State detection method and system for mass production of laser radars

Abstract

The invention relates to the technical field of intelligent sensors, and discloses a state detection method and system for mass production of laser radars, and the method comprises the steps: carrying out the fusion processing of laser radar sensing data and environment sensor data through multi-modal data, and obtaining an operation feature set; extracting a space-time correlation mode by adopting a convolutional neural network, and determining a boundary distinguishing index of environment interference and state change of the equipment; adjusting the feature weight through an adaptive filtering algorithm, and obtaining an optimized feature fusion vector; analyzing the dynamic sequence change by using a long short-term memory network, judging a fault mode and obtaining state classification probability distribution; perceptual parameters are updated according to the fault positioning coordinate system, and a real-time adjustment instruction sequence is generated through environment compensation and abnormal propagation analysis. According to the method, the problem of insufficient laser radar state detection precision in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, and in particular to a method and system for state detection of mass-produced lidar. Background Technology

[0002] Currently, in the field of intelligent sensors, LiDAR serves as a core environmental perception sensor, and its operational reliability directly determines the performance and safety of the entire system. Especially in complex industrial environments, LiDAR needs to maintain stable operation amidst multiple interference factors such as temperature fluctuations, humidity changes, and dust accumulation, which places higher demands on the real-time monitoring and accurate diagnosis of equipment status.

[0003] In existing technologies, lidar status detection primarily relies on fixed threshold judgments or statistical analysis based on a single data source. The system collects the raw sensing signals from the lidar and compares them with preset static thresholds to determine the device status. For example, when the signal strength consistently falls below a certain fixed threshold, an alarm for device anomaly is triggered. Another approach attempts to introduce environmental sensors, but only performs simple comparisons or independent analyses of the sensing data and environmental data, failing to achieve deep fusion of multi-source information. When dealing with complex and changing working scenarios, these methods often struggle to effectively distinguish between similar abnormal behaviors caused by environmental interference (such as signal attenuation due to mirror condensation) and equipment malfunctions (such as performance degradation due to laser aging). Existing technologies, lacking effective fusion and deep feature correlation analysis of multimodal data, blur the boundary between environmental interference and changes in the device's own status, making accurate fault tracing and status classification difficult. Furthermore, static threshold mechanisms and single-dimensional analysis are ill-suited to dynamically changing working environments, resulting in insufficient accuracy in status assessment and limited timeliness of fault warnings.

[0004] Therefore, existing technologies suffer from insufficient accuracy in lidar state detection. Summary of the Invention

[0005] This invention provides a method and system for state detection of mass-produced lidar, in order to solve the problem of insufficient state detection accuracy of lidar in the prior art.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a state detection method for mass-produced lidar, comprising: Spatial information is acquired through lidar, environmental data is acquired through environmental sensors, the acquired data is integrated and anomaly correction and consistency verification are performed to obtain a set of operational characteristics. Spatiotemporal variation features are extracted from the set of operational features and matched with a pre-established environmental interference feature library to obtain interference tags. Based on the interference tags, the boundary between environmental interference and the equipment's own state changes is determined and the boundary distinction index is calculated. If the boundary discrimination index exceeds the preset discrimination index threshold, then a subset of environmental features is extracted from the running feature set and recombined to obtain a feature fusion vector; A high-dimensional feature subset is obtained based on the feature fusion vector. Key node data is extracted from the high-dimensional feature subset and compared with a pre-established fault mode library. The fault mode category is determined based on the comparison result and the state classification probability distribution is calculated. Based on the state classification probability distribution, low-confidence regions are divided, feature subsets of the low-confidence regions are obtained, and deviation correction and spatial transformation are performed to determine the fault location coordinate system. Based on the fault location coordinate system, update the sensing parameters of the lidar and re-collect data to obtain the calibrated operating state vector. If the operating status vector shows an abnormal deviation, compensation is made for the abnormal deviation in response to environmental influences. After compensation, the operating status vector is updated and a status detection is performed to obtain the status detection result. Anomalies are obtained from the status detection results, and a comprehensive fault location path is determined by comparing signals and simulating paths from the anomalies. Based on the comprehensive fault location path, environmental data of the fault location is extracted to obtain an environmental feature set. Adjustment instructions are generated based on the environmental feature set, and the instructions are subjected to multi-scenario adaptation verification and hierarchical processing to obtain a real-time adjustment instruction sequence.

[0007] Secondly, the present invention provides a condition detection system for mass-produced lidar, comprising: The runtime feature acquisition module is used to acquire spatial information through lidar, acquire environmental data through environmental sensors, integrate the acquired data and perform anomaly correction and consistency verification to obtain a set of runtime features. The boundary differentiation module is used to extract spatiotemporal variation features from the set of operating features and match them with a pre-established environmental interference feature library to obtain interference tags. Based on the interference tags, the boundary between environmental interference and the changes in the device's own state is determined and the boundary differentiation index is calculated. The weight allocation module is used to extract a subset of environmental features from the running feature set and reorganize them to obtain a feature fusion vector if the boundary discrimination index exceeds a preset discrimination index threshold. The fault mode recognition module is used to obtain a high-dimensional feature subset based on the feature fusion vector, extract key node data from the high-dimensional feature subset and compare it with a pre-established fault mode library, determine the fault mode category based on the comparison result and calculate the state classification probability distribution. The fault coordinate localization module is used to divide the low-confidence region according to the state classification probability distribution, obtain the feature subset of the low-confidence region and perform deviation correction and spatial transformation to determine the fault localization coordinate system. The perception parameter update module is used to update the perception parameters of the lidar and re-acquire data according to the fault location coordinate system to obtain the calibrated operating status vector. An environmental compensation module is used to compensate for abnormal deviations caused by environmental factors if the operating state vector shows abnormal deviations. After compensation, the operating state vector is updated and a state detection is performed to obtain the state detection result. The anomaly analysis module is used to obtain anomalies from the status detection results, and determine the comprehensive fault location path by comparing the signals and simulating the paths of the anomalies. The instruction generation module is used to extract environmental data of the fault location based on the comprehensive fault location path, obtain an environmental feature set, generate adjustment instructions based on the environmental feature set, and perform multi-scenario adaptation verification and hierarchical processing on the instructions to obtain a real-time adjustment instruction sequence.

[0008] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a state detection method for mass-produced lidar as described in any one of the above.

[0009] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform any of the above-described state detection method for mass-produced lidar.

[0010] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention collects real-time sensing data from lidar and environmental sensor data, and uses a multi-modal data fusion method for preliminary processing. It can comprehensively analyze and filter environmental interference factors such as temperature changes, humidity effects and dust accumulation, thereby improving the integrity of the equipment operation feature set and providing a more reliable data foundation for subsequent status identification.

[0011] (2) This invention utilizes convolutional neural networks to extract spatiotemporal correlation patterns in signal attenuation and operating characteristics, thereby determining the boundary distinguishing index between environmental interference and equipment state changes. This helps to more clearly distinguish between external environmental influences and equipment internal state changes in complex operating data, thereby improving the accuracy of state analysis.

[0012] (3) When the boundary differentiation index exceeds the preset threshold, the present invention dynamically adjusts the weight allocation of the running features through an adaptive filtering algorithm to obtain an optimized feature fusion vector, thereby making the feature expression more adaptable to environmental changes and enhancing the stability of state representation.

[0013] (4) This invention uses long short-term memory network to analyze the dynamic sequence changes of high-dimensional feature subsets, judge potential fault modes and obtain state classification probability distribution, which can capture the temporal evolution law of equipment state and improve the timeliness of fault warning and the comprehensiveness of state identification.

[0014] (5) The present invention generates a real-time adjustment instruction sequence based on the comprehensive fault location path, and through multi-scenario adaptation verification and hierarchical processing, enables the lidar to maintain stable operation configuration in complex environments, thereby improving the overall adaptive capability and reliability of the system. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a state detection method for mass-produced lidar provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a state detection system for mass-produced lidar provided in the second embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Reference Figure 1 The first embodiment of the present invention provides a state detection method for mass-produced lidar, comprising the following steps: S11: Spatial information is acquired through lidar, environmental data is acquired through environmental sensors, the acquired data is integrated and anomaly correction and consistency verification are performed to obtain a set of operational characteristics; S12, extract spatiotemporal variation features from the set of operating features and match them with a pre-established environmental interference feature library to obtain interference tags. Based on the interference tags, determine the boundary between environmental interference and the changes in the equipment's own state and calculate the boundary distinction index. S13, if the boundary discrimination index exceeds the preset discrimination index threshold, then extract the environmental feature subset from the running feature set and reorganize it to obtain the feature fusion vector; S14. Obtain a high-dimensional feature subset based on the feature fusion vector, extract key node data from the high-dimensional feature subset and compare it with a pre-established fault mode library, determine the fault mode category based on the comparison result and calculate the state classification probability distribution. S15, based on the state classification probability distribution, divide the low confidence region, obtain the feature subset of the low confidence region and perform deviation correction and spatial transformation to determine the fault location coordinate system; S16. Based on the fault location coordinate system, update the sensing parameters of the lidar and re-collect data to obtain the calibrated operating state vector. S17, If the running state vector shows an abnormal deviation, then the abnormal deviation is compensated for the environmental impact, the running state vector is updated after compensation and state detection is performed to obtain the state detection result. S18, Obtain anomalies from the status detection results, and determine the comprehensive fault location path by comparing signals and simulating paths of the anomalies; S19. Based on the comprehensive fault location path, extract the environmental data of the fault location to obtain an environmental feature set. Generate adjustment instructions based on the environmental feature set and perform multi-scenario adaptation verification and hierarchical processing on the instructions to obtain a real-time adjustment instruction sequence.

[0018] In step S11, spatial information is acquired through lidar, environmental data is acquired through environmental sensors, the acquired data is integrated and anomaly correction and consistency verification are performed to obtain a set of operational features, including: Spatial information is acquired through lidar, environmental data is acquired through environmental sensors, and the initial operational characteristics are obtained by integrating them. If the initial operating characteristics exceed the preset characteristic threshold range, then the abnormal characteristic data is marked, and the type of environmental interference corresponding to the abnormal characteristic data is determined. Based on the type of environmental interference, the corresponding correction parameters are obtained from the pre-established environmental interference database, and the abnormal feature data is adjusted to obtain an adjusted feature data set. A consistency check is performed on the adjusted feature data set. If the consistency check passes, the adjusted feature data set is determined as the running feature set.

[0019] First, spatial information in the form of point clouds is acquired using LiDAR, while environmental data is collected simultaneously using temperature, humidity, and dust sensors, together forming the raw perception data. The LiDAR scanning frequency is set to 10Hz, a frequency determined based on the balance between the obstacle update rate requirements of typical autonomous driving applications (usually >5Hz) and hardware processing capabilities. The sampling frequency of the environmental sensors is set to 1Hz, a frequency chosen because changes in environmental physical quantities (such as temperature and humidity) are typically much slower than mechanical scanning frequencies, aiming to reduce the system's computational load. Second, the raw perception data undergoes preprocessing and feature extraction. First, the point cloud density and effective detection distance are calculated from the LiDAR point cloud. Point cloud density is obtained by counting the number of points per unit area (1m²); the effective detection distance is determined by the distance to the farthest effective reflection point in the point cloud data. Temperature (unit: °C), humidity (unit: %RH), and dust concentration (unit: μg / m³) are directly read from the environmental sensors. These five physical quantities together constitute a 5-dimensional feature vector, which is the initial operational feature.

[0020] Then, the values ​​of each dimension in the initial operating features are compared with their respective preset feature threshold ranges. These preset feature threshold ranges are based on historical data from 300 hours of continuous operation of the lidar under standard experimental conditions, with their 95% confidence intervals considered as normal ranges. For example, the point cloud density threshold is [1500, 3500] points / m², and the temperature threshold is [-10, 45] ℃. If data in a certain dimension exceeds its corresponding threshold, that dimension is marked as abnormal feature data, and the corresponding environmental interference type is determined based on the physical quantity represented by that dimension. For example, exceeding the temperature limit corresponds to "high temperature interference," and an abnormal decrease in point cloud density accompanied by an increase in dust concentration corresponds to "dust interference."

[0021] Next, the correction parameters corresponding to the environmental interference type are queried from a pre-established environmental interference database. The environmental interference database is constructed through controlled environmental experiments. In a temperature-controlled experimental chamber, the lidar is placed under a standard background environment, and a single environmental variable is systematically changed (e.g., the temperature is gradually increased from -10℃ to 60℃). At each stable temperature point, operational characteristic data such as point cloud density and effective detection distance are collected. By analyzing the functional relationship between specific environmental variables (e.g., temperature) and operational characteristics (e.g., point cloud density), a correction parameter lookup table or fitting curve is established. For example, for "high-temperature interference," the database stores a lookup table of compensation coefficients for point cloud density at different temperatures. The queried correction parameters are used to adjust abnormal characteristic data, forming an adjusted characteristic data set. For example, if the current temperature is 50℃ and the lookup table shows a compensation coefficient of 1.2, the original point cloud density value is multiplied by 1.2, while other characteristic dimensions that do not exceed the threshold remain unchanged, collectively forming the adjusted characteristic data set.

[0022] Finally, a logistic regression model was used to verify the consistency of the adjusted feature dataset. This logistic regression model was trained using 2000 sets of historical feature data labeled "stable" or "unstable," and the model parameters were solved using the maximum likelihood estimation method. The decision rule for consistency verification relies on a preset stability probability threshold. This threshold was determined by evaluating the model's performance using an independent validation set (containing 500 sets of data not used in training) after training. Precision and recall were calculated at different decision thresholds, and a precision-recall curve was plotted. A suitable probability cutoff point was selected from this curve as the stability probability threshold, ensuring that the model's precision for classifying "stable" states is higher than 0.90 while maintaining an acceptable recall level (e.g., not lower than 0.70). Using this method, the stability probability threshold was set to 0.85. During verification, the adjusted feature dataset was input into the trained logistic regression model to obtain the probability value of being classified as "stable." If the probability value is below 0.85, the verification fails. In this case, the adjusted data set is discarded, and the most recently verified running feature set is used as the final running feature set for the current moment to ensure the continuity of system output data. If the probability value is greater than or equal to 0.85, the verification passes, and the current adjusted feature data set is determined as the final running feature set.

[0023] In step S12, spatiotemporal variation features are extracted from the operational feature set and matched with a pre-established environmental interference feature library to obtain interference tags. Based on the interference tags, the boundary between environmental interference and the equipment's own state changes is determined, and the boundary distinction index is calculated, including: Signal data is obtained based on the set of operational features, and features are extracted from the signal data to obtain spatiotemporal variation features; The spatiotemporal variation features are matched with a pre-established environmental interference feature library to obtain the feature matching degree. If the feature matching degree exceeds the preset feature matching threshold, it is determined to be a dominant feature of environmental interference, and an interference label is obtained. Based on the interference markers, the boundary between environmental interference and changes in the equipment's own state is determined, and preliminary boundary delineation data is obtained; Spatiotemporal correlation analysis was performed on the preliminary boundary delineation data to calculate the boundary differentiation index.

[0024] First, signal data is extracted from the operational feature set, such as a point cloud density sequence of 50 consecutive sampling periods. A convolutional neural network is used to extract features from this signal data. This network structure includes a one-dimensional convolutional layer (using the ReLU activation function) and a max-pooling layer to capture the local correlation and periodic patterns of the signal in the time dimension. The model is trained using supervised learning, with training data derived from historical operational data of the LiDAR under known interference environments. The collected time-series signal data is used as input, and the corresponding experimentally verified interference type labels are used as the target output. Cross-entropy loss is used as the loss function, and the Adam optimizer is used for iterative training until the model converges. Through this training process, the model can extract discriminative spatiotemporal variation features from the input signal data.

[0025] Secondly, the extracted spatiotemporal variation features are matched with a pre-established environmental interference feature library. The environmental interference feature library is constructed as follows: in a controlled experimental environment, various typical interference conditions (such as lens condensation and strong light glare) are simulated, and the operating data of the lidar under various interferences are collected. Its temporal features are extracted and averaged to form a standardized feature template library. The matching process is completed by calculating the cosine similarity between the current feature and each template in the feature library. The preset feature matching threshold is set to 0.75. This threshold is determined by plotting an ROC curve on the validation set and selecting the point that achieves the optimal balance between the true positive rate and the false positive rate. If the feature matching degree exceeds this threshold, it is judged as a dominant environmental interference feature, and the successfully matched interference type and matching degree value are recorded, generating an interference labeling result containing the interference type and confidence level.

[0026] Next, based on the interference labeling results, the feature dimensions that are strongly correlated with the successfully matched interference types in the running feature set are assigned to the environmental interference part, and the remaining feature dimensions are assigned to the device's own state part, thereby determining the boundary between environmental interference and changes in the device's own state and obtaining preliminary boundary division data.

[0027] Finally, a spatiotemporal correlation analysis was performed on the preliminary boundary delineation data. The autocorrelation coefficients of the environmental interference component and the equipment's own state component were calculated at three time scales: 1 second, 5 seconds, and 10 seconds. The cross-correlation coefficients between the two data components were also calculated. Boundary discrimination index B was quantified using the following formula: in, The calculated boundary discrimination index, This represents the correlation coefficient between the environmental interference component and the equipment's own state component. The intensity amplitude representing the characteristics of environmental disturbance is obtained by calculating the L2 norm of this characteristic. and The weighting coefficient is determined through grid search. The larger the value of this index, the clearer the boundary between environmental interference and changes in the equipment's own state.

[0028] In step S13, if the boundary discrimination index exceeds a preset discrimination index threshold, a subset of environmental features is extracted from the running feature set and recombined to obtain a feature fusion vector, including: If the boundary discrimination index exceeds the preset discrimination index threshold, the data content of the running feature set is decomposed hierarchically to obtain an environmental feature subset related to environmental adaptation. Calculate the initial feature weights of the environmental feature subset and dynamically adjust them to obtain the adjusted weight distribution data; The environmental feature subset is reorganized using the weighted distribution data to obtain a feature fusion vector.

[0029] First, the boundary discrimination index threshold is set to 0.5. This threshold is determined by calculating the maximum value of the Youden index based on ROC curves plotted from 500 sets of historical operational data, which optimally balances the true positive rate and the false positive rate. If the calculated boundary discrimination index B exceeds this threshold, the operational feature set is stratified. The stratified decomposition uses principal component analysis (PCA). The specific steps are as follows: standardize the feature data in the operational feature set to eliminate the influence of dimensions; calculate the covariance matrix of the standardized features; solve for the eigenvalues ​​and eigenvectors of the covariance matrix; sort the eigenvalues ​​from largest to smallest and select the top three principal components with eigenvalues ​​greater than 1; calculate the cumulative variance contribution rate to ensure it reaches over 85%. These three principal components are defined as a subset of environmental features related to environmental adaptation. For example, the loading matrix of the first principal component shows that it has a high weight on point cloud density (loading 0.72) and temperature change (loading 0.68), mainly reflecting the changes in sensing performance under the influence of thermal environment; the second principal component has a high loading on dust concentration (loading 0.75) and humidity (loading 0.71), reflecting the coupled influence of particulate matter and humidity; the third principal component mainly characterizes the variation pattern of effective detection distance (loading 0.82) under different environmental conditions.

[0030] Then, the initial feature weights of the environmental feature subset are calculated. These initial weights are set as the proportion of variance contribution rate corresponding to each dimension of the data. If the feature weights of the environmental feature subset exceed a preset feature weight threshold, the feature weights are dynamically adjusted to obtain the adjusted weight distribution data. Specifically, the preset feature weight threshold is set to 0.4, which is determined based on the 70th quantile of the feature weight distribution in historical data, ensuring that only significantly important features are dynamically adjusted. If the initial weight of a feature subset is greater than this threshold, an adaptive filtering algorithm is used for dynamic adjustment. An LMS adaptive filter is established, with its step size parameter μ set to 0.05 according to stability requirements. The mean feature value obtained after 100 hours of continuous operation under standard conditions is used as the desired signal, and the weights are iteratively updated using the least mean square algorithm. The update formula is: in, For output error, For the input feature vector, and The first The second iteration and the first The weight vector at the next iteration. The iteration process continues until the weight change is less than 0.001 or the maximum number of iterations (100) is reached, finally yielding the adjusted weight distribution data.

[0031] Finally, based on the adjusted weight distribution data, a weighted linear combination of the environmental feature subsets is performed to reorganize and generate a 3-dimensional feature fusion vector. Each dimension of this vector represents the weighted fusion result of different environmental adaptation features.

[0032] In step S14, a high-dimensional feature subset is obtained based on the feature fusion vector. Key node data is extracted from the high-dimensional feature subset and compared with a pre-established fault mode library. The fault mode category is determined based on the comparison result, and the state classification probability distribution is calculated, including: A high-dimensional feature subset is obtained based on the feature fusion vector. The dynamic sequence changes of the high-dimensional feature subset are analyzed. If the change amplitude exceeds the preset change amplitude range, it is determined to be key node data. The key node data is compared with a pre-established fault mode library. Based on the comparison results, the fault mode category is determined and the state classification probability distribution is calculated.

[0033] First, a high-dimensional feature subset is obtained based on the feature fusion vector. Then, kernel principal component analysis is used to map the 3D feature fusion vector to a higher-dimensional feature space. The radial basis function is chosen as the kernel function, and its expression is: in and It is a feature fusion vector. This represents the inner product of two vectors in the kernel space. These are kernel parameters, set according to the feature dimension and data distribution characteristics. , For training dataset The variance of the features is calculated. This mapping yields a 10-dimensional high-dimensional feature subset, with orthogonality maintained between the features of each dimension.

[0034] Secondly, the dynamic sequence changes of the high-dimensional feature subset are analyzed. A high-dimensional feature subset sequence of 20 consecutive time steps is input into a Long Short-Term Memory (LSTM) network for time-series analysis. The LSTM network structure consists of an input layer (10 nodes), a hidden layer (50 nodes), and an output layer (6 nodes). During the training phase, a dataset containing 1000 labeled sequences is used, and the cross-entropy loss function is minimized using the Adam optimizer. The training cycle is set to 200 epochs. The feature change trajectory is output through the LSTM network, and the difference in feature change between adjacent time steps is calculated. The preset change range is determined based on the 95th quantile of the normal state change range in historical data, defined as [-0.1, 0.1]. If the feature change difference exceeds this range, the time point is marked as a key node.

[0035] Then, the key node data is compared with a pre-established fault mode library. The fault mode library is constructed by simulating six typical fault states in the laboratory, including fault modes such as "laser power attenuation," "scanning motor jitter," and "optical window contamination." Each fault mode stores 100 sets of standard feature sequences. The comparison process uses a dynamic time warping algorithm to calculate the minimum path distance as a sequence similarity measure, and the calculated minimum path distance is converted into a similarity score using an exponential function. in, To calculate the similarity score, The scale parameter is set to 0.1. The minimum path distance is used. A similarity threshold of 0.8 is set, determined through ROC curve analysis on the validation set, which effectively distinguishes between normal and fault states. If the similarity exceeds this threshold, the current device is determined to be in the corresponding fault mode category. The state classification probability distribution is then calculated. The raw scores from the LSTM network output layer are then used. The Softmax function is used to calculate the probability distribution of each category. For a given input sample... It is classified as the first One failure mode (total) The probability of (6 patterns) The calculation formula is: in, The weight matrix of the last layer of the LSTM network and bias vector Through linear transformation Calculations show that This is the output of the LSTM hidden layer. The final result is a state classification probability distribution containing the probability values ​​of six fault modes, with the sum of all probability values ​​being 1.

[0036] In step S15, based on the state classification probability distribution, low-confidence regions are divided, a feature subset of the low-confidence regions is obtained and bias correction and spatial transformation are performed to determine the fault location coordinate system, including: The feature subset distribution is obtained from the state classification probability distribution, and the feature subset distribution is processed in a hierarchical manner to obtain the core feature data; Environmental interference factors are separated from the core feature data to obtain a clean data set; If the similarity between the clean data group and the fault mode category is lower than a preset similarity threshold, it is determined as a low-confidence region, and the feature subset of the low-confidence region is clustered to obtain candidate data clusters. The candidate data clusters are subjected to deviation correction and spatial transformation to obtain the fault location coordinate system.

[0037] First, the feature subset distribution is obtained from the state classification probability distribution. The probability threshold is set to 0.7, and the original feature dimensions corresponding to the categories with probability values ​​lower than this threshold are extracted as the feature subset distribution.

[0038] Secondly, the feature subset distribution is stratified to obtain core feature data. The K-means clustering algorithm is used to divide the feature subset distribution into three clusters. The number of clusters is determined by the elbow rule, which calculates the sum of squares within each cluster corresponding to different numbers of clusters and selects the value corresponding to the inflection point as the optimal number of clusters. From the three obtained clusters, the cluster with the largest number of samples is selected, and its centroid coordinates are calculated as the core feature data.

[0039] Then, environmental interference factors are separated from the core feature data to obtain a clean data set. Principal component analysis is used to extract the components most correlated with environmental interference from the core feature data as interference factors. Projection operations are then used to remove these interference factors from the core feature data; the specific calculation formula is as follows: in, For core feature data, This is the transpose of the environmental disturbance feature vector, which can be obtained by extracting the first principal component vector from historical environmental data. The core feature data after cleaning, This represents the dot product of vectors.

[0040] The mean cosine similarity between the clean data set and the centroids of each category in the fault mode library is calculated, with a preset similarity threshold of 0.75. This threshold is determined based on the statistical analysis of the similarity distribution of a large number of samples. If the average similarity is lower than this threshold, the region is identified as a low-confidence region, and the DBSCAN clustering algorithm is used to group the feature subsets of the low-confidence region. The DBSCAN parameters are set as follows: neighborhood radius eps = 0.5, minimum number of samples min_samples = 5. These parameters are optimized through parameter scanning experiments to obtain candidate data clusters.

[0041] Finally, bias correction and spatial transformation are performed on the candidate data clusters to determine the fault location coordinate system. The correction amount is then calculated. : in, For the centroid of the data cluster, The reference values ​​for this type of feature under normal conditions are determined through the statistical characteristics of long-term operating data of the equipment under standard operating conditions, generally taking the mean or median. Based on the calculated correction amount, the candidate data clusters are corrected. Principal component analysis is performed on the corrected data clusters, and the first two principal components are selected to form a transformation matrix. The cumulative variance contribution rate of these two principal components must reach more than 80% to ensure that the information loss is within an acceptable range. A two-dimensional fault location coordinate system is established with the direction of the first principal component as the horizontal axis (PCA1) and the direction of the second principal component as the vertical axis (PCA2). The origin of this coordinate system is set as the feature reference point under normal operating conditions, so that the distribution of fault features in this coordinate system can intuitively reflect the degree and direction of its deviation from the normal state.

[0042] In step S16, the sensing parameters of the lidar are updated and data is re-acquired according to the fault location coordinate system to obtain the calibrated operating state vector.

[0043] First, based on the established fault location coordinate system, its physical meaning is further analyzed. Principal component load analysis is performed on historical fault data to determine the correspondence between the two principal component axes and the actual fault types. For example, the PCA1 axis has a load of 0.85 on the laser emission power-related characteristics, and is therefore defined as the power attenuation characteristic axis; the PCA2 axis has a load of 0.78 on the optical transmittance-related characteristics, and is therefore defined as the window contamination characteristic axis. This coordinate system, with the normal operating condition characteristic reference point as its origin, can accurately characterize the degree and direction of equipment status deviating from the normal range.

[0044] Secondly, the current equipment operating characteristics are projected onto the fault location coordinate system, and the coordinate values ​​(xpca1, xpca2) are calculated. The lidar sensing parameters are updated using a predefined parameter mapping table, which is established through systematic offline calibration experiments. In the calibration experiments, the optimal parameter configurations at different coordinate system positions are recorded by precisely controlling the laser power attenuation and window contamination levels.

[0045] The specific parameter update rule is as follows: when When the absolute value is greater than 0.5, an abnormal laser power is determined. The laser emission power is adjusted using linear interpolation, and the calculation formula is as follows: in, This indicates the calibrated laser emission power, measured in watts. This indicates the original power setting value, in watts. This represents the proportionality coefficient, which was determined to be -0.1 through a power-coordinate calibration experiment. This indicates the coordinate value of the device on the PCA1 axis. When... When the absolute value of 2 is greater than 0.3, optical window contamination is considered to exist. The echo signal strength threshold is adjusted using a lookup table; the calculation formula is as follows: in, This represents the calibrated echo signal strength threshold, in volts. This represents the original threshold setting value, in volts. Indicates the amount of compensation, measured in volts. When... When 2 is greater than 0.3, the compensation amount is calculated using the following formula: in This represents the pollution compensation ratio, calibrated to 2.5 volts through pollution experiments.

[0046] The parameter update operation is performed, and the calculated P_new and T_new parameters are written to the corresponding registers through the low-level control interface of the lidar. Specifically, the power control word is written to register 0x1A and the threshold setting value is written to register 0x2B via the SPI communication protocol.

[0047] Immediately after the parameters are updated, a cycle of sensing data is collected, and key operational indicators, including point cloud density, effective detection range, and noise ratio, are extracted to form a three-dimensional vector [Density, Range, NoiseRatio] as the calibrated operational status vector. The baseline value of this vector [Density_0, Range_0, NoiseRatio_0] is determined by averaging 100 consecutive measurements under factory standard conditions, where point cloud density is in points / m², effective detection range is in meters, and noise ratio is in percentage.

[0048] In step S17, if the operating state vector shows an abnormal deviation, compensation is made for the abnormal deviation to account for environmental influences. After compensation, the operating state vector is updated and state detection is performed to obtain the state detection result, including: Abnormal deviation data is obtained from the running state vector. If the abnormal deviation data exceeds the preset deviation threshold, it is determined that an abnormal deviation has occurred and the abnormal deviation data is labeled to obtain the labeled deviation dataset. Environmental impact data is obtained from the deviation dataset, and the deviation dataset is compensated based on the environmental impact data to obtain a compensated dataset; Based on the compensation dataset, the running state vector is updated, and state detection is performed on the updated running state vector to obtain the state detection result.

[0049] First, abnormal deviation data is obtained from the running state vector. The running state vector V is a three-dimensional vector [Density, Range, NoiseRatio]. The Euclidean distance D between it and the standard running state vector V_std is calculated as the deviation data. A preset deviation threshold is set, which is determined based on the 99th percentile of normal state deviations in historical data. If the deviation data exceeds the preset deviation threshold, it is determined that an abnormal deviation has occurred. The contribution of each dimension to the total deviation is calculated, and the data of the top 10% of dimensions with the highest contribution are marked as abnormal deviation data, forming the labeled deviation dataset.

[0050] Secondly, environmental impact data is obtained from the deviation dataset. Based on the pre-established environmental disturbance feature library, dimensional data related to known environmental disturbance types are extracted as environmental impact data E, and E is normalized to its maximum and minimum values. The maximum and minimum values ​​of E are determined based on the statistical extreme values ​​of environmental impact data in historical data.

[0051] Then, based on the environmental impact data compensation deviation dataset, an S-shaped compensation function is used to generate a compensation vector, calculated as follows: in, The calculated compensation vector, The slope parameter is set to 5, and this value was determined through optimization experiments on the compensation effect. The baseline level for environmental disturbance is set to 0.5, a value determined based on the average level of environmental disturbance in historical data. The original operating state vector V is multiplied element-wise using the compensation vector C to obtain the compensated operating state vector, forming the compensation dataset.

[0052] Finally, the running state vector is updated based on the compensated dataset, and state detection is performed. The compensated running state vector is input into a support vector machine classifier, which uses a radial basis function as the kernel function and is trained on historical data containing 3000 labeled samples. The kernel function parameter γ is optimized using a grid search method. The classifier classifies the running state into three categories: "normal," "warning," and "fault." The classification decision is based on the distance from the sample to the hyperplane, and the distance threshold is determined through cross-validation. The output of the classifier is the final state detection result, which includes the state category and the corresponding confidence score.

[0053] In step S18, anomalies are obtained from the state detection results, and signal comparison and path simulation are performed on the anomalies to determine the comprehensive fault location path, including: Anomalies are obtained from the state detection results, and these anomalies are classified to obtain an anomaly classification set. Based on the anomaly point classification set, and combined with environmental interference factors, the signal of the anomaly points is compared to obtain the signal attenuation source region. Based on the signal attenuation source region, a path simulation is performed to obtain a simulated propagation path. The simulated propagation paths are then prioritized to obtain a set of positioning paths. A consistency check is performed on the set of location paths. If the consistency check result meets the preset verification standard, the comprehensive fault location path is determined.

[0054] First, outliers are identified from the state detection results and classified to obtain an outlier classification set. Specifically, all data points with a state category of "fault" and a confidence score greater than 0.8 are selected as outliers from the state detection results. The DBSCAN clustering algorithm is used to automatically classify the outliers, with the algorithm parameters set to a neighborhood radius eps = 0.5 and a minimum sample size min_samples = 5. These parameters are determined through silhouette coefficient analysis. The clustering process is based on the two-dimensional coordinates (PCA1, PCA2) of the outliers in the fault location coordinate system, grouping outliers with a spatial distance less than eps into the same category to form the outlier classification set.

[0055] Secondly, based on the anomaly point classification set and environmental interference factors, the signal attenuation source areas of the anomaly points are compared. The spatial distribution data of environmental interference comes from temperature, humidity, and dust concentration data collected by an environmental sensor network. An environmental interference raster map with a resolution of 0.1m × 0.1m is generated using Kriging interpolation. The Pearson correlation coefficient between the average signal attenuation rate and the environmental interference intensity of each anomaly point classification set is calculated. Continuous areas with correlation coefficients exceeding a preset threshold of 0.72 are marked as signal attenuation source areas. This threshold is determined based on the 95th percentile of the correlation coefficient distribution of real fault cases in historical data.

[0056] Then, path simulation is performed based on the signal attenuation source region to obtain simulated propagation paths. These simulated propagation paths are then prioritized to obtain a set of positioning paths. Path simulation employs a ray tracing algorithm, considering the actual layout of the internal optical components of the lidar, including key components such as the laser emitter, lens group, and scanning galvanometer. The simulation process is based on geometric optics principles, calculating all possible optical paths from the laser source to the signal attenuation source region. A comprehensive score is calculated for each simulated propagation path using the following formula: in Indicates path length. This represents the maximum path length among all simulated paths. The quantitative metric for path complexity is calculated by dividing the number of path turns by the total number of nodes in the path. This indicates the frequency of the path in historical fault data. w1, w2, and w3 are weighting coefficients set to 0.4, 0.3, and 0.3 respectively. These weighting coefficients are determined using the entropy weighting method, calculating the weights based on the information entropy of each evaluation indicator. The paths are sorted from highest to lowest score, and the top 10 paths are selected to form a location path set.

[0057] Finally, a consistency check is performed on the location path set. If the consistency check result meets the preset verification criteria, the comprehensive fault location path is determined. Verification is then performed by calculating the matching degree between each path and the actual fault paths in historical fault cases. The matching degree calculation formula is as follows: in, To calculate the matching degree, This represents the number of nodes that the path traverses along with historical cases. This represents the total number of nodes in the path. The Jaccard similarity coefficient, which represents the shape of a path, is calculated as the ratio of the number of intersection nodes to the number of union nodes. and The weighting coefficients were set to 0.6 and 0.4, respectively, and were determined using principal component analysis, a method that assigns weights based on the contribution of each factor to the path matching degree in historical data. A matching degree threshold of 0.65 was set, which, based on historical data analysis, effectively distinguishes between real fault paths and random paths. When the path matching degree exceeds this threshold, it is considered to have passed the consistency check. All paths that pass the check are integrated, and a topology network is constructed based on the connectivity of the path nodes. A depth-first search algorithm is used to determine the set of shortest paths covering all key nodes, forming the final comprehensive fault location path.

[0058] In step S19, environmental data of the fault location is extracted based on the comprehensive fault location path to obtain an environmental feature set. Adjustment instructions are generated based on the environmental feature set, and the instructions undergo multi-scenario adaptation verification and hierarchical processing to obtain a real-time adjustment instruction sequence, including: Based on the comprehensive fault location path, feature extraction is performed on the environmental data of the fault location to obtain an environmental feature set. If the feature values ​​in the environmental feature set exceed the preset environmental threshold, an adjustment instruction is generated, and an instruction sequence is determined. If the instruction sequence passes the multi-scenario adaptation verification, then the optimized instruction combination is obtained; The optimized combination of instructions is layered to obtain a layered result. If the layered result meets the requirements for adapting to complex scenarios, a real-time adjustment instruction sequence is obtained.

[0059] First, based on the physical location indicated by the comprehensive fault location path, environmental data at that location is obtained, including real-time temperature, humidity, and dust concentration data collected by an environmental sensor network deployed in the fault area, as well as historical environmental data of that location over a past period of time extracted from the equipment operation logs.

[0060] Next, feature extraction was performed on the environmental data. First, Z-score normalization was applied to the real-time and historical environmental data to eliminate the influence of dimensions. The mean (μ) and standard deviation (σ) used for normalization were calculated based on the historical environmental data. Subsequently, principal component analysis was used to reduce the dimensionality of the normalized multidimensional environmental data, and the top three principal components with a cumulative variance contribution rate exceeding 85% were extracted, forming a set of environmental features that can comprehensively characterize the environmental state.

[0061] Next, adjustment instructions are generated based on the environmental feature set. The scores of each principal component in the environmental feature set are compared with preset environmental thresholds determined based on historical operational data statistics. For example, the threshold for temperature-related principal components is set to 2.0 (standardized score) corresponding to the 95th percentile of historical data, the threshold for humidity-related principal components is set to 1.8, and the threshold for dust concentration-related principal components is set to 2.2. When any principal component score exceeds its corresponding threshold, a corresponding equipment adjustment instruction is generated according to the type and magnitude of the exceeding principal component, using predefined instruction mapping rules. For example, when the temperature principal component score exceeds 2.0, an instruction to "proportionally reduce the laser drive current" is generated; when the dust concentration principal component score exceeds 2.2, an instruction to "activate the positive pressure dust removal system and increase the cooling fan speed" is generated. All generated instructions are sorted according to their preset execution urgency (e.g., safety-related, performance-related, maintenance-related) to form an initial instruction sequence.

[0062] Then, the initial instruction sequence is verified for adaptation across multiple scenarios. In the industrial application context of this embodiment, the verification scenarios are set as typical industrial manufacturing environments, including three representative scenarios: a stamping workshop (high-frequency vibration, metal dust), a spraying workshop (high humidity, chemical aerosols), and an assembly workshop (standard operating conditions). A corresponding system simulation model is constructed for each scenario. The initial instruction sequence is executed within the model, and key performance indicators are monitored, including: calculating the point cloud density retention rate by dividing the point cloud density after instruction execution by the standard point cloud density; calculating the effective detection distance change rate by dividing the absolute difference between the detection distance after instruction execution and the standard detection distance by the standard detection distance; and calculating the system power consumption growth rate by dividing the difference between the system power consumption after instruction execution and the standard power consumption by the standard power consumption. The verification pass criteria are: point cloud density retention rate ≥ 90%, effective detection distance change rate ≤ ±8%, and system power consumption growth rate ≤ 20%. Only instruction sequences that meet the above criteria in all specified verification scenarios can be considered to have passed the adaptation verification, forming an optimized instruction combination.

[0063] Finally, the optimized combination of instructions is processed in layers to generate the final instruction sequence. Based on the urgency of instruction execution and its impact on system safety / performance, the instructions are divided into three layers: Layer 1 (Emergency Instruction Layer): Instructions involving emergency protection or safe shutdown of equipment, requiring a response and execution within 100 milliseconds; Layer 2 (Optimization Instruction Layer): Instructions involving dynamic adjustment of operating parameters to maintain performance, requiring a response and execution within 1 second; Layer 3 (Maintenance Instruction Layer): Instructions involving preventative maintenance recommendations or calibration of non-critical parameters, with no strict real-time requirements.

[0064] By analyzing the execution effects and resource consumption of instructions at each level in various industrial scenario simulation models, it was confirmed that the layering results meet the adaptation requirements for complex industrial scenarios. Finally, a real-time adjustment instruction sequence is output according to the priority order of the first layer > the second layer > the third layer.

[0065] In summary, this invention discloses a state detection method for mass-produced lidar. It obtains an operational feature set by fusing lidar sensing data and environmental sensor data through multimodal data fusion; it uses a convolutional neural network to extract spatiotemporal correlation patterns and determine the boundary distinguishing index between environmental interference and equipment state changes; it adjusts feature weights using an adaptive filtering algorithm to obtain an optimized feature fusion vector; it utilizes a long short-term memory network to analyze dynamic sequence changes, determine fault modes, and obtain a state classification probability distribution; it updates sensing parameters based on the fault location coordinate system, and generates a real-time adjustment command sequence through environmental compensation and anomaly propagation analysis, thereby solving the problem of insufficient lidar state detection accuracy in existing technologies.

[0066] Reference Figure 2 The second embodiment of the present invention provides a state detection system for mass-produced lidar, comprising: The runtime feature acquisition module is used to acquire spatial information through lidar, acquire environmental data through environmental sensors, integrate the acquired data and perform anomaly correction and consistency verification to obtain a set of runtime features. The boundary differentiation module is used to extract spatiotemporal variation features from the set of operating features and match them with a pre-established environmental interference feature library to obtain interference tags. Based on the interference tags, the boundary between environmental interference and the changes in the device's own state is determined and the boundary differentiation index is calculated. The weight allocation module is used to extract a subset of environmental features from the running feature set and reorganize them to obtain a feature fusion vector if the boundary discrimination index exceeds a preset discrimination index threshold. The fault mode recognition module is used to obtain a high-dimensional feature subset based on the feature fusion vector, extract key node data from the high-dimensional feature subset and compare it with a pre-established fault mode library, determine the fault mode category based on the comparison result and calculate the state classification probability distribution. The fault coordinate localization module is used to divide the low-confidence region according to the state classification probability distribution, obtain the feature subset of the low-confidence region and perform deviation correction and spatial transformation to determine the fault localization coordinate system. The perception parameter update module is used to update the perception parameters of the lidar and re-acquire data according to the fault location coordinate system to obtain the calibrated operating status vector. An environmental compensation module is used to compensate for abnormal deviations caused by environmental factors if the operating state vector shows abnormal deviations. After compensation, the operating state vector is updated and a state detection is performed to obtain the state detection result. The anomaly analysis module is used to obtain anomalies from the status detection results, and determine the comprehensive fault location path by comparing the signals and simulating the paths of the anomalies. The instruction generation module is used to extract environmental data of the fault location based on the comprehensive fault location path, obtain an environmental feature set, generate adjustment instructions based on the environmental feature set, and perform multi-scenario adaptation verification and hierarchical processing on the instructions to obtain a real-time adjustment instruction sequence.

[0067] It should be noted that the state detection system for mass-produced lidar provided in this embodiment of the invention is used to execute all the process steps of the state detection method for mass-produced lidar in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0068] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a state detection program for mass-produced lidar. When the processor executes the computer program, it implements the steps described in the various state detection method embodiments for mass-produced lidar, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments, such as the data acquisition module.

[0069] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0070] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0071] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0072] The memory can be used to store the computer program or module. The processor implements various functions of the electronic device by running or executing the computer program or module stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0073] If the modules integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0074] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the system embodiment drawings provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0075] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A state detection method for mass-produced lidar, characterized in that, include: Spatial information is acquired through lidar, environmental data is acquired through environmental sensors, the acquired data is integrated and anomaly correction and consistency verification are performed to obtain a set of operational characteristics. Spatiotemporal variation features are extracted from the set of operational features and matched with a pre-established environmental interference feature library to obtain interference tags. Based on the interference tags, the boundary between environmental interference and the equipment's own state changes is determined and the boundary distinction index is calculated. If the boundary discrimination index exceeds the preset discrimination index threshold, then a subset of environmental features is extracted from the running feature set and recombined to obtain a feature fusion vector; A high-dimensional feature subset is obtained based on the feature fusion vector. Key node data is extracted from the high-dimensional feature subset and compared with a pre-established fault mode library. The fault mode category is determined based on the comparison result and the state classification probability distribution is calculated. Based on the state classification probability distribution, low-confidence regions are divided, feature subsets of the low-confidence regions are obtained, and deviation correction and spatial transformation are performed to determine the fault location coordinate system. Based on the fault location coordinate system, update the sensing parameters of the lidar and re-collect data to obtain the calibrated operating state vector. If the operating status vector shows an abnormal deviation, compensation is made for the abnormal deviation in response to environmental influences. After compensation, the operating status vector is updated and a status detection is performed to obtain the status detection result. Anomalies are obtained from the status detection results, and a comprehensive fault location path is determined by comparing signals and simulating paths from the anomalies. Based on the comprehensive fault location path, environmental data of the fault location is extracted to obtain an environmental feature set. Adjustment instructions are generated based on the environmental feature set, and the instructions are subjected to multi-scenario adaptation verification and hierarchical processing to obtain a real-time adjustment instruction sequence.

2. The state detection method for mass-produced lidar according to claim 1, characterized in that, The process involves acquiring spatial information via lidar, obtaining environmental data via environmental sensors, integrating the acquired data, and performing anomaly correction and consistency verification to obtain a set of operational characteristics, including: Spatial information is acquired through lidar, environmental data is acquired through environmental sensors, and the initial operational characteristics are obtained by integrating them. If the initial operating characteristics exceed the preset characteristic threshold range, then the abnormal characteristic data is marked, and the type of environmental interference corresponding to the abnormal characteristic data is determined. Based on the type of environmental interference, the corresponding correction parameters are obtained from the pre-established environmental interference database, and the abnormal feature data is adjusted to obtain an adjusted feature data set. A consistency check is performed on the adjusted feature data set. If the consistency check passes, the adjusted feature data set is determined as the running feature set.

3. The state detection method for mass-produced lidar according to claim 1, characterized in that, The process involves extracting spatiotemporal variation features from the operational feature set and matching them with a pre-established environmental interference feature library to obtain interference tags. Based on these interference tags, the boundary between environmental interference and changes in the equipment's own state is determined, and a boundary distinction index is calculated, including: Signal data is obtained based on the set of operational features, and features are extracted from the signal data to obtain spatiotemporal variation features; The spatiotemporal variation features are matched with a pre-established environmental interference feature library to obtain the feature matching degree. If the feature matching degree exceeds the preset feature matching threshold, it is determined to be a dominant feature of environmental interference, and an interference label is obtained. Based on the interference markers, the boundary between environmental interference and changes in the equipment's own state is determined, and preliminary boundary delineation data is obtained; Spatiotemporal correlation analysis was performed on the preliminary boundary delineation data to calculate the boundary differentiation index.

4. The state detection method for mass-produced lidar according to claim 1, characterized in that, If the boundary discrimination index exceeds a preset discrimination index threshold, then an environmental feature subset is extracted from the running feature set and recombined to obtain a feature fusion vector, including: If the boundary discrimination index exceeds the preset discrimination index threshold, the data content of the running feature set is decomposed hierarchically to obtain an environmental feature subset related to environmental adaptation. Calculate the initial feature weights of the environmental feature subset and dynamically adjust them to obtain the adjusted weight distribution data; The environmental feature subset is reorganized using the weighted distribution data to obtain a feature fusion vector.

5. The state detection method for mass-produced lidar according to claim 1, characterized in that, The process of obtaining a high-dimensional feature subset based on the feature fusion vector, extracting key node data from the high-dimensional feature subset and comparing it with a pre-established fault mode library, determining the fault mode category based on the comparison results, and calculating the state classification probability distribution includes: A high-dimensional feature subset is obtained based on the feature fusion vector. The dynamic sequence changes of the high-dimensional feature subset are analyzed. If the change amplitude exceeds the preset change amplitude range, it is determined to be key node data. The key node data is compared with a pre-established fault mode library. Based on the comparison results, the fault mode category is determined and the state classification probability distribution is calculated.

6. The state detection method for mass-produced lidar according to claim 1, characterized in that, The step of dividing the low-confidence region according to the state classification probability distribution, obtaining the feature subset of the low-confidence region and performing bias correction and spatial transformation to determine the fault location coordinate system includes: The feature subset distribution is obtained from the state classification probability distribution, and the feature subset distribution is processed in a hierarchical manner to obtain the core feature data; Environmental interference factors are separated from the core feature data to obtain a clean data set; If the similarity between the clean data group and the fault mode category is lower than a preset similarity threshold, it is determined as a low-confidence region, and the feature subset of the low-confidence region is clustered to obtain candidate data clusters. The candidate data clusters are subjected to deviation correction and spatial transformation to obtain the fault location coordinate system.

7. The state detection method for mass-produced lidar according to claim 1, characterized in that, If the operating state vector shows an abnormal deviation, compensation is made for the abnormal deviation to account for environmental influences. After compensation, the operating state vector is updated and state detection is performed to obtain the state detection result, including: Abnormal deviation data is obtained from the running state vector. If the abnormal deviation data exceeds the preset deviation threshold, it is determined that an abnormal deviation has occurred and the abnormal deviation data is labeled to obtain the labeled deviation dataset. Environmental impact data is obtained from the deviation dataset, and the deviation dataset is compensated based on the environmental impact data to obtain a compensated dataset; Based on the compensation dataset, the running state vector is updated, and state detection is performed on the updated running state vector to obtain the state detection result.

8. The state detection method for mass-produced lidar according to claim 1, characterized in that, The step of obtaining anomalies from the state detection results, performing signal comparison and path simulation on the anomalies, and determining the comprehensive fault location path includes: Anomalies are obtained from the state detection results, and these anomalies are classified to obtain an anomaly classification set. Based on the anomaly point classification set, and combined with environmental interference factors, the signal of the anomaly points is compared to obtain the signal attenuation source region. Based on the signal attenuation source region, a path simulation is performed to obtain a simulated propagation path. The simulated propagation paths are then prioritized to obtain a set of positioning paths. A consistency check is performed on the set of location paths. If the consistency check result meets the preset verification standard, the comprehensive fault location path is determined.

9. The state detection method for mass-produced lidar according to claim 1, characterized in that, The process involves extracting environmental data of the fault location based on the comprehensive fault location path to obtain an environmental feature set, generating adjustment instructions based on the environmental feature set, and performing multi-scenario adaptation verification and hierarchical processing on the instructions to obtain a real-time adjustment instruction sequence, including: Based on the comprehensive fault location path, feature extraction is performed on the environmental data of the fault location to obtain an environmental feature set. If the feature values ​​in the environmental feature set exceed the preset environmental threshold, an adjustment instruction is generated, and an instruction sequence is determined. If the instruction sequence passes the multi-scenario adaptation verification, then the optimized instruction combination is obtained; The optimized combination of instructions is layered to obtain a layered result. If the layered result meets the requirements for adapting to complex scenarios, a real-time adjustment instruction sequence is obtained.

10. A condition detection system for mass-produced lidar, characterized in that, include: The runtime feature acquisition module is used to acquire spatial information through lidar, acquire environmental data through environmental sensors, integrate the acquired data and perform anomaly correction and consistency verification to obtain a set of runtime features. The boundary differentiation module is used to extract spatiotemporal variation features from the set of operating features and match them with a pre-established environmental interference feature library to obtain interference tags. Based on the interference tags, the boundary between environmental interference and the changes in the device's own state is determined and the boundary differentiation index is calculated. The weight allocation module is used to extract a subset of environmental features from the running feature set and reorganize them to obtain a feature fusion vector if the boundary discrimination index exceeds a preset discrimination index threshold. The fault mode recognition module is used to obtain a high-dimensional feature subset based on the feature fusion vector, extract key node data from the high-dimensional feature subset and compare it with a pre-established fault mode library, determine the fault mode category based on the comparison result and calculate the state classification probability distribution. The fault coordinate localization module is used to divide the low-confidence region according to the state classification probability distribution, obtain the feature subset of the low-confidence region and perform deviation correction and spatial transformation to determine the fault localization coordinate system. The perception parameter update module is used to update the perception parameters of the lidar and re-acquire data according to the fault location coordinate system to obtain the calibrated operating status vector. An environmental compensation module is used to compensate for abnormal deviations caused by environmental factors if the operating state vector shows abnormal deviations. After compensation, the operating state vector is updated and a state detection is performed to obtain the state detection result. The anomaly analysis module is used to obtain anomalies from the status detection results, and determine the comprehensive fault location path by comparing the signals and simulating the paths of the anomalies. The instruction generation module is used to extract environmental data of the fault location based on the comprehensive fault location path, obtain an environmental feature set, generate adjustment instructions based on the environmental feature set, and perform multi-scenario adaptation verification and hierarchical processing on the instructions to obtain a real-time adjustment instruction sequence.

Images