Rotating mechanical equipment bearing fault prediction method

Abstract

The invention relates to a rotating mechanical equipment bearing fault prediction method, and relates to the technical field of bearing fault prediction, and the method comprises the steps: carrying out the interaction of bearing fault big data based on the scene feature information of a target scene, and obtaining a fault record set; performing clustering division on the fault record set based on fault categories, and analyzing a clustering division result to obtain a multi-working-condition prediction reference index set; according to the multi-working-condition prediction reference index set, performing prediction resource allocation of the multiple working conditions, and obtaining a prediction inclination coefficient set of the multiple working conditions; in combination with the fault record set and the prediction inclination coefficient set, a bearing fault prediction model is constructed and trained, the bearing fault prediction model comprises a plurality of prediction sub-networks, and the plurality of prediction sub-networks have working condition marks; and performing fault prediction based on the bearing real-time monitoring data through the bearing fault prediction model to obtain a fault prediction result. Therefore, the technical effects of online monitoring, complete monitoring and good early warning capability are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault detection, and in particular to a method for predicting bearing faults of rotating mechanical equipment. Background Art

[0002] Rotating machinery and equipment are widely used in various industries, including manufacturing, transportation, aviation, energy and other fields. As a key component of rotating machinery, the operating status of bearings directly affects the safety, reliability and operating efficiency of the equipment. Accurately and timely predicting bearing failures, especially early detection of potential problems during equipment operation, is an important task to improve equipment reliability, reduce maintenance costs and ensure production safety.

[0003] At present, most common bearing fault prediction methods rely on traditional signal processing technology and basic model analysis methods, such as vibration analysis, temperature monitoring and noise monitoring. These methods analyze the physical signals of the equipment to determine whether the equipment has a fault. There are technical problems such as reliance on expert experience, lack of adaptability, and poor performance when dealing with complex problems with multiple working conditions and multiple fault types. Summary of the invention

[0004] The present invention aims to solve the technical problems in the prior art that the prior art relies on expert experience, lacks adaptability, and has poor performance when dealing with complex problems with multiple working conditions and multiple fault types, and provides a method for predicting bearing faults of rotating mechanical equipment.

[0005] The technical solution of the present invention to solve the above technical problems is as follows: In a first aspect, the present invention provides a method for predicting bearing faults of rotating mechanical equipment.

[0006] Based on the scene feature information of the target scene, interactive bearing fault big data is used to obtain the fault record set.

[0007] The fault record set is clustered based on the fault category, and frequency dimension analysis, intensity dimension analysis and breadth dimension analysis are performed based on the clustering result to obtain a multi-operating condition prediction reference index set.

[0008] According to the multi-operating condition prediction reference indicator set, prediction resource allocation for multiple operating conditions is performed to obtain a prediction tilt coefficient set for multiple operating conditions, wherein the prediction tilt coefficient set includes sample tilt coefficients, call tilt coefficients and scale tilt coefficients for multiple operating conditions.

[0009] Combining the fault record set with the predicted tilt coefficient set, a bearing fault prediction model is constructed and trained, wherein the bearing fault prediction model includes a plurality of prediction sub-networks, and a plurality of the prediction sub-networks have operating condition labels.

[0010] The bearing fault prediction model is used to perform fault prediction based on the real-time monitoring data of the bearing and obtain a fault prediction result.

[0011] In a feasible implementation, the fault record set includes a source fault record set based on historical fault records of a target scenario and a homologous fault record set based on historical fault records of a homologous scenario of the target scenario.

[0012] In a feasible implementation, the fault record set is clustered based on the fault category, and frequency dimension analysis, intensity dimension analysis and breadth dimension analysis are performed based on the clustering result to obtain a multi-condition prediction reference index set, including: Taking multiple fault categories as the division targets, clustering and division of the fault record set is performed to generate multiple working condition fault record sets.

[0013] Traverse the plurality of working condition fault record sets, count the fault record frequencies, perform frequency dimension analysis, and obtain a frequency reference index set.

[0014] Traverse the multiple working condition fault record sets, obtain typical prediction accuracy rates of multiple fault categories for strength dimension, and obtain a strength reference indicator set.

[0015] Traverse the plurality of operating condition fault record sets, evaluate the fault impacts of the plurality of fault categories to perform breadth dimension analysis, and obtain a breadth reference indicator set.

[0016] The frequency reference index set, the intensity reference index set and the breadth reference index set are output as the multi-operating condition prediction reference index set.

[0017] In a feasible implementation, according to the multi-operating condition prediction reference indicator set, prediction resource allocation for multiple operating conditions is performed to obtain a prediction tilt coefficient set for multiple operating conditions, wherein the prediction tilt coefficient set includes sample tilt coefficients, call tilt coefficients, and scale tilt coefficients for multiple operating conditions, including: The sample tilt coefficients of multiple working conditions are determined according to the frequency reference indicator set and the breadth reference indicator set.

[0018] The scale tilt coefficients of multiple working conditions are determined according to the frequency reference indicator set and the intensity reference indicator set.

[0019] The calling tilt coefficients of multiple working conditions are determined according to the strength reference indicator set and the breadth reference indicator set.

[0020] In a feasible implementation, the fault record set and the predicted tilt coefficient set are combined to construct and train a bearing fault prediction model, including: Constructing a plurality of discriminant networks based on machine learning, wherein the plurality of discriminant networks are constructed based on different machine learning models.

[0021] Based on the call tilt coefficient, the proportion of the number of networks corresponding to multiple working conditions is defined, and the multiple discrimination networks are divided into multiple working condition discrimination network groups.

[0022] Based on the scale tilt coefficient, network depth adjustment is performed on the plurality of working condition discrimination network groups.

[0023] Based on the sample inclination coefficient, the sample capacities of the plurality of operating condition discrimination network groups are defined, and a plurality of operating condition sample sets are extracted corresponding to the fault record set.

[0024] The supervised training of the working condition discrimination network group after multiple network depth adjustments is performed through multiple working condition sample sets to obtain multiple prediction sub-networks with working condition labels, and the bearing fault prediction model is generated through an integrated learning method.

[0025] In a feasible implementation, the bearing fault prediction model is used to perform fault prediction based on the real-time monitoring data of the bearing to obtain a fault prediction result. After that, the method further includes: The fault prediction results are recorded and a prediction record library is constructed.

[0026] According to the prediction record library, an accumulator is configured to accumulate false fault prediction results of multiple working conditions, wherein the false fault prediction results include false fault prediction results and missed fault prediction results.

[0027] When the accumulated result of the accumulator is greater than or equal to the preset false alarm adjustment threshold, multiple false fault prediction results corresponding to the accumulated result are called as a false fault record set to optimize the bearing fault prediction model.

[0028] In a feasible implementation, multiple false fault prediction results corresponding to the accumulated results are called as a false fault record set to optimize the bearing fault prediction model, including: Fault handling information of the plurality of false fault prediction results is obtained.

[0029] Prediction corrections are performed on a plurality of false fault prediction results according to the fault handling information to obtain an improved fault record set.

[0030] The bearing fault prediction model is trained with enhanced supervision using the improved fault record set.

[0031] In a feasible implementation, the bearing fault prediction model is subjected to enhanced supervision training using the improved fault record set, including: Based on the false fault record set, a first false fault condition is extracted, and based on the improved fault record set, a corresponding first improved fault record subset is extracted.

[0032] Call multiple first operating condition prediction subnetworks corresponding to the first false fault condition, and perform model mutation and parameter exchange based on the multiple first operating condition prediction subnetworks to obtain N first supplementary prediction subnetworks, where N is a positive integer greater than or equal to 2.

[0033] Using the first improved fault record subset as training data, multiple first operating condition prediction subnetworks and N first supplementary prediction subnetworks are subjected to enhanced training to obtain multiple first operating condition enhanced prediction subnetworks and evaluate prediction performance.

[0034] According to the prediction performance evaluation result, the last N first working condition enhanced prediction sub-networks with the worst prediction performance are excluded, and the multiple first working condition enhanced prediction sub-networks retained are synchronized to the bearing fault prediction model.

[0035] In a second aspect, the present invention further provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing a method for predicting bearing faults of rotating mechanical equipment provided by the present invention when executing the executable instructions stored in the memory.

[0036] In a third aspect, the present invention further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for predicting bearing faults of rotating mechanical equipment provided by the present invention.

[0037] The beneficial effects of the present invention are as follows: a fault record set is obtained by combining the scene feature information of the target scene with the interactive bearing fault big data. The fault record set is clustered based on the fault category, and the frequency, intensity and breadth dimensions are analyzed according to the clustering results to obtain a multi-condition prediction reference index set. According to the multi-condition prediction reference index set, prediction resources are allocated for multiple working conditions to obtain a prediction tilt coefficient set, which includes a sample tilt coefficient, a call tilt coefficient and a scale tilt coefficient. Combining the fault record set with the prediction tilt coefficient set, a bearing fault prediction model is constructed and trained, the model includes multiple prediction sub-networks, and the working condition is marked for each sub-network. Through the bearing fault prediction model, fault prediction is performed based on the real-time monitoring data of the bearing to generate a fault prediction result. The technical effect of improving the adaptability of bearing fault prediction and improving the performance when handling multiple working conditions and multiple fault types is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of a flow chart of a method for predicting bearing failure of a rotating mechanical device provided by the present invention; Figure 2A schematic diagram of a process for constructing and training a bearing fault prediction model in a method for predicting bearing faults of rotating mechanical equipment provided by the present invention; Figure 3 A schematic diagram of the structure of an electronic device provided by the present invention; Figure 4 A schematic diagram of the structure of a computer-readable storage medium provided by the present invention.

[0039] In the accompanying drawings, the components represented by the reference numerals are listed as follows: Electronic device 500 , memory 510 , processor 520 , first computer program 511 , computer-readable storage medium 600 , second computer program 611 . DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0041] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0042] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or description". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention. Embodiment 1

[0043] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting bearing faults of rotating mechanical equipment.

[0044] Based on the scene feature information of the target scene, interactive bearing fault big data is used to obtain the fault record set.

[0045] Specifically, the target scenario refers to the operating state under the actual environment or conditions that need to be analyzed and processed in a specific application. In bearing fault prediction, the target scenario is a specific bearing and its corresponding operating environment. Among them, the scenario feature information refers to the data describing the relevant factors in the target scenario, including: environmental data, such as temperature, humidity, air pressure; mechanical parameters, such as speed, load, vibration frequency; operating parameters, such as working status, operating cycle, etc.

[0046] Specifically, the bearing failure big data includes historical data sets of bearing failures under multiple interacting factors (such as mechanical operating status, environmental conditions, operating conditions, etc.). In other words, the bearing failure big data includes bearing operation data from different time points and under different working conditions, and contains rich failure modes and fault signals.

[0047] Specifically, the fault record set is a set of historical fault data, which is extracted from the above-mentioned bearing fault big data and contains a complex mapping relationship between the bearing operating status data and the fault prediction results. It is used to train, test and evaluate the fault diagnosis model, and then predict future fault events.

[0048] In some embodiments, the fault record set includes a source fault record set based on historical fault records of a target scenario and a homologous fault record set based on historical fault records of a homologous scenario of the target scenario.

[0049] Furthermore, the fault record set includes a source fault record set and a homologous fault record set, wherein the source fault record set refers to a historical fault record based on a target scenario, including fault data occurring under the target scenario, such as record data of faults occurring under the same temperature and load conditions for bearings of the same model; the homologous fault record set refers to a historical fault record of a homologous scenario based on the target scenario. A homologous scenario is a scenario with similar working environments or operating conditions as the target scenario, that is, the failure mode in the target scenario may be highly similar to the failure mode in these homologous scenarios. Exemplarily, data with similar operating conditions and failure modes are classified by data clustering methods (such as K-means, DBSCAN, etc.), and fault records of similar scenarios are extracted therefrom as a homologous fault record set.

[0050] By combining original fault records and homologous fault records, the diversity of the training set can be expanded, overfitting can be reduced, and the adaptability of the model in different operating environments can be improved. At the same time, it helps to better understand the similarities between different fault modes and causes, thereby improving the accuracy of fault diagnosis and prediction.

[0051] The fault record set is clustered based on the fault category, and frequency dimension analysis, intensity dimension analysis and breadth dimension analysis are performed based on the clustering result to obtain a multi-operating condition prediction reference index set.

[0052] Specifically, the fault events in the fault record set are divided into different categories according to their similarities, so that each type of fault event has high similarity in characteristics. Through clustering, different types of fault modes, characteristics and laws can be identified, providing a basis for subsequent analysis and prediction.

[0053] Specifically, through frequency dimension analysis, intensity dimension analysis and breadth dimension analysis, a comprehensive multi-condition prediction reference index set is obtained, which helps to better understand the characteristics and impacts of different types of faults. This multi-condition prediction reference index set is used to guide the construction of the prediction model in the future, so that the prediction model can adapt to the fault characteristics of the target scenario and improve the fault prediction accuracy and response capability under different working conditions.

[0054] In some embodiments, the fault record set is clustered based on the fault category, and frequency dimension analysis, intensity dimension analysis and breadth dimension analysis are performed based on the clustering result to obtain a multi-operating condition prediction reference index set, including: Taking multiple fault categories as the division targets, clustering and dividing the fault record set are performed to generate multiple working condition fault record sets; traversing the multiple working condition fault record sets, counting the fault record frequencies to perform frequency dimension analysis, and obtaining a frequency reference index set; traversing the multiple working condition fault record sets, obtaining typical prediction accuracy rates of multiple fault categories to perform intensity dimension analysis, and obtaining an intensity reference index set; traversing the multiple working condition fault record sets, evaluating the fault impacts of multiple fault categories to perform breadth dimension analysis, and obtaining a breadth reference index set; outputting the frequency reference index set, the intensity reference index set, and the breadth reference index set as the multi-working condition prediction reference index set.

[0055] Specifically, firstly, by clustering the fault record sets, similar types of faults are divided into the same fault prediction working condition, and multiple working condition fault record sets are generated. In other words, each working condition fault record set represents fault record data under a fault category.

[0056] Specifically, frequency dimension analysis focuses on the probability or frequency of occurrence of different fault categories, thereby analyzing the frequency of each type of fault and generating a frequency reference index set, which characterizes the prevalence and commonness of the fault category. For example, for each operating condition fault record set, statistical methods such as frequency statistics and proportion statistics are used to calculate the number of occurrences and frequency of each fault category in the set as the strength reference index of multiple operating condition fault record sets.

[0057] Specifically, the intensity dimension analysis focuses on the difficulty of predicting different fault categories. In other words, the intensity reference index set is a quantitative representation of the difficulty of predicting different fault categories. Fault categories with greater prediction difficulty often involve more complex fault mechanisms and require deeper neural network structures for modeling. Among them, the difficulty of fault prediction is closely related to the complexity of the fault category and the characteristics of the data, so the prediction difficulty can be quantified based on the characteristic complexity of the fault. For example, some fault categories may contain multiple sub-fault modes, which are difficult to predict. Through intensity dimension analysis, it is possible to select appropriate network depth and structure for prediction tasks of different types of faults.

[0058] Specifically, breadth dimension analysis focuses on the impact of different fault categories. The impact involves the scope and severity of the fault's impact on the system or equipment. Fault categories with high severity and wide impact have greater impact and need to be given more attention in the prediction model. For example, first, the fault labels in the historical fault record set are used to analyze the scope of the fault's impact (for example, whether it causes equipment downtime, whether it affects other components) and severity (for example, whether the fault causes equipment damage, repair costs, etc.); then, the weighted average method, comprehensive scoring method, etc. are used, combined with data from different dimensions, to calculate the equivalent impact intensity of each fault category, and generate a comprehensive score of the impact intensity as a breadth reference indicator set.

[0059] Optionally, through breadth dimension analysis, priorities can be set for different fault categories. For example, fault categories with large impact and severity (such as faults that cause complete equipment damage) should be allocated more prediction resources to reduce potential economic losses.

[0060] In general, through multi-level analysis of frequency, intensity and breadth, different fault categories in the fault record set can be analyzed and predicted more accurately. This not only helps to better understand the characteristics and impacts of different types of faults, but also provides clear guidance for model design and resource allocation, thereby improving fault prediction accuracy and response capabilities.

[0061] According to the multi-operating condition prediction reference indicator set, prediction resource allocation for multiple operating conditions is performed to obtain a prediction tilt coefficient set for multiple operating conditions, wherein the prediction tilt coefficient set includes sample tilt coefficients, call tilt coefficients and scale tilt coefficients for multiple operating conditions.

[0062] Specifically, based on the frequency reference indicator set, intensity reference indicator set and breadth reference indicator set in the multi-condition prediction reference indicator set, the inclination levels of prediction resources for different conditions are defined, namely, the sample inclination coefficient, the call inclination coefficient and the scale inclination coefficient.

[0063] Specifically, the sample tilt coefficient quantifies the degree of tilt of the sample distribution of each working condition, which helps to make adjustments for the sample imbalance problem and avoid the model being biased towards the majority class samples; the call tilt coefficient quantifies the model attention corresponding to each working condition, ensuring that high-impact and complex tasks have a high model attention. In other words, for working conditions with higher call tilt coefficients, more computing resources or more efficient model call strategies (such as a larger number of prediction sub-networks) are preferentially allocated; the scale tilt coefficient defines the depth of the prediction model for each working condition. In other words, the larger the scale tilt coefficient, the greater the prediction difficulty of the model, and a more complex prediction model is required to better express the occurrence characteristics of the fault, thereby improving the prediction accuracy.

[0064] In some embodiments, according to the multi-operating condition prediction reference indicator set, prediction resource allocation for multiple operating conditions is performed to obtain a prediction tilt coefficient set for multiple operating conditions, wherein the prediction tilt coefficient set includes sample tilt coefficients, call tilt coefficients, and scale tilt coefficients for multiple operating conditions, including: According to the frequency reference indicator set and the breadth reference indicator set, the sample tilt coefficient of multiple working conditions is determined; according to the frequency reference indicator set and the intensity reference indicator set, the scale tilt coefficient of multiple working conditions is determined; according to the intensity reference indicator set and the breadth reference indicator set, the call tilt coefficient of multiple working conditions is determined.

[0065] Specifically, the sample tilt coefficient is evaluated and obtained based on the relationship between the frequency of fault occurrence and the scope of fault impact. A condition with a higher frequency reference index means that the frequency of fault occurrence of this category is higher, and undersampling is required to appropriately reduce the amount of sample data, balance the imbalance between classes, and prevent the model from being biased towards the majority class. A higher breadth reference index means that the impact of this fault category is greater, and it may be necessary to improve the data quality of the sample data accordingly.

[0066] Specifically, the scale tilt coefficient mainly reflects the difficulty of predicting the fault type and its requirements for the model depth. By combining the fault intensity reference index (i.e., difficulty and complexity) and the frequency reference index (i.e., frequency of occurrence), it is determined how deep the model is required for effective prediction of the working condition. For example, if the fault types of certain working conditions occur frequently but are difficult to predict, deeper models (such as deep neural networks, LSTM, CNN, etc.) may be required to process in order to improve the accuracy of fault prediction; if the frequency of a certain working condition is high but the prediction difficulty is low, then the working condition does not need to be configured with an overly deep model structure to reduce the demand for computing resources.

[0067] Specifically, the call tilt coefficient reflects the prediction attention of the fault category, that is, which working conditions require more computing resources or more efficient calling strategies in a multi-tasking environment. For example, if the fault type of certain working conditions occurs at a low frequency but is more difficult to predict, while using a deeper model network for it, the number of prediction sub-networks for this working condition in the integrated learning model is reduced to avoid excessive occupation of prediction resources.

[0068] Constructing a prediction tilt coefficient set by combining the frequency reference index set, the intensity reference index set, and the breadth reference index set can help achieve optimal resource allocation and dynamically adjust the resource configuration of the model according to the characteristics of different working conditions, thereby improving the accuracy and efficiency of the overall prediction.

[0069] Combining the fault record set with the predicted tilt coefficient set, a bearing fault prediction model is constructed and trained, wherein the bearing fault prediction model includes a plurality of prediction sub-networks, and a plurality of the prediction sub-networks have operating condition labels.

[0070] Optionally, based on an integrated learning method, a bearing fault prediction model including multiple prediction sub-networks is constructed and trained by combining a fault record set and a prediction tilt coefficient set. The multiple prediction sub-networks correspond to different operating condition categories, and the corresponding relationship is characterized by an operating condition label.

[0071] In some embodiments, Figure 2 As shown, combining the fault record set and the predicted tilt coefficient set, constructing and training a bearing fault prediction model includes: Construct multiple discriminant networks based on machine learning, wherein the multiple discriminant networks are constructed based on different machine learning models; based on the call tilt coefficient, define the proportion of the number of networks corresponding to multiple working conditions, and divide the multiple discriminant networks into multiple working condition discriminant network groups; based on the scale tilt coefficient, adjust the network depth of the multiple working condition discriminant network groups; based on the sample tilt coefficient, define the sample capacity of the multiple working condition discriminant network groups, and extract multiple working condition sample sets corresponding to the fault record set; supervised training of the working condition discriminant network groups after multiple network depth adjustments is performed through multiple working condition sample sets to obtain multiple prediction sub-networks with working condition labels, and generate the bearing fault prediction model through an integrated learning method.

[0072] Specifically, first, multiple discriminant networks based on machine learning are constructed for fault prediction. The structures of the multiple discriminant networks can be different machine learning models, such as support vector machine (SVM), decision tree, deep neural network, etc., to adapt to different working conditions.

[0073] Specifically, based on the call tilt coefficient, the ratio of the number of networks corresponding to different working conditions is defined, the computing resources and task complexity of the model are reasonably allocated, the resource utilization of the model is optimized, and then multiple corresponding working condition discrimination network groups are generated. Among them, defining the ratio of the number of networks corresponding to each working condition is to allocate a corresponding number of discrimination networks to each working condition. For example, a working condition with a higher call tilt coefficient may be allocated multiple networks to ensure that it receives sufficient attention during the training and prediction process.

[0074] Specifically, according to the scale tilt coefficient of each working condition, different working condition discrimination network groups are deeply adjusted to ensure that more complex fault modes have sufficient network levels and depths for effective prediction. For example, for working conditions with higher scale tilt coefficients, the number of layers (i.e., network depth) and nodes of the discrimination network are increased, or more complex models such as deep convolutional neural networks or long short-term memory networks are used. For working conditions with lower scale tilt coefficients, shallow networks or other lightweight models can be used to reduce computing resource consumption.

[0075] Specifically, sample data is allocated to different working conditions according to the sample tilt coefficient to avoid sample imbalance problems, thereby improving the generalization ability of the model. For example, for fault types (working conditions) with fewer samples (relatively rare occurrences) but with important impacts, oversampling and other techniques are used to increase the diversity of training data to ensure that the network can obtain sufficient sample support.

[0076] Furthermore, supervised training of the working condition discrimination network group after network depth adjustment is performed according to different working condition sample sets. Each working condition discrimination network group learns the failure mode of a specific working condition according to its corresponding working condition sample set, and obtains multiple prediction sub-networks with working condition labels. Then, these prediction sub-networks are combined through ensemble learning of multiple networks (such as weighted voting, stacking method, etc.) to generate a bearing fault prediction model. The bearing fault prediction model can perform fault prediction under different working conditions and provide reliable fault diagnosis support.

[0077] The bearing fault prediction model is used to perform fault prediction based on the real-time monitoring data of the bearing and obtain a fault prediction result.

[0078] Specifically, firstly, the real-time monitoring data of the bearing is obtained. The real-time monitoring data is collected through sensors or monitoring systems, including but not limited to temperature, vibration, pressure, speed, noise and other parameters related to the operating status of the bearing; then, the collected real-time data is preprocessed and key features that are helpful for fault prediction are extracted; then, the extracted features are input into the bearing fault prediction model for fault prediction to obtain the fault prediction results. Optionally, the fault prediction results include fault probability, fault level, fault type, etc.

[0079] Furthermore, by using the bearing fault prediction model, fault prediction based on the real-time monitoring data of the bearing is performed to obtain a fault prediction result. After that, the method further includes: Record the fault prediction results and build a prediction record library; according to the prediction record library, configure an accumulator to accumulate false fault prediction results of multiple working conditions, wherein the false fault prediction results include false fault prediction results and missed fault prediction results; when the accumulated result of the accumulator is greater than or equal to a preset false alarm adjustment threshold, call the multiple false fault prediction results corresponding to the accumulated result as a false fault record set to optimize the bearing fault prediction model.

[0080] Specifically, after obtaining the fault prediction result, the fault prediction result is stored in a constructed prediction record library, which includes multiple historical fault prediction results; then, for the false fault prediction results of different working conditions, an accumulator is configured to accumulate them in order to detect the error of the prediction model and prepare for subsequent optimization.

[0081] Specifically, the false fault record set includes false fault prediction results (false positive examples) and missed fault prediction results (false negative examples), such as false faults (the model incorrectly predicts the occurrence of a fault, but in fact the bearing has not failed) and missed faults (the model fails to predict the actual fault).

[0082] Specifically, a false alarm adjustment threshold is set. If the cumulative number of false fault prediction results of a certain working condition reaches the threshold, it means that the prediction results of the working condition have significant false alarm or missed alarm problems and need to be optimized. The corresponding false fault prediction results (including false alarm and missed alarm prediction results) are used as the false fault record set.

[0083] Specifically, the prediction model is retrained or fine-tuned through a set of false fault records to reduce the probability of false reporting and missed reporting and enhance the generalization and robustness of the model. Optionally, the bearing fault prediction model optimization can be implemented in ways such as incremental learning, model architecture adjustment, and loss function improvement.

[0084] In some embodiments, calling a plurality of false fault prediction results corresponding to the accumulated result as a false fault record set to optimize the bearing fault prediction model includes: Obtain fault handling information of the plurality of false fault prediction results; perform prediction corrections on the plurality of false fault prediction results according to the fault handling information to obtain an improved fault record set; and perform enhanced supervised training on the bearing fault prediction model using the improved fault record set.

[0085] Specifically, firstly, the fault handling information corresponding to multiple false fault prediction results is obtained. The fault handling information is the response and handling record of the false fault prediction results, including eliminating false alarms, supplementing missed alarms, etc. Then, the false fault prediction results are corrected according to the fault handling information. Including adjusting the predicted fault type, fault degree, fault time, etc. The corrected false fault prediction results are the improved fault record set, which can be considered as the correct fault prediction record data; then, the improved fault record set is used to perform enhanced supervision training on the bearing fault prediction model, including retraining the model with new training data, or optimizing the model with a reinforcement learning algorithm, and adjusting the parameters of the model to improve the accuracy of fault prediction.

[0086] In some implementations, performing enhanced supervised training on the bearing fault prediction model using the improved fault record set includes: Based on the false fault record set, extract the first false fault condition, and based on the improved fault record set, extract the corresponding first improved fault record subset; call multiple first condition prediction subnetworks corresponding to the first false fault condition, and perform model mutation and parameter exchange based on the multiple first condition prediction subnetworks to obtain N first supplementary prediction subnetworks, wherein N is a positive integer greater than or equal to 2; use the first improved fault record subset as training data to perform reinforcement training on multiple first condition prediction subnetworks and N first supplementary prediction subnetworks to obtain multiple first condition reinforced prediction subnetworks and evaluate the prediction performance; according to the prediction performance evaluation result, exclude the last N first condition reinforced prediction subnetworks with the worst prediction performance, and synchronize the retained multiple first condition reinforced prediction subnetworks to the bearing fault prediction model.

[0087] Specifically, a first false fault condition is first randomly extracted from the false fault record set, and a corresponding first improved fault record subset is extracted from the improved fault record set, wherein the first improved fault record set is a data set that is optimized or corrected for the false fault record set, that is, additional annotations or corrections are made on the basis of the false fault record set to more realistically reflect the fault mode under the working condition.

[0088] Optionally, use statistical analysis methods, such as calculating indicators such as false alarm rate and missed alarm rate, to select working conditions that show a higher probability of false alarm or missed alarm as the first false fault working condition, to ensure that working conditions with poor model prediction effect are improved first.

[0089] Specifically, the multiple first operating condition prediction sub-networks are multiple initial networks that have been trained for the first false fault condition. By performing network mutation (such as modifying the network structure, changing the activation function, etc.) and parameter exchange (such as sharing weights between different network layers, changing the initialization method, etc.) on the multiple first operating condition prediction sub-networks, N first supplementary prediction sub-networks are generated. The purpose of the N first supplementary prediction sub-networks is to increase the diversity of the network and prevent the model from falling into a local optimal solution.

[0090] Specifically, the first improved fault record subset is used as training data to perform reinforcement training on multiple first operating condition prediction subnetworks and first supplementary prediction subnetworks, that is, the weights of the networks are adjusted according to the reinforcement learning method to enable them to better cope with false fault conditions, thereby improving the prediction capabilities of multiple first operating condition prediction subnetworks and first supplementary prediction subnetworks for false fault conditions.

[0091] Furthermore, after the enhanced training, the prediction performance of each first-operating condition enhanced prediction subnetwork is evaluated through a validation set or cross-validation. For example, the performance of each network is evaluated based on indicators such as accuracy, recall rate, and F1 score. Then, the N first-operating condition enhanced prediction subnetworks with the worst performance are excluded, and the first-operating condition enhanced prediction subnetwork with the best performance is retained and integrated into the bearing fault prediction model.

[0092] Optionally, based on the above-mentioned enhanced supervised training and ensemble learning method, the false fault record set is traversed to perform model optimization for multiple working conditions, thereby comprehensively improving the model's adaptability to false fault conditions.

[0093] The reinforced supervised training process of the above-mentioned bearing fault prediction model is based on a set of false fault records, and uses a modified set of fault records to enhance the training effect of the model. This process combines multiple network mutations, parameter exchange, reinforced training, and performance evaluation to optimize the model, ultimately improving the accuracy and stability of fault prediction.

[0094] Through the above-mentioned method steps, the accuracy and robustness of the bearing fault prediction model in the face of false fault conditions can be effectively improved, thereby ensuring efficient fault prediction in complex environments.

[0095] A method for predicting bearing failures of rotating mechanical equipment provided by an embodiment of the present invention has at least the following technical effects: The fault record set is obtained by combining the scene feature information of the target scene with the interactive bearing fault big data. The fault record set is clustered based on the fault category, and the frequency, intensity and breadth dimensions are analyzed according to the clustering results to obtain a multi-condition prediction reference index set. According to the multi-condition prediction reference index set, prediction resources are allocated for multiple working conditions to obtain a prediction tilt coefficient set, which includes a sample tilt coefficient, a call tilt coefficient and a scale tilt coefficient. Combining the fault record set with the prediction tilt coefficient set, a bearing fault prediction model is constructed and trained. The model includes multiple prediction sub-networks, and the working condition is marked for each sub-network. Through the bearing fault prediction model, fault prediction is performed based on the real-time monitoring data of the bearing to generate a fault prediction result. Thereby, the adaptability of bearing fault prediction is improved and the performance when processing multiple working conditions and multiple fault types is improved. Embodiment 2

[0096] See also Figure 3 , Figure 3 Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a first computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the first computer program 511, the following steps are implemented: based on the scene feature information of the target scene, interactive bearing fault big data is used to obtain a fault record set; the fault record set is clustered based on the fault category, and frequency dimension analysis, intensity dimension analysis, and breadth dimension analysis are performed based on the clustering result to obtain a multi-condition prediction reference index set; according to the multi-condition prediction reference index set, prediction resources are allocated for multiple conditions to obtain a prediction tilt coefficient set for multiple conditions, wherein the prediction tilt coefficient set includes sample tilt coefficients, call tilt coefficients, and scale tilt coefficients for multiple conditions; combining the fault record set and the prediction tilt coefficient set, constructing and training a bearing fault prediction model, wherein the bearing fault prediction model includes multiple prediction subnetworks, and multiple prediction subnetworks have condition tags; through the bearing fault prediction model, fault prediction based on real-time bearing monitoring data is performed to obtain a fault prediction result. Embodiment 3

[0097] See also Figure 4 , Figure 4 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. Figure 4As shown, this embodiment provides a computer-readable storage medium 600, on which a second computer program 611 is stored. When the second computer program 611 is executed by the processor, the following steps are implemented: based on the scene feature information of the target scene, interactive bearing fault big data, and obtaining a fault record set; clustering the fault record set based on the fault category, and performing frequency dimension analysis, intensity dimension analysis and breadth dimension analysis based on the clustering result to obtain a multi-condition prediction reference index set; according to the multi-condition prediction reference index set, prediction resources are allocated for multiple conditions to obtain a prediction tilt coefficient set for multiple conditions, wherein the prediction tilt coefficient set includes sample tilt coefficients, call tilt coefficients and scale tilt coefficients for multiple conditions; combining the fault record set and the prediction tilt coefficient set, constructing and training a bearing fault prediction model, wherein the bearing fault prediction model includes multiple prediction sub-networks, and multiple prediction sub-networks have condition tags; through the bearing fault prediction model, fault prediction based on real-time bearing monitoring data is performed to obtain a fault prediction result.

[0098] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0099] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0101] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0103] Although preferred embodiments of the present invention have been described, additional changes and modifications may occur to these embodiments once those skilled in the art understand the basic inventive concepts.

[0104] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the present invention and its equivalent technologies, the present invention is also intended to include these changes and variations.

Claims

1. A method for predicting bearing failure of rotating mechanical equipment, characterized in that: The method comprises: Based on the scene feature information of the target scene, interactive bearing fault big data is used to obtain the fault record set; The fault record set is clustered based on the fault category, and frequency dimension analysis, intensity dimension analysis and breadth dimension analysis are performed based on the clustering result to obtain a multi-operating condition prediction reference indicator set; According to the multi-operating condition prediction reference indicator set, prediction resource allocation for multiple operating conditions is performed to obtain a prediction tilt coefficient set for multiple operating conditions, wherein the prediction tilt coefficient set includes sample tilt coefficients, call tilt coefficients, and scale tilt coefficients for multiple operating conditions; Combining the fault record set with the predicted tilt coefficient set, constructing and training a bearing fault prediction model, wherein the bearing fault prediction model includes a plurality of prediction subnetworks, and a plurality of the prediction subnetworks have operating condition labels; The bearing fault prediction model is used to perform fault prediction based on the real-time monitoring data of the bearing and obtain a fault prediction result.

2. A method for predicting bearing failure of rotating mechanical equipment according to claim 1, characterized in that: The fault record set includes a source fault record set based on historical fault records of a target scenario and a homologous fault record set based on historical fault records of a homologous scenario of the target scenario.

3. A method for predicting bearing failure of rotating mechanical equipment according to claim 2, characterized in that: The fault record set is clustered based on the fault category, and frequency dimension analysis, intensity dimension analysis and breadth dimension analysis are performed based on the clustering result to obtain a multi-condition prediction reference index set, including: Taking multiple fault categories as the division targets, clustering and dividing the fault record set to generate multiple working condition fault record sets; Traversing the plurality of working condition fault record sets, counting the fault record frequencies to perform frequency dimension analysis, and obtaining a frequency reference index set; Traversing multiple working condition fault record sets, obtaining typical prediction accuracy rates of multiple fault categories for strength dimension, and obtaining a strength reference index set; Traversing multiple working condition fault record sets, evaluating the fault impacts of multiple fault categories to perform breadth dimension analysis, and obtaining a breadth reference indicator set; The frequency reference index set, the intensity reference index set and the breadth reference index set are output as the multi-operating condition prediction reference index set.

4. A method for predicting bearing failure of rotating mechanical equipment according to claim 3, characterized in that: According to the multi-operating condition prediction reference indicator set, prediction resource allocation for multiple operating conditions is performed to obtain prediction tilt coefficient sets for multiple operating conditions, wherein the prediction tilt coefficient sets include sample tilt coefficients, call tilt coefficients, and scale tilt coefficients for multiple operating conditions, including: Determining the sample tilt coefficients of multiple working conditions according to the frequency reference indicator set and the breadth reference indicator set; Determining the scale tilt coefficients of multiple working conditions according to the frequency reference index set and the intensity reference index set; The calling tilt coefficients of multiple working conditions are determined according to the strength reference indicator set and the breadth reference indicator set.

5. A method for predicting bearing failure of rotating mechanical equipment according to claim 4, characterized in that: Combining the fault record set with the predicted tilt coefficient set, constructing and training a bearing fault prediction model, including: Constructing a plurality of discriminant networks based on machine learning, wherein the plurality of discriminant networks are constructed based on different machine learning models; Based on the call tilt coefficient, define the network quantity ratio corresponding to multiple working conditions, and divide the multiple discrimination networks into multiple working condition discrimination network groups; Based on the scale tilt coefficient, adjusting the network depth of the plurality of working condition discrimination network groups; Based on the sample tilt coefficient, define the sample capacity of the plurality of working condition discrimination network groups, and extract a plurality of working condition sample sets corresponding to the fault record set; The supervised training of the working condition discrimination network group after multiple network depth adjustments is performed through multiple working condition sample sets to obtain multiple prediction sub-networks with working condition labels, and the bearing fault prediction model is generated through an integrated learning method.

6. A method for predicting bearing failure of rotating mechanical equipment according to claim 1, characterized in that: By using the bearing fault prediction model, fault prediction based on the real-time monitoring data of the bearing is performed to obtain a fault prediction result. After that, the method further includes: Recording the fault prediction results and building a prediction record library; According to the prediction record library, an accumulator is configured to accumulate false fault prediction results of multiple working conditions, wherein the false fault prediction results include false fault prediction results and missed fault prediction results; When the accumulated result of the accumulator is greater than or equal to the preset false alarm adjustment threshold, multiple false fault prediction results corresponding to the accumulated result are called as a false fault record set to optimize the bearing fault prediction model.

7. A method for predicting bearing failure of rotating mechanical equipment according to claim 6, characterized in that: Calling multiple false fault prediction results corresponding to the accumulated results as a false fault record set to optimize the bearing fault prediction model includes: Obtaining fault handling information of a plurality of false fault prediction results; According to the fault handling information, prediction correction is performed on a plurality of false fault prediction results to obtain an improved fault record set; The bearing fault prediction model is trained with enhanced supervision using the improved fault record set.

8. A method for predicting bearing failure of rotating mechanical equipment according to claim 7, characterized in that: The bearing fault prediction model is subjected to enhanced supervision training using the improved fault record set, including: Based on the false fault record set, extracting a first false fault condition, and based on the improved fault record set, extracting a corresponding first improved fault record subset; Calling multiple first operating condition prediction subnetworks corresponding to the first false fault operating condition, and performing model mutation and parameter exchange based on the multiple first operating condition prediction subnetworks to obtain N first supplementary prediction subnetworks, where N is a positive integer greater than or equal to 2; Using the first improved fault record subset as training data, performing enhanced training on a plurality of the first operating condition prediction subnetworks and N of the first supplementary prediction subnetworks, obtaining a plurality of first operating condition enhanced prediction subnetworks and evaluating prediction performance; According to the prediction performance evaluation result, the last N first working condition enhanced prediction sub-networks with the worst prediction performance are excluded, and the multiple first working condition enhanced prediction sub-networks retained are synchronized to the bearing fault prediction model.

9. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program, thereby implementing a method for predicting bearing faults of rotating mechanical equipment as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer software program, and when the computer software program is executed by a processor, a method for predicting bearing faults of rotating mechanical equipment according to any one of claims 1 to 8 is implemented.

Images