Speed reducer fault monitoring method and system

Abstract

The invention relates to the technical field of speed reducers, and particularly discloses a speed reducer fault monitoring method and system, and the method comprises the steps: obtaining multi-source data and working condition parameters; preprocessing the multi-source data; extracting a multi-dimensional feature vector from the preprocessed multi-source data according to the working condition parameters; calculating the deviation degree between the collaborative change condition of the multi-dimensional feature vector and the reference baseline; judging whether early abnormality occurs according to the deviation degree; when the early abnormality occurs, analyzing the type of an abnormal feature inducing deviation in the multi-dimensional feature vector, and determining an abnormal type and / or an abnormal part according to the type of the abnormal feature; according to the method, multi-source heterogeneous data acquisition and multi-dimensional feature extraction based on working condition parameters are combined, and a deviation degree calculation method based on a multi-dimensional feature collaborative change reference baseline is introduced to detect early abnormality, so that the problem that weak early fault signals are difficult to recognize in a traditional method is solved.

Description

Technical Field

[0001] This application relates to the field of speed reducer technology, and more specifically, to a speed reducer fault monitoring method and system. Background Technology

[0002] In modern industrial automated production, speed reducers, as key transmission equipment, are crucial to production line efficiency and product quality. Currently, the industry widely adopts multi-sensor networks for condition monitoring, including vibration, temperature, and acoustic sensors, to collect operational data in real time. With technological advancements, monitoring systems are developing towards integration and intelligence, aiming to improve equipment reliability and preventative maintenance capabilities to meet the continuous demands of industrial automation.

[0003] However, early fault signals in gearboxes are typically extremely weak and easily masked by background noise and interference generated during normal operation, making it difficult for traditional monitoring methods based on single sensor parameter thresholds to effectively identify these anomalies. Traditional early fault monitoring methods generally rely on fixed threshold alarms for single sensor parameters, failing to capture fault characteristics with low amplitude and low signal-to-noise ratios. Furthermore, different types of early faults exhibit complex and subtle patterns of coordinated change across multidimensional data, which simple multi-parameter correlation logic cannot distinguish. Therefore, in complex industrial environments, how to integrate multi-source sensor data, accurately extract weak fault characteristics, and differentiate specific fault types has become a key technical challenge in the field of gearbox condition monitoring.

[0004] There is currently no effective technical solution to the above problems. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for monitoring gearbox faults, so as to solve the problem that traditional methods are difficult to identify weak early fault signals.

[0006] In a first aspect, this application provides a method for monitoring gearbox faults, used to identify early faults in the gearbox, the method comprising the following steps:

[0007] S1. Collect multi-source data during the operation of the reducer based on multi-source heterogeneous sensors, and simultaneously obtain the operating parameters of the reducer;

[0008] S2. Preprocess the multi-source data;

[0009] S3. Extract multi-dimensional feature vectors from the preprocessed multi-source data based on the operating parameters.

[0010] S4. Calculate the deviation between the coordinated change of the multidimensional feature vector and the reference baseline, wherein the reference baseline is the distribution parameters of the coordinated change of multidimensional features under normal conditions, which are pre-constructed.

[0011] S5. Determine whether an early abnormality has occurred based on the deviation.

[0012] S6. When an early abnormality occurs, analyze the types of abnormal features that induce deviation in the multidimensional feature vector, and determine the abnormal type and / or abnormal location based on the types of abnormal features.

[0013] The method in this application combines multi-source heterogeneous data acquisition with multi-dimensional feature extraction based on operating parameters, and introduces a deviation calculation method based on the reference baseline of multi-dimensional feature co-change to detect early anomalies. This overcomes the problem that traditional methods are difficult to identify weak early fault signals. This monitoring process can effectively integrate multi-source heterogeneous sensor data, overcome the limitations of a single data source, and can perform adaptive feature extraction based on operating parameters to improve the ability to capture weak early fault features. Furthermore, by analyzing the deviation of the co-change of multi-dimensional features, it can identify early abnormal signals that are difficult to detect by traditional methods, effectively improving the accuracy and refinement of fault monitoring and providing strong support for predictive maintenance of reducers.

[0014] The aforementioned method for monitoring gearbox faults includes a preprocessing step that sequentially performs time alignment processing and data cleaning processing.

[0015] The aforementioned method for monitoring gearbox faults, wherein step S3 includes:

[0016] S31. Determine the current operating status of the reducer based on the operating parameters;

[0017] S32. Determine the feature extraction strategy based on the current operating status;

[0018] S33. Extract multidimensional feature vectors from the preprocessed multi-source data based on the feature extraction strategy.

[0019] Through the above processing, the feature extraction process is no longer static, but can be dynamically adjusted according to the actual operation of the equipment, thereby more accurately capturing weak signals of early faults, improving the sensitivity and robustness of features to early faults, and providing more reliable input for subsequent fault judgment.

[0020] The aforementioned method for monitoring gearbox faults, wherein step S4 includes:

[0021] S41. Analyze the data quality of different data in the multi-source data and obtain the quality assessment results;

[0022] S42. Determine the weight of each feature component in the multidimensional feature vector based on the quality assessment results;

[0023] S43. Based on the operating parameters, obtain the reference baseline corresponding to the current operating state;

[0024] S44. Calculate the deviation based on the multidimensional feature vector, the weights of each feature component, and the reference baseline.

[0025] The aforementioned method for monitoring gearbox faults, wherein the reference baseline includes the mean vector and covariance matrix of the multidimensional feature vector under normal conditions, and step S44 includes:

[0026] S441. Calculate the weighted Mahalanobis distance between the multidimensional feature vector and the mean vector based on the multidimensional feature vector, the weights of each feature component, the mean vector, and the covariance matrix, and use this distance as the deviation.

[0027] The aforementioned method for monitoring gearbox faults, wherein step S5 includes:

[0028] S51. Obtain a first threshold that matches the current operating state based on the operating condition parameters;

[0029] S52. Obtain the threshold compensation value based on the quality assessment results;

[0030] S53. Based on the threshold compensation value, compensate the first threshold to obtain the second threshold;

[0031] S54. Compare the deviation and the second threshold to determine whether an early anomaly has occurred.

[0032] The aforementioned method for monitoring gearbox faults, wherein step S6 includes:

[0033] S61. When an early anomaly is determined to occur, analyze the contribution of each feature component in the multidimensional feature vector to the deviation.

[0034] S62. Based on the degree of contribution, identify several feature components that significantly contribute to the deviation as the abnormal features;

[0035] S63. Determine the abnormal type and / or the abnormal location based on the abnormal characteristics.

[0036] The aforementioned method for monitoring gearbox faults, wherein step S62 includes:

[0037] S621. Obtain the feature components whose contribution level exceeds the preset contribution threshold, and use them as the abnormal features.

[0038] The aforementioned method for monitoring gearbox faults, wherein step S63 includes:

[0039] S631. Based on the abnormal features, search a pre-established fault knowledge base to determine the abnormal type and / or the abnormal location. The fault knowledge base includes different abnormal features or combinations of abnormal features and their matching relationships with the abnormal type and / or the abnormal location.

[0040] Secondly, this application also provides a speed reducer fault monitoring system for identifying early-stage speed reducer faults, characterized in that the system comprises:

[0041] The acquisition module is used to collect multi-source data during the operation of the reducer based on multi-source heterogeneous sensors, and at the same time acquire the operating parameters of the reducer;

[0042] The preprocessing module is used to preprocess the multi-source data;

[0043] The feature extraction module is used to extract multi-dimensional feature vectors from the preprocessed multi-source data based on the operating condition parameters.

[0044] The deviation calculation module is used to calculate the deviation between the coordinated change of the multidimensional feature vector and the reference baseline, wherein the reference baseline is a pre-constructed distribution parameter of the coordinated change of multidimensional features under normal conditions.

[0045] An anomaly detection module is used to determine whether an early anomaly has occurred based on the deviation.

[0046] An anomaly localization module is used to analyze the types of abnormal features that induce deviations in the multidimensional feature vector when early anomalies occur, and to determine the anomaly type and / or anomaly location based on the types of abnormal features.

[0047] The system of this application combines multi-source heterogeneous data acquisition with multi-dimensional feature extraction based on operating parameters, and introduces a deviation calculation method based on multi-dimensional feature collaborative change reference baseline to detect early anomalies, thereby overcoming the problem that traditional systems have difficulty in identifying weak early fault signals.

[0048] As can be seen from the above, this application provides a method and system for monitoring gearbox faults. The method of this application combines multi-source heterogeneous data acquisition with multi-dimensional feature extraction based on operating parameters, and introduces a deviation calculation method based on the reference baseline of multi-dimensional feature co-change to detect early anomalies. This overcomes the problem that traditional methods are difficult to identify weak early fault signals. This monitoring process can effectively integrate multi-source heterogeneous sensor data, overcome the limitations of a single data source, and can perform adaptive feature extraction based on operating parameters to improve the ability to capture weak early fault features. It can also identify early abnormal signals that are difficult to detect by traditional methods by analyzing the deviation of the co-change of multi-dimensional features, effectively improving the accuracy and refinement of fault monitoring and providing strong support for predictive maintenance of gearboxes. Attached Figure Description

[0049] Figure 1 A flowchart of a speed reducer fault monitoring method provided in an embodiment of this application.

[0050] Figure 2 This is a schematic diagram of the structure of the speed reducer fault monitoring system provided in the embodiments of this application.

[0051] Figure reference numerals: 201, acquisition module; 202, preprocessing module; 203, feature extraction module; 204, deviation calculation module; 205, anomaly detection module; 206, anomaly localization module. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0053] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0054] Firstly, please refer to Figure 1This application provides a method for monitoring gearbox faults in some embodiments, used to identify early faults in the gearbox, characterized in that the method includes the following steps:

[0055] S1. Collect multi-source data during the operation of the reducer based on multi-source heterogeneous sensors, and simultaneously obtain the operating parameters of the reducer;

[0056] S2. Preprocess the multi-source data;

[0057] S3. Extract multi-dimensional feature vectors from the preprocessed multi-source data based on the operating parameters;

[0058] S4. Calculate the deviation between the coordinated change of the multidimensional feature vector and the reference baseline. The reference baseline is the distribution parameters of the coordinated change of the multidimensional features under the normal state, which are pre-constructed.

[0059] S5. Determine whether early abnormalities have occurred based on the degree of deviation;

[0060] S6. When early abnormalities occur, analyze the types of abnormal features that induce deviations in the multidimensional feature vector, and determine the abnormal type and / or abnormal location based on the types of abnormal features.

[0061] Specifically, multi-source heterogeneous sensors refer to various types or principles of sensors that collect data on the operating status of equipment. These can include vibration sensors, temperature sensors, acoustic sensors, current sensors, pressure sensors, speed sensors, etc. Their main purpose is to obtain comprehensive, multi-dimensional information reflecting the operating status of the reducer, overcoming the limitations of a single sensor data source. Operating parameters refer to parameters reflecting the current operating status of the reducer. These can be implemented using parameters such as speed, load torque, input power, output power, and operating time. Their main purpose is to characterize the specific working conditions of the equipment for state-based analysis.

[0062] More specifically, a multidimensional feature vector refers to a vector composed of multiple features of different types or dimensions extracted from preprocessed multi-source data. It can include time-domain features, frequency-domain features, time-frequency-domain features, statistical features, etc. Its main purpose is to transform the original sensor signals into numerical representations that better reflect the equipment status and potential fault information.

[0063] More specifically, extracting multidimensional feature vectors based on operating condition parameters refers to selecting or adjusting the feature extraction method or parameters according to the current operating state of the reducer. This can be achieved by selecting a suitable frequency band for vibration feature extraction at a specific speed, adjusting the calculation window of the temperature change rate according to the load, etc. The main purpose is to make the extracted features more sensitive to capture early fault information under the current operating conditions.

[0064] More specifically, the deviation of the co-variation from the reference baseline refers to the degree of difference between the joint variation pattern reflected by the current multidimensional feature vector and the pre-established pattern under normal conditions. It can be calculated by statistical distance (such as Mahalanobis distance), probability density ratio, anomaly score based on machine learning model, etc. It is mainly to capture the weak, non-linear co-variation between multidimensional features, which may be associated with early failures.

[0065] More specifically, a reference baseline refers to parameters or models that are constructed in advance by collecting a large amount of data under normal conditions to characterize the coordinated change pattern of multidimensional features. It can be implemented using the mean and covariance matrix of multidimensional feature vectors, multivariate probability distribution models (such as Gaussian mixture models), feature trajectory models under normal conditions, etc. Its main purpose is to provide a reference standard under normal conditions to quantify the degree of abnormality of the current state.

[0066] More specifically, abnormal features refer to characteristics that indicate anomalies. These can be identified using methods such as feature contribution analysis, sensitivity analysis, and feature importance ranking. The primary purpose is to pinpoint the source or manifestation of the abnormal signal, providing a basis for subsequent fault diagnosis. Determining the abnormal type and / or location refers to inferring the specific type of fault (e.g., gear pitting, bearing wear) or the specific location of the fault (e.g., primary gear, input bearing) that the gearbox may be experiencing based on the identified abnormal features or combinations of abnormal features. This can be achieved using rule-based reasoning, fault tree analysis, machine learning classifiers, and consulting fault knowledge bases. The main purpose is to provide specific diagnostic results, guiding maintenance personnel to conduct targeted inspections and repairs.

[0067] Specifically, the method of this application first collects multi-source data during the operation of the reducer using multi-source heterogeneous sensors, and simultaneously acquires operating parameters reflecting the current working state of the equipment. Then, step S2 preprocesses the collected multi-source data. Next, step S3 extracts multi-dimensional feature vectors reflecting the health status of the equipment from the preprocessed multi-source data based on the current operating parameters. The feature extraction strategy here is adjusted according to the operating conditions to better capture abnormal information under specific states. Then, step S4 calculates the deviation between the feature coordination change pattern reflected by the currently extracted multi-dimensional feature vector and a pre-constructed reference baseline for multi-dimensional feature coordination change under normal conditions. This reference baseline characterizes the joint distribution or coordination change pattern between multi-dimensional features under normal operating conditions. By calculating the deviation, the degree of difference between the current state and the normal state can be quantified. Step S5 compares the calculated deviation with a preset threshold to determine whether an early anomaly has occurred. If the deviation exceeds the threshold, an early anomaly is determined to exist. Upon determining the presence of an early anomaly, step S6 further analyzes which feature components in the multidimensional feature vector contribute most to the calculated deviation, thereby identifying the abnormal features that induce the deviation. Finally, based on the type or combination of the identified abnormal features, combined with pre-established fault knowledge or diagnostic rules, the specific type and / or location of the possible anomaly in the reducer is determined, completing the fault diagnosis. The entire process forms a closed loop, from data acquisition to final diagnosis, achieving comprehensive monitoring and diagnosis of early-stage faults in the reducer.

[0068] The method in this application combines multi-source heterogeneous data acquisition with multi-dimensional feature extraction based on operating parameters, and introduces a deviation calculation method based on the reference baseline of multi-dimensional feature co-change to detect early anomalies. This overcomes the problem that traditional methods are difficult to identify weak early fault signals. This monitoring process can effectively integrate multi-source heterogeneous sensor data, overcome the limitations of a single data source, and can perform adaptive feature extraction based on operating parameters to improve the ability to capture weak early fault features. Furthermore, by analyzing the deviation of the co-change of multi-dimensional features, it can identify early abnormal signals that are difficult to detect by traditional methods, effectively improving the accuracy and refinement of fault monitoring and providing strong support for predictive maintenance of reducers.

[0069] In some preferred embodiments, preprocessing includes time alignment and data cleaning processes performed sequentially.

[0070] Specifically, time alignment refers to adjusting data from different sensors, which may have different sampling rates or timestamps, to a unified time reference. This can be achieved using techniques such as resampling, interpolation, or synchronous acquisition. Its main purpose is to ensure the temporal synchronization of multi-source data, laying the foundation for subsequent collaborative analysis. Data cleaning refers to processing the raw acquired data to remove or correct noise, outliers, missing values, or erroneous data. This can be achieved using methods such as filtering, outlier detection and removal, and data interpolation. Its main purpose is to improve data quality and reliability, and reduce the impact of interference on fault identification.

[0071] In some preferred embodiments, step S3 includes:

[0072] S31. Determine the current operating status of the reducer based on the operating parameters;

[0073] S32. Determine the feature extraction strategy based on the current operating status;

[0074] S33. Extract multidimensional feature vectors from preprocessed multi-source data based on feature extraction strategies.

[0075] Specifically, the operating state refers to the stable or dynamic working mode of the reducer under a specific combination of operating parameters, such as high-speed heavy load, low-speed light load, no load, start-up, and shutdown. It can be classified and identified based on preset operating parameter thresholds or through data-driven methods such as cluster analysis.

[0076] More specifically, feature extraction strategies refer to a series of rules, algorithms, and parameter configurations followed in calculating or transforming a set of numerical features that can characterize the health status of a device from raw or preprocessed sensor data. Examples include time-domain statistics calculation methods, frequency-domain analysis parameters (such as FFT window type, length, and averaging method), time-frequency analysis methods (such as wavelet basis function selection and decomposition level), filter types and parameters, and extraction methods for specific signal components (such as harmonics and envelopes). These strategies can be pre-established with a strategy library corresponding to different states or dynamically generated based on the state.

[0077] Specifically, step S31 uses the collected operating parameters to identify the current operating state of the reducer. This is because the reducer generates vibration, temperature, acoustic, and current data with different baseline characteristics and noise levels under different operating states, and the manifestations and intensities of early faults also differ significantly in these data. Accurately identifying the current state provides crucial contextual information for subsequent feature extraction. Next, step S32 selects or generates a feature extraction strategy best suited for the identified current operating state. For example, for high-speed operating states, a high sampling rate and envelope analysis focusing on high-frequency vibration signals can be used; for low-speed operating states, a longer acquisition time and focus on the slow temperature change trend or specific low-frequency noise can be used. This dynamic adjustment of the strategy based on the state allows the feature extraction process to better adapt to the varying operating conditions of the reducer, avoiding the "under-capture" problem of weak fault features in certain states with a fixed strategy. Finally, step S33 applies the determined feature extraction strategy optimized for the current operating state to the preprocessed multi-source data to extract multi-dimensional feature vectors. These feature vectors are therefore able to more effectively reflect the true health status of the reducer under current operating conditions, improving the sensitivity and robustness of the features to early failures.

[0078] More specifically, through the above processing, the feature extraction process is no longer static, but can be dynamically adjusted according to the actual operation of the equipment, thereby more accurately capturing weak signals of early faults, improving the sensitivity and robustness of features to early faults, and providing more reliable input for subsequent fault judgment.

[0079] In some preferred embodiments, the multi-source data includes vibration data, temperature data, acoustic data, and current data;

[0080] The feature extraction strategy includes setting the frequency band range to be analyzed and filter parameters for vibration data; the feature extraction strategy also includes setting the calculation method for the temperature change rate for temperature data.

[0081] The feature extraction strategy also includes the calculation method for the frequency band range of the spectrum to be analyzed and the sound pressure level energy set for the acoustic data;

[0082] Feature extraction strategies also include methods for calculating the amplitude of current harmonics based on current data;

[0083] The multidimensional feature vectors include: gear meshing energy features and envelope impact features extracted from vibration data, temperature change rate features calculated from temperature data, high-frequency energy features calculated from acoustic data, and harmonic amplitude features calculated from current data.

[0084] Specifically, the method of this application, after collecting and preprocessing multi-source data during the operation of the reducer, determines a corresponding feature extraction strategy based on the current operating parameters. This strategy specifically defines the feature extraction methods and parameters to be used for different types of data (vibration, temperature, acoustics, and current). For example, for vibration data, the strategy specifies the frequency range to be focused on and the parameters used for filtering, in order to separate frequency components related to potential faults such as gear meshing and bearing impact from complex vibration signals. For temperature data, the strategy defines a specific algorithm for calculating the rate of temperature change, to capture dynamic temperature changes rather than just absolute values. For acoustic data, the strategy specifies the spectral range to be analyzed and the method for calculating sound pressure level energy, in order to extract energy features in areas prone to early fault signals, such as the high-frequency band. For current data, the strategy determines a method for calculating the current harmonic amplitude to obtain electrical characteristics reflecting the state of the motor or drivetrain. Based on these defined strategies, corresponding feature values ​​are extracted from the preprocessed multi-source data, and these feature values ​​are combined to form a multi-dimensional feature vector. This multidimensional feature vector includes gear meshing energy features and envelope impact features extracted from vibration data, temperature change rate features calculated from temperature data, high-frequency energy features calculated from acoustic data, and harmonic amplitude features calculated from current data. These features characterize the reducer's operating state from different dimensions, including mechanical, thermal, acoustic, and electrical aspects. Combining these multidimensional features provides a more comprehensive reflection of the reducer's state, overcoming the limitations of single features.

[0085] More specifically, through the above design, the method of this application can comprehensively capture the operating status information of the reducer from multiple physical dimensions. By adopting specific feature extraction strategies for different types of data, it can more effectively extract weak features related to early faults from complex signals. By combining these multi-dimensional features into vectors, it can comprehensively reflect the equipment status, improve the detection capability of early weak faults, and provide richer and more accurate information input for subsequent fault judgment, thereby improving the sensitivity and accuracy of reducer fault monitoring.

[0086] In some preferred embodiments, step S4 includes:

[0087] S41. Analyze the data quality of different data in multi-source data and obtain the quality assessment results;

[0088] S42. Determine the weights of each feature component in the multidimensional feature vector based on the quality assessment results;

[0089] S43. Based on the operating parameters, obtain the reference baseline corresponding to the current operating state;

[0090] S44. Calculate the deviation based on the multidimensional feature vector, the weights of each feature component, and the reference baseline.

[0091] Specifically, data quality refers to the reliability and effectiveness of the reducer operating status information reflected by multi-source data, which can be evaluated using indicators such as signal-to-noise ratio, completeness, consistency, and timeliness. The quality assessment result refers to the quantitative or qualitative description obtained after analyzing the quality of multi-source data, which can be expressed as a quality score, grade, or reliability coefficient for each data source or each feature component. The weight of a feature component refers to a coefficient assigned to each feature component in the multi-dimensional feature vector when calculating the deviation, used to adjust the contribution of that feature component to the total deviation. It can be represented as a vector equal to the number of feature components or a diagonal matrix.

[0092] Specifically, step S41 first performs quality analysis on the collected multi-source data to obtain quality assessment results reflecting the reliability of different data sources or feature components. Based on these quality assessment results, step S42 assigns a weight to each feature component in the multi-dimensional feature vector, assigning higher weights to feature components with high data quality and lower weights to feature components with low data quality. Simultaneously, step S43 obtains a reference baseline corresponding to the normal operating state under the current operating conditions of the reducer. Finally, when calculating the deviation between the current multi-dimensional feature vector and the reference baseline, step S44 uses the previously determined weights of each feature component for weighted calculation. In this way, feature components with higher data quality have a greater impact on the calculated deviation, while feature components with lower data quality have a smaller impact, effectively reducing the interference of low-quality data on the deviation calculation results. This scheme, as a key step in the entire fault monitoring method, can more accurately reflect the degree of deviation between the actual operating state of the reducer and the normal baseline, providing a more reliable basis for subsequent judgment of whether early anomalies have occurred. By incorporating data quality into the deviation calculation, the method in this application improves the accuracy and reliability of early anomaly detection and solves the problem caused by differences in the quality of multi-source data.

[0093] As a preferred implementation, the solution of this application is specifically implemented as follows: First, quality analysis is performed on the collected multi-source data such as vibration, temperature, acoustics, and current. For example, it can be checked whether the vibration data is discontinuous, saturated, or has strong interference; whether the temperature data exceeds a reasonable range or changes abnormally; whether the acoustic data has sudden noise; and whether the current data is stable. Based on these analyses, a quality score is calculated for each data source or feature component extracted from it (such as gear meshing energy, temperature change rate, high-frequency energy, current harmonic amplitude, etc.). For example, the score ranges from 0 to 1, with a higher score indicating better quality. Then, based on these quality scores, a linear mapping or nonlinear function can be used to convert the quality scores into weights, thereby determining the weight of each feature component. This forms a weight vector with the same dimensions as the multi-dimensional feature vector.

[0094] In some preferred embodiments, the reference baseline includes the mean vector and covariance matrix of the multidimensional eigenvectors under normal conditions, and step S44 includes:

[0095] S441. Based on the multidimensional eigenvector, the weights of each eigencomponent, the mean vector, and the covariance matrix, calculate the weighted Mahalanobis distance between the multidimensional eigenvector and the mean vector, which is used as the deviation.

[0096] Specifically, the mean vector refers to the average value of multiple multidimensional feature vectors collected under normal operating conditions, representing the central position of the multidimensional feature vectors under normal conditions; the covariance matrix refers to the covariance matrix of multiple multidimensional feature vectors collected under normal operating conditions, describing the interrelationships and range of variation among the components of the multidimensional feature vectors under normal conditions; the weighted Mahalanobis distance is a distance metric that considers data covariance and feature weights, used to measure the degree of deviation of a point (the current multidimensional feature vector) from a distribution (the normal state distribution described by the mean vector and covariance matrix), while also considering the importance and reliability of different feature components.

[0097] Specifically, in calculating the deviation, step S441 uses weighted Mahalanobis distance. This distance metric not only considers the distance between the current multidimensional feature vector and the normal state center (mean vector), but also utilizes the covariance matrix to reflect the correlation between features, and combines the weights of each feature component (which can reflect data quality or importance) to adjust the contribution of different features. This calculation method can effectively measure the degree of deviation of the overall distribution of the current multidimensional feature vector from the normal state distribution, especially when there are complex correlations between multidimensional features and inconsistent data quality, providing a more accurate deviation index. Step S441, combined with the steps of analyzing the quality of multi-source data and determining the weights of feature components, allows the deviation calculation to simultaneously consider the intrinsic relationships between features and the influence of external data quality, thereby improving the accuracy of early anomaly identification.

[0098] In some preferred embodiments, step S5 includes:

[0099] S51. Obtain the first threshold that matches the current operating state based on the operating condition parameters;

[0100] S52. Obtain the threshold compensation value based on the quality assessment results;

[0101] S53. Compensate the first threshold based on the threshold compensation value to obtain the second threshold;

[0102] S54. Compare the deviation and the second threshold to determine if an early abnormality has occurred.

[0103] Specifically, the first threshold matching the current operating state refers to the upper limit of the typical range of deviation of normal operating data from the reference baseline under a specific operating state. It can be obtained by consulting a pre-established threshold table or through statistical analysis of historical normal data. The threshold compensation value is a value calculated based on the quality assessment results to adjust the first threshold. It can be achieved by mapping the quality assessment results to a compensation value using a preset function; for example, the worse the data quality, the larger the compensation value. Compensating the first threshold based on the threshold compensation value means applying the threshold compensation value to the first threshold to obtain a new judgment threshold. This can be achieved using addition or multiplication operations.

[0104] Specifically, in the step of determining whether an early anomaly has occurred, the method of this application obtains a basic judgment threshold, namely a first threshold, based on the current operating parameters or running state. Simultaneously, using the quality assessment results obtained in the previous steps, a threshold compensation value is calculated, which reflects the reliability of the current multi-source data. Then, this threshold compensation value is applied to the first threshold to obtain a dynamically adjusted second threshold. Finally, the calculated deviation is compared with this second threshold, which considers data reliability. If the deviation exceeds the second threshold, an early anomaly is determined to have occurred. In this way, the scheme uses data quality assessment information not only for weighted calculation of the deviation but also for dynamically adjusting the anomaly judgment threshold, enabling the anomaly judgment process to adapt to fluctuations in data quality. This reduces false alarms and false negatives caused by data quality fluctuations, thereby improving the accuracy and robustness of early anomaly judgment for the reducer.

[0105] It should be noted that in some other implementations, in order to avoid the effects of the threshold compensation value and the deviation calculated based on the weight canceling each other out or being excessively amplified, step S5 can determine whether an early anomaly has occurred by comparing the deviation and the first threshold.

[0106] In some preferred embodiments, step S6 includes:

[0107] S61. When determining the occurrence of early anomalies, analyze the contribution of each feature component in the multidimensional feature vector to the degree of deviation.

[0108] S62. Based on the degree of contribution, identify several feature components that significantly contribute to the deviation as anomalous features;

[0109] S63. Based on the abnormal characteristics, determine the abnormal type and / or the abnormal location.

[0110] Specifically, in step S61, analyzing the contribution of each feature component in the multidimensional feature vector to the deviation refers to quantifying the influence of each individual feature dimension (i.e., feature component) in the multidimensional feature vector on the calculated total deviation value. This can be achieved using various mathematical or statistical methods, such as distance-based decomposition, sensitivity analysis, or statistical modeling.

[0111] More specifically, step S62, identifying several feature components that significantly contribute to the deviation, refers to selecting those feature components that have a greater impact on the total deviation based on the calculated degree of contribution. This can be achieved by setting a threshold, performing sorting and selection, or using statistical significance testing. Abnormal features refer to those feature components that are identified as significantly contributing to the total deviation during the identification process.

[0112] More specifically, in step S63, determining the anomaly type and / or anomaly location refers to inferring the specific fault mode (anomaly type) and the specific location (anomaly location) that caused the anomaly based on the identified anomaly features. This can be achieved using methods such as pattern matching, rule reasoning, classification algorithms, or consulting a pre-established knowledge base.

[0113] Specifically, steps S61-S63 work by first analyzing the contribution of each feature component in the multidimensional feature vector to the currently calculated total deviation. This analysis quantifies the degree of abnormality of each feature and its impact on the overall abnormal state. Then, based on established criteria, features with higher contributions are selected from all feature components and identified as key abnormal features leading to early anomalies. Finally, using these identified abnormal features, combined with existing fault knowledge or diagnostic models, the specific type of anomaly in the reducer and / or the location of the anomaly is inferred. This entire process forms a complete chain from early anomaly detection to anomaly feature localization and fault type and location diagnosis, enabling rapid and accurate problem localization after detecting weak early anomaly signals, providing a basis for subsequent maintenance decisions.

[0114] In some preferred embodiments, step S61 includes:

[0115] S611. When an early anomaly is detected, obtain the mean vector and covariance matrix of the reference baseline.

[0116] S612. Calculate the difference vector between the multidimensional feature vector and the mean vector;

[0117] S613. Based on the covariance matrix, calculate and obtain the inverse covariance matrix, and based on the inverse covariance matrix and the difference vector, calculate the contribution vector of each feature component in the multidimensional feature vector to the deviation.

[0118] S614. Determine the degree of contribution of each feature component in the multidimensional feature vector to the deviation based on the contribution vector.

[0119] Specifically, step S61 can further incorporate the weights of the feature components obtained from the previous analysis to optimize the difference vector. Step S61 can then be transformed into including:

[0120] S611′ When an early anomaly is detected, obtain the mean vector and covariance matrix of the reference baseline, as well as the weights of the feature components.

[0121] S612' Calculate the difference vector between the multidimensional feature vector and the mean vector, and use the weights of the feature components to weight the difference vector;

[0122] S613' Calculate the inverse covariance matrix based on the covariance matrix, and calculate the contribution vector of each feature component in the multidimensional feature vector to the deviation based on the inverse covariance matrix and the weighted difference vector.

[0123] S614' Determine the degree of contribution of each feature component in the multidimensional feature vector to the deviation based on the contribution vector.

[0124] More specifically, the inverse covariance matrix is ​​the inverse of the covariance matrix, used for standardization and decorrelation of data in a multidimensional space. The contribution vector is a vector whose each component quantifies the magnitude of the influence of the corresponding feature component in the multidimensional feature vector on the overall deviation.

[0125] Specifically, the working principle of the above-mentioned contribution level acquisition process is as follows: First, upon detecting an early anomaly, the mean vector and covariance matrix, which describe the statistical characteristics of the normal state, are acquired, along with weights reflecting the reliability or importance of each feature. Next, the difference between the current multidimensional feature vector and the normal mean vector is calculated, and these differences are adjusted using weights to consider the data quality or evaluation emphasis of different features. Then, the inverse matrix of the covariance matrix is ​​calculated, containing correlation information between features, used to eliminate the influence of linear dependencies between features when calculating contributions. Finally, combining the inverse covariance matrix and the weighted difference vector, the overall deviation is decomposed into each feature dimension using a principle similar to Mahalanobis distance decomposition, resulting in a contribution vector. Each component value of this vector represents the contribution of the corresponding feature component to the overall deviation. Through this series of steps, the method of this application can quantitatively reveal which feature component changes mainly caused the overall deviation, thereby achieving accurate identification of anomalous features. This method utilizes the reference baseline parameters (mean, covariance) and weights already used in the deviation calculation, and performs decomposition calculations based on these parameters. This ensures that the identification of abnormal features and the detection of abnormalities are consistent in terms of data and principles, improving the accuracy of diagnosis. It enables the system to accurately identify the key features that lead to early abnormalities, providing a reliable quantitative basis for subsequent diagnosis of fault types and locations.

[0126] In some preferred embodiments, step S62 includes:

[0127] S621. Obtain feature components whose contribution level exceeds the preset contribution threshold as abnormal features.

[0128] Specifically, the preset contribution threshold refers to a pre-defined numerical limit used to screen for abnormal features, which can be determined based on historical operational data analysis, expert experience, or statistical methods.

[0129] More specifically, after identifying early anomalies and analyzing the contribution of each feature component in the multidimensional feature vector to the deviation, the method in this application does not rely on subjective or vague standards to determine which features contribute "significantly." Instead, it compares the contribution of each feature component with a pre-set quantification threshold. This threshold-based screening mechanism provides a clear, repeatable, and easily automated standard for identifying anomalous features. In this way, the scheme can accurately locate the key features that have the greatest impact on the overall deviation from numerous feature components, thus providing a focused and reliable basis for subsequently determining the anomaly type or location. This objective identification method, combined with the previous contribution calculation step, forms a complete and automated anomaly feature identification process, avoiding the uncertainty caused by subjective judgment, improving the accuracy and efficiency of anomaly feature identification, and thus providing a more reliable foundation for subsequent fault type and location diagnosis.

[0130] In some preferred embodiments, step S63 includes:

[0131] S631. Based on the abnormal characteristics, search the pre-established fault knowledge base to determine the abnormal type and / or abnormal location. The fault knowledge base includes different abnormal characteristics or combinations of abnormal characteristics and their matching relationships with the abnormal type and / or abnormal location.

[0132] Specifically, a fault knowledge base refers to a collection of knowledge related to equipment fault diagnosis, which can be implemented using a database, lookup table, rule base, or ontology model. Matching relationships refer to the rules, mappings, or association information stored in the fault knowledge base used to associate specific abnormal features or combinations of abnormal features with corresponding abnormal types and / or abnormal locations. These can be implemented using condition-result rules, lookup table entries, or probabilistic associations. Abnormal types and / or abnormal locations refer to the specific fault categories that the equipment may experience (e.g., gear wear, bearing spalling) and the specific location where the fault occurs (e.g., input shaft, output shaft, a specific gear), which is the target output of the diagnostic process.

[0133] Specifically, the method of this application, after determining that an early abnormality has occurred in the reducer and identifying the abnormal features that induce deviation, utilizes a pre-built fault knowledge base for diagnosis. This knowledge base stores a large number of known abnormal features or combinations of abnormal features and their correspondences with specific abnormal types and / or abnormal locations. By comparing the currently detected abnormal features with the information in the knowledge base, the system can find the fault type and location that best matches the current abnormal features. For example, if the identified abnormal features are increased high-frequency vibration energy and enhanced envelope impact characteristics, the system will search for entries corresponding to these features in the knowledge base, thereby determining that the possible fault is bearing outer ring spalling, occurring at the input bearing location. This knowledge base-based search method systematically maps complex abnormal feature patterns to specific fault causes and locations, overcoming the limitations of relying solely on experience or simple rules for judgment. By combining this with the specific abnormal features identified in previous steps, this scheme can more accurately locate and classify early faults, improving the reliability of diagnosis.

[0134] Through the above processing, the method of this application provides a structured and systematic early fault diagnosis method, which can effectively map complex abnormal features extracted from multi-source data to specific fault types and locations, improve the accuracy and reliability of early fault diagnosis of reducers, and help to take timely maintenance measures to avoid the expansion of faults.

[0135] Secondly, please refer to Figure 2 Some embodiments of this application also provide a speed reducer fault monitoring system for identifying early speed reducer faults, characterized in that the system includes:

[0136] The acquisition module 201 is used to collect multi-source data during the operation of the reducer based on multi-source heterogeneous sensors, and at the same time acquire the operating parameters of the reducer.

[0137] Preprocessing module 202 is used to preprocess the multi-source data;

[0138] Feature extraction module 203 is used to extract multi-dimensional feature vectors from preprocessed multi-source data based on the operating condition parameters;

[0139] The deviation calculation module 204 is used to calculate the deviation between the coordinated change of the multidimensional feature vector and the reference baseline, wherein the reference baseline is a pre-constructed distribution parameter of the coordinated change of multidimensional features under normal conditions;

[0140] Anomaly detection module 205 is used to determine whether an early anomaly has occurred based on the deviation.

[0141] The anomaly localization module 206 is used to analyze the types of abnormal features that induce deviation in the multidimensional feature vector when an early anomaly occurs, and determine the anomaly type and / or anomaly location based on the types of abnormal features.

[0142] The system in this application combines multi-source heterogeneous data acquisition with multi-dimensional feature extraction based on operating parameters, and introduces a deviation calculation method based on the reference baseline of multi-dimensional feature co-change to detect early anomalies. This overcomes the problem that traditional systems have difficulty identifying weak early fault signals. The monitoring process can effectively integrate multi-source heterogeneous sensor data, overcome the limitations of a single data source, and can perform adaptive feature extraction based on operating parameters to improve the ability to capture weak early fault features. Furthermore, by analyzing the deviation of the co-change of multi-dimensional features, it can identify early abnormal signals that are difficult for traditional systems to detect, effectively improving the accuracy and refinement of fault monitoring and providing strong support for predictive maintenance of reducers.

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

[0144] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0145] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0146] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring gearbox faults, used to identify early-stage faults in a gearbox, characterized in that, The method includes the following steps: S1. Collect multi-source data during the operation of the reducer based on multi-source heterogeneous sensors, and simultaneously obtain the operating parameters of the reducer; S2. Preprocess the multi-source data; S3. Extract multidimensional feature vectors from the preprocessed multi-source data according to the operating condition parameters; S4. Calculate the deviation between the coordinated change of the multidimensional feature vectors and the reference baseline, where the reference baseline is a pre-constructed distribution parameter of the coordinated change of multidimensional features under normal conditions; S5. Determine whether an early anomaly has occurred based on the deviation. S6. When an early abnormality occurs, analyze the types of abnormal features that induce deviation in the multidimensional feature vector, and determine the abnormal type and / or abnormal location based on the types of abnormal features.

2. The method for monitoring gearbox faults according to claim 1, characterized in that, Preprocessing includes time alignment and data cleaning, which are performed sequentially.

3. The method for monitoring gearbox faults according to claim 1, characterized in that, Step S3 includes: S31. Determine the current operating status of the reducer based on the operating parameters; S32. Determine the feature extraction strategy based on the current operating status; S33. Extract multidimensional feature vectors from the preprocessed multi-source data based on the feature extraction strategy.

4. The method for monitoring gearbox faults according to claim 1, characterized in that, Step S4 includes: S41. Analyze the data quality of different data in the multi-source data and obtain the quality assessment results; S42. Determine the weight of each feature component in the multidimensional feature vector based on the quality assessment results; S43. Based on the operating parameters, obtain the reference baseline corresponding to the current operating state; S44. Calculate the deviation based on the multidimensional feature vector, the weights of each feature component, and the reference baseline.

5. The method for monitoring gearbox faults according to claim 4, characterized in that, The reference baseline includes the mean vector and covariance matrix of the multidimensional feature vector under normal conditions. Step S44 includes: S441. Calculate the weighted Mahalanobis distance between the multidimensional feature vector and the mean vector based on the multidimensional feature vector, the weights of each feature component, the mean vector, and the covariance matrix, and use this distance as the deviation.

6. The method for monitoring gearbox faults according to claim 4, characterized in that, Step S5 includes: S 51. Obtain a first threshold that matches the current operating state based on the operating parameters; S 52. Obtain a threshold compensation value based on the quality assessment result; S53. Compensate the first threshold based on the threshold compensation value to obtain the second threshold; S54. Compare the deviation and the second threshold to determine whether an early abnormality has occurred.

7. The method for monitoring gearbox faults according to claim 1, characterized in that, Step S6 includes: S61. When an early anomaly is determined to occur, analyze the contribution of each feature component in the multidimensional feature vector to the deviation. S62. Based on the degree of contribution, identify several feature components that significantly contribute to the deviation as the abnormal features; S 63. Based on the abnormal characteristics, determine the abnormal type and / or the abnormal location.

8. The method for monitoring gearbox faults according to claim 7, characterized in that, Step S62 includes: S621. Obtain the feature components whose contribution level exceeds the preset contribution threshold, and use them as the abnormal features.

9. The method for monitoring gearbox faults according to claim 7, characterized in that, Step S63 includes: S631. Based on the abnormal features, search a pre-established fault knowledge base to determine the abnormal type and / or the abnormal location. The fault knowledge base includes different abnormal features or combinations of abnormal features and their matching relationships with the abnormal type and / or the abnormal location.

10. A speed reducer fault monitoring system for identifying early-stage speed reducer faults, characterized in that, The system includes: The acquisition module is used to collect multi-source data during the operation of the reducer based on multi-source heterogeneous sensors, and at the same time acquire the operating parameters of the reducer; The preprocessing module is used to preprocess the multi-source data; The feature extraction module is used to extract multi-dimensional feature vectors from the preprocessed multi-source data based on the operating condition parameters. The deviation calculation module is used to calculate the deviation between the coordinated change of the multidimensional feature vector and the reference baseline, wherein the reference baseline is a pre-constructed distribution parameter of the coordinated change of multidimensional features under normal conditions. An anomaly detection module is used to determine whether an early anomaly has occurred based on the deviation degree; an anomaly location module is used to analyze the types of abnormal features that induce deviation in the multidimensional feature vector when an early anomaly occurs, and determine the anomaly type and / or anomaly location based on the types of abnormal features.

Images