Bearing noise processing method for bearing fault diagnosis

Abstract

The invention discloses a bearing noise processing method for bearing fault diagnosis, and relates to the technical field of data processing, and the method comprises the steps: obtaining a target model of a target bearing, carrying out the retrieval in a historical fault database based on the target model, and determining a historical bearing fault record set; determining a plurality of differentiated historical bearing fault record sets; obtaining a plurality of noise feature set values and a plurality of noise feature associated bandwidths; obtaining a plurality of noise monitoring feature sets; and performing cross-scale interactive fusion to obtain a target interactive fusion noise monitoring feature set, and taking the target interactive fusion noise monitoring feature set as a bearing noise processing result. According to the method, the technical problem of low bearing fault diagnosis reliability caused by insufficient bearing noise analysis depth in the prior art is solved, and the technical effects of deeply analyzing the bearing noise data, accurately obtaining the bearing noise data and improving the bearing fault diagnosis accuracy are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a bearing noise processing method for bearing fault diagnosis. Background Art

[0002] Bearings are key components in many mechanical equipment, and their fault diagnosis is crucial to the stable operation of the equipment. Traditional bearing fault diagnosis methods mainly analyze noise data directly, but noise data is often interfered by the external environment, making it difficult to accurately extract fault features. In recent years, bearing fault diagnosis methods based on neural network models have gradually been applied. Through feature analysis of noise data, bearing faults can be detected earlier and more accurately. However, traditional noise processing methods usually use the same receptive field for feature extraction, ignoring the impact of different fault types on noise signal characteristics, resulting in low fault diagnosis accuracy. Traditional bearing noise processing methods use a single-scale feature extraction method, which fails to fully consider the differences in noise characteristics corresponding to different fault types, and easily causes the generation of redundant data and information loss.

[0003] The existing technology has the technical problem that the bearing noise analysis is not deep enough, resulting in low reliability of bearing fault diagnosis. Summary of the invention

[0004] The present application provides a bearing noise processing method for bearing fault diagnosis, which is used to solve the technical problem in the prior art that the bearing noise analysis is not deep enough, resulting in low reliability of bearing fault diagnosis.

[0005] In view of the above problems, the present application provides a bearing noise processing method for bearing fault diagnosis, the method comprising: Obtaining a target model of a target bearing, searching a historical fault database based on the target model, and determining a set of historical bearing fault records; Performing heterogeneous differentiation on the historical bearing fault record set to determine a plurality of differentiated historical bearing fault record sets; Traversing the plurality of distinguished historical bearing fault record sets to perform concentrated noise feature extraction, and obtaining a plurality of noise feature concentrated values ​​and a plurality of noise feature associated bandwidths; The multiple noise feature association bandwidths are used as multiple receptive fields of multiple noise processing network layers, and the configured multiple noise processing network layers are used to perform multi-scale feature analysis on the noise monitoring data sequence of the target bearing within a preset monitoring window to obtain multiple noise monitoring feature sets; The multiple noise monitoring feature sets are interactively fused across scales to obtain a target interactive fusion noise monitoring feature set, and the target interactive fusion noise monitoring feature set is used as a bearing noise processing result.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application obtains the target model of the target bearing, searches the historical fault database based on the target model, determines the historical bearing fault record set, and then distinguishes the historical bearing fault record set to determine multiple distinguished historical bearing fault record sets; traverses multiple distinguished historical bearing fault record sets to extract noise features, obtain multiple noise feature concentration values ​​and multiple noise feature associated bandwidths, and then uses the multiple noise feature associated bandwidths as multiple receptive fields of multiple noise processing network layers, and uses the configured multiple noise processing network layers to perform multi-scale feature analysis on the noise monitoring data sequence of the target bearing within the preset monitoring window to obtain multiple noise monitoring feature sets, and then performs cross-scale interactive fusion on the multiple noise monitoring feature sets to obtain the target interactive fusion noise monitoring feature set, and uses the target interactive fusion noise monitoring feature set as the bearing noise processing result. The technical effect of ensuring the reliability of the bearing fault diagnosis results and improving the depth and accuracy of the bearing noise data analysis is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A flow chart of a bearing noise processing method for bearing fault diagnosis provided in an embodiment of the present application; Figure 2 A schematic diagram of a flow chart of determining multiple noise feature concentration values ​​in a bearing noise processing method for bearing fault diagnosis provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] The present application provides a bearing noise processing method for bearing fault diagnosis, which is used to solve the technical problem in the prior art that the bearing noise analysis is not deep enough, resulting in low reliability of bearing fault diagnosis.

[0010] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0011] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0012] Examples, such as Figure 1 As shown, the present application provides a bearing noise processing method for bearing fault diagnosis, wherein the method comprises: S100: Obtain a target model of a target bearing, and search a historical fault database based on the target model to determine a historical bearing fault record set; In a possible embodiment, the target bearing refers to a bearing that needs to be diagnosed for a fault, usually a bearing used in a specific device. The target model refers to the specific model information of the target bearing, including the specifications, type, design parameters, etc. of the bearing. This information is important for retrieving historical fault data, because different models of bearings may have different fault modes and fault characteristics. The historical fault database is a data set containing a variety of bearing models and their historical fault records. The data in the database usually includes information such as bearing fault type, occurrence time, fault mode, noise data, etc.

[0013] Then, the target model is used as an index to search in the historical fault database, and the historical bearing fault record set is obtained according to the search results. The historical bearing fault record set refers to the fault records related to the bearing of this model found in the historical fault database. The set contains various fault instances that have occurred in the past for the bearing of this model. By obtaining the historical bearing fault record set, the technical effect of providing data support for the subsequent in-depth analysis of bearing noise data is achieved.

[0014] S200: performing heterogeneous differentiation on the historical bearing fault record set to determine a plurality of differentiated historical bearing fault record sets; Further, the historical bearing fault record set is heterogeneously distinguished to determine a plurality of distinguished historical bearing fault record sets. Step S200 of the embodiment of the present application further includes: Randomly extracting a plurality of historical bearing fault records from the historical bearing fault record set; Enumerate the multiple historical bearing fault records in pairs to obtain multiple enumeration combinations; Determine whether there is an enumeration combination whose record similarity among the multiple enumeration combinations is greater than a preset record similarity threshold, and if not, use the multiple historical bearing fault records as multiple differentiation targets; Based on the multiple differentiation targets, the historical bearing fault record set is heterogeneously differentiated according to a preset record similarity threshold to obtain the multiple differentiated historical bearing fault record sets, wherein each differentiated historical bearing fault record set corresponds to a differentiation target.

[0015] In one embodiment, by classifying different fault data in the historical bearing fault record set, fault modes with significant differences are identified, and record sets belonging to different fault types are distinguished, thereby achieving heterogeneous distinction of the historical bearing fault record set. This process helps to ensure that data of specific fault types can be focused on separately during in-depth analysis of noise data.

[0016] The plurality of differentiated historical bearing fault record sets are obtained by performing heterogeneous differentiation on the historical bearing fault record set, wherein each differentiated historical bearing fault record set refers to a record set obtained after heterogeneous differentiation, and each set contains historical bearing fault records with similar fault characteristics, representing bearing records of a fault type.

[0017] Optionally, multiple historical bearing fault records are randomly selected from the historical fault record set, and multiple enumeration combinations are formed by enumerating them in pairs. Then, the similarity between each pair of enumeration combinations is calculated to obtain the record similarity. The record similarity is a metric used to measure the similarity between two historical fault record combinations, which can be calculated using the cosine similarity calculation formula.

[0018] If the similarity of a pair of combinations exceeds the preset record similarity threshold (a value pre-set by a person skilled in the art to determine whether two historical fault record combinations are similar enough to be classified into the same category), it is considered that these records represent the same type of fault. At this time, the randomly selected multiple historical bearing fault records cannot be used as the distinction target, and multiple historical bearing fault records need to be randomly selected again. If it does not exist, the randomly selected multiple historical bearing fault records belong to different types of faults, and then these records are used as multiple distinction targets to distinguish the historical bearing fault record set.

[0019] Next, based on these differentiation targets, all historical fault records are further heterogeneously differentiated according to a preset record similarity threshold, that is, the historical bearing fault records in the historical bearing fault record set are respectively compared with the multiple differentiation targets, and the record similarity calculation formula is used to calculate the record similarity, and then the historical bearing fault records are added to the set corresponding to the differentiation target whose record similarity calculation result is greater than or equal to the preset record similarity threshold, so as to obtain the multiple differentiated historical bearing fault record sets. Optionally, when there is any historical bearing fault record whose record similarity calculation result with more than two differentiation targets is greater than or equal to the preset record similarity threshold, it is added to the set of differentiation targets corresponding to the maximum value of the record similarity calculation result. Each differentiated historical bearing fault record set in the multiple differentiated historical bearing fault record sets represents a different fault type or fault mode.

[0020] By distinguishing the heterogeneity of historical fault records, the interference of noise data and redundant information can be effectively reduced, providing more accurate training data for subsequent noise feature analysis and fault diagnosis. Heterogeneous distinction helps ensure that the data used in the analysis of specific fault types is more consistent, improving the accuracy and reliability of diagnosis.

[0021] S300: traversing the plurality of distinguished historical bearing fault record sets to perform concentrated noise feature extraction, and obtaining a plurality of concentrated noise feature values ​​and a plurality of noise feature associated bandwidths; Further, the plurality of distinguished historical bearing fault record sets are traversed to perform concentrated noise feature extraction, and a plurality of noise feature concentration values ​​and a plurality of noise feature associated bandwidths are obtained. Step S300 of the embodiment of the present application further includes: Traversing the plurality of distinguishing historical bearing fault record sets to extract noise monitoring data, and obtaining a plurality of distinguishing historical bearing noise monitoring data sequence sets; Extracting instantaneous noise monitoring data from the plurality of distinguished historical bearing noise monitoring data series respectively, and determining a plurality of historical instantaneous noise monitoring data sets, wherein each historical instantaneous noise monitoring data is monitoring data at a time when the maximum data change fluctuation occurs in each distinguished historical bearing noise monitoring data series; Extracting features from the multiple historical instantaneous noise monitoring data sets to obtain multiple historical instantaneous noise feature sets, and performing centralized analysis on the multiple historical instantaneous noise feature sets to determine multiple noise feature centralized values; Using the multiple historical instantaneous noise feature sets as indexes, feature-associated bandwidth diffusion identification is performed on the multiple distinguishing historical bearing noise monitoring data sequence sets to determine multiple historical instantaneous noise feature-associated bandwidth sets, wherein each historical instantaneous noise feature-associated bandwidth reflects the duration of data associated with the historical instantaneous noise feature at the time of fault occurrence in a distinguishing historical bearing noise monitoring data sequence; The mean values ​​of the plurality of historical instantaneous noise feature associated bandwidth sets are respectively calculated to obtain the plurality of noise feature associated bandwidths.

[0022] In an embodiment of the present application, noise feature concentration extraction is performed on each of the multiple sets of distinguishing historical bearing fault records, and the most representative noise feature concentration value is determined to obtain the multiple noise feature concentration values. Feature association analysis is performed based on the data in the multiple sets of distinguishing historical bearing fault records to determine the multiple noise feature association bandwidths. The noise feature association bandwidth is the time period of noise feature association when a bearing fault occurs.

[0023] Optionally, noise monitoring data is extracted from multiple sets of distinguished historical bearing fault record records, thereby obtaining multiple sets of distinguished historical bearing noise monitoring data series, which record the noise changes of the bearings in different states in the form of time series. Then, monitoring data analysis is performed on each noise monitoring data series to obtain data with large instantaneous changes in the data at these historical time points, thereby obtaining multiple sets of historical instantaneous noise monitoring data. Among them, each historical instantaneous noise monitoring data is the monitoring data at the moment when the maximum data change fluctuation occurs in each distinguished historical bearing noise monitoring data series.

[0024] Feature extraction is performed on multiple historical instantaneous noise monitoring data sets, and the noise data is converted into a set of representative values ​​to capture the main change characteristics of the noise data (including time domain characteristics, frequency domain characteristics, instantaneous amplitude, instantaneous phase and instantaneous frequency, etc.), so as to obtain the multiple historical instantaneous noise feature sets. The multiple noise feature concentration values ​​are obtained by performing centralized analysis on the multiple historical instantaneous noise feature sets and analyzing the representative feature concentration values ​​of each historical instantaneous noise feature set. The multiple noise feature concentration values ​​represent the main characteristics of the noise corresponding to different fault types.

[0025] On this basis, the historical instantaneous noise feature set is used as an index to further identify the feature-associated bandwidth diffusion of multiple sets of distinguishing historical bearing noise monitoring data sequences. The goal of this process is to identify the diffusion range of the noise signal in different frequency bands, that is, to determine the noise feature-associated bandwidth and obtain the multiple sets of historical instantaneous noise feature-associated bandwidths. Among them, each historical instantaneous noise feature-associated bandwidth reflects the duration of data associated with the noise feature at the time of fault occurrence in a distinguishing historical bearing noise monitoring data sequence.

[0026] Then, the mean of each historical instantaneous noise feature associated bandwidth set is calculated, and finally multiple noise feature associated bandwidths are obtained. These bandwidth values ​​can reflect the noise spectrum characteristics of the bearing under different working conditions, and thus provide valuable data support for the processing of bearing noise data.

[0027] This process is crucial to the entire bearing fault diagnosis system because by analyzing the noise feature concentration value and the noise feature correlation bandwidth, the system can understand the noise characteristics of the bearing under different fault conditions from multiple dimensions, help accurately distinguish different fault types, and thus provide a more reliable basis for subsequent diagnosis.

[0028] Further, such as Figure 2 As shown, the instantaneous feature extraction is performed on the multiple historical instantaneous noise monitoring data sets to obtain multiple historical instantaneous noise feature sets, and the multiple historical instantaneous noise feature sets are centrally analyzed to determine multiple noise feature concentration values. Step S300 of the embodiment of the present application also includes: Using a noise feature extraction network layer to extract instantaneous features from the plurality of historical instantaneous noise monitoring data sets, to obtain the plurality of historical instantaneous noise feature sets; Traversing multiple historical instantaneous noise feature sets to perform feature mean calculations and determine multiple historical instantaneous noise feature means; Iterating the mean values ​​of the multiple historical instantaneous noise features in the multiple historical instantaneous noise feature sets according to a preset iteration bandwidth to obtain multiple iterative historical instantaneous noise features; When the aggregation amount of the multiple iterative historical instantaneous noise features is less than or equal to the aggregation amount of the multiple historical instantaneous noise feature mean values, the multiple historical instantaneous noise feature mean values ​​are used as the centralized values ​​of the multiple noise features.

[0029] Furthermore, step S300 of the embodiment of the present application further includes: When the aggregation amount of the multiple iterative historical instantaneous noise features is greater than the aggregation amount of the multiple historical instantaneous noise feature means, determine whether the aggregation amount difference between the multiple iterative historical instantaneous noise features and the multiple historical instantaneous noise feature means is greater than or equal to a preset aggregation amount difference threshold, and if so, continue to iterate based on the multiple iterative historical instantaneous noise features until the maximum number of iterations is met, and use the multiple iterative historical instantaneous noise features obtained in the last iteration as the multiple noise feature concentration values; If not, the iteration is stopped, and the multiple iterative historical instantaneous noise features are used as the centralized values ​​of the multiple noise features.

[0030] In a possible embodiment, a plurality of sample noise monitoring data and a plurality of sample noise features are obtained as training data, and the framework constructed based on the feedforward neural network is supervised and trained using the training data to learn a one-to-one mapping relationship between the noise monitoring data and the noise features until the training meets the maximum number of training times (e.g., 500 times, 600 times, etc.), thereby obtaining the noise feature extraction network layer that has completed the training.

[0031] Optionally, the noise feature extraction network layer is used to perform instantaneous feature analysis on the multiple historical instantaneous noise monitoring data sets to obtain the multiple historical instantaneous noise feature sets, wherein each historical instantaneous noise feature reflects the instantaneous information of the corresponding historical instantaneous noise monitoring data change, including instantaneous frequency, instantaneous amplitude, instantaneous phase and other features.

[0032] Then, characteristic mean values ​​are calculated for the multiple historical instantaneous noise feature sets respectively to determine multiple historical instantaneous noise characteristic mean values, wherein the multiple historical instantaneous noise characteristic mean values ​​reflect the average characteristic conditions of the multiple historical instantaneous noise feature sets in consideration of accidental values ​​and error values.

[0033] The mean values ​​of the multiple historical instantaneous noise features are iterated in the multiple historical instantaneous noise feature sets according to the preset iteration bandwidth, that is, the multiple iterative historical instantaneous noise features are obtained according to the moving amplitude set by the preset iteration bandwidth (the moving amplitude in a single iteration pre-set by a person skilled in the art, that is, the historical instantaneous noise feature difference in a single movement).

[0034] In one embodiment, multiple central areas are constructed with the mean values ​​of the multiple historical instantaneous noise features as the center and the preset iteration bandwidth as the radius, and the number of historical instantaneous noise features included in the multiple central areas is counted to obtain the aggregation amount of the multiple historical instantaneous noise feature mean values. The aggregation amount of the multiple historical instantaneous noise feature mean values ​​reflects the number of historical instantaneous noise features that are aggregated around the multiple historical instantaneous noise feature mean values. The more the number of aggregations, the more the corresponding historical instantaneous noise feature mean values ​​can represent the characteristic conditions of the multiple historical instantaneous noise feature sets. Based on the same principle of obtaining the aggregation amount of the multiple historical instantaneous noise feature mean values, the aggregation amount of the multiple iterative historical instantaneous noise features is calculated.

[0035] When the aggregation amount of the multiple iterative historical instantaneous noise features is less than or equal to the aggregation amount of the multiple historical instantaneous noise feature means, it indicates that the multiple historical instantaneous noise feature means can better represent the situation of the multiple historical instantaneous noise feature sets. Therefore, the multiple historical instantaneous noise feature means are used as the concentrated value of the multiple noise features.

[0036] When the aggregation amount of the multiple iterative historical instantaneous noise features is greater than the aggregation amount of the multiple historical instantaneous noise feature means, it indicates that the multiple iterative historical instantaneous noise features are more representative of the characteristics of the multiple historical instantaneous noise feature sets than the multiple historical instantaneous noise feature means. At this time, further judgment is needed to determine whether to continue iteration.

[0037] Optionally, determine whether the aggregation difference between the multiple iterative historical instantaneous noise features and the mean of the multiple historical instantaneous noise features is greater than or equal to a preset aggregation difference threshold. If so, it indicates that although the multiple iterative historical instantaneous noise features are not only better able to represent the characteristics of the multiple historical instantaneous noise feature sets than the mean of the multiple historical instantaneous noise features, but also to a deeper degree, then continue to iterate based on the multiple iterative historical instantaneous noise features until the maximum number of iterations (the maximum number of iterations set by technicians in this field, such as 30 times, 45 times, etc.) is met, and the multiple iterative historical instantaneous noise features obtained in the last iteration are used as the concentrated values ​​of the multiple noise features.

[0038] If not, it means that although the multiple iterative historical instantaneous noise features can better represent the characteristics of the multiple historical instantaneous noise feature sets than the multiple historical instantaneous noise feature means, the difference between the two is not far, and has reached a level that can better represent the multiple historical instantaneous noise feature sets. At this time, the iteration is stopped and the multiple iterative historical instantaneous noise features are used as the concentrated values ​​of the multiple noise features.

[0039] Further, taking the multiple historical instantaneous noise feature sets as indexes, the multiple distinguishing historical bearing noise monitoring data sequence sets are subjected to feature-associated bandwidth diffusion identification, and multiple historical instantaneous noise feature-associated bandwidth sets are determined. Step S300 of the embodiment of the present application further includes: Using the noise feature extraction network layer, the noise features of the plurality of distinguishing historical bearing noise monitoring data sequence sets are extracted to obtain a plurality of distinguishing historical bearing noise feature sequence sets; Randomly extracting a historical instantaneous noise feature from the plurality of historical instantaneous noise feature sets as a first historical instantaneous noise feature, and matching a corresponding first distinguishing historical bearing noise feature sequence from the plurality of distinguishing historical bearing noise feature sequence sets; According to a preset approximate correlation scale, a neighbor search is performed on the first historical instantaneous noise feature in the first distinguishing historical bearing noise feature sequence to obtain a first historical instantaneous noise feature neighborhood; Counting the duration of the first historical instantaneous noise feature neighborhood, and using the statistical result as the first historical instantaneous noise feature associated bandwidth; According to the preset nearest neighbor association scale, feature association bandwidth diffusion identification is performed on the multiple historical instantaneous noise feature sets in the corresponding multiple sets of distinguished historical bearing noise monitoring data sequences to determine multiple historical instantaneous noise feature association bandwidth sets.

[0040] In one embodiment of the present application, a plurality of distinguishing historical bearing noise monitoring data sequences are first analyzed using a noise feature extraction network layer, and the noise features of each data sequence are extracted to obtain the plurality of distinguishing historical bearing noise feature sequence sets. Each distinguishing historical bearing noise feature sequence contains the noise feature changes of a type of bearing fault generated during the operation of the bearing. Through the automatic feature learning capability of the deep learning network, the system can identify important noise features related to bearing faults.

[0041] Firstly, a feature is randomly selected from a plurality of historical instantaneous noise feature sets as the first historical instantaneous noise feature, and a feature sequence matching the feature is searched from a plurality of feature sequence sets for distinguishing historical bearing noises to obtain a first feature sequence for distinguishing historical bearing noises.

[0042] In the matched first distinguishing historical bearing noise feature sequence, the system will perform a neighbor search on the first historical instantaneous noise feature according to a preset approximate association scale (the minimum similarity when the features are associated, which is preset by a person skilled in the art). Through the neighbor search, the system can find multiple first distinguishing historical bearing noise features associated with the first historical instantaneous noise feature, and obtain the first historical instantaneous noise feature neighborhood.

[0043] Optionally, by calculating the similarity of the first historical instantaneous noise feature with the first differentiated historical bearing noise feature of its neighbor in the first differentiated historical bearing noise feature sequence, when the calculation result satisfies a preset approximate association scale, the first differentiated historical bearing noise feature of the neighbor is added to the neighborhood of the first historical instantaneous noise feature, and the similarity between the first differentiated historical bearing noise feature of the second neighbor and the first historical instantaneous noise feature is further analyzed to see whether it satisfies the preset neighbor association scale. If it does not satisfy the preset neighbor association scale, the neighbor retrieval is stopped to obtain the neighborhood of the first historical instantaneous noise feature.

[0044] For the first historical instantaneous noise feature neighborhood found, the time points at the left and right ends of the neighborhood are counted to obtain the duration of the neighborhood, which is then used as the first historical instantaneous noise feature associated bandwidth.

[0045] Based on the same principle of obtaining the first historical instantaneous noise feature correlation bandwidth, the plurality of historical instantaneous noise feature sets are identified by feature correlation bandwidth diffusion in the corresponding plurality of distinguished historical bearing noise monitoring data sequence sets according to the preset nearest neighbor correlation scale, and a plurality of historical instantaneous noise feature correlation bandwidth sets are determined, thereby achieving the technical effect of providing a basis for subsequent multi-scale noise data processing.

[0046] S400: using the multiple noise feature association bandwidths as multiple receptive fields of multiple noise processing network layers, and using the configured multiple noise processing network layers to perform multi-scale feature analysis on the noise monitoring data sequence of the target bearing within a preset monitoring window to obtain multiple noise monitoring feature sets; In one embodiment, multiple noise feature association bandwidths will be used as receptive fields (Receptive Fields) of multiple noise processing network layers. Receptive field refers to the input area that a neuron can "perceive" in a neural network. It defines the range of data that the network can process when performing feature extraction. Using noise feature association bandwidth as the receptive field can help the network extract features of noise data at different scales. Each receptive field corresponds to a noise feature association bandwidth, which means that the network will perform feature extraction in different time periods or frequency bands for different noise features.

[0047] The noise processing network layer is a hierarchical structure built by deep learning networks (such as convolutional neural networks) that is specifically used to process and analyze noise data. Each layer of the network analyzes the input noise data through different receptive fields and extracts features at different scales. The advantage of using multiple noise processing network layers is that the system can obtain different perspectives on the bearing state through different levels of feature extraction, such as short-term noise changes (high-frequency signals) and long-term changes (low-frequency signals).

[0048] Multiscale feature analysis refers to analyzing the characteristics of noise data at multiple scales (time scale or frequency scale). This is very important because bearing faults often show different noise patterns at different time scales. For example, high-frequency noise may indicate tiny cracks or surface damage, while low-frequency noise may be related to the overall operating condition of the bearing or large-scale damage. Through multiscale feature analysis, the system is able to comprehensively consider signals at different time / frequency scales to more comprehensively diagnose bearing faults.

[0049] Within the preset monitoring window, the noise monitoring data sequence of the target bearing is collected, and the obtained noise monitoring data sequence will be input into multiple noise processing network layers. The input of the data sequence will be passed layer by layer, and feature extraction will be performed through each noise processing network layer. Each layer can process and identify noise patterns at different scales to obtain the multiple noise monitoring feature sets. These feature sets contain noise features extracted from different scales. Each noise monitoring feature set includes time domain features: such as amplitude, average value, peak value, root mean square value, etc.; frequency domain features: such as frequency components, spectral density, etc.; time-frequency domain features: features extracted by methods such as wavelet transform, combining information in the time domain and frequency domain.

[0050] By using multiple noise processing network layers to perform multi-scale noise feature analysis, it is ensured that effective features of noise data are extracted at different scales and ranges. This allows for more comprehensive data analysis and processing of noise monitoring data, thereby achieving the technical effect of improving the accuracy of bearing fault detection.

[0051] S500: Perform cross-scale interactive fusion on the multiple noise monitoring feature sets to obtain a target interactive fusion noise monitoring feature set, and use the target interactive fusion noise monitoring feature set as a bearing noise processing result.

[0052] Further, the multiple noise monitoring feature sets are interactively fused across scales to obtain a target interactively fused noise monitoring feature set. Step S500 of the embodiment of the present application further includes: randomly extracting a first noise monitoring feature set and a second noise monitoring feature set from the plurality of noise monitoring feature sets; Calculating the first noise monitoring feature set and the second noise monitoring feature set for similarity identification to determine a first feature similarity set; Normalizing the first feature similarity set to obtain a first feature similarity normalized value set; Performing convolution calculation on the first feature similarity normalization value set and the second noise monitoring feature set to obtain a first interactive fusion noise monitoring feature set; randomly extracting a third noise monitoring feature set from the multiple noise monitoring feature sets, and performing cross-scale interactive fusion on the third noise monitoring feature set with the first interactive fusion noise monitoring feature set to obtain a second interactive fusion noise monitoring feature set; After multiple cross-scale interactive fusions, until all noise monitoring features in the multiple noise monitoring feature sets are fused, the target interactive fusion noise monitoring feature set is obtained.

[0053] Furthermore, step S500 of the embodiment of the present application further includes: Acquire multiple sample feature similarity normalized value sets, multiple sample noise monitoring feature sets, and multiple sample interactive fusion noise monitoring feature sets as training data; Use the training data to supervise the network layer built based on the convolutional neural network until the training converges to obtain a trained convolutional network layer; The convolutional network layer is used to perform convolution calculation on the first feature similarity normalization value set and the second noise monitoring feature set to obtain the first interactive fusion noise monitoring feature set.

[0054] In one embodiment of the present application, multiple noise monitoring feature sets are interactively fused across scales to obtain a target interactive fusion noise monitoring feature set. Interactive fusion across scales refers to fusing noise monitoring features from different scales (time scale, frequency scale, etc.). By fusing information from each scale, richer noise features can be obtained. This method can integrate noise patterns at different scales to help the system make more accurate diagnoses.

[0055] In this step, multiple noise monitoring feature sets will be gradually fused into a target interactive fusion noise monitoring feature set through similarity calculation, normalization, convolution and other operations. Through multiple interactive fusions, it is possible to more comprehensively integrate features of different scales, further improve the comprehensiveness and reliability of noise monitoring data processing, and obtain high-quality bearing noise processing results. By obtaining the bearing noise processing results, high-quality analysis data is provided for bearing fault diagnosis, thereby achieving the technical effect of improving the depth of bearing noise processing and providing data support for improving the reliability of bearing fault diagnosis.

[0056] In one embodiment, a first noise monitoring feature set and a second noise monitoring feature set are randomly extracted from the plurality of noise monitoring feature sets. Then, the first noise monitoring feature set and the second noise monitoring feature set are calculated using a cosine similarity calculation formula for similarity identification to determine a first feature similarity set. Each first feature similarity reflects the degree of similarity between the first noise monitoring feature and the second noise monitoring feature.

[0057] Optionally, the first feature similarity set is normalized by using a minimum-maximum normalization formula to obtain a first feature similarity normalization value set. The minimum-maximum normalization formula is X norm =(XX min ) / (X max -X min ), X norm is the normalized value of the first feature similarity after normalization, X max is the maximum value in the first feature similarity set, X min is the minimum value in the first feature similarity set.

[0058] Then, a neural network is used to perform convolution calculation on the first feature similarity normalization value set and the second noise monitoring feature set to obtain a first interactive fusion noise monitoring feature set. Convolution operation is a common technique in deep learning, which can extract local features in input data and fuse correlation information between different features.

[0059] Then, a third noise monitoring feature set is randomly extracted from the multiple noise monitoring feature sets, and the third noise monitoring feature set is cross-scale interactively fused with the first interactively fused noise monitoring feature set to obtain a second interactively fused noise monitoring feature set, thereby realizing interactive fusion of features at different scales. After multiple cross-scale interactive fusions, until all the noise monitoring features in the multiple noise monitoring feature sets are fused, the target interactively fused noise monitoring feature set is obtained. By repeatedly fusing different noise monitoring feature sets, the expressive power of the features is gradually enhanced. After each fusion, the generated new feature set will contain more comprehensive noise information.

[0060] In order to perform supervised learning, it is necessary to collect training data. Optionally, multiple sample feature similarity normalization value sets, multiple sample noise monitoring feature sets, and multiple sample interactive fusion noise monitoring feature sets are obtained as training data.

[0061] By using the training data to supervise the network layer built based on the convolutional neural network, the complex relationship between different noise features can be learned, and the parameters of the convolutional network layer can be optimized until the network converges and accurate feature fusion capabilities are obtained. The training is continued until convergence, and a trained convolutional network layer is obtained. Furthermore, the trained convolutional network layer is used to perform convolution calculations on the first feature similarity normalization value set and the second noise monitoring feature set to obtain more accurate interactive fusion features, that is, the first interactive fusion noise monitoring feature set. The technical effect of interactively fusing multi-scale noise monitoring features and deeply analyzing bearing noise data is achieved.

[0062] In summary, the embodiments of the present application have at least the following technical effects: This application obtains the target interactive fusion noise monitoring feature set by performing heterogeneous differentiation and concentrated noise feature extraction on the historical bearing fault record set, combining multi-scale analysis and cross-scale interactive fusion, and uses the target interactive fusion noise monitoring feature set as the bearing noise processing result. This achieves the technical effect of improving the accuracy of noise data analysis and providing reliable data for bearing fault diagnosis.

[0063] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0064] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0065] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A bearing noise processing method for bearing fault diagnosis, characterized in that: The method comprises: Obtaining a target model of a target bearing, searching a historical fault database based on the target model, and determining a set of historical bearing fault records; Performing heterogeneous differentiation on the historical bearing fault record set to determine a plurality of differentiated historical bearing fault record sets; Traversing the plurality of distinguished historical bearing fault record sets to perform concentrated noise feature extraction, and obtaining a plurality of noise feature concentrated values ​​and a plurality of noise feature associated bandwidths; The multiple noise feature association bandwidths are used as multiple receptive fields of multiple noise processing network layers, and the configured multiple noise processing network layers are used to perform multi-scale feature analysis on the noise monitoring data sequence of the target bearing within a preset monitoring window to obtain multiple noise monitoring feature sets; The multiple noise monitoring feature sets are interactively fused across scales to obtain a target interactive fusion noise monitoring feature set, and the target interactive fusion noise monitoring feature set is used as a bearing noise processing result.

2. A bearing noise processing method for bearing fault diagnosis according to claim 1, characterized in that: The plurality of distinguished historical bearing fault record sets are traversed to perform concentrated noise feature extraction, and a plurality of noise feature concentration values ​​and a plurality of noise feature associated bandwidths are obtained, including: Traversing the plurality of distinguishing historical bearing fault record sets to extract noise monitoring data, and obtaining a plurality of distinguishing historical bearing noise monitoring data sequence sets; Extracting instantaneous noise monitoring data from the plurality of distinguished historical bearing noise monitoring data series respectively, and determining a plurality of historical instantaneous noise monitoring data sets, wherein each historical instantaneous noise monitoring data is monitoring data at a time when the maximum data change fluctuation occurs in each distinguished historical bearing noise monitoring data series; Extracting features from the multiple historical instantaneous noise monitoring data sets to obtain multiple historical instantaneous noise feature sets, and performing centralized analysis on the multiple historical instantaneous noise feature sets to determine multiple noise feature centralized values; Using the multiple historical instantaneous noise feature sets as indexes, feature-associated bandwidth diffusion identification is performed on the multiple sets of distinguished historical bearing noise monitoring data sequences to determine multiple sets of historical instantaneous noise feature-associated bandwidths, wherein each historical instantaneous noise feature-associated bandwidth reflects the duration of data associated with the historical instantaneous noise feature at the time of fault occurrence in a distinguished historical bearing noise monitoring data sequence; The mean values ​​of the plurality of historical instantaneous noise feature associated bandwidth sets are respectively calculated to obtain the plurality of noise feature associated bandwidths.

3. A bearing noise processing method for bearing fault diagnosis according to claim 2, characterized in that: Extracting instantaneous features from the plurality of historical instantaneous noise monitoring data sets to obtain a plurality of historical instantaneous noise feature sets, and centrally analyzing the plurality of historical instantaneous noise feature sets to determine a plurality of noise feature central values, including: Using a noise feature extraction network layer to extract instantaneous features from the plurality of historical instantaneous noise monitoring data sets, to obtain the plurality of historical instantaneous noise feature sets; Traversing multiple historical instantaneous noise feature sets to perform feature mean calculations and determine multiple historical instantaneous noise feature means; Iterating the mean values ​​of the multiple historical instantaneous noise features in the multiple historical instantaneous noise feature sets according to a preset iteration bandwidth to obtain multiple iterative historical instantaneous noise features; When the aggregation amount of the multiple iterative historical instantaneous noise features is less than or equal to the aggregation amount of the multiple historical instantaneous noise feature mean values, the multiple historical instantaneous noise feature mean values ​​are used as the centralized values ​​of the multiple noise features.

4. A bearing noise processing method for bearing fault diagnosis according to claim 3, characterized in that: include: When the aggregation amount of the multiple iterative historical instantaneous noise features is greater than the aggregation amount of the multiple historical instantaneous noise feature means, determine whether the aggregation amount difference between the multiple iterative historical instantaneous noise features and the multiple historical instantaneous noise feature means is greater than or equal to a preset aggregation amount difference threshold, and if so, continue to iterate based on the multiple iterative historical instantaneous noise features until the maximum number of iterations is met, and use the multiple iterative historical instantaneous noise features obtained in the last iteration as the multiple noise feature concentration values; If not, the iteration is stopped, and the multiple iterative historical instantaneous noise features are used as the centralized values ​​of the multiple noise features.

5. A bearing noise processing method for bearing fault diagnosis according to claim 4, characterized in that: Using the multiple historical instantaneous noise feature sets as indexes, performing feature-associated bandwidth diffusion identification on the multiple sets of distinguished historical bearing noise monitoring data sequences, and determining multiple historical instantaneous noise feature-associated bandwidth sets, including: Using the noise feature extraction network layer, the noise features of the plurality of distinguishing historical bearing noise monitoring data sequence sets are extracted to obtain a plurality of distinguishing historical bearing noise feature sequence sets; Randomly extracting a historical instantaneous noise feature from the plurality of historical instantaneous noise feature sets as a first historical instantaneous noise feature, and matching a corresponding first distinguishing historical bearing noise feature sequence from the plurality of distinguishing historical bearing noise feature sequence sets; According to a preset approximate correlation scale, a neighbor search is performed on the first historical instantaneous noise feature in the first distinguishing historical bearing noise feature sequence to obtain a first historical instantaneous noise feature neighborhood; Counting the duration of the first historical instantaneous noise feature neighborhood, and using the statistical result as the first historical instantaneous noise feature associated bandwidth; According to the preset nearest neighbor association scale, feature association bandwidth diffusion identification is performed on the multiple historical instantaneous noise feature sets in the corresponding multiple sets of distinguished historical bearing noise monitoring data sequences to determine multiple historical instantaneous noise feature association bandwidth sets.

6. A bearing noise processing method for bearing fault diagnosis according to claim 1, characterized in that: Performing cross-scale interactive fusion on the multiple noise monitoring feature sets to obtain a target interactive fusion noise monitoring feature set includes: randomly extracting a first noise monitoring feature set and a second noise monitoring feature set from the plurality of noise monitoring feature sets; Calculating the first noise monitoring feature set and the second noise monitoring feature set for similarity identification to determine a first feature similarity set; Normalizing the first feature similarity set to obtain a first feature similarity normalized value set; Performing convolution calculation on the first feature similarity normalization value set and the second noise monitoring feature set to obtain a first interactive fusion noise monitoring feature set; randomly extracting a third noise monitoring feature set from the multiple noise monitoring feature sets, and performing cross-scale interactive fusion on the third noise monitoring feature set with the first interactive fusion noise monitoring feature set to obtain a second interactive fusion noise monitoring feature set; After multiple cross-scale interactive fusions, until all noise monitoring features in the multiple noise monitoring feature sets are fused, the target interactive fusion noise monitoring feature set is obtained.

7. A bearing noise processing method for bearing fault diagnosis according to claim 6, characterized in that: include: Acquire multiple sample feature similarity normalized value sets, multiple sample noise monitoring feature sets, and multiple sample interactive fusion noise monitoring feature sets as training data; Use the training data to supervise the network layer built based on the convolutional neural network until the training converges to obtain a trained convolutional network layer; The convolutional network layer is used to perform convolution calculation on the first feature similarity normalization value set and the second noise monitoring feature set to obtain the first interactive fusion noise monitoring feature set.

8. A bearing noise processing method for bearing fault diagnosis according to claim 1, characterized in that: The historical bearing fault record set is heterogeneously distinguished to determine a plurality of distinguished historical bearing fault record sets, including: Randomly extracting a plurality of historical bearing fault records from the historical bearing fault record set; Enumerate the multiple historical bearing fault records in pairs to obtain multiple enumeration combinations; Determine whether there is an enumeration combination whose record similarity among the multiple enumeration combinations is greater than a preset record similarity threshold, and if not, use the multiple historical bearing fault records as multiple differentiation targets; Based on the multiple differentiation targets, the historical bearing fault record set is heterogeneously differentiated according to a preset record similarity threshold to obtain the multiple differentiated historical bearing fault record sets, wherein each differentiated historical bearing fault record set corresponds to a differentiation target.

Images