A device status monitoring and abnormal warning method based on multi-source data fusion

Abstract

The present invention discloses a device status monitoring and abnormality early warning method based on multi-source data fusion, specifically relating to the field of monitoring and early warning technology. The method involves segmenting the original monitoring data according to sampling frequency differences to generate time series subsets, extracting transient features from the time series subsets and constructing asynchronous interference identification indicators. The transient features are evaluated and corrected for time domain misalignment to obtain a synchronous feature set. The synchronous feature set is holographically converted to construct an initial holographic feature matrix, which is then filtered through a feature sparse discrimination mechanism to obtain a sparse feature matrix. A multimodal mapping model is trained using historical data from target device fault diagnosis, and an abnormality risk score is output. The target device's status label and corresponding early warning result are generated based on a preset abnormality threshold, achieving high-precision, low-latency prediction and alarming of device failures under complex non-steady-state conditions.

Description

Technical Field

[0001] The present invention relates to the field of monitoring and early warning technology, and more specifically, to a device status monitoring and abnormality early warning method based on multi-source data fusion. Background Art

[0002] Existing equipment condition monitoring methods often rely on single sensor signals or feature extraction strategies under steady-state conditions, making it difficult to adapt to the temporal inconsistencies and feature expression complexity of multi-source heterogeneous data during non-steady-state operation. In multi-source data fusion scenarios, differences in sampling frequency and response delay between different sensors often lead to misalignment of fault information in the time domain, making it difficult to align transient features and causing severe fusion distortion. Furthermore, while current spectral feature construction methods can integrate multi-source information, the effective feature components are sparsely distributed, and critical fault information is easily overwhelmed by redundant features, resulting in reduced diagnostic performance and wasted computing resources.

[0003] In order to solve the above problems, a technical solution is now provided. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an equipment status monitoring and abnormality early warning method based on multi-source data fusion to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A device status monitoring and abnormality early warning method based on multi-source data fusion includes the following steps:

[0007] Collect the original monitoring data of the target equipment under non-steady-state conditions, segment it according to the sampling frequency difference, and obtain multiple time series subsets;

[0008] According to the operating state characteristics of the target equipment under non-steady-state conditions, the transient features of each time series subset are extracted to construct a multi-scale transient feature set;

[0009] Establish asynchronous interference identification indicators, perform time domain misalignment evaluation and correction on each transient feature, and generate a corrected synchronous transient feature set;

[0010] Performing holographic spectrum conversion on the synchronous transient feature set to obtain the holographic spectrum feature initial matrix, and using the feature sparsity discrimination mechanism to evaluate the sparsity of the holographic spectrum feature initial matrix and remove redundant features to form a sparse feature matrix;

[0011] Based on the target equipment fault diagnosis history data, a multimodal mapping model between the sparse feature matrix and the target equipment status is established. The sparse feature matrix is input to obtain the abnormal risk score value.

[0012] Based on the comparison between the abnormal risk score value and the preset abnormal threshold, the status label of the target device is determined, and the abnormal status warning result of the target device is given based on the status label of the target device.

[0013] In a preferred embodiment, the original monitoring data of the target device under non-steady-state conditions is collected and segmented according to the sampling frequency difference to obtain multiple time series subsets, specifically:

[0014] Obtain the original monitoring data of multiple different types of sensors of the target equipment under non-steady-state conditions, and determine the sampling interval of the time series based on the sampling frequency of each sensor;

[0015] The original monitoring data are time-series segmented according to the sampling interval to form independent time series subsets corresponding to the sampling frequency of each sensor;

[0016] Each independent time series subset is marked according to a unified time base to obtain multiple time series subsets.

[0017] In a preferred embodiment, based on the operating state characteristics of the target device under non-steady-state conditions, the transient features of each time series subset are extracted to construct a multi-scale transient feature set, specifically:

[0018] Each time series subset is subjected to sliding slicing processing according to a fixed window length to generate multiple time window segments;

[0019] In each time window segment, the time domain variation features including signal fluctuation mutation, edge jump, short-term peak and periodic disturbance are extracted to construct a single-scale transient feature vector;

[0020] Multiple single-scale transient feature vectors generated under different window lengths are uniformly encoded and combined according to the time window scale level to generate a multi-scale transient feature set.

[0021] In a preferred embodiment, an asynchronous interference identification index is established, and time domain misalignment evaluation and correction are performed on each transient feature to generate a corrected synchronous transient feature set, specifically:

[0022] Calculating the time domain offset between each single-scale transient characteristic vector and a preset synchronization reference, and determining the degree of asynchronous interference according to the time domain offset;

[0023] An asynchronous interference identification index is established according to the degree of asynchronous interference, and the asynchronous interference identification index is used to perform time domain misalignment correction on each single-scale transient feature vector;

[0024] All single-scale transient feature vectors that have completed time domain misalignment correction are reconstructed in the time domain to generate a set of synchronized transient features that have completed correction.

[0025] In a preferred embodiment, the synchronous transient feature set is converted into a holographic spectrum to obtain an initial holographic spectrum feature matrix, and the sparsity evaluation and redundant feature removal of the initial holographic spectrum feature matrix are performed using a feature sparsity discrimination mechanism to form a sparse feature matrix, specifically:

[0026] Performing holographic spectrum conversion on the corrected single-scale transient feature vectors in the synchronous transient feature set to obtain a frequency spectrum corresponding to each corrected single-scale transient feature vector;

[0027] Expand each frequency spectrum according to a unified frequency axis coordinate to form an initial high-dimensional frequency feature space, generating an initial matrix of holographic spectrum features;

[0028] Calculating the sparsity of each spectrum characteristic component in the initial matrix of holographic spectrum characteristics, and determining the contribution value of each spectrum characteristic component in the initial matrix of holographic spectrum characteristics;

[0029] A sparsity threshold is set, and the redundant features in the initial matrix of holographic spectrum features are screened and removed using the sparsity threshold, and the spectrum feature components with contribution values higher than the sparsity threshold are extracted to form a sparse feature matrix.

[0030] In a preferred embodiment, a multimodal mapping model of a sparse feature matrix and the target device state is established based on the target device fault diagnosis history data, and the sparse feature matrix is input to obtain an abnormal risk score value, specifically:

[0031] Construct a data set containing the historical operating status data of the target device and the corresponding status labels, and standardize the historical operating status data to form a historical sparse feature sample set;

[0032] Pair the historical sparse feature sample set with the corresponding state label to establish a corresponding data pair set between the sparse features and the state label;

[0033] Constructing a multimodal mapping model based on a set of historical sparse feature samples and a set of corresponding relationship data pairs; the input of the multimodal mapping model is a sparse feature matrix, and the output of the multimodal mapping model is an abnormal risk score value;

[0034] The sparse feature matrix is input into the multimodal mapping model, and the mapping inference operation is performed to obtain the abnormal risk score value corresponding to the real-time operating status of the target device.

[0035] In a preferred embodiment, based on the comparison between the abnormal risk score value and the preset abnormal threshold, the status label of the target device is determined, and the abnormal status warning result of the target device is output based on the status label of the target device, specifically:

[0036] A first abnormality threshold and a second abnormality threshold are preset; the first abnormality threshold is smaller than the second abnormality threshold;

[0037] When the abnormal risk score value is less than or equal to the first abnormal threshold, determining that the state label of the target device is normal;

[0038] When the score value is greater than the first abnormality threshold and less than the second abnormality threshold, the status label of the target device is determined to be a warning state;

[0039] When the score value is greater than or equal to the second abnormality threshold, determining that the state label of the target device is an abnormal state;

[0040] The corresponding warning signal is triggered according to the status tag of the target device. The normal state does not trigger the warning, the warning state triggers the primary alarm prompt, and the abnormal state triggers the emergency stop command and generates a fault report.

[0041] The technical effects and advantages of the device status monitoring and abnormality early warning method based on multi-source data fusion of the present invention are as follows:

[0042] Through time series segmentation based on sampling frequency differences, multi-scale transient feature extraction, and correction based on asynchronous interference identification indicators, the system achieves precise time-domain alignment of multi-sensor non-steady-state fault signals, reducing diagnostic delays and misjudgments caused by signal misalignment. Furthermore, by combining holographic spectrum conversion with a sparse discrimination mechanism, it effectively compresses the high-dimensional spectrum space, retains highly sensitive frequency components, reduces redundant information, and efficiently captures critical faults. Based on a multimodal mapping model, it establishes a mapping relationship between sparse features and device status, outputs anomaly risk scores, and combines them with preset anomaly thresholds to determine status labels and generate graded warnings, achieving automation from raw data to anomaly warnings. This system improves the accuracy and robustness of equipment fault identification, reduces computational burden, and offers excellent real-time and scalability, making it suitable for online monitoring and early warning under a variety of complex operating conditions. It also has the ability to respond to both sudden and progressive faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a schematic diagram of an equipment status monitoring and abnormality early warning method based on multi-source data fusion in the present invention. DETAILED DESCRIPTION

[0044] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0045] Example:

[0046] Figure 1 The present invention provides a device status monitoring and abnormality early warning method based on multi-source data fusion, which includes the following steps:

[0047] Collect the original monitoring data of the target equipment under non-steady-state conditions, segment it according to the sampling frequency difference, and obtain multiple time series subsets;

[0048] According to the operating state characteristics of the target equipment under non-steady-state conditions, the transient features of each time series subset are extracted to construct a multi-scale transient feature set;

[0049] Establish asynchronous interference identification indicators, perform time domain misalignment evaluation and correction on each transient feature, and generate a corrected synchronous transient feature set;

[0050] Performing holographic spectrum conversion on the synchronous transient feature set to obtain the holographic spectrum feature initial matrix, and using the feature sparsity discrimination mechanism to evaluate the sparsity of the holographic spectrum feature initial matrix and remove redundant features to form a sparse feature matrix;

[0051] Based on the target equipment fault diagnosis history data, a multimodal mapping model between the sparse feature matrix and the target equipment status is established. The sparse feature matrix is input to obtain the abnormal risk score value.

[0052] Based on the comparison between the abnormal risk score value and the preset abnormal threshold, the status label of the target device is determined, and the abnormal status warning result of the target device is given based on the status label of the target device.

[0053] Specifically, the original monitoring data of the target equipment under non-steady-state conditions is collected and segmented according to the sampling frequency difference to obtain multiple time series subsets, including:

[0054] Obtain the original monitoring data of multiple different types of sensors of the target equipment under non-steady-state conditions, and determine the sampling interval of the time series based on the sampling frequency of each sensor;

[0055] Specifically, various types of sensors are deployed to collect real-time data on the working status of target equipment in non-steady-state conditions. Sensor types include but are not limited to:

[0056] Structural vibration sensor, used to monitor acceleration changes on the surface of the target equipment;

[0057] Temperature sensors to monitor heat buildup during operation;

[0058] Current sensor, used to record the current dynamics of the drive circuit;

[0059] Stress and strain sensors are used to obtain the mechanical response characteristics of materials in key areas;

[0060] Displacement sensors are used to monitor the relative movement of structural components.

[0061] Each type of sensor has its own sampling mechanism. The sampling frequency refers to the number of data points generated by the sensor per unit time. Different types of sensors have different sampling frequency settings.

[0062] For example: structural vibration sensors use high-frequency sampling, and the single sampling interval is a short time period; temperature sensors use low-frequency sampling due to the slow data change speed, and the sampling interval is relatively long; current sensors use medium-frequency sampling, and the sampling interval is between low-frequency sampling and high-frequency sampling.

[0063] The sampling interval for each sensor type's time series is calculated based on the reciprocal relationship between sampling frequency and sampling interval. The sampling interval can be calculated as follows: If the sampling frequency is the number of data points per unit time, then the sampling interval is the unit time divided by the number of data points. This assigns a precise time interval parameter to each type of sensor data.

[0064] The original monitoring data are time-series segmented according to the sampling interval to form independent time series subsets corresponding to the sampling frequency of each sensor;

[0065] Specifically, the raw monitoring data is segmented and processed along the time axis according to the fixed sampling intervals for each sensor type. The data collected by each sensor type is divided into several time segments, each representing a set of observations within a continuous time window. The number of data points within a time window is automatically determined based on the sensor's sampling interval, forming a uniformly distributed sequence of data points.

[0066] Exemplarily, the sampling interval of the structural vibration sensor is a short time period, and the monitoring data is segmented into high-density time window segments;

[0067] The sampling interval of the temperature sensor is relatively long, and the divided time segments contain fewer data points.

[0068] The segmentation operation ensures that each type of sensor forms a time series subset with clear structure and stable sampling interval, which helps to construct multi-scale windows in the transient feature extraction stage.

[0069] Mark each independent time series subset according to a unified time base to obtain multiple time series subsets;

[0070] Specifically, in order to ensure that each independent time series subset has a unified time index structure, each independent time series subset needs to be marked with a unified time reference, specifically:

[0071] For each independent time series subset, a recursive calculation method is used to generate a corresponding theoretical timestamp sequence based on the sampling start time and sampling interval of the corresponding sensor. Each theoretical timestamp is generated by continuously accumulating a fixed sampling interval from the sampling start time as the initial time to obtain the theoretical occurrence time of each data point in sequence, resulting in a single-channel time index list in order of the sampling point number.

[0072] A unified time base is constructed, using the monitoring start time of the target device's non-steady-state operation phase as the unified starting point to map the theoretical timestamps of each sensor. The mapping conversion process includes: adding the start offset time of each theoretical timestamp to the corresponding channel to convert it into a global timestamp in a unified reference coordinate; using the global timestamp, the time index of the time series subsets of different sensors is consistent.

[0073] After mapping, each time series subset retains the original sensor properties but is comparable and time-aligned with other sensor data along the time axis. Ultimately, multiple time series subsets with a unified time-stamping structure are formed, providing a structurally consistent data foundation for multimodal fusion, feature alignment, and interference correction.

[0074] Specifically, according to the operating state characteristics of the target equipment under non-steady-state conditions, the transient features of each time series subset are extracted to construct a multi-scale transient feature set, including:

[0075] Each time series subset is subjected to sliding slicing processing according to a fixed window length to generate multiple time window segments;

[0076] Specifically, for each subset of time series that has been uniformly time-stamped, data slicing is performed according to a pre-set sliding window strategy. This sliding window strategy consists of two key parameters: the window length, which defines the time span covered by each segment, and the sliding step size, which defines the time interval between the start and end of two adjacent segments.

[0077] The sliding window operation is performed as follows: starting from the first data point of each time series subset, a number of consecutive data points are extracted as the first window segment, and then the window is slid forward by one step, and data of the same length are extracted to form the second window segment until the end of the time series.

[0078] The window length is typically determined based on the typical periodicity of the sensor data. If the structural response reflected by the vibration data has multiple periodic components, multiple window lengths of varying sizes are required to accommodate varying dynamic response rates. The time window segments generated by sliding have local temporal continuity, enabling the capture of local response variations during unsteady processes.

[0079] In each time window segment, the time domain variation features including signal fluctuation mutation, edge jump, short-term peak and periodic disturbance are extracted to construct a single-scale transient feature vector;

[0080] Specifically, in each time window segment, local transient feature extraction processing is carried out based on the time domain structure:

[0081] Signal fluctuation mutation characteristics: Calculate the difference sequence between adjacent data points in the current window segment, record the maximum difference amplitude, average fluctuation amplitude and the number of fluctuation direction changes to reflect the degree of local sudden disturbance.

[0082] Edge transition feature: Detects changes in edge slope in a data sequence and evaluates the direction and strength of the transition trend by measuring the rising and falling gradients between two or more consecutive points.

[0083] Short-term peak characteristics: Count the number of extreme points exceeding the preset threshold within the time window segment, including the number of positive peaks and negative valleys, and record the amplitude and position index separately to identify abnormal fluctuation points.

[0084] Periodic disturbance characteristics: The envelope analysis method of the signal within the window is used to measure the main frequency band position and energy distribution trend of the periodic disturbance, reflecting the existence of short-period characteristics.

[0085] All of the above transient change features are uniformly encoded into a vector structure containing fixed-length data items, defined as a single-scale transient feature vector. Each single-scale transient feature vector fully corresponds to a time window segment, preserving the original time index and channel identification information.

[0086] Multiple single-scale transient feature vectors generated under different window lengths are uniformly encoded and combined according to the time window scale level to generate a multi-scale transient feature set;

[0087] Specifically, in order to fully capture the non-steady-state operation characteristics at different time granularities, repeated sliding slicing and transient feature extraction are performed on the same time series subset using multiple different window lengths. The feature vectors extracted under different window lengths are complementary in scale:

[0088] A smaller window length can capture short-period shock features and is suitable for identifying small disturbances;

[0089] Larger window lengths can extract stable trend changes and low-frequency variations, making them suitable for identifying slow aging or wear patterns.

[0090] For all single-scale transient feature vectors obtained under different window lengths, unified dimension encoding is performed to ensure that the single-scale transient feature vectors under all window lengths have a consistent structural format. A multi-level feature structure with scale labels is constructed, namely a multi-scale transient feature set. The unified encoding operation includes but is not limited to: maintaining consistency in the number of dimensions of all single-scale transient feature vectors, the order of the various transient features, and the mapping method of the feature value range.

[0091] Specifically, an asynchronous interference identification index is established, and time domain misalignment evaluation and correction are performed on each transient feature to generate a corrected synchronous transient feature set, including:

[0092] Calculating the time domain offset between each single-scale transient characteristic vector and a preset synchronization reference, and determining the degree of asynchronous interference according to the time domain offset;

[0093] Specifically, based on multiple time series subsets with unified time tags and the single-scale transient feature vectors generated therefrom, a sensor channel with high temporal resolution is selected as a synchronization reference channel. For example, a high sampling frequency channel composed of a vibration sensor is selected as the reference channel.

[0094] For each non-reference channel's single-scale transient feature vector, a sliding comparison is performed with the single-scale transient feature vector of the reference channel in chronological order. By calculating the similarity trend between the single-scale transient feature vectors of the non-reference and reference channels in adjacent time periods, the location where local feature alignment occurs is identified. The time offset value corresponding to this location is the time domain offset of the non-reference channel relative to the reference channel.

[0095] The time domain offset is represented by an integer sliding window displacement value, reflecting the degree of response delay between channels when observing the same fault signature. After completing the time domain offset measurement for all channels, the time domain offsets of each channel are summarized to construct an asynchronous interference degree matrix. The asynchronous interference degree quantifies the relative offset between multi-channel signals. Larger time offset values indicate more severe asynchronous interference between channels. Fault signatures refer to the cross-channel transient response characteristics caused by a local fault or disturbance of the target device during the same physical time period when multiple different channels record the operating data of the target device.

[0096] An asynchronous interference identification index is established according to the degree of asynchronous interference, and the asynchronous interference identification index is used to perform time domain misalignment correction on each single-scale transient feature vector;

[0097] Specifically, based on the asynchronous interference degree matrix, a corresponding asynchronous interference identification index is generated for each channel. The asynchronous interference identification index is defined as the product of each channel's time domain offset and the channel sampling period, indicating the time value by which the current channel data needs to be shifted forward or backward on the unified time base.

[0098] Each single-scale transient eigenvector is used as a processing unit, and its time index information is corrected. This means that its original timestamp is adjusted by the time value required to be shifted forward or backward according to the asynchronous interference identification index. If the time value to be shifted is positive, the time value is shifted backward by the corresponding number of sampling intervals; if the time value to be shifted is negative, the time value is adjusted forward by the corresponding time step. This ensures that the single-scale transient eigenvectors of all channels are aligned on the time axis to the greatest extent possible when the same physical event occurs.

[0099] The time domain misalignment correction operation does not change the original eigenvalue structure, but only shifts the time label, retaining the original eigenvector content and sampling source information.

[0100] Perform time domain reconstruction on all single-scale transient feature vectors that have completed time domain misalignment correction to generate a set of synchronized transient features after correction;

[0101] Specifically, all single-scale transient feature vectors that have completed the time domain misalignment correction are reorganized into a unified time axis, and the time series is reconstructed according to their new corrected time index. The specific reconstruction process is:

[0102] Establish a global time index structure with a unified time base as the horizontal axis;

[0103] Attach each single-scale transient feature vector to the corresponding time node according to the corrected timestamp;

[0104] The single-scale transient feature vectors from different channels and different window lengths at the same time node are numbered and sorted to construct a complete set of synchronous feature set units.

[0105] After reconstruction, the positions of all single-scale transient feature vectors on the time axis have been unified, forming a set of synchronized transient features with consistent structure, comparable time, and corresponding channels, providing strictly aligned data input conditions for spectral transformation and fusion modeling.

[0106] Specifically, the synchronous transient feature set is converted into a holographic spectrum to obtain an initial holographic spectrum feature matrix, and the sparsity of the initial holographic spectrum feature matrix is evaluated and redundant features are removed using a feature sparsity discrimination mechanism to form a sparse feature matrix, including:

[0107] Performing holographic spectrum conversion on the corrected single-scale transient feature vectors in the synchronous transient feature set to obtain a frequency spectrum corresponding to each corrected single-scale transient feature vector;

[0108] Specifically, a frequency conversion operation is performed on each single-scale transient eigenvector that has undergone time-domain alignment, mapping it from a time-domain representation to a frequency-domain representation. This conversion process uses a preset window function to perform boundary processing on the eigenvectors, and then extracts their energy distribution characteristics along the frequency dimension through frequency response mapping. Each frequency response output corresponds to a frequency spectrogram, where each item represents the amplitude response at that frequency position.

[0109] The frequency spectrum reflects the response intensity of single-scale transient characteristics to different frequency components in the frequency domain, and can reveal weak oscillations or periodic disturbances in the time domain.

[0110] Expand each frequency spectrum according to a unified frequency axis coordinate to form an initial high-dimensional frequency feature space, generating an initial matrix of holographic spectrum features;

[0111] Specifically, each frequency spectrum is linearly expanded along a predefined frequency axis, and the response amplitudes corresponding to each frequency point are arranged in sequence as a vector structure. All frequency spectrum vectors are kept consistent in length and the frequency coordinates are aligned, forming a frequency response matrix with a unified coordinate system.

[0112] After unfolding all frequency spectra, they are arranged row by row according to the time node and channel sequence to form a frequency feature matrix. The number of rows in the frequency feature matrix is equal to the number of single-scale transient eigenvectors, and the number of columns is equal to the number of frequency components covered by the frequency spectra. This frequency feature matrix is the initial matrix of holographic spectrum features and serves as the input basis for sparse feature discrimination and redundant information elimination.

[0113] Calculating the sparsity of each spectrum characteristic component in the initial matrix of holographic spectrum characteristics, and determining the contribution value of each spectrum characteristic component in the initial matrix of holographic spectrum characteristics;

[0114] Specifically, each frequency component column in the initial matrix of the holographic spectrum characteristics corresponds to a spectrum characteristic component. For each frequency component column in the initial matrix of the holographic spectrum characteristics, the distribution density of non-zero response amplitudes is calculated. The specific processing method is to count the number of data points with non-zero response amplitudes in each frequency component column and calculate the ratio of this value to the total number of samples to obtain the response density value of the current spectrum characteristic component.

[0115] The amplitude variance and concentration of each column of spectral feature components are jointly measured to calculate the contribution of each column of spectral feature components to the feature distribution difference in the overall spectral space. Specifically, for each column of frequency components, the response amplitude variance across all samples is first calculated. The response concentration of the frequency component under different target device state labels is then calculated, for example, by calculating the entropy value. The response amplitude variance and entropy value are normalized and then weighted and summed to obtain the feature discrimination contribution value of the frequency component in the spectral space. The higher the feature discrimination contribution value, the stronger the ability of the response change at the frequency point to distinguish between different target device states.

[0116] The response density of the spectral feature components and the feature distinction contribution value are combined to form a sparse discriminant index sequence for feature screening operations.

[0117] Setting a sparsity threshold, using the sparsity threshold to screen and remove redundant features in the initial matrix of holographic spectrum features, extracting spectrum feature components with contribution values higher than the sparsity threshold, and forming a sparse feature matrix;

[0118] Specifically, a sparsity threshold is set to determine which spectral feature components possess sufficient diagnostic value. The sparsity threshold is determined based on the distribution of historical data samples for the target device and is typically the median of the feature discrimination contribution values. The sparsity threshold is set based on the principle of retaining spectral feature components that are representative of the target device's state. Spectral feature components in the sparse discrimination indicator sequence whose feature discrimination contribution values exceed the sparsity threshold are retained, while the remaining spectral feature components are considered redundant and removed.

[0119] The retained spectral feature components form a new frequency feature sub-matrix, which is a sparse feature matrix, which can effectively improve the modeling efficiency and discrimination ability in the modeling of multimodal mapping models.

[0120] Specifically, a multimodal mapping model between a sparse feature matrix and the target device status is established based on the target device fault diagnosis history data. The sparse feature matrix is input to obtain an abnormal risk score value, including:

[0121] Construct a data set containing the historical operating status data of the target device and the corresponding status labels, and standardize the historical operating status data to form a historical sparse feature sample set;

[0122] Specifically, the system collects historical fault diagnosis data from target devices over multiple typical operating cycles, including samples of normal, warning, and abnormal states. The system then performs time series subsetting, multi-scale transient feature extraction, asynchronous interference identification and correction, spectrum conversion, and sparse feature extraction on the historical fault diagnosis data to generate a historical sparse feature matrix.

[0123] Normalize the frequency components of each column of the historical sparse feature matrix. This is done by subtracting the mean of the response amplitudes for all samples in each column from the mean of the corresponding column, and then dividing the result by the standard deviation of the response amplitudes for that column. The normalized data becomes the historical sparse feature sample set.

[0124] Pair the historical sparse feature sample set with the corresponding state label to establish a corresponding data pair set between the sparse features and the state label;

[0125] Specifically, for each sample entry in the historical sparse feature sample set, the corresponding actual operating status record in the historical equipment monitoring records is searched. The actual operating status is obtained by expert annotation or automatic identification from the operation log, and specifically includes but is not limited to status labels such as normal, warning, and abnormal.

[0126] Each sparse feature vector is paired with its corresponding state label to construct a set of sample-label correspondence pairs. Each data pair contains two fields: the normalized sparse feature vector and the target device's state classification label at the corresponding moment. This set of correspondence pairs serves as the training dataset for the multimodal mapping model and contains both state distribution and sparse response feature information.

[0127] Constructing a multimodal mapping model based on a set of historical sparse feature samples and a set of corresponding relationship data pairs; the input of the multimodal mapping model is a sparse feature matrix, and the output of the multimodal mapping model is an abnormal risk score value;

[0128] Specifically, the larger the abnormality risk score value, the greater the possibility that the target device has abnormality risks.

[0129] The sparse feature matrix is input into the multimodal mapping model, and the mapping inference operation is performed to obtain the abnormal risk score value corresponding to the real-time operating status of the target device;

[0130] Specifically, a mapping inference operation is performed, that is, the sparse feature matrix is input into the mapping function set in the multimodal mapping model inference stage, and the abnormal risk score value corresponding to the real-time operating status of the target device is output according to the structural parameters and response rules in the multimodal mapping model.

[0131] Specifically, based on the comparison between the abnormal risk score value and the preset abnormal threshold, the status label of the target device is determined, and the abnormal status warning result of the target device is output based on the status label of the target device, including:

[0132] A first abnormality threshold and a second abnormality threshold are preset; the first abnormality threshold is smaller than the second abnormality threshold;

[0133] Specifically, the setting of the first abnormality threshold and the second abnormality threshold is based on the statistical distribution characteristics of the historical abnormality risk score values corresponding to the target device in the historical operation data, specifically: all historical abnormality risk score values are counted, and the historical abnormality risk score values are classified and grouped according to the status label; the upper quartile of the historical abnormality risk score values corresponding to the historical normal state is selected as the first abnormality risk score threshold, and the lower quartile of the historical abnormality risk score values corresponding to the historical abnormal state is selected as the second abnormality risk score threshold.

[0134] When the abnormal risk score value is less than or equal to the first abnormal threshold, determining that the state label of the target device is normal;

[0135] When the score value is greater than the first abnormality threshold and less than the second abnormality threshold, the status label of the target device is determined to be a warning state;

[0136] When the score value is greater than or equal to the second abnormality threshold, determining that the state label of the target device is an abnormal state;

[0137] The corresponding warning signal is triggered according to the status tag of the target device. The normal state does not trigger the warning, the warning state triggers the primary alarm prompt, and the abnormal state triggers the emergency stop command and generates a fault report.

[0138] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0139] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0140] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0141] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0142] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0143] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0144] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0145] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0146] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0147] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A device status monitoring and abnormality early warning method based on multi-source data fusion, characterized in that: The steps include: Collect the original monitoring data of the target equipment under non-steady-state conditions, segment it according to the sampling frequency difference, and obtain multiple time series subsets; According to the operating state characteristics of the target equipment under non-steady-state conditions, the transient features of each time series subset are extracted to construct a multi-scale transient feature set; Establish asynchronous interference identification indicators, perform time domain misalignment evaluation and correction on each transient feature, and generate a corrected synchronous transient feature set; Performing holographic spectrum conversion on the synchronous transient feature set to obtain the holographic spectrum feature initial matrix, and using the feature sparsity discrimination mechanism to evaluate the sparsity of the holographic spectrum feature initial matrix and remove redundant features to form a sparse feature matrix; Based on the target equipment fault diagnosis history data, a multimodal mapping model between the sparse feature matrix and the target equipment status is established. The sparse feature matrix is input to obtain the abnormal risk score value. Determine the status label of the target device based on the comparison between the abnormal risk score value and the preset abnormal threshold value, and issue an abnormal warning result for the status of the target device based on the status label of the target device; According to the operating state characteristics of the target equipment under non-steady-state conditions, the transient features of each time series subset are extracted and a multi-scale transient feature set is constructed, specifically: Each time series subset is subjected to sliding slicing processing according to a fixed window length to generate multiple time window segments; In each time window segment, the time domain variation features including signal fluctuation mutation, edge jump, short-term peak and periodic disturbance are extracted to construct a single-scale transient feature vector; Multiple single-scale transient feature vectors generated under different window lengths are uniformly encoded and combined according to the time window scale level to generate a multi-scale transient feature set; The synchronous transient feature set is converted into a holographic spectrum to obtain the initial holographic spectrum feature matrix. The sparsity of the initial holographic spectrum feature matrix is evaluated and redundant features are removed using the feature sparse discrimination mechanism to form a sparse feature matrix, which is specifically: Performing holographic spectrum conversion on the corrected single-scale transient feature vectors in the synchronous transient feature set to obtain a frequency spectrum corresponding to each corrected single-scale transient feature vector; Expand each frequency spectrum according to a unified frequency axis coordinate to form an initial high-dimensional frequency feature space, generating an initial matrix of holographic spectrum features; Calculating the sparsity of each spectrum characteristic component in the initial matrix of holographic spectrum characteristics, and determining the contribution value of each spectrum characteristic component in the initial matrix of holographic spectrum characteristics; A sparsity threshold is set, and the redundant features in the initial matrix of holographic spectrum features are screened and removed using the sparsity threshold, and the spectrum feature components with contribution values higher than the sparsity threshold are extracted to form a sparse feature matrix.

2. The device status monitoring and abnormality early warning method based on multi-source data fusion according to claim 1 is characterized in that: The original monitoring data of the target equipment under non-steady-state conditions is collected and segmented according to the sampling frequency difference to obtain multiple time series subsets, specifically: Obtain the original monitoring data of multiple different types of sensors of the target equipment under non-steady-state conditions, and determine the sampling interval of the time series based on the sampling frequency of each sensor; The original monitoring data are time-series segmented according to the sampling interval to form independent time series subsets corresponding to the sampling frequency of each sensor; Each independent time series subset is marked according to a unified time base to obtain multiple time series subsets.

3. The device status monitoring and abnormality early warning method based on multi-source data fusion according to claim 2 is characterized in that: Establish asynchronous interference identification indicators, perform time domain misalignment evaluation and correction on each transient feature, and generate a corrected synchronous transient feature set, specifically: Calculating the time domain offset between each single-scale transient characteristic vector and a preset synchronization reference, and determining the degree of asynchronous interference according to the time domain offset; An asynchronous interference identification index is established according to the degree of asynchronous interference, and the asynchronous interference identification index is used to perform time domain misalignment correction on each single-scale transient feature vector; All single-scale transient feature vectors that have completed time domain misalignment correction are reconstructed in the time domain to generate a set of synchronized transient features that have completed correction.

4. The device status monitoring and abnormality early warning method based on multi-source data fusion according to claim 3 is characterized in that: Based on the target equipment fault diagnosis history data, a multimodal mapping model between the sparse feature matrix and the target equipment status is established. The sparse feature matrix is input to obtain the abnormal risk score value, which is specifically: Construct a data set containing the historical operating status data of the target device and the corresponding status labels, and standardize the historical operating status data to form a historical sparse feature sample set; Pair the historical sparse feature sample set with the corresponding state label to establish a corresponding data pair set between the sparse features and the state label; Constructing a multimodal mapping model based on a set of historical sparse feature samples and a set of corresponding relationship data pairs; the input of the multimodal mapping model is a sparse feature matrix, and the output of the multimodal mapping model is an abnormal risk score value; The sparse feature matrix is input into the multimodal mapping model, and the mapping inference operation is performed to obtain the abnormal risk score value corresponding to the real-time operating status of the target device.

5. The device status monitoring and abnormality early warning method based on multi-source data fusion according to claim 4 is characterized in that: Based on the comparison between the abnormal risk score and the preset abnormal threshold, the status label of the target device is determined, and the abnormal status warning result of the target device is output based on the status label of the target device, specifically: A first abnormality threshold and a second abnormality threshold are preset; the first abnormality threshold is smaller than the second abnormality threshold; When the abnormal risk score value is less than or equal to the first abnormal threshold, determining that the state label of the target device is normal; When the score value is greater than the first abnormality threshold and less than the second abnormality threshold, the status label of the target device is determined to be a warning state; When the score value is greater than or equal to the second abnormality threshold, determining that the state label of the target device is an abnormal state; The corresponding warning signal is triggered according to the status tag of the target device. The normal state does not trigger the warning, the warning state triggers the primary alarm prompt, and the abnormal state triggers the emergency stop command and generates a fault report.

Images