronic imaging component monitoring, the probabilistic modeling characteristics of the variational autoencoder algorithm are particularly suitable for processing the complex working state changes of optoelectronic equipment, and the reparameterization technique ensures the stability and interpretability of the feature learning process. The multi-modal fusion algorithm is specially optimized for the internal relationship between the temperature, vibration, electrical and image physical quantities of the optoelectronic imaging equipment, and the multi-head self-attention mechanism of the time series analysis network can simultaneously focus on the short-term fluctuations and long-term trends of the optoelectronic equipment. The organic combination of these algorithm characteristics enables the present application to have significant technical advantages in the field of optoelectronic imaging component monitoring, not only improving the monitoring accuracy and prediction accuracy, but also realizing intelligent operation and maintenance decision support. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor based on these drawings.

[0018] Figure 1 is a schematic diagram of one embodiment of the working state remote monitoring method for optoelectronic imaging components in the embodiments of the present application.

[0019] Figure 2 is a schematic diagram of one embodiment of the working state remote monitoring system for optoelectronic imaging components in the embodiments of the present application.

[0020] Figure 3 is a structural schematic block diagram of the working state remote monitoring device for optoelectronic imaging components in the embodiments of the present application. DETAILED DESCRIPTION

[0021] The embodiment of the present application provides a kind of photoelectric imaging assembly working state remote monitoring method and system.The terms "first", "second", "third", "fourth" and the like (if exist) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence.It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein.In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0022] For ease of understanding, the specific process of the embodiment of the present application is described below, please refer to Figure 1 One embodiment of the photoelectric imaging assembly working state remote monitoring method in the embodiment of the present application includes:

[0023] Step S101, the temperature data, vibration data, electrical parameter data and image quality data of photoelectric imaging assembly are structured, and standard monitoring data set is obtained;

[0024] Step S102, according to the variational autoencoder algorithm, the standard monitoring data set is extracted and encoded, and the photoelectric imaging assembly working state feature vector is obtained;

[0025] Step S103, the photoelectric imaging assembly working state feature vector is processed by multi-modal fusion algorithm to identify the state, and the real-time working state identification of photoelectric imaging assembly is obtained;

[0026] Step S104, according to time series analysis network, the real-time working state identification of photoelectric imaging assembly is processed for predictive fault diagnosis, and photoelectric imaging assembly fault warning information is obtained;

[0027] Step S105, through distributed data processing architecture, photoelectric imaging assembly fault warning information is processed for remote transmission and decision support, and photoelectric imaging assembly remote monitoring result is obtained.

[0028] It can be understood that the execution subject of the present application can be a photoelectric imaging assembly working state remote monitoring system, and can also be a terminal or a server, and the specific place is not limited.The embodiment of the present application takes the server as the execution subject for example.

[0029] Specifically, the structured processing procedure adopts a 3σ criterion for outlier detection on the temperature monitoring point data of CCD and CMOS sensors, specifically by calculating the mean and standard deviation of the temperature data sequence, identifying abnormal data points that exceed the range of the mean plus or minus three times the standard deviation, and forming temperature anomaly identification data. Based on this anomaly identification, the three-axis vibration acceleration data of the optical lens assembly is subjected to missing value interpolation, and the ARIMA model of the time series is used to analyze the historical vibration pattern, and the vibration value at the missing time is predicted through the autoregressive and moving average parameters, generating a vibration data sequence. Next, the voltage and current monitoring values of the filter system are processed by the adaptive Kalman filter algorithm, which dynamically adjusts the filter gain according to the real-time statistical characteristics of the signal to remove high-frequency noise interference and generate clean electrical parameter data. The signal-to-noise ratio data of the image sensor and the clean electrical parameter data are jointly input into the standardization algorithm, and the Min-Max normalization is used to map the data of different dimensions to the interval [0, 1], and finally the standardized monitoring data set is formed.

[0030] The feature extraction process of the variational autoencoder algorithm inputs the standardized monitoring data set into the encoder network, which contains three fully connected layers that gradually compress high-dimensional input data into low-dimensional latent feature space data. In the latent space, the algorithm calculates the mean vector and variance vector of the data distribution, which describe the probability distribution characteristics of the monitoring data in the feature space. The reparameterization sampling process generates a sample feature vector by sampling random noise from a standard normal distribution combined with the mean and variance parameters, which ensures that the probability distribution of the feature vector meets the theoretical requirements of variational inference. The decoder network receives the sample feature vector and reconstructs the original data dimension, and the reconstruction error reflects the quality of feature extraction, with smaller error indicating that the feature vector better preserves the key information of the original data.

[0031] The multi-modal fusion algorithm first separates the working state feature vectors of the optoelectronic imaging assembly into four independent feature subsets: temperature modal features, vibration modal features, electrical modal features, and image modal features. The attention weight calculation uses a softmax function to quantify the correlation between temperature modal features and vibration modal features, and determines the importance weight of each feature by calculating the dot product similarity between features, realizing intra-modal fusion to form physical state fusion features. The cross-attention mechanism establishes the association mapping between electrical modal features and image modal features, while considering the influence of physical state fusion features, and calculates the interaction strength between modalities through matrix multiplication operations to generate performance state association features. The support vector machine classifier maps the performance state association features to a high-dimensional space, finds the optimal separating hyperplane to classify different working states, and outputs the probability distribution containing normal, attention, warning, and failure states.

[0032] The time series analysis network adopts a Transformer encoder to process the historical sequence of the real-time working state identification of the photoelectric imaging assembly. The encoder retains the time sequence information through a positional encoding mechanism and extracts the time sequence pattern features of the state changes. The multi-head self-attention mechanism calculates the dependency between the states at different times in the historical state sequence. Each attention head focuses on different time scale patterns, and the time sequence correlation weight matrix is formed by weighted summation. Based on the weight matrix, the state evolution of the future time window is predicted, and the prediction state probability sequence is generated in a sequence-to-sequence mapping manner. The fault type matching process calculates the similarity between the prediction probability sequence and the predefined fault mode template. When the matching degree exceeds the set threshold, the corresponding fault type recognition is triggered, and the fault occurrence probability evaluation result is formed.

[0033] The distributed data processing architecture adopts a consistent hashing algorithm to perform data sharding on the photoelectric imaging assembly fault early warning information. The algorithm calculates the hash value based on the device identifier and the timestamp, and uniformly distributes the data to the storage nodes on the hash ring. The network quality evaluation calculates the network quality index by comprehensively considering the bandwidth, delay, packet loss rate and other indicators, selects the optimal transmission path according to the index value, and generates the transmission strategy. The cloud collaborative computing node receives the distributed storage data block, distributes the computing task to different processing nodes through the load balancing algorithm, and avoids single point overload. The knowledge graph reasoning engine maintains the semantic relationship network of device-fault-maintenance measures, matches the corresponding maintenance decision rules according to the current fault type and device state, and generates specific maintenance suggestions. The visualization rendering module converts the maintenance decision suggestions into various display forms such as charts and reports, forming the remote monitoring results.

[0034] In a specific embodiment, the process of step S101 can specifically include the following steps:

[0035] Anomaly value detection processing is performed on the temperature monitoring point data of the CCD sensor and the CMOS sensor to obtain temperature anomaly identification data;

[0036] According to the temperature anomaly identification data, the three-axis vibration acceleration data of the optical lens assembly is subjected to missing value interpolation processing to obtain a vibration data sequence;

[0037] Based on the vibration data sequence, noise filtering processing is performed on the voltage and current monitoring values of the optical filter system to obtain cleaned electrical parameter data;

[0038] The signal-to-noise ratio data of the image sensor and the cleaned electrical parameter data are input into a standardization algorithm for numerical normalization processing to obtain standardized quality parameters;

[0039] The temperature anomaly identification data, the vibration data sequence, the cleaned electrical parameter data and the standardized quality parameters are subjected to data fusion processing to obtain a standardized monitoring data set.

[0040] Specifically, the temperature monitoring point data of the CCD sensor and the CMOS sensor are subjected to an outlier detection process using a 3σ criterion as the core detection algorithm. The criterion is based on the normal distribution theory, which considers that the probability of data points falling outside the range of the mean plus or minus three times the standard deviation is extremely small, and therefore marks such data points as outliers. The specific data processing flow includes collecting the continuous temperature values of the sensor temperature monitoring points within a time window, calculating the arithmetic mean of these values as the temperature mean, calculating the variance of the temperature data relative to the mean and taking the square root to obtain the standard deviation, then comparing each temperature data point with the upper and lower boundaries of the mean plus or minus three times the standard deviation, and marking the data points outside the boundary range as outliers and generating the corresponding abnormal identification bit. The abnormal identification bit of normal data points is set to 0, and the abnormal identification bit of abnormal data points is set to 1, finally forming a temperature abnormal identification data sequence corresponding to the original temperature data. The time series interpolation technique is used to perform missing value interpolation processing on the three-axis vibration acceleration data of the optical lens assembly based on the temperature abnormal identification data. The technique predicts the vibration values at the missing time by analyzing the historical vibration patterns. The data processing process first checks the position of the missing points in the three-axis vibration acceleration data sequence, then judges the environmental background of the missing occurrence according to the temperature abnormal identification data, when the temperature abnormal identification is 1, it indicates that the sensor is in an abnormal state, at this time the missing value interpolation needs to consider the influence of temperature abnormality on vibration measurement, and the weighted interpolation method is used to reduce the influence weight of the data before and after the abnormal period, when the temperature abnormal identification is 0, it indicates that the sensor is working normally, the standard linear interpolation or cubic spline interpolation method is used, the vibration trend curve before and after the missing point is fitted to predict the vibration value at the missing position, finally a continuous vibration data sequence without missing is generated. The adaptive filtering algorithm is used to perform noise filtering processing on the voltage and current monitoring values of the filter system based on the vibration data sequence. The algorithm dynamically adjusts the filtering parameters according to the change characteristics of the vibration data sequence. The data processing logic is based on the correlation between vibration and electrical parameter noise. When the vibration amplitude is large, the electrical connection is easy to produce contact noise, therefore the filtering strength needs to be enhanced, when the vibration amplitude is small, the electrical signal is relatively stable, and light filtering is adopted to avoid signal distortion. The specific algorithm calculates the root mean square value of the vibration data sequence as the vibration intensity indicator, sets the cutoff frequency and gain parameters of the filter according to the vibration intensity, performs frequency domain filtering processing on the voltage and current monitoring values, removes the high frequency noise components and retains the effective electrical parameter information, and outputs the purified electrical parameter purification data. The signal-to-noise ratio data of the image sensor and the electrical parameter purification data are input into the standardization algorithm for numerical normalization processing using a piecewise standardization strategy. The strategy uses corresponding normalization methods for different types of data. The signal-to-noise ratio data is subjected to logarithmic standardization processing, because the signal-to-noise ratio is usually logarithmically distributed in decibels, the data is mapped to a linear space through logarithmic transformation, and then scaled to the 0 to 1 interval using maximum and minimum value normalization.The electrical parameter purification data is processed by Z-score standardization, the mean and standard deviation of the data are calculated, each data point is subtracted from the mean and divided by the standard deviation, so that the data conforms to the standard normal distribution, and finally the two types of standardized data are combined to form a standardized quality parameter. The feature matrix construction method is used for data fusion processing of the temperature anomaly identification data, vibration data sequence, electrical parameter purification data and standardized quality parameter. This method organizes different types of data into a unified data structure according to the time alignment principle. The data fusion process first establishes a time reference axis, synchronizes all data according to the sampling time stamp, then constructs a multi-dimensional feature matrix, the rows of the matrix represent time series, and the columns represent different feature dimensions, including temperature anomaly identification, three-axis vibration acceleration, voltage and current purification value, standardized signal-to-noise ratio and other features, then the feature matrix is subjected to integrity check and consistency verification to ensure accurate data dimension matching and time alignment, and finally a structured standardized monitoring data set is output.

[0041] In a specific embodiment, the process of performing step S102 can specifically include the following steps:

[0042] The standardized monitoring data set is input into the encoder network of the variational autoencoder for dimension compression processing to obtain latent feature space data;

[0043] The latent feature space data is subjected to mean and variance calculation processing to obtain feature distribution parameters;

[0044] The reparameterization sampling process is subjected to probability sampling processing according to the feature distribution parameters to obtain a sampling feature vector;

[0045] The decoder network is subjected to reconstruction error calculation processing based on the sampling feature vector to obtain a feature quality evaluation result;

[0046] The sampling feature vector and the feature quality evaluation result are subjected to comprehensive encoding processing to obtain an optoelectronic imaging component working state feature vector.

[0047] Specifically, the standardized monitoring data set is input into the encoder network of the variational autoencoder for dimension compression processing. A multi-layer neural network structure is used to realize low-dimensional representation of high-dimensional data. The encoder network includes three fully connected layers, and the number of neurons in each layer decreases successively to form a funnel-shaped structure. The data processing process first inputs the standardized monitoring data set as an input vector into the first fully connected layer, which includes 512 neurons. Each neuron performs nonlinear transformation on the input data through weighted summation and a ReLU activation function. The ReLU activation function sets negative values to zero and retains positive values, effectively avoiding the problem of gradient vanishing. The second layer includes 256 neurons, which receive the output of the first layer and further compress the data dimension. The third layer includes 128 neurons, which finally compress the high-dimensional input data into 128-dimensional latent feature space data. The dimension compression process learns the internal structure and patterns of the data, removes redundant information and retains key features, so that the latent feature space data can represent the essential characteristics of the original monitoring data in a more compact form. The mean and variance calculation process of the latent feature space data uses statistical methods to quantify the data distribution characteristics. This process treats the latent feature space data as a sample of probability distribution, and describes the distribution law of the data by calculating its statistical parameters. The specific calculation process includes calculating the mean of each dimension of the latent feature space data. The mean parameter is obtained by summing all data points in this dimension and dividing by the total number of data points. The variance parameter is calculated by squaring the difference between each data point and the mean, summing the squared differences, and dividing by the total number of data points. The mean parameter reflects the center position of the latent feature in that dimension, and the variance parameter reflects the dispersion degree of the data in that dimension. The two parameters together constitute the feature distribution parameter, which describes the probability distribution characteristics of the photoelectric imaging component working state in the latent feature space.

[0048] According to the characteristic distribution parameters, the reparameterization sampling process is processed by probability sampling. The reparameterization technique is used to realize differentiable random sampling. The technique separates randomness from the sampling process, enabling end-to-end gradient propagation of the entire network. The reparameterization sampling process first generates a random noise vector from a standard normal distribution, where each element of the noise vector follows a normal distribution with a mean of 0 and a variance of 1. Then, the mean and variance in the characteristic distribution parameters are used to linearly transform the noise vector. The specific transformation formula is to multiply the noise vector by the square root of the variance parameter and add the mean parameter. This transformation ensures that the sampling result follows the specified mean and variance of the normal distribution. The sampling feature vector generated by the sampling process not only maintains randomness but also can be accurately controlled through the mean and variance parameters, so that the sampling result can reflect the real distribution state of the optoelectronic imaging assembly in the latent feature space. Based on the sampling feature vector, the decoder network is reconstructed to calculate the error. The compressed features are expanded to the original data dimension using a symmetric neural network structure. The decoder network is opposite to the encoder network structure and includes three fully connected layers with increasing number of neurons. The data processing flow inputs the sampling feature vector into the first layer of the decoder, which contains 256 neurons. The low-dimensional features are mapped to a medium-dimensional space through weighted summation and an activation function. The second layer contains 512 neurons to further expand the feature dimension. The third layer has the same number of neurons as the original input data dimension, and outputs the reconstructed data. The reconstruction error is obtained by calculating the mean square error between the reconstructed data and the original standardized monitoring data set. The error value reflects the degree of information retention of the sampling feature vector. The smaller the error, the higher the information fidelity of the feature vector, forming a feature quality evaluation result.

[0049] The sampling feature vector and the feature quality evaluation result are comprehensively encoded to generate the final working state feature vector using a weighted fusion strategy. This process considers both the information content of the feature vector and the quality evaluation result. The comprehensive encoding process first calculates the quality weight based on the feature quality evaluation result. The quality weight is inversely proportional to the reconstruction error. The smaller the reconstruction error, the larger the quality weight. Then, each element of the sampling feature vector is multiplied by the corresponding quality weight to achieve quality-based feature weighting. Next, the weighted feature vector is concatenated with the quality evaluation result to form an extended feature vector containing both feature information and quality information. The final optoelectronic imaging assembly working state feature vector contains key feature information that has been quality-screened, as well as a quantitative indicator of feature reliability. This comprehensive encoding method ensures that the feature vector accurately reflects the working state of the optoelectronic imaging assembly while providing a reliability assessment of the feature quality.

[0050] In a specific embodiment, the process of performing step S103 can specifically include the following steps:

[0051] The modal separation processing is performed on the working state feature vector of the photoelectric imaging assembly to obtain temperature modal features, vibration modal features, electrical modal features and image modal features.

[0052] The intra-modal fusion processing is performed on the temperature modal features and the vibration modal features based on the attention weight calculation to obtain physical state fusion features.

[0053] The inter-modal correlation processing is performed on the electrical modal features and the image modal features and the physical state fusion features according to the cross-attention mechanism to obtain performance state correlation features.

[0054] The performance state correlation features are input into a support vector machine classifier for state classification processing to obtain a state category probability distribution.

[0055] The confidence evaluation and label mapping processing are performed on the state category probability distribution to obtain a real-time working state identification of the photoelectric imaging assembly.

[0056] Specifically, the modal separation processing of the photoelectric imaging component working state feature vector adopts a feature dimension index mapping method, and the dimension position of different data sources in the feature vector is classified and divided. The modal separation process is based on the dimension arrangement rule when the feature vector is constructed, and the 129-dimensional photoelectric imaging component working state feature vector is divided into four independent modal subsets according to the data source. The temperature modal feature corresponds to the first 32 dimensions of the feature vector, which includes the temperature distribution, temperature gradient, temperature change rate and other related features extracted from the CCD sensor and CMOS sensor temperature monitoring points. The vibration modal feature corresponds to the 33rd to 64th dimension, which includes the frequency domain features, time domain statistical features and vibration mode recognition features of the three-axis vibration acceleration of the optical lens assembly. The electrical modal feature corresponds to the 65th to 96th dimension, which includes the stability features, fluctuation amplitude features and power spectral density features of the voltage and current monitoring values of the optical filter system. The image modal feature corresponds to the 97th to 128th dimension, which includes the image sensor signal-to-noise ratio, modulation transfer function, image quality evaluation and other imaging performance features. The last dimension is the feature quality evaluation result. The modal intra-fusion processing of the temperature modal feature and the vibration modal feature based on the attention weight calculation adopts a self-attention mechanism to quantify the importance distribution of the two modal internal features. The attention weight calculation realizes the feature importance evaluation by calculating the similarity matrix between the intra-modal features. The specific calculation process performs dot product operation between each dimension in the temperature modal feature and other dimensions to generate a 32x32 similarity matrix. The attention weight distribution within the temperature modal is obtained by performing softmax normalization on the similarity matrix. Similarly, the 32x32 attention weight matrix of the vibration modal feature is calculated. Then the attention weight is weighted and summed with the corresponding feature value. The temperature modal fusion feature is obtained by multiplying the 32 temperature features by their corresponding attention weights and then summing them to obtain a single temperature representative value. The vibration modal fusion feature is obtained by the same method. Finally, the temperature representative value and the vibration representative value are spliced to form the physical state fusion feature.

[0057] According to the cross attention mechanism, the electrical modal features, the image modal features and the physical state fusion features are processed for inter-modal correlation. A cross-modal interaction calculation method is adopted. The cross attention mechanism quantifies the correlation degree between different modalities by calculating the interaction intensity between different modalities. In the inter-modal correlation processing process, the cross attention weight between the electrical modal features and the image modal features is calculated first. The 32-dimensional features of the electrical modal are taken as the query vector, and the 32-dimensional features of the image modal are taken as the key value vector. The similarity between the query vector and the key value vector is calculated by matrix multiplication to obtain a 32x32 cross attention matrix. Then, the cross attention weight between the physical state fusion features and the electrical modal features is calculated. The physical state fusion features are taken as the query vector to interact with the electrical modal features. At the same time, the cross attention weight between the physical state fusion features and the image modal features is calculated. This three-way interaction calculation establishes the correlation mapping between the temperature vibration physical state and the electrical image performance state. The performance state correlation features are obtained by weighted fusion of the three cross attention calculation results. The features comprehensively reflect the mutual influence relationship between different working modalities of the optoelectronic imaging assembly. The performance state correlation features are input into a support vector machine classifier for state classification processing. A kernel function mapping and optimal separating hyperplane construction method is adopted. The support vector machine classifier finds the optimal decision boundary in the high-dimensional space by mapping the input features to the high-dimensional space to realize multi-class classification. In the state classification processing process, the performance state correlation features are taken as the input vector and input into the trained support vector machine model. The support vector machine uses a radial basis function as the kernel function to map the input features to a high-dimensional feature space. In the high-dimensional space, a separating hyperplane is constructed to classify different working states. The classifier outputs score values of four state categories including normal operation, performance decline, abnormal fluctuation and precursor of failure. The score values are converted into probability distribution by a softmax function. Each state category corresponds to a probability value, and the sum of all probability values is equal to 1, forming a state category probability distribution.

[0058] The state category probability distribution is processed for confidence evaluation and label mapping. A probability threshold judgment and confidence quantification method is adopted. The confidence evaluation analyzes the concentration degree of the probability distribution to judge the reliability of the classification result. In the confidence evaluation process, the difference between the maximum probability value and the second maximum probability value in the state category probability distribution is calculated as the confidence index. The larger the difference, the more certain the classification result and the higher the confidence. At the same time, the entropy value of the probability distribution is calculated to quantify the uncertainty of the distribution. The smaller the entropy value, the more concentrated the probability distribution and the more reliable the classification result. The label mapping process determines the final state label according to the state category corresponding to the maximum probability value in the probability distribution. When the confidence exceeds the preset threshold, the corresponding state label is directly output. When the confidence is lower than the threshold, an uncertain state label is output and an artificial review mechanism is triggered. Finally, the real-time working state label of the optoelectronic imaging assembly is obtained.

[0059] In a specific embodiment, the process of performing step S104 can specifically include the following steps:

[0060] The historical state sequence composed of the real-time working state identification of the photoelectric imaging assembly is input into a Transformer encoder for time sequence pattern extraction processing to obtain state evolution pattern features;

[0061] Based on the multi-head self-attention mechanism, the state evolution pattern features are analyzed and processed for long-term and short-term dependency relationship to obtain a time sequence correlation weight matrix;

[0062] According to the time sequence correlation weight matrix, the state change trend of the future time window is calculated and processed for prediction to obtain a predicted state probability sequence;

[0063] The predicted state probability sequence is processed for fault type matching and probability threshold judgment to obtain a fault occurrence probability evaluation result;

[0064] The fault occurrence probability evaluation result is compared and analyzed with a preset warning level threshold to obtain photoelectric imaging assembly fault warning information.

[0065] Specifically, the history state sequence composed of real-time working state identifiers of the photoelectric imaging assembly is input into a Transformer encoder for time series pattern extraction processing using sequence encoding and position embedding technology. The Transformer encoder captures global dependencies in time series through a self-attention mechanism. The time series pattern extraction process first converts each state identifier in the history state sequence into a corresponding numerical code, with normal operation state encoded as 1, performance degradation state encoded as 2, abnormal fluctuation state encoded as 3, and fault precursor state encoded as 4, forming a numerical state sequence. Then, the state sequence is subjected to position encoding processing. The position encoding generates a unique position vector for each time position in the sequence through sine and cosine functions, ensuring that the model can distinguish state information at different times. The position encoding vector is added to the state encoding vector to form an input vector containing time information. The Transformer encoder adopts a multi-layer structure, each layer containing a self-attention sublayer and a feedforward neural network sublayer. The self-attention sublayer calculates the correlation strength between each time in the sequence and all other times, and the feedforward network sublayer performs nonlinear transformation on the attention output. Residual connection and layer normalization are used to ensure training stability. Finally, the state evolution pattern feature is output, which captures the time evolution law and pattern change trend in the history state sequence. Based on the multi-head self-attention mechanism, long and short-term dependency analysis processing of the state evolution pattern feature is performed using a parallel computing method with multiple attention heads. Each attention head focuses on capturing dependency relationships at different time scales. The multi-head self-attention mechanism projects the state evolution pattern feature into three spaces: query, key, and value. Each attention head uses a different projection matrix to generate independent query, key, and value vectors. The first attention head focuses on short-term dependencies, calculating the direct influence between adjacent time states. The second attention head focuses on medium-term dependencies, analyzing the state correlation between time intervals. The remaining attention heads focus on dependency patterns at longer time scales.

[0066] The dependency analysis process obtains attention weights by calculating the dot product similarity of the query vector and the key-value vector, and then normalizing by softmax. The attention weights reflect the influence degree of different time states on the current prediction. The attention weights and the numerical vectors are weighted and summed to obtain the output of each attention head. The outputs of all attention heads are spliced and linearly transformed to form a time correlation weight matrix. The rows of the matrix represent different time positions, and the columns represent the contribution weights of each time to the prediction. The greater the weight value, the more important the state of that time to the future prediction. The state change trend of the future time window is calculated and processed according to the time correlation weight matrix using a weighted prediction and probability reasoning method. The prediction calculation infers the occurrence probability of the future state based on the historical state pattern and the correlation weight. The prediction calculation process first determines the length of the prediction time window, which is usually set to a time range of 7 to 30 days in the future. Then, according to the weight distribution in the time correlation weight matrix, the historical time with the greatest impact on the prediction is determined. The state information of these key time points is weighted and combined with the corresponding weight values to calculate the occurrence probability of each state at each future time. The specific calculation method is to perform matrix multiplication operation on the one-hot encoding of the historical state and the time correlation weight to obtain the state probability vector of each time in the future time window. The probability vector contains four elements corresponding to the occurrence probability of the four working states. The probability vectors of all times form a prediction state probability sequence, which describes the state evolution trend and probability distribution of the optoelectronic imaging assembly in the future time period.

[0067] The fault type matching and probability threshold judgment process of the predicted state probability sequence adopts pattern recognition and probability statistics method, and the fault type matching is realized by similarity comparison between the predicted probability sequence and the predefined fault evolution pattern. The fault type matching process establishes the probability sequence templates of eight typical fault evolution patterns, including sensor overheating failure pattern, optical lens contamination pattern, filter aging pattern, circuit parameter drift pattern, etc. Each pattern corresponds to a specific state probability change trajectory. By calculating the cosine similarity between the predicted state probability sequence and various fault pattern templates, the most matched fault type is determined, and the pattern with the highest similarity is identified as the possible fault type. The probability threshold judgment process sets the trigger threshold for different states. When the probability of the fault precursor state in the predicted sequence exceeds the preset threshold, the cumulative probability of the fault type is calculated. The cumulative probability is obtained by weighted summation of the fault probabilities at all times within the prediction time window, and the weight is set according to the time distance, i.e. the closer the time, the greater the weight. Finally, the fault occurrence probability evaluation result is obtained. The comparison and analysis between the fault occurrence probability evaluation result and the preset warning level threshold adopts the hierarchical warning and risk assessment method. The comparison and analysis determines the warning level and response measures by comparing the fault probability with different level thresholds. The comparison and analysis process sets three warning level thresholds, i.e. yellow warning corresponds to fault probability of 30%-50%, orange warning corresponds to 50%-70%, and red warning corresponds to more than 70%. When the fault occurrence probability evaluation result exceeds a certain level threshold, the warning information of the corresponding level is triggered to generate. The warning information includes fault type, expected occurrence time, impact degree and recommended maintenance measures, etc. At the same time, the warning priority is adjusted according to the severity of the fault type and the importance of the equipment, and the final photoelectric imaging component fault warning information is formed.

[0068] In a specific embodiment, the process of performing step S105 can specifically include the following steps:

[0069] Based on the consistency hash algorithm, the photoelectric imaging component fault warning information is subjected to data fragmentation processing to obtain distributed storage data blocks;

[0070] According to the network quality evaluation result, the distributed storage data blocks are subjected to transmission path optimization processing to obtain the optimal transmission strategy;

[0071] The distributed storage data blocks and the optimal transmission strategy are input into the cloud collaborative computing node for load balancing processing to obtain the computing task allocation result;

[0072] Based on the knowledge graph reasoning engine, the computing task allocation result is subjected to decision rule matching processing to obtain the maintenance decision suggestion;

[0073] The maintenance decision suggestion is subjected to visual rendering and report generation processing to obtain the photoelectric imaging component remote monitoring result.

[0074] Specifically, the data sharding processing of the photoelectric imaging component fault early warning information based on the consistent hashing algorithm adopts the hash ring mapping and virtual node technology. The consistent hashing algorithm realizes the uniform distribution and dynamic expansion of data by mapping the data and the storage node to the same hash ring. The data sharding processing process firstly splices the device identifier and the timestamp of the photoelectric imaging component fault early warning information to form a unique data key value, and then uses the MD5 hash function to calculate the hash of the data key value to obtain a 32-bit hexadecimal hash value. The hash value is converted into a 32-bit unsigned integer as the position of the data on the hash ring. The size of the hash ring is set to 2 raised to the power of 32, and the storage node is also determined by hash calculation to determine the position on the ring. Each physical storage node creates multiple virtual nodes on the ring to improve the uniformity of data distribution. The data is distributed to the nearest storage node in the clockwise direction. When the data hash value is determined, the algorithm finds the first storage node along the hash ring in the clockwise direction as the main storage location of the data, and selects the next two nodes as the backup storage location to realize the three-copy redundant storage of the data. Each data shard contains the original early warning information, node identifier, copy identifier and check code, and finally forms a distributed storage data block. The transmission path optimization processing of the distributed storage data block according to the network quality evaluation result adopts the multi-path selection and dynamic adjustment strategy. The network quality evaluation calculates the network quality index by comprehensively analyzing the bandwidth, delay, packet loss rate and signal strength and other indicators. The transmission path optimization process firstly evaluates the quality of the available transmission paths, including 4G / 5G cellular network, WiFi wireless network, wired Ethernet and satellite communication and other transmission modes. The network quality index of each transmission mode is calculated by weighted calculation. The bandwidth weight accounts for 40%, the delay weight accounts for 30%, the packet loss rate weight accounts for 20%, and the signal strength weight accounts for 10%.

[0075] The path optimization algorithm allocates paths according to the priority of the distributed storage data block and the network quality index. The emergency early warning data preferentially selects the transmission path with the highest quality index, the important early warning data selects the path with the second highest quality index, and the regular monitoring data is allocated to multiple paths in a load balancing manner. The transmission cost and reliability factors are considered to select the main transmission path and the backup transmission path for each data block. When the main path fails, it automatically switches to the backup path to form an optimal transmission strategy including path selection, transmission order, retransmission strategy and switching conditions. The distributed storage data block and the optimal transmission strategy are input into the cloud collaborative computing node for load balancing processing using a task scheduling and resource allocation algorithm. Load balancing achieves reasonable allocation of tasks by monitoring the CPU usage, memory occupancy and network bandwidth utilization of each computing node. The load balancing process establishes a computing node resource state monitoring mechanism. Each node regularly reports its resource usage, including CPU core number, memory capacity, storage space and network bandwidth, as well as dynamic indicators such as current CPU load, memory usage, disk I / O rate and network traffic. The task allocation algorithm allocates tasks according to the computational complexity and resource requirements of the distributed storage data block, combined with the resource state of each node and the path information of the optimal transmission strategy, using a target function of minimizing the total completion time to allocate tasks. The data processing tasks with high computational complexity are allocated to powerful computing nodes, and the tasks with low computational complexity are allocated to lightly loaded nodes. The balance between data transmission time and computation time is considered to avoid network transmission as a bottleneck. Finally, the computing task allocation result including task allocation mapping, execution order, resource reservation and load monitoring is obtained.

[0076] The decision rule matching process of the computing task allocation result based on the knowledge graph reasoning engine adopts semantic reasoning and rule matching technology. The knowledge graph contains entities such as optoelectronic imaging components, fault types, maintenance measures, technical personnel, and spare parts inventory, as well as the relationships between them. The decision rule matching process first maps the fault information in the computing task allocation result to the corresponding entities in the knowledge graph, establishing an association between fault types and optoelectronic imaging component models. Then, through a graph traversal algorithm, it searches for maintenance rule nodes related to the current fault. Each maintenance rule node contains attributes such as fault trigger conditions, maintenance steps, required tools, estimated time, and skill requirements. The reasoning engine uses a forward reasoning method, starting from the current fault fact and gradually matching applicable maintenance rules. When the fault type is sensor overheating, the reasoning engine matches the heat dissipation system inspection rule, temperature calibration rule, and sensor replacement rule. According to the fault severity and equipment importance, it determines the priority order, while considering the availability of maintenance resources, including the skill level, workload, and geographic location of technical personnel, the inventory quantity, procurement cycle, and cost budget of spare parts, and finally generates a maintenance decision suggestion containing maintenance plan, execution steps, resource demand, and time arrangement. The visualization rendering and report generation process of the maintenance decision suggestion adopts graphical interface design and automated report generation technology. The visualization rendering displays monitoring information at different levels of detail through multiple levels of interface. The visualization rendering process converts the maintenance decision suggestion into an intuitive graphical interface, including device status overview, fault distribution heat map, maintenance timeline, and resource allocation Gantt chart. The device status overview uses color coding to display the real-time status of all optoelectronic imaging components, with green indicating normal, yellow indicating attention, orange indicating warning, and red indicating fault. The fault distribution heat map displays the frequency and severity distribution of faults for different device types and geographic locations.

[0077] The report generation process adopts templating design and automated content filling technology. The report template contains standard sections such as execution summary, fault analysis, maintenance suggestion, resource demand, risk assessment, and follow-up plan. Automated content filling extracts key information from the maintenance decision suggestion and organizes it into report content according to the predetermined format. The report supports multiple output formats including PDF document, HTML webpage, and Excel table, and generates a simplified version adapted for mobile devices. Finally, the optoelectronic imaging component remote monitoring result is obtained, which contains complete monitoring analysis results, decision suggestions, and execution plans.

[0078] In a specific embodiment, the data sharding process of the optoelectronic imaging component fault warning information based on the consistent hash algorithm can specifically include the following steps:

[0079] Hash value calculation is performed on the device identifier and timestamp of the optoelectronic imaging component fault warning information to obtain the data sharding key value.

[0080] According to the data shard key-value pair, the node position mapping processing is performed on the hash ring space to obtain a storage node allocation result.

[0081] Based on the storage node allocation result, the data content of the fault warning information is blocked and cut to obtain a data shard unit.

[0082] The data shard unit is subjected to redundancy backup and check code generation processing to obtain a fault-tolerant data block.

[0083] The fault-tolerant data block and the node identification information are encapsulated and combined to obtain a distributed storage data block.

[0084] Specifically, the device identification and timestamp of the photoelectric imaging component fault warning information are subjected to hash value calculation processing, which adopts string splicing and hash function mapping technology. The hash value calculation generates a fixed-length hash value through mathematical transformation after combining the device unique identifier and the precise timestamp. The hash value calculation process first extracts the device identification field in the fault warning information, which contains a unique identifier composed of device model, serial number and deployment location, and then obtains the precise timestamp when the fault warning information is generated. The timestamp is represented in Unix time format, which represents the number of seconds since January 1, 1970, ensuring time accuracy and uniqueness. Then the device identification and timestamp are combined into a composite key value through string splicing operation, and the splicing format is device identification plus underscore plus timestamp, forming a combined string similar to "DEVICE001_1640995200". Then the SHA-256 hash algorithm is used to calculate the hash value of the combined string. The SHA-256 algorithm converts an input string of any length into a 256-bit hash value through a series of logical operations and bit operations. The hash value is represented in 64-bit hexadecimal string form, ensuring that the same input produces the same output and different inputs produce different outputs, and finally obtaining a unique data shard key value. According to the data shard key-value pair, the node position mapping processing is performed on the hash ring space to obtain a storage node allocation result.

[0085] Meanwhile, the hash ring position calculation is performed for all available storage nodes, the hash value of the storage node is obtained by performing SHA-256 hash calculation on the IP address and port number of the node, and each physical storage node creates multiple virtual nodes on the hash ring to improve the uniformity of data distribution, and the hash value of the virtual node is generated by adding a serial number suffix to the node identifier and then performing hash calculation. The node mapping algorithm starts from the hash ring position of the data, finds the first encountered storage node in the clockwise direction as the primary storage node, continues to find the next node in the clockwise direction as the first backup node, and then finds the next node as the second backup node, realizes the three-copy storage strategy of the data, and finally obtains the storage node allocation result containing the information of the primary node and the backup node. Based on the storage node allocation result, the data content of the fault warning information is processed by block cutting, and the fixed size block and boundary alignment technology are used. The data block cutting is performed by dividing the original data into multiple independent data segments according to the predetermined size. In the block cutting process, the standard size of the data block is first determined, which is usually set to 64KB or 128KB to balance the storage efficiency and transmission performance, then the complete data content of the fault warning information is sequentially read in the form of byte stream, and a data segment unit is created when the standard block size is reached. When the original data size cannot be divided by the block size, the last data block is padded to the standard size, and the padding content uses a specific padding byte mode to ensure data integrity and format consistency. Each data segment unit contains segment number, original data content, data length and boundary identifier, the segment number uses incremental numbering to ensure the correct order during data recombination, the original data content maintains the exact consistency with the source data, the data length records the effective data byte number of the segment, and the boundary identifier marks the start and end positions of the segment in the original data, and finally a structured data segment unit set is obtained.

[0086] The redundancy backup and check code generation process of the data slice unit adopts the erasure code technology and the cyclic redundancy check algorithm. The redundancy backup improves the fault tolerance and recovery performance of the data by generating additional redundant data blocks. The redundancy backup process adopts the Reed-Solomon erasure code algorithm, which regards the original data slices as the coefficients of a polynomial and generates redundant check slices through polynomial operation. Specifically, a plurality of data slice units form a coding group, which usually contains 4 data slices and 2 check slices. The check slices are generated by linear combination operation on the data slices. Even if any 2 slices are lost, the original data can be completely recovered through the remaining slices. The check code generation adopts the CRC-32 cyclic redundancy check algorithm to calculate the check code of the content of each data slice unit. The CRC-32 algorithm generates a 32-bit check code through polynomial division operation, which can detect single-bit errors and burst errors occurring in the data transmission or storage process. The backup process also includes metadata redundancy, which creates metadata records containing slice information, check code, timestamp and version number for each data slice. The metadata also adopts multi-copy storage to ensure the reliability of the management information. Finally, the fault-tolerant data block containing the original data, redundant check data and integrity check information is obtained. The encapsulation and combination of the fault-tolerant data block and the node identification information adopt the data encapsulation protocol and identification mapping technology. The encapsulation and combination package the data content and control information into a unified storage format. The encapsulation and combination process first creates a data block header structure, which contains magic number identification, version number, data block type, compression identification, encryption identification and integrity check fields. The magic number identification uses a fixed byte sequence to mark the start position of the data block. The version number records the version information of the data format for subsequent compatibility processing. The data block type distinguishes between original data blocks and redundant check blocks.

[0087] The node identification information includes the IP address, port number, node weight and load state of the main storage node, the identification information and storage path of the backup node, and the cluster identification and consistency version number of the distributed storage system. The encapsulation process organizes the content part of the fault-tolerant data block, the header information and the node identification together through a specific data structure, adopts the TLV encoding format, i.e. the triple structure of type-length-value, to ensure the self-descriptiveness and scalability of the data. The final distributed storage data block contains complete data content, redundancy protection information, node allocation information and system metadata, supporting the functional requirements of distributed storage, fault-tolerant recovery, load balancing and consistency maintenance.

[0088] The above describes the method for remotely monitoring the working state of the photoelectric imaging assembly in the embodiments of the application, and the following describes the system for remotely monitoring the working state of the photoelectric imaging assembly in the embodiments of the application. Please refer to Figure 2The working state remote monitoring system for the photoelectric imaging assembly in the embodiment of the present application comprises the following modules:

[0089] The processing module is configured to perform structural processing on the temperature data, vibration data, electrical parameter data and image quality data of the photoelectric imaging assembly to obtain a standardized monitoring data set.

[0090] The encoding module is configured to perform feature extraction and encoding processing on the standardized monitoring data set according to a variational autoencoder algorithm to obtain a working state feature vector of the photoelectric imaging assembly.

[0091] The recognition module is configured to perform state recognition processing on the working state feature vector of the photoelectric imaging assembly by a multi-modal fusion algorithm to obtain a real-time working state identification of the photoelectric imaging assembly.

[0092] The diagnosis module is configured to perform predictive fault diagnosis processing on the real-time working state identification of the photoelectric imaging assembly according to a time series analysis network to obtain fault early warning information of the photoelectric imaging assembly.

[0093] The transmission module is configured to perform remote transmission and decision support processing on the fault early warning information of the photoelectric imaging assembly through a distributed data processing architecture to obtain a remote monitoring result of the photoelectric imaging assembly.

[0094] The above Figure 2 The working state remote monitoring system for the photoelectric imaging assembly in the embodiment of the present application is described in detail from the perspective of modular functional entities, and the working state remote monitoring device for the photoelectric imaging assembly in the embodiment of the present application is described in detail from the perspective of hardware processing.

[0095] Referring to Figure 3 The embodiment of the present application also provides a working state remote monitoring device for a photoelectric imaging assembly. The working state remote monitoring device for the photoelectric imaging assembly can be a server, and the internal structure thereof can be as shown in Figure 3 The working state remote monitoring device for the photoelectric imaging assembly comprises a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer is configured to provide computing and control capabilities. The memory of the working state remote monitoring device for the photoelectric imaging assembly comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The database of the working state remote monitoring device for the photoelectric imaging assembly is configured to store corresponding data in the embodiment. The network interface of the working state remote monitoring device for the photoelectric imaging assembly is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0096] Those skilled in the art can understand that, Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the working state remote monitoring device for the photoelectric imaging assembly to which the scheme of the present application is applied.

[0097] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium, and can also be a volatile computer readable storage medium, and the computer readable storage medium stores instructions, and when the instructions run on a computer, the computer executes the steps of the working state remote monitoring method for the photoelectric imaging assembly.

[0098] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0099] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical scheme of the present application or the part of the prior art that essentially contributes or the whole or part of the technical scheme can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing an optical imaging assembly working state remote monitoring device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0100] The above embodiments are only used to illustrate the technical scheme of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical scheme recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical scheme deviate from the spirit and scope of the technical scheme of each embodiment of the present application.

Claims

1. A method for remote monitoring of the operating state of an electrophotographic imaging assembly, characterized in that, The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The real-time working state identifier of the photoelectric imaging assembly is predicted and diagnosed in real time according to a time sequence analysis network to obtain photoelectric imaging assembly fault early warning information, including: inputting a historical state sequence formed by the real-time working state identifier of the photoelectric imaging assembly into a Transformer encoder for time sequence mode extraction processing to obtain state evolution mode features; performing long-short term dependency relationship analysis processing on the state evolution mode features based on a multi-head self-attention mechanism to obtain a time sequence correlation weight matrix; performing prediction calculation processing on a state change trend of a future time window according to the time sequence correlation weight matrix to obtain a predicted state probability sequence; performing fault type matching and probability threshold judgment processing on the predicted state probability sequence to obtain a fault occurrence probability evaluation result; and performing comparison analysis processing on the fault occurrence probability evaluation result and a preset early warning level threshold to obtain the photoelectric imaging assembly fault early warning information; The photoelectric imaging assembly fault early warning information is transmitted and decision supported remotely through a distributed data processing architecture to obtain photoelectric imaging assembly remote monitoring results.

2. The method for remote monitoring of the operating state of an optoelectronic imaging assembly according to claim 1, characterized in that, The standardized monitoring data set is feature extracted and encoded according to a variational autoencoder algorithm to obtain a photoelectric imaging assembly working state feature vector, including: The standardized monitoring data set is input into an encoder network of a variational autoencoder for dimension compression processing to obtain latent feature space data; The latent feature space data is subjected to mean-variance calculation processing to obtain feature distribution parameters; The feature distribution parameters are subjected to probability sampling processing on a reparameterization sampling process to obtain a sampling feature vector; The sampling feature vector is subjected to reconstruction error calculation processing on a decoder network to obtain a feature quality evaluation result; The sampling feature vector and the feature quality evaluation result are subjected to comprehensive encoding processing to obtain the photoelectric imaging assembly working state feature vector.

3. The method for remote monitoring of operating conditions of optoelectronic imaging assemblies according to claim 1, characterized in that, The photoelectric imaging assembly fault early warning information is transmitted and decision supported remotely through a distributed data processing architecture to obtain photoelectric imaging assembly remote monitoring results, including: The photoelectric imaging assembly fault early warning information is subjected to data sharding processing based on a consistent hashing algorithm to obtain distributed storage data blocks; The distributed storage data blocks are subjected to transmission path optimization processing according to network quality evaluation results to obtain an optimal transmission strategy; The distributed storage data blocks and the optimal transmission strategy are input into cloud collaborative computing nodes for load balancing processing to obtain computing task allocation results; The computing task allocation results are subjected to decision rule matching processing based on a knowledge graph reasoning engine to obtain maintenance decision suggestions; The maintenance decision suggestions are subjected to visual rendering and report generation processing to obtain the photoelectric imaging assembly remote monitoring results.

4. The method for remote monitoring of the operating state of an optoelectronic imaging assembly according to claim 3, characterized in that, The photoelectric imaging assembly fault early warning information is subjected to data sharding processing based on a consistent hashing algorithm to obtain distributed storage data blocks, including: Hash value calculation processing is performed on device identifiers and time stamps of the photoelectric imaging assembly fault early warning information to obtain data sharding key values; According to the data slice key-value pair, a node position mapping process is performed on the hash ring space to obtain a storage node allocation result; Based on the storage node allocation result, a data content of a fault early warning information is blocked and cut to obtain a data slice unit; A redundancy backup and a check code generation process are performed on the data slice unit to obtain a fault-tolerant data block; The fault-tolerant data block and node identification information are encapsulated and combined to obtain the distributed storage data block.

5. A system for remote monitoring of the operating state of an electrophotographic imaging assembly, characterized in that, The working state remote monitoring method for the photoelectric imaging assembly according to any one of claims 1-4, the working state remote monitoring system for the photoelectric imaging assembly comprises: The processing module is configured to perform structured processing on temperature data, vibration data, electrical parameter data and image quality data of the photoelectric imaging assembly to obtain a standardized monitoring data set, including: performing outlier detection processing on temperature monitoring point data of a CCD sensor and a CMOS sensor to obtain temperature anomaly identification data; performing missing value interpolation processing on three-axis vibration acceleration data of an optical lens assembly according to the temperature anomaly identification data to obtain a vibration data sequence, when the temperature anomaly identification is 1, it indicates that the sensor is in an abnormal state, at this time, the missing value interpolation needs to consider the influence of temperature anomaly on vibration measurement, a weighted interpolation method is used to reduce the influence weight of data before and after the abnormal period, when the temperature anomaly identification is 0, it indicates that the sensor is working normally, a standard linear interpolation or cubic spline interpolation method is used to generate a vibration data sequence without missing values by fitting the vibration trend curve before and after the missing point to predict the vibration value at the missing position; based on the vibration data sequence, noise filtering processing is performed on voltage and current monitoring values of a filter system to obtain electrical parameter purification data, wherein the data processing logic is based on the correlation between vibration and electrical parameter noise, when the vibration amplitude is large, contact noise is easily generated in electrical connection, therefore, the filtering strength needs to be enhanced, when the vibration amplitude is small, the electrical signal is relatively stable, a light filtering is used to avoid signal distortion; the signal-to-noise ratio data of the image sensor and the electrical parameter purification data are input into a standardization algorithm for numerical normalization processing to obtain standardized quality parameters; the temperature anomaly identification data, the vibration data sequence, the electrical parameter purification data and the standardized quality parameters are subjected to data fusion processing to obtain the standardized monitoring data set; The encoding module is configured to perform feature extraction and encoding processing on the standardized monitoring data set according to a variational autoencoder algorithm to obtain a photoelectric imaging assembly working state feature vector. The recognition module is configured to perform state recognition processing on the optoelectronic imaging assembly working state feature vector through a multi-modal fusion algorithm to obtain an optoelectronic imaging assembly real-time working state identifier, including: performing modal separation processing on the optoelectronic imaging assembly working state feature vector to obtain temperature modal features, vibration modal features, electrical modal features, and image modal features; performing intra-modal fusion processing on the temperature modal features and the vibration modal features based on attention weight calculation to obtain physical state fusion features; performing inter-modal correlation processing on the electrical modal features, the image modal features, and the physical state fusion features according to a cross-attention mechanism to obtain performance state correlation features; inputting the performance state correlation features into a support vector machine classifier to perform state classification processing to obtain a state category probability distribution; performing confidence evaluation and label mapping processing on the state category probability distribution to obtain the optoelectronic imaging assembly real-time working state identifier; The diagnosis module is configured to perform predictive fault diagnosis processing on the optoelectronic imaging assembly real-time working state identifier according to a time series analysis network to obtain optoelectronic imaging assembly fault early warning information, including: inputting a historical state sequence formed by the optoelectronic imaging assembly real-time working state identifier into a Transformer encoder to perform time series mode extraction processing to obtain state evolution mode features; performing long-short term dependency relationship analysis processing on the state evolution mode features based on a multi-head self-attention mechanism to obtain a time series correlation weight matrix; performing prediction calculation processing on a state change trend of a future time window according to the time series correlation weight matrix to obtain a predicted state probability sequence; performing fault type matching and probability threshold judgment processing on the predicted state probability sequence to obtain a fault occurrence probability evaluation result; performing comparative analysis processing on the fault occurrence probability evaluation result and a preset early warning level threshold to obtain the optoelectronic imaging assembly fault early warning information; The transmission module is configured to perform remote transmission and decision support processing on the optoelectronic imaging assembly fault early warning information through a distributed data processing architecture to obtain an optoelectronic imaging assembly remote monitoring result.

6. A working condition remote monitoring apparatus for an electrophotographic imaging assembly, characterized by, The computer program is configured to enable the processor to perform the optoelectronic imaging assembly working state remote monitoring method according to any one of claims 1 to 4 when the computer program is executed on the processor.

7. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is configured to enable the processor to perform the optoelectronic imaging assembly working state remote monitoring method according to any one of claims 1 to 4 when the computer program is executed on the processor.

Images