Multi-source data intelligent measurement and abnormality identification method for robot patrol

Abstract

This invention discloses a multi-source data intelligent measurement and anomaly identification method for robot inspection, belonging to the field of data anomaly identification technology. The method includes the following steps: temporally aligning multi-source data to obtain unified temporal data; spatially aligning the unified temporal data to obtain unified coordinate data; performing a reliability assessment based on the unified coordinate data to obtain weighted measurement data; performing multi-dimensional feature decomposition based on the weighted measurement data to obtain combined feature data; determining temporal consistency based on the combined feature data to obtain a stable state result; and performing a health assessment based on the stable state result to obtain an anomaly determination result. This invention defines a reliability level for each type of data through reliability assessment, making the data comparable, improving adaptability to complex environments, realizing differentiated expression of multi-source data, improving the reliability of multi-source data fusion, and enhancing the accuracy of anomaly identification.

Description

Technical Field

[0001] This invention relates to the field of data anomaly identification technology, specifically to a method for intelligent measurement and anomaly identification of multi-source data for robot inspection. Background Technology

[0002] As industrial facilities and infrastructure continue to expand, the objects to be inspected are increasingly characterized by diversified equipment types, complex operating states, and dynamic changes in environmental conditions. Against this backdrop, robot-based inspection tasks are becoming the mainstream approach. By deploying multiple types of sensors, continuous monitoring of target equipment and its surrounding environment is achieved, and this technology is widely used in scenarios such as power systems, petrochemical plants, and rail transportation.

[0003] Existing technologies have shortcomings in data processing: They typically use a uniform processing method for multi-source measurement data, without distinguishing data quality differences or using simple threshold methods to remove outlier data. They also fail to assess the reliability of data from different sources, resulting in low-quality data still being included in the calculation. The fusion results are easily affected by low-quality data, thus impacting the accuracy and stability of anomaly identification. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for intelligent measurement and anomaly identification of multi-source data for robot inspection, thereby solving the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides a method for intelligent measurement and anomaly identification of multi-source data for robot inspection, comprising the following steps:

[0007] S1. Time alignment of multi-source data to obtain unified time data;

[0008] S2. Perform spatial alignment based on unified time data to obtain unified coordinate data;

[0009] S3. Conduct a credibility assessment based on the unified coordinate data to obtain weighted measurement data;

[0010] S4. Perform multidimensional feature decomposition based on the weighted measurement data to obtain combined feature data;

[0011] S5. Determine the temporal consistency based on the combined feature data to obtain the stable state result;

[0012] S6. Based on the stable state results, conduct a health assessment to obtain the abnormality determination results.

[0013] To further optimize this technical solution, the time alignment in step S1 includes:

[0014] A unified time reference is determined, a time series is constructed based on the raw data from multiple sources, a time mapping relationship is established for each time point in the time series, linear interpolation calculation is performed to achieve time alignment, and unified time data is obtained.

[0015] To further optimize this technical solution, the spatial alignment in step S2 includes:

[0016] Based on the obtained unified time data, spatial location information of multi-source data and robot pose information at each time point are obtained. Based on the unified spatial coordinate system, coordinate transformation and spatial mapping are performed on the data at each time point to obtain unified coordinate data.

[0017] To further optimize this technical solution, the credibility assessment in step S3 includes:

[0018] Based on the obtained unified coordinate data, the reliability score is generated by calculating the signal quality index, noise intensity index, and environmental adaptability index of various measurement signals, and weights are assigned to various measurement signals to obtain weighted measurement data.

[0019] To further optimize this technical solution, the credibility score includes:

[0020]

[0021] in:

[0022] : No. Signal type at time Credibility score;

[0023] : No. Signal type at time Signal quality metrics;

[0024] : No. Signal type at time The noise intensity index;

[0025] : No. Signal type at time The degree of environmental adaptability;

[0026] : No. The signal quality of the signal affects the weights;

[0027] : No. Noise suppression weights for signal types;

[0028] : No. Environmental adaptation weights for signal types;

[0029] A credibility score is obtained by weighting the signal quality, noise intensity, and environmental adaptability.

[0030] To further optimize this technical solution, the multidimensional feature decomposition in step S4 includes:

[0031] Based on the obtained weighted measurement data, statistical features, frequency domain features, and image texture features of the data are extracted by constructing a time analysis window. The features are then standardized and combined to obtain combined feature data.

[0032] To further optimize this technical solution, the timing consistency determination in step S5 includes:

[0033] Based on the obtained combined feature data, a time sliding window is constructed for the current time point. By performing initial state determination and state distribution statistics for each time point within the window, and based on the state distribution, a state consistency judgment is made to obtain a stable state result.

[0034] To further optimize this technical solution, the health assessment in step S6 includes:

[0035] Based on the obtained stable state results, combined with the combined feature data, the degree of feature deviation is determined based on the health benchmark features, the health status is calculated, and the abnormality judgment result is obtained.

[0036] To further optimize this technical solution, the degree of feature deviation includes:

[0037]

[0038] in:

[0039] :time The degree of deviation of the characteristics at that time;

[0040] :time The first time Class feature values;

[0041] : No. Health benchmarks for class characteristics;

[0042] The number of feature types;

[0043] A small constant used to prevent the denominator from being zero and to improve computational stability;

[0044] The degree of deviation of the features is calculated by comparing the results of multidimensional features with the health benchmark.

[0045] To further optimize this technical solution, the health calculation includes:

[0046]

[0047] in:

[0048] :time Health status at that time;

[0049] Maximum permissible deviation;

[0050] The health score is calculated using an exponential function based on the degree of feature deviation and the maximum permissible deviation.

[0051] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the multi-source data intelligent measurement and anomaly identification method for robot inspection as described in the first aspect of the present invention are implemented.

[0052] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the multi-source data intelligent measurement and anomaly identification method for robot inspection as described in the first aspect of the present invention are implemented.

[0053] Compared with existing technologies, this invention provides a multi-source data intelligent measurement and anomaly identification method for robot inspection, which has the following beneficial effects:

[0054] This intelligent measurement and anomaly identification method for multi-source data in robot inspection defines the reliability level of each type of data through credibility assessment, making the data comparable, improving the adaptability to complex environments, realizing the differentiated expression of multi-source data, improving the reliability of multi-source data fusion, and enhancing the accuracy of anomaly identification. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1This is a flowchart illustrating the multi-source data intelligent measurement and anomaly identification method for robot inspection proposed in this invention. Detailed Implementation

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0058] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0059] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0060] Example 1:

[0061] Reference Figure 1 This is the first embodiment of the present invention, which provides a method for intelligent measurement and anomaly identification of multi-source data for robot inspection, including the following steps:

[0062] S1. Perform time alignment on multi-source data to obtain unified time data.

[0063] In this embodiment, the time alignment includes:

[0064] In robotic inspection scenarios, the fundamental differences in the working mechanisms of different types of sensors lead to significant inconsistencies in the output data over time. Specifically, this manifests in differences in sampling frequency (e.g., vibration signals are sampled at high frequencies, while images are sampled at low frequencies), making it impossible to establish a temporal correspondence between different data points. Inconsistent timestamps and clock offsets within different sensors cause time misalignments of the same physical event across different signals, resulting in varying event response delays. Different physical quantities respond differently to the same anomaly; for example, changes in vibration signals occur earliest, while temperature changes lag behind. Without a unified timeline, causal relationships can be misjudged. Furthermore, inconsistent data structures—e.g., vibration and acoustic signals are continuous signals, while images are discrete frames—necessary to map multi-source data to a unified time frame are also problematic.

[0065] A unified time benchmark is established, a time series is constructed based on the raw data from multiple sources, a time mapping relationship is established for each time point in the time series, linear interpolation calculation is performed to achieve time alignment, and unified time data is obtained. This establishes a unified time reference system and the time correspondence between multiple source data, supporting subsequent fusion and analysis calculations.

[0066] The time alignment steps include:

[0067] Establish a unified time reference: Select the system time as the unified reference time, define the start time of the inspection task as the start time, and define the unified sampling period as the time interval. The time interval is less than or equal to the minimum value among all sensor sampling intervals, and meets the sampling requirements of the highest frequency signal (usually vibration signal). Construct a unified time series to provide a unified alignment standard for all sensor data.

[0068] Read raw data from each sensor: Read raw data for each type of sensor, including the timestamp corresponding to each sampling point and the measurement value of each sampling point, to form multiple data sequences, including temperature, vibration, acoustic signals, electrical parameters and visual images, to provide complete input for subsequent time mapping.

[0069] Establish time mapping relationship: For each time point in the unified time series, find the two sampling points that are closest to that time point in each type of sensor data. The two sampling points are located before and after the target time point, respectively. That is, the time of the previous sampling point is slightly less than the target time, and the time of the subsequent sampling point is slightly greater than the target time, which provides a basis for subsequent interpolation calculation.

[0070] Linear interpolation calculation: For each type of sensor, the corresponding value is calculated at each time point. If the target time point is exactly equal to a certain sampling point, the value of that sampling point is used directly. If the target time point is between two sampling points, it is calculated according to the linear trend. By calculating the time difference between the target time point and the previous sampling point, the time interval between the previous and previous sampling points is calculated to realize the calculation of the target value based on the proportional relationship. For image data, the image frame with the closest time is selected as the image at that time point, and no interpolation calculation is performed. This maps data with different sampling frequencies to a unified time point, ensuring that each time point has a corresponding value.

[0071] Construct a unified time series dataset: After completing the time alignment of all sensors, summarize all signal values ​​corresponding to each time point, construct data records containing multiple physical quantities, and form a unified data structure.

[0072] Handling missing and abnormal time points: When a certain type of sensor has no valid data at a certain time point, or the data interval is too large to be interpolated, if the time interval is within the allowable range, interpolation is still performed; if it exceeds the allowable range, the data at that time point is marked as invalid. Invalid data is recorded but not included in subsequent calculations, thereby ensuring data integrity and reliability.

[0073] S2. Spatial alignment is performed based on unified time data to obtain unified coordinate data.

[0074] In this embodiment, the spatial alignment includes:

[0075] In industrial inspection scenarios, spatial alignment is a crucial and essential step. Inconsistent spatial coordinates can lead to variations in sensor installation locations, resulting in measurement data at the same time corresponding to different spatial points. Different data types also have different spatial representations, making it impossible to establish a unified spatial reference and perform fusion calculations. Furthermore, coordinate changes caused by the movement of the inspection robot can make it difficult to determine the changing trends of the same equipment's position. Additionally, all types of data must be expressed within the same spatial reference system for multi-source fusion to be successful. Therefore, it is necessary to uniformly map data to the same spatial coordinate system.

[0076] Based on the obtained unified time data, spatial location information of multi-source data and robot pose information at each time point are acquired. Based on the unified spatial coordinate system, coordinate transformation and spatial mapping are performed on the data at each time point to obtain unified coordinate data, thereby establishing a unified spatial reference system and spatial correspondence, and realizing unified dynamic measurement position.

[0077] The steps to achieve spatial alignment include:

[0078] Determine a unified spatial coordinate system: First, select a unified spatial reference system, usually the equipment body coordinate system (based on the equipment) or the inspection environment coordinate system (based on the scene). The coordinate system must be fixed and can describe the position of all measurement points. After determining the coordinate system, all data must be converted to this system.

[0079] Obtain the installation position parameters of each sensor: For each sensor, obtain its spatial position information, including its installation position in the robot body and its relative positional relationship with the target device, i.e., three-dimensional position coordinates and installation direction information. This information is usually obtained in advance through calibration and used as fixed parameters to establish a fixed relationship between the sensor coordinate system and the robot coordinate system.

[0080] Obtaining robot pose information: Since the robot is in motion, it is necessary to obtain its position and orientation information at each point in time, including spatial position and orientation angle, to describe the robot's position in a unified spatial coordinate system. This can be obtained by fusing multiple positioning data (such as odometry and inertial measurement) through an extended Kalman filter.

[0081] Coordinate transformation calculation: For the data at each time point, the position of the sensor in the robot coordinate system is transformed by position translation and spatial rotation. Combined with the current pose of the robot, it is mapped to a unified spatial coordinate system to obtain the actual position of the sensor in the unified coordinate system, thereby realizing a unified expression of the spatial position of each sensor.

[0082] Image data spatial mapping processing: For image data, additional processing is required. A perspective transformation algorithm is used to establish the geometric relationship between the image and the actual space. By obtaining the intrinsic and extrinsic parameters of the camera, the image pixel coordinates are mapped to the actual spatial position. Then, the mapping result is transformed to a unified spatial coordinate system, thereby mapping the image data from a two-dimensional plane to a three-dimensional space.

[0083] Constructing a unified coordinate data set: After completing the spatial transformation of all data, for each time point, all sensor data are marked according to unified coordinates to construct a unified structure of time, space and multi-source data combination, resulting in data that is consistent in time, space and complete in data type.

[0084] Spatial consistency verification: The conversion results are checked, including whether data from the same device location is clustered and whether there are obvious spatial offset anomalies. If an anomaly is found, the data point is marked and not included in subsequent processing, thereby improving the reliability of spatial data and preventing error accumulation.

[0085] S3. Conduct a reliability assessment based on the unified coordinate data to obtain weighted measurement data.

[0086] In this embodiment, the credibility assessment includes:

[0087] In industrial inspection environments, multi-source data are not always equivalent, and their reliability varies with environmental changes. This is because sensors are affected differently by the environment. For example, temperature sensors are greatly affected by ambient temperature fluctuations, and acoustic signals are easily interfered with by background noise. Therefore, at the same point in time, different data have different levels of reliability. Noise levels change dynamically. For example, noise may originate from mechanical interference, electromagnetic interference, or environmental noise. When noise increases, it can cause data to deviate from the true value, and the signal quality becomes unstable. Signal quality (such as signal-to-noise ratio) changes over time, being higher under normal operating conditions and potentially decreasing under abnormal operating conditions. In subsequent fusion steps, if the original data is used directly, low-quality data will affect the overall results and may even mask real anomalies. Therefore, it is necessary to differentiate and process the data.

[0088] Based on the obtained unified coordinate data, the reliability score is generated by calculating the signal quality index, noise intensity index, and environmental adaptability index of various measurement signals. The weights of various measurement signals are then assigned to obtain weighted measurement data, thereby defining the reliability of each type of data. This makes the data comparable, improves the adaptability to complex environments, realizes the differentiated expression of multi-source data, improves the reliability of multi-source data fusion, provides a more accurate data foundation for subsequent feature extraction, and enhances the accuracy of anomaly identification.

[0089] Credibility assessment methods include:

[0090] Read unified coordinate data: Obtain the multi-source measurement values ​​corresponding to each time point from the unified coordinate data output in step S2. Each data point has a unified time and space identifier, thereby establishing the data set for the current time point.

[0091] Calculate the quality index of various types of signals: For each type of sensor data, calculate its signal quality index. The overall signal level is reflected by averaging the signal amplitude before and after the current time point. The signal amplitude range is obtained by calculating the difference between the maximum and minimum values, which reflects the intensity of signal changes. The average deviation is obtained by calculating the deviation of each sample value from the mean and averaging it, which reflects the signal fluctuation. The signal quality index is calculated by ratioing the average deviation to the amplitude range. This index indicates that the larger the average deviation, the lower the signal quality. Each type of signal corresponds to a quality index.

[0092] Calculate the noise intensity index: For each type of signal, assess its noise level, such as whether there are sudden changes or abnormal fluctuations. By calculating the amount of change between adjacent sampling points and the mean of the amount of change, the overall trend of change is reflected. Then, the degree of deviation of the amount of change from the mean is statistically analyzed, that is, the degree of fluctuation. Normalization is used to obtain the noise intensity index. The greater the deviation, the more drastic the change and the higher the noise intensity.

[0093] Assess environmental adaptability: For different sensors, analyze their applicability in the current environment. For example, image adaptability decreases in low light environment, and acoustic adaptability decreases in high noise environment. The degree of environmental adaptability is calculated based on the deviation of the current environmental conditions (such as light intensity, noise intensity, etc.) from the sensor's optimal adaptability range. For example, if the current ambient light intensity deviates from the median value of the sensor's optimal adaptability range by 20%, then the environmental adaptability is 80%.

[0094] Generate a credibility score: Based on the obtained quality index, noise intensity index and environmental adaptability, a comprehensive score is generated for each type of signal to describe the credibility of the current signal. The higher the signal quality, the higher the score; the greater the noise, the lower the score; and the better the environmental adaptability, the higher the score.

[0095] Weighting is performed: Based on the credibility scores of various signals, the scores are converted into weight values. The sum of all weights is a fixed value to ensure overall consistency, and signals with higher scores have greater weights.

[0096] Calculate weighted measurement data: Combine various types of data according to their corresponding weights, and sum the data at the same time point in a weighted manner to obtain weighted measurement data, which represents the result after comprehensively considering various types of data, and is used for subsequent feature extraction and anomaly identification.

[0097] Furthermore, the signal quality indicators include:

[0098]

[0099] in:

[0100] Signal quality index: This indicates the clarity of the signal. The higher the value, the more stable the signal and the higher the quality; the lower the value, the more cluttered the signal and the lower the quality.

[0101] Average deviation represents the signal fluctuation, and is obtained by averaging the deviations of each sampling point from the signal mean.

[0102] Signal amplitude range: This represents the intensity of signal changes and is obtained by calculating the difference between the maximum and minimum signal amplitude.

[0103] : Small constants to prevent the denominator from being zero.

[0104] Furthermore, the credibility score includes:

[0105]

[0106] in:

[0107] : No. Signal type at time The credibility score represents the reliability of the signal under the combined effect of multiple factors. The larger the value, the more reliable the signal. It is used to normalize the weight parameters of the signal.

[0108] : No. Signal type at time The signal quality index measures the effectiveness of the signal itself, ranging from 0 to 1. The higher the value, the higher the signal quality.

[0109] : No. Signal type at time The noise intensity index measures the degree of interference in a signal, ranging from 0 to 1. The higher the value, the higher the noise intensity.

[0110] : No. Signal type at time The degree of environmental adaptability measures the applicability of a signal under the current environmental conditions. It indicates whether a certain type of sensor is suitable for working in the current environment. The range is from 0 to 1, and the larger the value, the more suitable the environment is for the sensor to work.

[0111] : No. The signal quality influence weight of the signal type is used to adjust the degree of influence of signal quality on credibility. It is determined by empirical setting or historical data statistics, and ranges from 0 to 1. The sum of the three weights is 1.

[0112] : No. The noise suppression weights for signal types are used to adjust the degree of influence of noise on credibility. They are determined through empirical settings or historical data statistics and range from 0 to 1.

[0113] : No. The environmental adaptation weight of the signal is used to adjust the degree of influence of the environment on the credibility. It is determined by empirical setting or historical data statistics and ranges from 0 to 1.

[0114] A credibility score is obtained by weighting the signal quality, noise intensity, and environmental adaptability.

[0115] Furthermore, the weighted measurement data includes:

[0116]

[0117] in:

[0118] :time The weighted measurement data at the time represents the comprehensive measurement result after weighting the multi-source data, and serves as the input data for subsequent feature extraction and anomaly identification;

[0119] : No. Signal type at time The original measurement parameters, i.e. the sensor's measurement values ​​under unified time and unified spatial coordinates, are obtained through time unification and spatial unification in steps S1 and S2;

[0120] : No. Signal type at time The weight parameters determine the contribution ratio of each type of signal in the fusion, ranging from 0 to 1, and the sum of the weight parameters of all signals is 1;

[0121] The number of signal categories;

[0122] Weighted measurement data is obtained by weighting and synthesizing multi-range signal data.

[0123] S4. Perform multidimensional feature decomposition based on the weighted measurement data to obtain combined feature data.

[0124] In this embodiment, the multidimensional feature decomposition includes:

[0125] The weighted measurement data obtained in step S3 only reflects the measurement value at the current moment and cannot directly reflect signal change trends, periodic characteristics, and structural anomalies. Anomaly information is usually implicit in features; for example, equipment failure is often manifested as an increase in specific frequency components, and mechanical wear is manifested as an increase in vibration variance. This information cannot be directly reflected by a single numerical value, and different dimensions of information are complementary. For example, the time dimension reflects change trends, and the spatial dimension reflects image texture. A single dimension is insufficient to comprehensively reflect the equipment status. Therefore, feature decomposition is required to extract multidimensional features that can reflect the equipment status from the numerical data.

[0126] Based on the obtained weighted measurement data, statistical features, frequency domain features, and image texture features of the data are extracted by constructing a time analysis window. The features are then standardized and combined to obtain combined feature data, thereby extracting key state information, reducing the complexity of data dimensions, improving computational efficiency, and making different states (normal, degraded, abnormal) significantly different in the feature space, providing an input basis for subsequent judgment.

[0127] The steps of multidimensional eigenvalue decomposition include:

[0128] Read weighted measurement data: Obtain weighted measurement data from step S3. This data is a continuous data sequence arranged in time. By reading the data in time order, an input data sequence is provided for feature analysis.

[0129] Constructing a time analysis window: In order to extract stable features, a time window mechanism is introduced. By selecting a data segment of a certain length before and after the current time point, a local analysis interval is formed to avoid the randomness of single-point data and improve the stability of feature calculation. The window length is determined according to the actual application. Too short a window will lead to unstable features, while too long a window will lead to response lag, thus providing a stable basis for statistical and frequency domain analysis.

[0130] Extracting statistical features: Within the time window, perform statistical analysis on the data to extract basic features, including average level (reflecting the overall state) and degree of fluctuation (reflecting stability). These features are obtained by calculating the average value of all data within the window and the degree of dispersion of the data around the average value, thereby describing the overall distribution characteristics of the signal for subsequent anomaly detection. For example, an increase in the average temperature may indicate that the equipment is overheating, and increased vibration fluctuations may indicate that there is mechanical loosening.

[0131] Frequency domain feature extraction: Frequency analysis is performed on the data within the time window to extract periodic features. By using the mature Fast Fourier Transform (FFT) method, the time series is converted into a frequency distribution, and typical frequency domain features are extracted to obtain the main frequency component (reflecting the main vibration frequency) and the spectral energy distribution (reflecting the distribution of energy at different frequencies). This allows the capture of periodic variation characteristics in equipment operation. Many equipment anomalies have obvious frequency characteristics. For example, bearing damage usually manifests as a specific frequency anomaly, and gear anomalies will produce periodic impact signals, etc.

[0132] Extracting image texture features: Spatial feature extraction is performed on the image data. By using the mature Gray-Level Co-occurrence Matrix (GLCM) technology, the spatial relationship between pixels is analyzed, and contrast (reflecting texture changes) and contrast (reflecting texture changes) features are extracted. This transforms the image information into calculable numerical features. For example, high-contrast images may have defects, and images with low uniformity may have anomalies on their surfaces.

[0133] Feature standardization: Since different features have different dimensions, it is necessary to convert all types of features into the same scale range. This can be achieved by normalizing according to the maximum and minimum values ​​or by standardizing according to the statistical distribution, so as to avoid a certain type of feature dominating subsequent calculations.

[0134] Constructing combined feature data: Integrating various features to form combined feature data, which is used to describe the current device status. The features at each time point include statistical features, statistical characteristics, and image texture features.

[0135] S5. Determine the temporal consistency based on the combined feature data to obtain the stable state result.

[0136] In this embodiment, the timing consistency determination includes:

[0137] Even after S4 processing to obtain combined feature data, features at a single point in time can still be affected by random noise, transient interference, and data fluctuations, leading to unstable anomaly detection results. Furthermore, industrial equipment states are continuous; actual equipment operating conditions typically change gradually rather than drastically in a very short time. Therefore, single-point anomalies often do not represent true faults. True anomalies usually occur continuously and have temporal continuity, requiring time verification; otherwise, random fluctuations can easily be misjudged as anomalies. Therefore, consistent feature assessment across the time dimension is necessary to improve the reliability of the judgment.

[0138] Based on the obtained combined feature data, a time sliding window is constructed for the current time point. By performing initial state determination and state distribution statistics for each time point within the window, and making a state consistency judgment based on the state distribution, a stable state result is obtained, thereby reducing misjudgments, improving the accuracy of anomaly identification, reducing sensitivity fluctuations, avoiding frequent anomaly triggers due to small fluctuations, and enhancing stability.

[0139] Methods for implementing timing consistency determination include:

[0140] Read the combined feature time series: Obtain the combined feature data from the output of step S4, where each time point corresponds to a set of feature data, and the data is arranged in chronological order, which is used to construct the basic data for time series analysis.

[0141] Construct a time sliding window: For the current time point, establish a time window that includes the current time point and several time points before it, forming a continuous time interval to collect feature information over a period of time. The length of the time window needs to cover short-term fluctuations without affecting the response speed.

[0142] Initial state determination: For each time point within the window, the combined features are judged to obtain the corresponding state result. The state result usually includes normal state, degenerate state and abnormal state. The judgment criteria can be determined based on historical data. Judgment methods include: judgment based on single feature threshold. For example, for features with clear physical meaning (such as vibration amplitude, temperature, etc.), if the feature value exceeds the threshold range, it is judged as abnormal. For multi-feature combinations, multiple conditions must be met simultaneously to judge as abnormal. For example, the combination of increased vibration and enhanced high frequency indicates that the bearing is abnormal.

[0143] State distribution within the statistical window: Within the time window, the state results at all time points are statistically analyzed to obtain the frequency of occurrence of each type of state, thereby determining whether the current state is stable and analyzing the trend of state changes. For example, if the proportion of abnormal states is high, there may be real anomalies, while if the anomalies are scattered, they may be noise.

[0144] Consistency judgment processing: Based on the state distribution within the window, a unified judgment is made, including judging whether there are continuous abnormal states or whether abnormal states are dominant. If the abnormal states appear continuously or the proportion of abnormal states exceeds the abnormal proportion threshold, the current state is judged as abnormal; otherwise, it is judged as a normal or degenerate state. The abnormal proportion threshold can be set based on operational experience and historical data by statistically analyzing the abnormal proportion under normal and actual abnormal states, and fine-tuning it in actual operation. It is usually set to 0.5 to 0.8.

[0145] Output stable state result: The consistency determination result is used as the final state output at the current time point. This result has eliminated the influence of single-point fluctuations and can reflect the real operating state.

[0146] Time series update processing: Move the time window as time progresses and repeat the above steps to achieve continuous state determination.

[0147] S6. Based on the stable state results, conduct a health assessment to obtain the abnormality determination results.

[0148] In this embodiment, the health assessment includes:

[0149] The stable state result obtained in step S5 is a discrete judgment result, which cannot reflect the degree of state change. The severity of two abnormal states may be completely different, and the degradation of industrial equipment is continuous, usually manifesting as a gradual deviation from the normal state with a continuous change process. If only discrete judgment is used, early degradation cannot be captured, and different equipment with different characteristics are difficult to compare directly. Therefore, it is necessary to unify the evaluation scale and transform the discrete state result into a continuous health status expression to support early warning and trend analysis.

[0150] Based on the obtained stable state results, combined with the combined feature data, the degree of feature deviation is determined based on the health benchmark features, the health status is calculated, and the anomaly judgment result is obtained. In this way, the equipment status is transformed into a continuous numerical expression, a quantitative index of equipment health is constructed to describe the degree of status deviation, anomaly classification judgment is realized, and subsequent trend analysis is supported.

[0151] Methods for health assessment include:

[0152] Read the steady-state results and combined feature data: Obtain the steady-state results from step S5 and obtain the combined feature data at the corresponding time points from step S4 as input data for health assessment.

[0153] Determine health baseline characteristics: Select the characteristics of the equipment under normal conditions as a reference baseline, such as the average value of the characteristics during normal operation or the stable characteristics during the equipment initialization phase. This baseline is used to represent the ideal health state of the equipment.

[0154] Determine the degree of deviation of features: Compare the current feature value of each feature with the health benchmark, comprehensively analyze the change of each feature relative to the normal state, calculate the overall degree of deviation, and use it to reflect the distance between the current state and the normal state.

[0155] Health score calculation: A health score index is constructed based on the degree of deviation. The smaller the deviation, the higher the health score. At the same time, a maximum allowable deviation parameter is introduced to limit the range of health score variation, thereby converting feature differences into health score values.

[0156] Health score normalization: The health score is normalized so that its value range is fixed between 0 and 1. Normal state is usually processed to be close to 1 and abnormal state is close to 0, so as to provide a uniform scale of health evaluation index, which is convenient for comparison between different devices.

[0157] Set health assessment thresholds: Based on experience or historical data, set two thresholds. The upper threshold is used to distinguish between normal and deteriorating states, and the lower threshold is used to distinguish between deteriorating and abnormal states. The thresholds can be set by combining historical data statistics with experience, thereby dividing continuous health into discrete levels.

[0158] Generate anomaly determination results: Based on the relationship between health level and threshold, output the final result. For example, if the health level is higher than the upper threshold, output the normal state; if it is between the upper and lower thresholds, output the degraded state; if the health level is lower than the lower threshold, output the abnormal state. This provides the final inspection conclusion for alarm or maintenance decision-making.

[0159] Furthermore, the degree of feature deviation includes:

[0160]

[0161] in:

[0162] :time The degree of deviation of the feature at time represents the comprehensive relative deviation of the multidimensional features. It is used to quantify the difference between the current state and the health state. The value is greater than 0, and the larger the value, the more serious the deviation.

[0163] :time The first time The class feature value, as the basic input for health calculation, reflects the current device status and is obtained through step S4;

[0164] : No. A health benchmark for class features is used as a comparison reference to calculate the degree of feature deviation.

[0165] The number of feature types, i.e., the total number of features contained in the combined feature data;

[0166] A small constant used to prevent the denominator from being zero and to improve computational stability;

[0167] The degree of deviation of the features is calculated by comparing the results of multidimensional features with the health benchmark.

[0168] Furthermore, the health score calculation includes:

[0169]

[0170] in:

[0171] :time The health status at that time indicates the current health level of the device, with a range of (0,1]. The closer the value is to 1, the more normal the data is; the closer the value is to 0, the more abnormal the data is.

[0172] Maximum allowable deviation: This indicates the maximum allowable range of feature deviation, used to control the rate of health decay. It can be set by combining historical abnormal data statistics with experience, for example, a value of 0.5 to 5. When the value is larger, the health changes slowly, and when the value is smaller, the health changes sensitively.

[0173] The health score is calculated using an exponential function based on the degree of feature deviation and the maximum permissible deviation.

[0174] Example 2:

[0175] In practical applications, this invention can be applied to the equipment inspection scenario of power substations, and is used to continuously monitor and identify anomalies in the operating status of high-voltage switchgear, transformers and busbar connection parts. The typical scenario is illustrated by the inspection of the main transformer area of ​​a 110kV outdoor substation.

[0176] In this scenario, the inspection robot performs periodic inspections of the main transformer and its auxiliary equipment along a preset track. The robot is equipped with temperature sensors, vibration sensors, an acoustic acquisition module, a current and voltage detection unit, and a visible light camera. Specifically, the temperature sensor acquires the temperature values ​​of the transformer windings and tank surface; the vibration sensor detects the amplitude of mechanical vibrations during equipment operation; the acoustic module collects abnormal discharge or mechanical noise signals; the electrical parameter detection unit acquires current and voltage changes; and the camera acquires image information of the equipment surface.

[0177] During the inspection, various sensors synchronously sample according to a unified time reference, and acquire the spatial location information of the equipment through a robot positioning system, mapping the measurement data corresponding to different sensors to a unified equipment coordinate system. For example, temperature measurement points, vibration measurement points, and image acquisition areas are uniformly mapped to the specific spatial location of the transformer casing, thereby forming a spatially consistent set of measurement data.

[0178] After obtaining unified coordinate data, reliability assessments are performed on various types of measurement data. By analyzing the short-term fluctuation amplitude of temperature data, the spectral stability of vibration signals, the continuity of energy distribution in acoustic signals, and the smoothness of electrical parameter changes, quality indices for each type of data are calculated. These are then combined with noise intensity and environmental adaptability indices to obtain a comprehensive reliability score. Based on these assessment results, corresponding weights are assigned to each type of measurement data, ensuring that data with high stability and good continuity occupy a higher proportion in subsequent processing, thus forming weighted measurement data.

[0179] Based on weighted measurement data, multidimensional features are extracted within a fixed time window, including average temperature and its variation amplitude, vibration dominant frequency and spectral energy distribution, acoustic signal energy concentration, and image texture contrast. All features are then processed to a uniform scale to form combined feature data for describing the equipment status.

[0180] The combined features within a continuous time period are used to determine the state. By classifying the state at each moment, the corresponding preliminary state results are obtained. Based on this, the state distribution is statistically analyzed through a sliding time window to determine consistency. When an abnormal state is detected to occur continuously within a continuous time interval or its proportion exceeds a set ratio, the current device is determined to be in a stable abnormal state, thereby avoiding misjudgment due to transient interference.

[0181] After obtaining the stable state results, the characteristics of the equipment during normal operation are selected as the health benchmark. The deviation between the current combined characteristics and the benchmark characteristics is calculated, and a health index is constructed. By normalizing the health index and combining it with a preset threshold, the equipment status is divided into three levels: normal, performance degradation, and abnormal. When the health index is lower than the abnormal threshold and the stable state result is also determined to be abnormal, the final abnormal determination result is output, and an alarm signal is triggered.

[0182] Through the above implementation methods, unified expression of multi-source data, credibility weighted processing, and stable anomaly identification can be achieved in the complex electromagnetic environment of substations, effectively improving the accuracy and reliability of inspection results and meeting the requirements of unmanned inspection for real-time performance and stability.

[0183] Example 3:

[0184] This embodiment also provides a computer device applicable to the intelligent measurement and anomaly identification method for multi-source data in robot inspection, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent measurement and anomaly identification method for multi-source data in robot inspection as proposed in the above embodiment.

[0185] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the multi-source data intelligent measurement and anomaly identification method for robot inspection proposed in the above embodiments.

[0186] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0187] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0188] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0189] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0190] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0191] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent measurement and anomaly identification of multi-source data for robot inspection, characterized in that, Includes the following steps: S1. Time alignment of multi-source data to obtain unified time data; S2. Perform spatial alignment based on unified time data to obtain unified coordinate data; S3. Conduct a credibility assessment based on the unified coordinate data to obtain weighted measurement data; S4. Perform multidimensional feature decomposition based on the weighted measurement data to obtain combined feature data; S5. Determine the temporal consistency based on the combined feature data to obtain the stable state result; S6. Based on the stable state results, conduct a health assessment to obtain the abnormality determination results.

2. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 1, characterized in that, The time alignment in step S1 includes: A unified time reference is determined, a time series is constructed based on the raw data from multiple sources, a time mapping relationship is established for each time point in the time series, linear interpolation calculation is performed to achieve time alignment, and unified time data is obtained.

3. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 1, characterized in that, The spatial alignment in step S2 includes: Based on the obtained unified time data, spatial location information of multi-source data and robot pose information at each time point are obtained. Based on the unified spatial coordinate system, coordinate transformation and spatial mapping are performed on the data at each time point to obtain unified coordinate data.

4. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 1, characterized in that, The credibility assessment in step S3 includes: Based on the obtained unified coordinate data, the reliability score is generated by calculating the signal quality index, noise intensity index, and environmental adaptability index of various measurement signals, and weights are assigned to various measurement signals to obtain weighted measurement data.

5. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 4, characterized in that, The credibility score includes: ； in: : No. Signal type at time Credibility score; : No. Signal type at time Signal quality metrics; : No. Signal type at time The noise intensity index; : No. Signal type at time The degree of environmental adaptability; : No. The signal quality of the signal affects the weights; : No. Noise suppression weights for signal types; : No. Environmental adaptation weights for signal types; A credibility score is obtained by weighting the signal quality, noise intensity, and environmental adaptability.

6. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 1, characterized in that, The multidimensional feature decomposition in step S4 includes: Based on the obtained weighted measurement data, statistical features, frequency domain features, and image texture features of the data are extracted by constructing a time analysis window. The features are then standardized and combined to obtain combined feature data.

7. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 1, characterized in that, The timing consistency determination in step S5 includes: Based on the obtained combined feature data, a time sliding window is constructed for the current time point. By performing initial state determination and state distribution statistics for each time point within the window, and based on the state distribution, a state consistency judgment is made to obtain a stable state result.

8. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 1, characterized in that, The health assessment in step S6 includes: Based on the obtained stable state results, combined with the combined feature data, the degree of feature deviation is determined based on the health benchmark features, the health status is calculated, and the abnormality judgment result is obtained.

9. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 8, characterized in that, The degree of deviation of the feature includes: ； in: :time The degree of deviation of the characteristics at that time; :time The first time Class feature values; : No. Health benchmarks for class characteristics; The number of feature types; A small constant used to prevent the denominator from being zero and to improve computational stability; The degree of deviation of the features is calculated by comparing the results of multidimensional features with the health benchmark.

10. The method for intelligent measurement and anomaly identification of multi-source data for robot inspection according to claim 8, characterized in that, The health score calculation includes: ； in: :time Health status at that time; Maximum permissible deviation; The health score is calculated using an exponential function based on the degree of feature deviation and the maximum permissible deviation.

Images