Bearing life prediction model training method, bearing life prediction method and device

Abstract

The invention discloses a bearing life prediction model training method and a bearing life prediction method and device, and relates to the technical field of artificial intelligence. The bearing life prediction model training method comprises the steps that in the working process of a sample bearing, multiple maximum amplitudes in vibration signals of the sample bearing are collected, the operation state and corresponding state parameters of the sample bearing are determined in combination with a preset amplitude threshold value, and therefore a health index sequence and a degradation index sequence are obtained; and determining a mixed kernel function model according to at least two preset kernel functions, and training the mixed kernel function model through the health index sequence and the degradation index sequence to obtain a bearing life prediction model. According to the method, the health index sequence and the degradation index sequence of the sample bearing are acquired, the hybrid kernel function model which at least comprises two preset kernel functions is trained by using the two kinds of sequence data, and the hybrid kernel function model is used for predicting the service life of the bearing, so that the prediction accuracy and robustness of the residual service life of the bearing are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a bearing life prediction model training method, a bearing life prediction method and a bearing life prediction device. Background Art

[0002] Bearings are an important component widely used in modern mechanical equipment. Their health status directly affects the operational stability and safety of mechanical equipment. Predicting the remaining life of bearings can optimize maintenance plans and reduce mechanical equipment downtime and maintenance costs.

[0003] In current bearing remaining life prediction schemes, bearing remaining life prediction is usually achieved through a data-driven approach, that is, using event data, condition monitoring data and machine learning technology to train the prediction model, and then using the trained prediction model to predict the remaining life of the bearing.

[0004] However, the above-mentioned bearing remaining life prediction scheme usually uses single-dimensional feature information for prediction, and the parameters of the prediction model usually rely on human selection and adjustment, which leads to insufficient accuracy and robustness of the bearing remaining life prediction. Summary of the Invention

[0005] The main purpose of this application is to propose a bearing life prediction model training method, a bearing life prediction method and a bearing life prediction device, aiming to improve the accuracy and robustness of bearing remaining life prediction.

[0006] In a first aspect, the present invention provides a bearing life prediction model training method, comprising:

[0007] During the operation of the sample bearing, collecting a plurality of maximum amplitudes of the vibration signal of the sample bearing;

[0008] Determining an operating state and corresponding state parameters of the sample bearing according to a plurality of maximum amplitudes and a preset amplitude threshold in the vibration signal of the sample bearing, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage;

[0009] Acquire a health indicator sequence and a degradation indicator sequence according to the operating state of the sample bearing and the corresponding state parameters;

[0010] A mixed kernel function model is determined according to at least two preset kernel functions, and the mixed kernel function model is trained by the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model, which is used to predict the bearing life.

[0011] In an optional embodiment, during the operation of the sample bearing, collecting a plurality of maximum amplitudes of the vibration signal of the sample bearing includes:

[0012] During the working process of the sample bearing, a plurality of maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle are collected according to a preset window length;

[0013] The step of determining the operating state and corresponding state parameters of the sample bearing according to the plurality of maximum amplitudes and a preset amplitude threshold in the vibration signal of the sample bearing comprises:

[0014] Calculating a mean and a standard deviation of the plurality of maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle;

[0015] determining whether each of the maximum amplitudes is abnormal according to the mean, the standard deviation, and the preset amplitude threshold;

[0016] If the maximum number of consecutive abnormalities in the current cycle reaches a preset number of consecutive abnormalities, then starting from the current cycle, the operating state of the sample bearing is the degradation stage. Before the current cycle, the operating state of the sample bearing is the healthy stage, and the state parameters corresponding to the healthy stage and the degradation stage are recorded respectively.

[0017] In an optional embodiment, after determining whether each of the maximum amplitudes is abnormal according to the mean, the standard deviation, and the preset amplitude threshold in sequence, the method further includes:

[0018] If the number of the maximum amplitudes of the consecutive anomalies in the current cycle does not reach the preset number of consecutive anomalies, multiple maximum amplitudes in the vibration signal of the sample bearing in the next cycle are collected according to the preset window length.

[0019] In an optional embodiment, obtaining a health indicator sequence and a degradation indicator sequence according to the operating state and the corresponding state parameter of the sample bearing includes:

[0020] Multimodal data feature extraction is adopted to perform feature extraction on the state parameters corresponding to the healthy stage and the degradation stage respectively, to obtain the health indicator sequence and the degradation indicator sequence.

[0021] In an optional embodiment, before determining the hybrid kernel function model according to at least two preset kernel functions, the method further includes:

[0022] Determine the output matrix of the hidden layer nodes according to the extreme learning machine activation function, and determine the extreme learning machine function model according to the output matrix of the hidden layer nodes and preset weights;

[0023] The step of determining a hybrid kernel function model according to at least two preset kernel functions includes:

[0024] Determine a kernel extreme learning machine function model according to a first preset kernel function and the extreme learning machine function model;

[0025] The hybrid kernel function model is determined according to a second preset kernel function and the kernel extreme learning machine function model.

[0026] In an optional embodiment, the training of the hybrid kernel function model using the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model includes:

[0027] Using the health indicator sequence as a training set and the degradation indicator sequence as a test set, training the hybrid kernel function model;

[0028] The fitting curve and error corresponding to the trained hybrid kernel function model are calculated. If both the fitting curve and the error meet the preset training judgment conditions, the trained hybrid kernel function model is used as the bearing life prediction model.

[0029] In an optional embodiment, if the fitting curve and / or the error do not meet the preset training judgment conditions, the method further includes:

[0030] Randomly generating a set of candidate solutions, wherein the set of candidate solutions includes multiple groups of candidate solutions, each group of candidate solutions represents a set of hyperparameters of the hybrid kernel function model;

[0031] Iterate the candidate solution set according to a preset optimization algorithm and obtain a set of optimal candidate solutions;

[0032] The hyperparameters of the hybrid kernel function model are optimized according to the optimal candidate solution, and the optimized hybrid kernel function model is used as a bearing life prediction model.

[0033] In a second aspect, the present invention provides a bearing life prediction method, comprising:

[0034] Collect and obtain the state parameters of the bearing to be predicted;

[0035] A bearing life prediction model obtained by training using the method described in any of the aforementioned implementations is used to obtain a bearing life prediction result based on the state parameters of the bearing to be predicted.

[0036] In a third aspect, the present invention provides a bearing life prediction model training device, comprising:

[0037] A first acquisition module is used to acquire a plurality of maximum amplitudes of the vibration signal of the sample bearing during operation of the sample bearing;

[0038] a determination module, configured to determine an operating state and corresponding state parameters of the sample bearing based on a plurality of maximum amplitudes and a preset amplitude threshold in the vibration signal of the sample bearing, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage;

[0039] an acquisition module, configured to acquire a health indicator sequence and a degradation indicator sequence according to the operating state of the sample bearing and the corresponding state parameters;

[0040] A training module is used to determine a mixed kernel function model based on at least two preset kernel functions, and train the mixed kernel function model through the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model, which is used to predict the bearing life.

[0041] In a fourth aspect, the present invention provides a bearing life prediction device, comprising:

[0042] The second acquisition module is used to acquire the state parameters of the bearing to be predicted;

[0043] A prediction module is used to obtain a bearing life prediction model trained by the bearing life prediction model training device described in the third aspect, and obtain a bearing life prediction result based on the state parameters of the bearing to be predicted.

[0044] In a fifth aspect, the present application provides an electronic device comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, the processor and the storage medium communicate via the bus, and the processor executes the machine-readable instructions to perform a method as described in any of the foregoing embodiments.

[0045] In a sixth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program, when executed by a processor, executes the method as described in any of the aforementioned embodiments.

[0046] The beneficial effects of this application are:

[0047] The bearing life prediction model training method provided in an embodiment of the present application includes: during the operation of a sample bearing, collecting multiple maximum amplitudes in the vibration signal of the sample bearing; determining the operating state and corresponding state parameters of the sample bearing based on the multiple maximum amplitudes in the vibration signal of the sample bearing and a preset amplitude threshold, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage; obtaining a health indicator sequence and a degradation indicator sequence based on the operating state and the corresponding state parameters of the sample bearing; determining a mixed kernel function model based on at least two preset kernel functions, and training the mixed kernel function model through the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model, wherein the bearing life prediction model is used to predict the bearing life. The method compares multiple maximum amplitudes of the vibration signals of the sample bearings collected during their operation with preset amplitude thresholds in sequence to divide the operating status of the sample bearings into a healthy stage and a degraded stage, and divides the collected status information of the sample bearings into status information corresponding to the healthy stage and status information corresponding to the degraded stage, namely, a health indicator sequence and a degradation indicator sequence. This method trains a hybrid kernel function model containing at least two preset kernel functions through the health indicator sequence and the degradation indicator sequence, and uses the trained hybrid kernel function model as a bearing life prediction model to predict the bearing life, thereby improving the accuracy and robustness of the bearing remaining life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0049] Figure 1 A flowchart of a bearing life prediction model training method provided in one embodiment of the present application;

[0050] Figure 2 A flowchart of a bearing life prediction model training method provided in another embodiment of the present application;

[0051] Figure 3 A flowchart of a bearing life prediction model training method according to another embodiment of the present application is provided;

[0052] Figure 4 A flowchart of a bearing life prediction model training method provided in yet another embodiment of the present application;

[0053] Figure 5A schematic diagram of the overall process of the bearing life prediction model training method provided in one embodiment of the present application;

[0054] Figure 6 A schematic flow chart of a bearing life prediction method provided in one embodiment of the present application;

[0055] Figure 7 A schematic structural diagram of a bearing life prediction model training device provided in an embodiment of the present application;

[0056] Figure 8 A schematic structural diagram of a bearing life prediction device provided in an embodiment of the present application;

[0057] Figure 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings of 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. Generally, the components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0059] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present application. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in the present application without creative work are within the scope of protection of the present application.

[0060] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures. The terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0061] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0062] In the current bearing remaining life prediction scheme, single-dimensional feature information derived from bearing event data or condition monitoring data is usually input into a trained machine learning model to obtain the output result of the recent learning model to realize the prediction of the bearing remaining life. The above-mentioned machine learning model usually does not contain a kernel function, and the hyperparameters of the machine learning model are usually manually selected and adjusted. Once the above-mentioned single-dimensional feature information is missing or there is a large error, it will have a greater impact on the prediction of the bearing remaining life. The prediction performance of the machine learning model that does not contain a kernel function and manually selects and adjusts the hyperparameters is also poor. It can be seen that the accuracy and robustness of the bearing remaining life prediction using the current bearing remaining life prediction scheme are insufficient. In this context, the main purpose of this application is to propose a bearing life prediction model training method, which aims to improve the accuracy and robustness of the bearing remaining life prediction.

[0063] Figure 1 This is a flow chart of a bearing life prediction model training method provided in one embodiment of the present application. The execution subject of the method may be, for example, a computer or other device with computing and processing capabilities, but is not limited thereto. Figure 1 As shown, the method may include:

[0064] S101. During the operation of a sample bearing, a plurality of maximum amplitude values ​​of a vibration signal of the sample bearing are collected.

[0065] Exemplarily, the above-mentioned collection of multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing can be achieved by, for example, a sensor arranged near the above-mentioned sample bearing. The sensor can be provided with one or more sensors. If multiple sensors are provided, the multiple sensors can be respectively arranged at different positions near the above-mentioned sample bearing. The above-mentioned sensor collects multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing, for example, it can refer to collection at a certain frequency, such as once every 1 second, once every 10 seconds, once every 30 seconds, etc., but the specific collection frequency can be adjusted and determined according to actual conditions, and is not limited to the frequencies listed above.

[0066] It is understandable that the specific device used to collect the multiple maximum amplitudes in the vibration signals of the above-mentioned sample bearings, the type and model of the above-mentioned sensors, and the specific number and setting positions of the above-mentioned sensors can all be selected and determined according to actual conditions and are not limited here.

[0067] S102. Determine the operating state and corresponding state parameters of the sample bearing based on the plurality of maximum amplitudes and a preset amplitude threshold in the vibration signal of the sample bearing, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage.

[0068] Exemplarily, determining the operating status of the sample bearing based on the multiple maximum amplitudes and the preset amplitude threshold in the vibration signal of the sample bearing may refer to, for example, comparing the multiple maximum amplitudes with the preset amplitude threshold respectively. When the cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches a preset value, or when the continuous cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches a preset value, or when the difference between a certain maximum amplitude and the preset amplitude threshold reaches a preset value, the time period corresponding to the multiple maximum amplitudes may be used as the starting point for the operating status of the sample bearing to enter the degradation stage. That is, from the time period corresponding to the multiple maximum amplitudes, the operating status of the sample bearing is the degradation stage, and before the time period corresponding to the multiple maximum amplitudes, the operating status of the sample bearing is the healthy stage.

[0069] For example, assuming that a sample bearing has been working normally for 24 hours (for ease of explanation, the normal working time of the sample bearing is, for example, 0:00:00 to 24:00:00), and multiple maximum amplitude values ​​in the vibration signal of the sample bearing are collected at a frequency of once every 10 seconds, then based on the above, it can be further assumed that the multiple maximum amplitude values ​​in the vibration signal of the sample bearing collected in the 80-second time period from 16:00:00 to 16:01:20 are as shown in Table 1 below:

[0070]

[0071] Table 1 Record of multiple maximum amplitudes in the vibration signal of a sample bearing

[0072] Please refer to Table 1 above. According to Table 1, there are 9 maximum amplitude values ​​in the vibration signals of the sample bearings collected within the 80-second time period from 16:00:00 to 16:01:20.

[0073] Taking "when the cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches a preset value, the starting points of the time periods corresponding to the multiple maximum amplitudes can be used as the starting points for the operating state of the sample bearing to enter the degradation stage" as an example, assuming that the preset amplitude threshold is 0.05 mm, and the cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches the preset value, which means that the cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches 5. According to the data in Table 1, the cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold is 5 (that is, the 5 maximum amplitudes from 16:00:40 to 16:01:20). When the preset value is reached, the starting points of the time periods corresponding to the 9 maximum amplitudes can be used as the starting points for the operating state of the sample bearing to enter the degradation stage. That is, the sample bearing enters the degradation stage at 16:00:00, the healthy stage of the sample bearing is 0:00:00 to 15:59:59, and the degradation stage of the sample bearing is 16:00:00 to 24:00:00.

[0074] Taking "when the continuous cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches a preset value, the starting point of the time period corresponding to the multiple maximum amplitudes can be used as the starting point of the running state of the sample bearing entering the degradation stage" as an example, assuming that the preset amplitude threshold is 0.05mm, and the continuous cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches the preset value means that the cumulative number of maximum amplitudes greater than or equal to the preset amplitude threshold reaches 3, then according to the data in Table 1, the maximum amplitude is greater than or equal to the preset If the continuous cumulative number of amplitude thresholds is 5 (i.e., the 5 maximum amplitudes from 16:00:40 to 16:01:20), reaching the above-mentioned preset value, the starting point of the time period corresponding to the above-mentioned 9 maximum amplitudes can be used as the starting point for the operating state of the sample bearing to enter the degradation stage. That is, the above-mentioned sample bearing enters the degradation stage at 16:00:00, the healthy stage of the sample bearing is 0:00:00 to 15:59:59, and the degradation stage of the sample bearing is 16:00:00 to 24:00:00.

[0075] Finally, taking "when the difference between a certain maximum amplitude and a preset amplitude threshold reaches a preset value, the starting point of the time period corresponding to the multiple maximum amplitudes can be used as the starting point for the operating state of the sample bearing to enter the degradation stage" as an example, assuming that the preset amplitude threshold is 0.05mm, and the difference between the maximum amplitude and the preset amplitude threshold reaches the preset value, which means that the difference between the maximum amplitude and the preset amplitude threshold reaches 0.03mm, then according to the data in Table 1, the maximum difference between the maximum amplitude and the preset amplitude threshold is 0.06mm-0.05mm=0.01mm, which does not reach the preset value. Therefore, it cannot be determined that the sample bearing has entered the degradation stage based on the data in Table 1.

[0076] It should be noted that the above contents are all possible examples. How to actually determine the operating status of the above sample bearings is not limited to the above examples. Data including but not limited to the length of the time period, preset amplitude threshold, normal working time of the bearing, etc. may actually be different from the above examples.

[0077] The state parameters corresponding to the operating state of the above-mentioned sample bearing may, for example, refer to the state parameters including but not limited to electrical parameters, motion parameters, setting parameters, fault record parameters, etc. of the sample bearing and the equipment where the sample bearing is located, collected by sensors, etc. while collecting multiple maximum amplitudes in the vibration signal of the above-mentioned sample bearing through sensors arranged near the above-mentioned sample bearing. By way of example, the above-mentioned electrical parameters may include parameters such as voltage and current, the above-mentioned motion parameters may include parameters such as vibration, noise, speed, torque, temperature, etc., the above-mentioned setting parameters may include parameters such as human-computer interaction setting parameters, and the above-mentioned fault record parameters may refer to historical fault records, etc. It can be understood that the collection of the above-mentioned state parameters may be achieved through multiple sensors of different types, models, and positions, and no specific limitation is made here.

[0078] Determining the state parameters corresponding to the operating state of the above-mentioned sample bearing may, for example, mean dividing the above-mentioned electrical parameters, motion parameters, setting parameters, fault record parameters and other state parameters into state parameters collected in the healthy stage of the sample bearing and state parameters collected in the degraded stage of the sample bearing after determining the healthy stage and the degraded stage of the motion state of the sample bearing. It should be noted that the above-mentioned electrical parameters, motion parameters, setting parameters and fault record parameters are only example parameters, and the actual parameter types and quantities may be the same as or different from the above-mentioned example parameters, and are not limited to the above-mentioned ranges.

[0079] S103: Obtain a health indicator sequence and a degradation indicator sequence according to the operating status of the sample bearing and the corresponding status parameters.

[0080] Exemplarily, the above-mentioned health indicator sequence and degradation indicator sequence are obtained based on the above-mentioned operating status of the above-mentioned sample bearing and the corresponding above-mentioned status parameters. For example, it can refer to the status parameters collected in the health stage of the above-mentioned sample bearing and the status parameters collected in the degradation stage of the sample bearing, which are respectively subjected to processing including but not limited to data denoising, data filling, feature extraction, etc., to form a health indicator sequence corresponding to the status parameters collected in the health stage of the sample bearing and a degradation indicator sequence corresponding to the sample parameters collected in the degradation stage of the sample bearing. The specific steps and methods of the above-mentioned data denoising, data filling, feature extraction and other processing can be selected and determined based on the actual type, quantity, characteristics, etc. of the status parameters, and are not specifically limited here.

[0081] Of course, the storage form of the above-mentioned health indicator sequence or degradation indicator sequence can include but is not limited to one or more matrices, one or more arrays, etc., and can be specifically selected and determined based on the actual type, quantity, characteristics, etc. of the state parameters, and is not limited to the storage form of matrices and arrays.

[0082] S104. Determine a hybrid kernel function model based on at least two preset kernel functions, and train the hybrid kernel function model using the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model. The bearing life prediction model is used to predict the bearing life.

[0083] Exemplarily, the above-mentioned determination of the hybrid kernel function model based on at least two preset kernel functions may, for example, refer to first selecting and determining a preset machine learning model as a basic model based on the actual type, quantity, characteristics, etc. of state parameters. The preset machine learning model may, for example, be a machine learning model including but not limited to ELM (Extreme Learning Machine) model, DLNs (Deep Linear Networks) model, DBMs (Deep Boltzmann Machines) model, etc., and then introducing at least two preset kernel functions into the above-mentioned preset machine learning model to form a hybrid kernel function model, wherein the above-mentioned at least two preset kernel functions may, for example, include a local kernel function and a global kernel function. Of course, the above-mentioned content is only a possible example. The specific hybrid kernel function model determination method, the specific type and quantity of preset kernel functions, etc. can all be adjusted and determined according to actual conditions and are not limited to the above-mentioned content.

[0084] The above-mentioned hybrid kernel function model is trained by the above-mentioned health indicator sequence and the above-mentioned degradation indicator sequence. For example, it can refer to using the health indicator sequence as training data and the degradation indicator sequence as test data to complete the training of the above-mentioned hybrid kernel function model. The above-mentioned bearing life prediction model can, for example, refer to the above-mentioned hybrid kernel function model that has been trained. It can be understood that the normal working time of the above-mentioned sample model, that is, the time when the sample bearing is installed in the working position, starts running normally until the sample bearing fails and stops running is known. Therefore, when the above-mentioned health indicator sequence is used as training data and the degradation indicator sequence is used as test data, the above-mentioned training data and test data should also be marked with the bearing life of the sample bearing, that is, how long the current sample bearing can still work normally.

[0085] For example, let's assume that a sample bearing has been working normally for 24 hours (for ease of explanation, the normal working hours of the sample bearing are from 0:00:00 to 24:00:00), and the state parameters of the sample bearing are collected at a frequency of once every 10 seconds. Based on the above, it can be further assumed that the health indicator sequence or degradation indicator sequence of the sample bearing in the 60-second time period from 16:00:00 to 16:01:00, which is marked with the bearing life of the sample bearing, is as shown in Table 2 below:

[0086]

[0087] Table 2 Example of part of the health indicator sequence or degradation indicator sequence marked with bearing life for a sample bearing

[0088] Please refer to Table 2 above. The "features" in Table 2 above may refer to feature data in a health indicator sequence or degradation indicator sequence obtained after the above-mentioned state parameters have been processed, including but not limited to data noise reduction, data gap filling, and feature extraction. It is understandable that the specific type, form, and quantity of the feature data in the "features" may depend on the type, quantity, characteristics, etc. of the state parameters, and are not specifically limited here. The duration data in the above-mentioned "bearing life" is obtained by subtracting the normal working time of the sample bearing from the actual working time. In this embodiment, the above-mentioned bearing life of 8:00:00 can be obtained by calculating 24:00:00-16:00:00=8:00:00. Of course, the normal working time of the sample bearing being 24 hours is only an assumption. The specific working time of the sample bearing should be based on actual conditions and is not a fixed 24 hours.

[0089] The bearing life prediction model training method provided in the embodiment of the present application includes: collecting multiple maximum amplitudes in the vibration signal of the sample bearing during the operation of the sample bearing. According to the multiple maximum amplitudes in the vibration signal of the sample bearing and the preset amplitude threshold, the operating state and corresponding state parameters of the sample bearing are determined, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage. According to the operating state of the sample bearing and the corresponding state parameters, a health indicator sequence and a degradation indicator sequence are obtained. A mixed kernel function model is determined according to at least two preset kernel functions, and the mixed kernel function model is trained by the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model, which is used to predict the bearing life. The method compares multiple maximum amplitudes of the vibration signals of the sample bearings collected during their operation with preset amplitude thresholds in sequence to divide the operating status of the sample bearings into a healthy stage and a degraded stage, and divides the collected status information of the sample bearings into status information corresponding to the healthy stage and status information corresponding to the degraded stage, namely, a health indicator sequence and a degradation indicator sequence. This method trains a hybrid kernel function model containing at least two preset kernel functions through the health indicator sequence and the degradation indicator sequence, and uses the trained hybrid kernel function model as a bearing life prediction model to predict the bearing life, thereby improving the accuracy and robustness of the bearing remaining life prediction.

[0090] Figure 2 A flow chart of a bearing life prediction model training method provided in another embodiment of the present application is provided. Figure 2 , optionally, in the above Figure 1 Based on the embodiment, during the operation of the sample bearing, collecting multiple maximum amplitude values ​​of the vibration signal of the sample bearing may include:

[0091] During the operation of the sample bearing, a plurality of maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle are collected according to a preset window length.

[0092] Exemplarily, the above-mentioned preset window length may refer to the time span of collecting multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing in the current cycle. Taking the content in Table 1 of the above-mentioned embodiment as an example, the preset window length corresponding to Table 1 may be 80 seconds (16:00:00~16:01:20). Therefore, the multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing in the current cycle are the multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing in the time period of 80 seconds (16:00:00~16:01:20). The above-mentioned sample bearing in the current cycle is collected according to the preset window length. The multiple maximum amplitudes in the vibration signal of the bearing may refer to those collected in the form of a sliding window according to a preset window length. Taking the collection of the multiple maximum amplitudes in the vibration signal of the sample bearing at a frequency of once every 10 seconds as an example, the multiple maximum amplitudes in the vibration signal of the sample bearing in the next cycle of the current cycle may be the multiple maximum amplitudes in the vibration signal of the sample bearing in the time period of 16:00:10 to 16:01:30, that is, there is an overlap in the maximum amplitudes between each cycle, and the difference between each cycle is the collection time difference corresponding to the two adjacent maximum amplitudes.

[0093] It should be noted that the above content is only an example of a specific collection method for collecting multiple maximum amplitude values ​​in the vibration signal of the above sample bearing in the current period according to a preset window length. The actual collection method may be the same as or different from the above example.

[0094] On this basis, determining the operating state and corresponding state parameters of the sample bearing according to the plurality of maximum amplitudes and the preset amplitude threshold in the vibration signal of the sample bearing may include:

[0095] S201. Calculate the mean and standard deviation of the plurality of maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle.

[0096] Exemplarily, the calculation of the average of the plurality of maximum amplitudes can be achieved, for example, by the following formula:

[0097]

[0098] Among them, the above u (t) is the average of the multiple maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle, the above N is the number of the multiple maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle, and the above X i is the i-th maximum amplitude. It can be understood that N≥i≥1.

[0099] The above calculation of the standard deviation of the plurality of maximum amplitudes can be achieved, for example, by the following formula:

[0100]

[0101] Among them, the above σ (t) is the standard deviation of multiple maximum amplitudes mentioned above in the vibration signal of the sample bearing in the current cycle.

[0102] S202 , determining whether each of the maximum amplitudes is abnormal based on the mean, the standard deviation, and the preset amplitude threshold.

[0103] For example, it is assumed that the maximum amplitude values ​​in the vibration signal of the sample bearing in the current period form a maximum amplitude sequence X MA (t), then the above-mentioned determination of whether each of the above-mentioned maximum amplitudes is abnormal according to the above-mentioned mean, the above-mentioned standard deviation and the above-mentioned preset amplitude threshold can be achieved, for example, by the following formula:

[0104] |X MA (t)-μ(t)|≥3σ(t),

[0105] That is, multiple maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle are combined into a maximum amplitude sequence X Mp Each maximum value in (t) is subtracted from the above u (t) Then compare it with 3σ(t). If the maximum amplitude is ≥ 3σ(t), the maximum amplitude is determined to be an abnormal value.

[0106] Of course, the above contents are only possible examples. The specific method of determining whether each of the above maximum values ​​is abnormal can be adjusted according to the actual situation and is not limited to the above examples.

[0107] S203. If the maximum number of consecutive abnormalities in the current cycle reaches a preset number of consecutive abnormalities, then starting from the current cycle, the operating state of the sample bearing is the degradation stage. Before the current cycle, the operating state of the sample bearing was the healthy stage. The state parameters corresponding to the healthy stage and the degradation stage are recorded respectively.

[0108] For example, the preset number of consecutive abnormalities may be, for example, 3, 4, 5, etc., but is not limited thereto. Specifically, a maximum amplitude sequence X may be formed based on a plurality of maximum amplitudes in the vibration signal of the sample bearing in the current period. MAThe maximum amplitude number in (t) is adjusted and determined. If the maximum amplitude number of consecutive abnormalities in the current cycle reaches the preset number of consecutive abnormalities, then from the current cycle onwards, the operating state of the sample bearing is the degradation stage. Before the current cycle, the operating state of the sample bearing is the healthy stage. The state parameters corresponding to the healthy stage and the degradation stage are recorded respectively. For example, the maximum amplitude sequence X composed of multiple maximum amplitudes in the vibration signal of the sample bearing in the current cycle can be referred to. MA If the number of maximum amplitudes continuously judged as abnormal values ​​in (t) reaches the above-mentioned preset number of continuous abnormalities, the starting time point of the current cycle is taken as the starting point for the sample bearing to enter the degradation stage, that is, the stage from the start of the sample bearing to the starting time point of the above-mentioned current cycle is the healthy stage of the sample bearing, and the stage from the starting time point of the above-mentioned current cycle to the stop of normal operation of the above-mentioned sample bearing is the degradation stage of the sample bearing.

[0109] The bearing life prediction model training method provided in the embodiment of the present application includes: during the operation of the sample bearing, collecting multiple maximum amplitudes in the vibration signal of the sample bearing in the current cycle according to a preset window length. Based on the multiple maximum amplitudes in the vibration signal of the sample bearing in the current cycle, calculating the mean and standard deviation of the multiple maximum amplitudes. According to the above mean, the above standard deviation and the above preset amplitude threshold, it is judged in turn whether each of the above maximum amplitudes is abnormal. If the number of the above maximum amplitudes that are continuously abnormal in the current cycle reaches the preset number of continuous abnormalities, then from the current cycle onwards, the above operating state of the above sample bearing is the above degradation stage, and before the current cycle, the above operating state of the above sample bearing is the above healthy stage, and the above state parameters corresponding to the above healthy stage and the above degradation stage are recorded respectively. The method calculates the mean and standard deviation of multiple maximum amplitudes in the vibration signal of the sample bearing in the current cycle, and determines whether each of the maximum amplitudes in the vibration signal of the sample bearing in the current cycle is abnormal through the mean and standard deviation, thereby dividing the healthy stage and the degraded stage of the sample bearing's operating status, thereby facilitating a more accurate division of the state parameters of the sample bearing, so as to train the model more accurately and improve the training effect of the bearing life prediction model.

[0110] Optionally, based on the above embodiment, after determining whether each of the maximum amplitudes is abnormal according to the mean, the standard deviation, and the preset amplitude threshold, the method may further include:

[0111] If the number of the maximum amplitude values ​​of the consecutive anomalies in the current cycle does not reach the preset number of consecutive anomalies, multiple maximum amplitude values ​​in the vibration signal of the sample bearing in the next cycle are collected according to the preset window length.

[0112] For example, if the number of the maximum amplitude values ​​of the continuous abnormalities in the current cycle does not reach the preset number of continuous abnormalities, the principle of collecting multiple maximum amplitude values ​​in the vibration signal of the sample bearing in the next cycle according to the preset window length can be, for example, the same as the above Figure 2 The same applies to the example in the embodiment of “collecting multiple maximum amplitude values ​​in the vibration signal of the sample bearing in the form of a sliding window according to a preset window length”.

[0113] It can be understood that the above-mentioned “collecting multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing in the form of a sliding window according to a preset window length” can, for example, refer to the moment when the sample bearing starts working, that is, gradually collecting multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing in the form of a sliding window according to the preset window length. On this basis, if the number of the above-mentioned maximum amplitude values ​​of consecutive abnormalities in the current cycle does not reach the above-mentioned preset number of consecutive abnormalities, it can be said that as of the current cycle, the operating status of the sample bearing is still in a healthy stage and has not yet entered a degradation stage. At this time, multiple maximum amplitude values ​​in the vibration signal of the above-mentioned sample bearing in the next cycle are collected according to the above-mentioned preset window length, and the above-mentioned steps are repeated. Figure 2 In the judgment process of the embodiment, until the maximum number of consecutive abnormalities within a certain period reaches a preset number of consecutive abnormalities, it can be indicated that the operating state of the sample bearing has entered the degradation stage, and the classification of the operating state of the sample bearing is completed.

[0114] Furthermore, based on any of the above embodiments, obtaining the health indicator sequence and the degradation indicator sequence according to the operating state of the sample bearing and the corresponding state parameters may include:

[0115] Multimodal data feature extraction is adopted to perform feature extraction on the state parameters corresponding to the healthy stage and the degradation stage respectively, to obtain the health indicator sequence and the degradation indicator sequence.

[0116] For example, since the state parameters corresponding to the health stage and the degradation stage can include multi-type and multi-dimensional data, feature extraction of the state parameters corresponding to the health stage and the degradation stage can extract more comprehensive and stable feature data compared to a single data source, thereby improving the data quality of the health indicator series and the degradation indicator series. It will be understood that the specific data type, format, and quantity of the feature data extracted using multimodal data features may depend on the type, quantity, and characteristics of the state parameters, and are not specifically limited here.

[0117] Figure 3 A flow chart of a bearing life prediction model training method according to another embodiment of the present application is shown as follows: Figure 3 As shown in the above Figure 1 or Figure 2 Based on the embodiment, before determining the hybrid kernel function model according to at least two preset kernel functions, the method may further include:

[0118] The output matrix of the hidden layer nodes is determined according to the extreme learning machine activation function, and the extreme learning machine function model is determined according to the output matrix of the hidden layer nodes and the preset weights.

[0119] Exemplarily, the extreme learning machine activation function may be, for example, the following function:

[0120] g(w i x i +b i ),

[0121] Among them, the above w i is the weight of the i-th node in the input layer, the above b i Represents the bias value of the i-th node in the hidden layer. The above x i The sample feature vector representing the input of the i-th node is the feature data in the health indicator sequence and the degradation indicator sequence obtained by performing feature extraction on the state parameters corresponding to the health stage and the degradation stage respectively.

[0122] On this basis, the output matrix of the hidden layer nodes determined by the extreme learning machine activation function can be the following matrix:

[0123]

[0124] Among them, the above H(x) is the output matrix of the above hidden layer nodes for the sample feature vector x. The output matrix of the above hidden layer nodes has n rows and m columns, w1~w m With the above w i Similarly, x1~x2 are the weights of the 1st to mth nodes in the input layer, and similarly, x1~x3 are the weights of the 1st to mth nodes in the input layer, and similarly, x1~x4 are the weights of the 1st to mth nodes in the input layer, and similarly, x1~x5 are the weights of the 1st to mth nodes in the input layer, and similarly, x1~x6 are the weights of the 1st to mth n are the sample feature vectors input to the 1st to nth nodes, b1~b m are the bias values ​​of the 1st to mth nodes in the hidden layer respectively.

[0125] On this basis, the above extreme learning machine function model can be expressed as the following function:

[0126] f(x)=βH(x)=Hβ=HH T (HH T +I / C) -1 T,

[0127] Among them, the above β is the weight between the hidden layer and the output layer, that is, the above preset weight, the above H is the hidden layer output matrix, the above T is the target output, the above I is the unit matrix, the above C is the regularization coefficient, I and C are both adjustable preset values, and the above HH T is the kernel matrix of the extreme learning machine function model, which can also be expressed as W ELM .

[0128] The above-mentioned determination of the hybrid kernel function model based on at least two preset kernel functions may include:

[0129] S301: Determine a kernel extreme learning machine function model according to a first preset kernel function and the extreme learning machine function model.

[0130] Based on the above examples, we can know K(x i ,x j ) is the first preset kernel function mentioned above, which represents the sample feature vector x of the i-th node input i and the sample feature vector x of the j-th node input j The inner product in the hidden space.

[0131] Then the above kernel extreme learning machine function model can be expressed as:

[0132]

[0133] Among them, the above y is the new input sample feature vector, and the above y1~y N is the sample feature vector in the training set, and the above K(y,y1)~K(y,y N ) represent the similarity measure between the new input sample feature vector and each sample feature vector in the training set in the kernel space.

[0134] S302: Determine the hybrid kernel function model according to the second preset kernel function and the kernel extreme learning machine function model.

[0135] Exemplarily, the second preset kernel function may refer to a mixed kernel function obtained by mixing at least two preset kernel functions in the aforementioned embodiment.

[0136] Taking the above second preset kernel function as an example, which is a hybrid kernel function obtained by mixing two preset kernel functions, and the two preset kernel functions are a local kernel function RBF (Radial Basis Function) and a global kernel function Poly (Polynomial Kernel Function), respectively:

[0137] The local kernel function RBF can be, for example, as follows:

[0138] K RBF (x i ,x j )=exp(-||x i -x j || 2 / a 2 ),

[0139] The global kernel function Poly can be as follows:

[0140] K Ploy (x i ,x j )=(x i ,x j +z) d ,

[0141] Among them, the above a, d, and z are kernel function parameters of the local kernel function RBF and the global kernel function Poly, and are adjustable preset values.

[0142] Then the second preset kernel function after mixing the two preset kernel functions can be expressed as:

[0143] K H (x i ,x j )=sK RBF (x i ,x j )+(1-s)K Ploy (x i ,x j ),

[0144] Among them, s is the weight of the mixture of two preset kernel functions, which is an adjustable preset value.

[0145] By bringing the second preset kernel function into the above-mentioned kernel extreme learning machine function model, the above-mentioned hybrid kernel function model can be obtained.

[0146] It should be noted that the above content is only one possible method for determining a hybrid kernel function model. The actual method for determining a hybrid kernel function model may be different from the content in the above example, and is not specifically limited here.

[0147] The bearing life prediction model training method provided in an embodiment of the present application includes: determining the output matrix of the hidden layer nodes based on the extreme learning machine activation function, and determining the extreme learning machine function model based on the output matrix of the hidden layer nodes and preset weights. Determining the core extreme learning machine function model based on a first preset kernel function and the extreme learning machine function model. Determining the above-mentioned hybrid kernel function model based on a second preset kernel function and the above-mentioned core extreme learning machine function model. This method determines the hybrid kernel function model by using a kernel function that is a mixture of at least two preset kernel functions on the basis of the core extreme learning machine function model, thereby achieving a layer-by-layer determination from the extreme learning machine function model, to the core extreme learning machine function model, and then to the hybrid kernel function model, thereby better introducing the hybrid kernel function into the extreme learning machine function model and improving the determination efficiency and reliability of the hybrid kernel function model.

[0148] In addition, in the above Figure 1 or Figure 2 Based on the embodiment, the above-mentioned training of the hybrid kernel function model by using the above-mentioned health indicator sequence and the above-mentioned degradation indicator sequence to obtain the bearing life prediction model may include:

[0149] The above health indicator sequence is used as a training set, and the above degradation indicator sequence is used as a test set to train the above hybrid kernel function model.

[0150] For example, since different feature data have different sensitivities to the trend of performance degradation, the health indicator sequence and the degradation indicator sequence are used as the training set and the test set respectively. The respective characteristics of the health indicator sequence and the degradation indicator sequence can be used to better train the hybrid kernel function model to improve the prediction performance of the hybrid kernel function model.

[0151] After the hybrid kernel function model training is completed, the fitting curve and error corresponding to the trained hybrid kernel function model are calculated. If the fitting curve and the error both meet the preset training judgment conditions, the trained hybrid kernel function model is used as the bearing life prediction model.

[0152] Exemplarily, the above-mentioned preset training judgment conditions may mean that the degree of overlap between the above-mentioned fitting curve and the preset optimal fitting curve reaches, for example, more than 95%, and the error between the prediction result of the above-mentioned bearing life prediction model and the actual life is less than, for example, 10 hours. Of course, the actual preset training judgment conditions can be adjusted and determined according to actual conditions, and are not limited to the above-mentioned examples.

[0153] Figure 4 This is a flow chart of a bearing life prediction model training method provided by another embodiment of the present application. Please refer to Figure 4 Based on the above embodiment, if the fitting curve and / or the error do not meet the preset training judgment conditions, the method may further include:

[0154] S401 . Randomly generate a set of candidate solutions, where the set of candidate solutions includes multiple groups of candidate solutions, and each group of candidate solutions represents a set of hyperparameters of the hybrid kernel function model.

[0155] Exemplarily, the above-mentioned randomly generated candidate solution set can be generated by a computer or other device according to preset conditions. The hyperparameters of the above-mentioned hybrid kernel function model may, for example, refer to the kernel function parameters of the local kernel function RBF and the global kernel function Poly including but not limited to the above-mentioned a, d, z, etc., and are not specifically limited here.

[0156] S402: update and iterate the candidate solution set according to a preset optimization algorithm, and obtain a set of optimal candidate solutions.

[0157] For example, the preset optimization algorithm may be a simulated annealing algorithm or a frost optimization algorithm. Taking the frost optimization algorithm as an example:

[0158] Then the process of updating and iterating the above candidate solution set by the frost optimization algorithm can be expressed as:

[0159]

[0160] Among them, the above Updates the new position of the particle for the frost optimization algorithm. q and w represent the wth particle of the qth frost factor. R bestw is the wth particle with the best frost factor in the frost population R. r1 is an algorithm parameter, which is a random number that meets r1∈(-1, 1). v is an environmental factor, which is an adjustable preset value. h is the adhesion, which is used to control the center distance between two frost particles and is an adjustable preset value. E is the adhesion coefficient, which affects the condensation probability of the frost factor and increases with the number of iterations. r2 is an algorithm parameter, which is a random number that meets r2∈(0, 1). Together with E, it controls whether the particles condense, that is, whether the particle positions are updated. θ is the phase angle that changes with the number of iterations and is used to adjust the search range of the frost optimization algorithm. Assuming the number of iterations is t, then:

[0161]

[0162] Among them, T is the maximum number of iterations, which is an adjustable preset value.

[0163] The above environmental factor v can be expressed as:

[0164]

[0165] The above w is the number of segments of the step function, which is an adjustable preset value. For example, w can be preset to 5, but is not limited thereto.

[0166] The above adhesion coefficient E can be expressed as:

[0167]

[0168] The frost optimization algorithm simulates the frost growth process and proposes a frost punching mechanism. This mechanism is used to replace frost particles. The frost punching mechanism can be expressed as:

[0169]

[0170] Among them, r3 is the algorithm parameter, which is a random number that meets r3∈(-1,1). normr Indicates the normalized value of the current frost factor fitness value.

[0171] The frost optimization algorithm will output the above optimal candidate solution after updating and iteration.

[0172] S403 , optimizing the hyperparameters of the hybrid kernel function model according to the optimal candidate solution, and using the optimized hybrid kernel function model as a bearing life prediction model.

[0173] Since each of the above-mentioned candidate solutions represents a set of hyperparameters of the above-mentioned hybrid kernel function model, the hyperparameters of the hybrid kernel function model represented by the above-mentioned optimal candidate solution can be used to directly replace the original hyperparameters of the hybrid kernel function model. The hybrid kernel function model after replacing the hyperparameters is the optimized hybrid kernel function model, which can be used as a bearing life prediction model for bearing life prediction.

[0174] To understand the complete solution, Figure 5 A schematic diagram of the overall process of a bearing life prediction model training method provided in one embodiment of the present application.

[0175] Please refer to Figure 5 When the sample bearing starts working, multiple maximum amplitudes of the vibration signal of the sample bearing are collected, and at the same time, the state parameters of the sample bearing are collected until the sample bearing stops working normally. Then, based on the multiple maximum amplitudes of the sample bearing during normal working, the operating state of the sample bearing during normal working is divided into a healthy stage and a degraded stage. Then, based on the state parameters corresponding to the healthy stage and the state parameters corresponding to the degraded stage of the sample bearing, a health index sequence and a degradation index sequence are obtained.

[0176] The health indicator sequence and degradation indicator sequence are used to train a hybrid kernel function model. The hybrid kernel function model is constructed by first determining an extreme learning machine function model, introducing a first preset kernel function into the extreme learning machine function model to obtain a kernel extreme learning machine function model, and then introducing a second preset kernel function into the kernel extreme learning machine function model to obtain a hybrid kernel function model. The second preset kernel function may be a hybrid kernel function comprising at least two kernel functions.

[0177] Finally, the trained hybrid kernel function model is used as a bearing life prediction model, or the trained hybrid kernel function model is optimized and used as a bearing life prediction model to predict bearing life.

[0178] It should be noted that Figure 5 The contents in the embodiment are only a general description for the convenience of understanding the complete scheme. The specific details of the steps corresponding to the above contents can be referred to Figures 1 to 4 The above content is only a possible example of the overall process of bearing life prediction model training. The actual overall process of bearing life prediction model training can be the same as Figure 5 The contents shown are different and are not Figure 5 The process shown is limited.

[0179] Figure 6 This is a flow chart of a bearing life prediction method provided in one embodiment of the present application. This method can be applied to the bearing life prediction model trained by the bearing life prediction model training method in the above embodiment. Please refer to Figure 6 , the method may include:

[0180] S601: Acquire state parameters of a bearing to be predicted.

[0181] The method for collecting the state parameters of the bearing to be predicted may be consistent with the method for collecting the state parameters of the sample bearing in the above embodiment, and will not be described in detail here.

[0182] S602: A bearing life prediction model is obtained by training using the bearing life prediction model training method in the above embodiment, and a bearing life prediction result is obtained based on the state parameters of the bearing to be predicted.

[0183] The bearing life prediction result can be used to optimize maintenance plans, reduce mechanical equipment downtime and maintenance costs, but is not limited to this.

[0184] Figure 7 This is a structural diagram of a bearing life prediction model training device provided in an embodiment of the present application. The bearing life prediction model training device can execute the above-mentioned bearing life prediction model training method. The device can be integrated into the above-mentioned computer or other equipment with computing and processing functions, such as Figure 7As shown, the device may include:

[0185] The first acquisition module 710 is configured to acquire a plurality of maximum amplitudes of the vibration signal of the sample bearing during operation of the sample bearing.

[0186] The determination module 720 is used to determine the operating state and corresponding state parameters of the sample bearing based on the multiple maximum amplitudes and preset amplitude thresholds in the vibration signal of the sample bearing, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage.

[0187] The acquisition module 730 is configured to acquire a health indicator sequence and a degradation indicator sequence according to the operating status of the sample bearing and the corresponding status parameters.

[0188] The training module 740 is used to determine a hybrid kernel function model based on at least two preset kernel functions, and train the hybrid kernel function model through the above health indicator sequence and the above degradation indicator sequence to obtain a bearing life prediction model, which is used to predict the bearing life.

[0189] The bearing life prediction model training method provided in the embodiment of the present application includes: collecting multiple maximum amplitudes in the vibration signal of the sample bearing during the operation of the sample bearing. According to the multiple maximum amplitudes in the vibration signal of the sample bearing and the preset amplitude threshold, the operating state and corresponding state parameters of the sample bearing are determined, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage. According to the operating state of the sample bearing and the corresponding state parameters, a health indicator sequence and a degradation indicator sequence are obtained. A mixed kernel function model is determined according to at least two preset kernel functions, and the mixed kernel function model is trained by the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model, which is used to predict the bearing life. The method compares multiple maximum amplitudes of the vibration signals of the sample bearings collected during their operation with preset amplitude thresholds in sequence to divide the operating status of the sample bearings into a healthy stage and a degraded stage, and divides the collected status information of the sample bearings into status information corresponding to the healthy stage and status information corresponding to the degraded stage, namely, a health indicator sequence and a degradation indicator sequence. This method trains a hybrid kernel function model containing at least two preset kernel functions through the health indicator sequence and the degradation indicator sequence, and uses the trained hybrid kernel function model as a bearing life prediction model to predict the bearing life, thereby improving the accuracy and robustness of the bearing remaining life prediction.

[0190] Optionally, the first acquisition module 710 is specifically configured to acquire a plurality of maximum amplitudes in the vibration signal of the sample bearing in a current cycle according to a preset window length during operation of the sample bearing.

[0191] Determination module 720 is specifically configured to calculate a mean and standard deviation of the maximum amplitude values ​​in the vibration signal of the sample bearing within the current cycle. Determine whether each maximum amplitude value is abnormal based on the mean, standard deviation, and the preset amplitude threshold. If the number of consecutively abnormal maximum amplitude values ​​within the current cycle reaches a preset number of consecutive abnormalities, the operating state of the sample bearing is determined to be the degraded stage starting from the current cycle. Prior to the current cycle, the operating state of the sample bearing was the healthy stage. The state parameters corresponding to the healthy and degraded stages are recorded.

[0192] Optionally, the first acquisition module 710 can also be used to collect multiple maximum amplitudes in the vibration signal of the sample bearing in the next cycle according to the preset window length if the number of the maximum amplitudes of the continuous abnormalities in the current cycle does not reach the preset number of continuous abnormalities.

[0193] Optionally, the acquisition module 730 is specifically configured to use multimodal data feature extraction to perform feature extraction on the state parameters corresponding to the healthy stage and the degradation stage, respectively, to obtain the health indicator sequence and the degradation indicator sequence.

[0194] Optionally, the training module 740 may also be configured to determine an output matrix of hidden layer nodes according to an extreme learning machine activation function, and determine an extreme learning machine function model according to the output matrix of the hidden layer nodes and preset weights.

[0195] The training module 740 is specifically configured to determine a kernel extreme learning machine function model based on a first preset kernel function and the extreme learning machine function model, and to determine the hybrid kernel function model based on a second preset kernel function and the kernel extreme learning machine function model.

[0196] Optionally, the training module 740 is specifically configured to train the hybrid kernel function model using the health indicator sequence as a training set and the degradation indicator sequence as a test set. A fitting curve and error corresponding to the trained hybrid kernel function model are calculated. If both the fitting curve and the error meet preset training judgment conditions, the trained hybrid kernel function model is used as the bearing life prediction model.

[0197] Optionally, the apparatus may further include an optimization module configured to randomly generate a set of candidate solutions, wherein the set includes multiple groups of candidate solutions, each group of candidate solutions representing a set of hyperparameters of the hybrid kernel function model. The set of candidate solutions is iteratively updated according to a preset optimization algorithm to obtain a set of optimal candidate solutions. The hyperparameters of the hybrid kernel function model are optimized based on the optimal candidate solutions, and the optimized hybrid kernel function model is used as the bearing life prediction model.

[0198] The above-mentioned device is used to execute the method provided in the above-mentioned embodiment. Its implementation principle and technical effect are similar and will not be repeated here.

[0199] Figure 8 The embodiment of the present application provides a bearing life prediction device, which can execute the above-mentioned bearing life prediction method. The device can be integrated into the above-mentioned computer or other equipment with computing and processing functions, such as Figure 8 As shown, the device includes:

[0200] The second acquisition module 810 is used to acquire the state parameters of the bearing to be predicted.

[0201] The prediction module 820 is configured to obtain a bearing life prediction result based on the state parameters of the bearing to be predicted by using the bearing life prediction model training device in the above embodiment to train the obtained bearing life prediction model.

[0202] The above-mentioned device is used to execute the method provided in the above-mentioned embodiment. Its implementation principle and technical effect are similar and will not be repeated here.

[0203] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may be a device with computing and processing functions such as the above-mentioned computer, server, etc. Figure 9 As shown, the device 900 includes:

[0204] Processor 910 , storage medium 920 and bus 930 . Processor 910 and storage medium 920 are communicatively connected via bus 930 .

[0205] The storage medium 920 stores machine-readable instructions executable by the processor 910. When the electronic device is running, the processor 910 executes the above-mentioned machine-readable instructions to perform the above-mentioned bearing life prediction model training method or bearing life prediction method.

[0206] It should be understood that Figure 9 The structure shown is only a schematic diagram of the structure of the electronic device. The electronic device may also include Figure 9 More or fewer components than shown, or with Figure 9 Different configurations shown. Figure 9 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0207] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program can be executed by a processor, it can implement the bearing life prediction model training method or the bearing life prediction method described in the above method embodiment.

[0208] The computer-readable storage medium can be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read-Only Memory), an EPROM, a hard disk, or a ROM. Alternatively, the computer-readable storage medium includes a non-transitory computer-readable storage medium. The computer-readable storage medium has storage space for program codes that execute any of the method steps of the above method. These program codes can be read from or written into one or more computer program products. The program code can be compressed, for example, in an appropriate form.

[0209] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the devices, methods, and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0210] In addition, the functional modules in each embodiment of the present 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.

[0211] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0212] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. All equivalent structural transformations made by using the contents of the present application description and drawings under the inventive concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A bearing life prediction model training method, characterized in that: include: During the operation of the sample bearing, collecting a plurality of maximum amplitudes of the vibration signal of the sample bearing; Determining an operating state and corresponding state parameters of the sample bearing according to a plurality of maximum amplitudes and a preset amplitude threshold in the vibration signal of the sample bearing, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage; Acquire a health indicator sequence and a degradation indicator sequence according to the operating state of the sample bearing and the corresponding state parameters; A mixed kernel function model is determined according to at least two preset kernel functions, and the mixed kernel function model is trained by the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model, which is used to predict the bearing life.

2. The method according to claim 1, characterized in that During the operation of the sample bearing, collecting a plurality of maximum amplitude values ​​of the vibration signal of the sample bearing includes: During the working process of the sample bearing, a plurality of maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle are collected according to a preset window length; The step of determining the operating state and corresponding state parameters of the sample bearing according to the plurality of maximum amplitudes and a preset amplitude threshold in the vibration signal of the sample bearing comprises: Calculating a mean and a standard deviation of the plurality of maximum amplitude values ​​in the vibration signal of the sample bearing in the current cycle; determining whether each of the maximum amplitudes is abnormal according to the mean, the standard deviation, and the preset amplitude threshold; If the maximum number of consecutive abnormalities in the current cycle reaches a preset number of consecutive abnormalities, then starting from the current cycle, the operating state of the sample bearing is the degradation stage. Before the current cycle, the operating state of the sample bearing is the healthy stage, and the state parameters corresponding to the healthy stage and the degradation stage are recorded respectively.

3. The method according to claim 2, characterized in that After determining whether each of the maximum amplitudes is abnormal according to the mean, the standard deviation, and the preset amplitude threshold, the method further includes: If the number of the maximum amplitudes of the consecutive anomalies in the current cycle does not reach the preset number of consecutive anomalies, multiple maximum amplitudes in the vibration signal of the sample bearing in the next cycle are collected according to the preset window length.

4. The method according to claims 1-3, characterized in that The acquiring of a health indicator sequence and a degradation indicator sequence according to the operating state of the sample bearing and the corresponding state parameters includes: Multimodal data feature extraction is adopted to perform feature extraction on the state parameters corresponding to the healthy stage and the degradation stage respectively, to obtain the health indicator sequence and the degradation indicator sequence.

5. The method according to any one of claims 1 to 3, characterized in that Before determining the hybrid kernel function model according to at least two preset kernel functions, the method further includes: Determine the output matrix of the hidden layer nodes according to the extreme learning machine activation function, and determine the extreme learning machine function model according to the output matrix of the hidden layer nodes and preset weights; The step of determining a hybrid kernel function model according to at least two preset kernel functions includes: Determine a kernel extreme learning machine function model according to a first preset kernel function and the extreme learning machine function model; The hybrid kernel function model is determined according to a second preset kernel function and the kernel extreme learning machine function model.

6. The method according to any one of claims 1 to 3, characterized in that The training of the hybrid kernel function model by using the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model includes: Using the health indicator sequence as a training set and the degradation indicator sequence as a test set, training the hybrid kernel function model; The fitting curve and error corresponding to the trained hybrid kernel function model are calculated. If both the fitting curve and the error meet the preset training judgment conditions, the trained hybrid kernel function model is used as the bearing life prediction model.

7. The method according to claim 6, characterized in that If the fitting curve and / or the error do not meet the preset training judgment condition, the method further includes: Randomly generating a set of candidate solutions, wherein the set of candidate solutions includes multiple groups of candidate solutions, each group of candidate solutions represents a set of hyperparameters of the hybrid kernel function model; Iterate the candidate solution set according to a preset optimization algorithm and obtain a set of optimal candidate solutions; The hyperparameters of the hybrid kernel function model are optimized according to the optimal candidate solution, and the optimized hybrid kernel function model is used as a bearing life prediction model.

8. A bearing life prediction method, characterized in that: include: Collect and obtain the state parameters of the bearing to be predicted; A bearing life prediction model obtained by training using the method according to any one of claims 1 to 7 is used to obtain a bearing life prediction result based on the state parameters of the bearing to be predicted.

9. A bearing life prediction model training device, characterized in that: include: A first acquisition module is used to acquire a plurality of maximum amplitudes of the vibration signal of the sample bearing during operation of the sample bearing; a determination module, configured to determine an operating state and corresponding state parameters of the sample bearing based on a plurality of maximum amplitudes and a preset amplitude threshold in the vibration signal of the sample bearing, wherein the operating state of the sample bearing includes: a healthy stage and a degraded stage; an acquisition module, configured to acquire a health indicator sequence and a degradation indicator sequence according to the operating state of the sample bearing and the corresponding state parameters; A training module is used to determine a mixed kernel function model based on at least two preset kernel functions, and train the mixed kernel function model through the health indicator sequence and the degradation indicator sequence to obtain a bearing life prediction model, which is used to predict the bearing life.

10. A bearing life prediction device, characterized in that: include: The second acquisition module is used to acquire the state parameters of the bearing to be predicted; A prediction module is used to obtain a bearing life prediction model trained by the bearing life prediction model training device according to claim 9, and obtain a bearing life prediction result based on the state parameters of the bearing to be predicted.

Images