Bearing fault simulation method and system

Abstract

The invention relates to the technical field of data simulation, in particular to a bearing fault simulation method and system. The method comprises the following steps: collecting multi-modal bearing data so as to construct a bearing distributed edge data set; extracting contact power data of the bearing distributed edge data set, and calculating a bearing rigidity change curve according to the contact power data; carrying out random parameter modeling based on the bearing distributed edge data set, and carrying out physical modeling benchmark reference on the bearing rigidity change curve to obtain a bearing fault physical-data hybrid model; therefore, by integrating multi-modal data acquisition, physical-data hybrid modeling and a dynamic hyper-parameter adjustment mechanism, the defects of a traditional bearing fault diagnosis method in the aspects of accuracy and real-time performance are overcome, and the fault prediction and early warning precision and the response speed are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data simulation, and in particular, to a bearing fault simulation method and system. Background Art

[0002] Bearing fault diagnosis technology has gradually evolved from relying on manual experience and vibration signal analysis to the application of multi-modal data fusion and intelligent algorithms. Although modern technologies can provide more comprehensive fault identification and prediction, they still face certain challenges in aspects such as data quality, model generalization ability, and real-time performance. Traditional fault diagnosis methods mostly rely on single-sensor data, such as vibration signals or temperature signals, making it difficult for them to comprehensively capture multi-dimensional fault information. Although some multi-modal data fusion methods have been proposed, many technologies still remain at the stage of simple data splicing or weighted fusion, ignoring the mutual relationship between different data sources, resulting in insufficient accuracy and sensitivity of diagnostic results. Secondly, most of the existing fault prediction models are based on statistical learning of historical data and lack adaptability to complex fault patterns. With the change of system operating conditions, traditional models require frequent manual intervention and data recalibration, restricting the adaptive ability of the models and their performance under new fault types. In addition, current fault diagnosis methods mainly focus on data-driven approaches and lack sufficient support from physical models, leading to diagnostic results lacking physical basis and making it difficult to fundamentally analyze the causes of faults. Summary of the Invention

[0003] Based on this, it is necessary to provide a bearing fault simulation method and system to solve at least one of the above technical problems.

[0004] To achieve the above object, a bearing fault simulation method, the method includes the following steps:

[0005] Step S1: Collect multi-modal bearing data to construct a bearing distributed edge data set;

[0006] Step S2: Extract the contact dynamic data of the bearing distributed edge data set, and calculate the bearing stiffness change curve according to the contact dynamic data; perform random parameter modeling based on the bearing distributed edge data set, and use the bearing stiffness change curve as a physical modeling reference to obtain a bearing fault physical-data hybrid model;

[0007] Step S3: Output the bearing fault model result data from the bearing fault physical-data hybrid model; perform time-domain comparative analysis on the bearing fault model result data, and conduct characteristic fault assessment to obtain bearing fault prediction assessment data; generate hyperparameters for the bearing fault prediction assessment data according to the learning rate index of the bearing fault physical-data hybrid model, so as to dynamically adjust the bearing fault physical-data hybrid model;

[0008] Step S4: Trace the cause of the failure based on the bearing fault prediction and evaluation data, and generate a bearing simulation report.

[0009] The beneficial effects of the present invention are as follows. Through the deployment of a multi-modal sensor array and the construction of a distributed edge data set, the contact dynamic data of the bearing are comprehensively collected, providing accurate data support for subsequent fault diagnosis. After data acquisition, the system can calculate the stiffness change curve by extracting the contact dynamic data, which can reflect the stiffness change of the bearing under different working conditions in real time, thereby providing key physical characteristic information for fault diagnosis. The combination of physical modeling based on the stiffness change curve and stochastic parameter modeling can effectively capture the randomness and complexity in the operation of the bearing. The generated physical-data hybrid model of bearing faults fully integrates the characteristics of physical theory and actual operation data, greatly improving the accuracy and reliability of fault diagnosis. Through time-domain comparative analysis of the results of the fault model and characteristic fault assessment, not only can the fault type of the bearing be accurately identified, but also the degree and occurrence time of the fault can be predicted, providing an accurate basis for the maintenance and replacement of the bearing. In addition, by dynamically adjusting the hyperparameters of the fault prediction and evaluation data using the learning rate index, the performance of the model is effectively optimized, enabling it to continuously improve the prediction accuracy according to the actual situation. Finally, through tracing the cause of the fault and constructing a simulation report, a detailed fault analysis report can be provided for engineers to help quickly locate the root cause of the problem and optimize the maintenance decision. Overall, this system can not only provide real-time bearing fault prediction and evaluation, but also improve the long-term stability and adaptability of the system through dynamic adjustment and optimization. Therefore, the present invention solves the deficiencies of traditional bearing fault diagnosis methods in terms of accuracy and real-time performance by integrating multi-modal data acquisition, physical-data hybrid modeling, and dynamic hyperparameter adjustment mechanisms, improving the accuracy and response speed of fault prediction and early warning.

[0010] Preferably, the construction of the bearing distributed edge data set includes:

[0011] Deploy a multi-modal sensor array on the bearing test bench, where the multi-modal sensor array includes a vibration sensor, a torque sensor, and an infrared thermal imager;

[0012] Use a MEMS accelerometer and a vibration sensor with a bandwidth of 0-20 kHz to collect bearing vibration signals;

[0013] Use a torque sensor with a parameter of 0.1 rpm resolution to collect bearing material property data;

[0014] Use an infrared thermal imager with a sampling rate of 30 Hz and an accuracy of ±0.5 °C to collect bearing infrared thermal image data;

[0015] Preprocess the bearing vibration signal, bearing material property data, and bearing infrared thermal image data to obtain a bearing distributed edge data set.

[0016] In the present invention, by deploying a multi-modal sensor array on a bearing test bench, multi-dimensional operating data of the bearing can be comprehensively collected, thereby providing rich and diverse data support for bearing health monitoring and fault diagnosis. Specifically, the combined use of three sensors, namely a vibration sensor, a torque sensor, and an infrared thermal imager, enables the system to comprehensively monitor the working state of the bearing from multiple angles. The vibration sensor uses a MEMS accelerometer with a bandwidth of 0 - 20 kHz, which can accurately capture the high-frequency vibration signals of the bearing. These signals are highly sensitive to minor faults of the bearing, such as rolling element damage or poor lubrication. The bearing material property data collected by the torque sensor can reflect the load changes, friction characteristics, and mechanical property changes of the bearing during operation, which is crucial for evaluating the mechanical behavior and health state of the bearing. The infrared thermal imager monitors the thermal imaging of the bearing at a sampling rate of 30 Hz and an accuracy of ±0.5 °C. By capturing the temperature changes on the bearing surface, it can effectively detect local overheating caused by friction, excessive load, or poor lubrication, thereby providing a basis for early fault warning. After preprocessing the above-collected data to remove noise, outliers, and redundant information, through effective data fusion, a multi-modal distributed edge data set is constructed, laying a foundation for subsequent data analysis, feature extraction, and model construction. By integrating data from different sensors, the system can not only capture the vibration characteristics of the bearing but also synchronously obtain its material properties and thermal state information, significantly improving the comprehensiveness and accuracy of fault diagnosis. The establishment of this data set provides more accurate and multi-dimensional support for the early diagnosis, condition monitoring, and health assessment of bearing faults, enhancing the adaptability and robustness of the system in complex working environments.

[0017] Preferably, the data preprocessing of the bearing vibration signal, bearing material property data, and bearing infrared thermal image data includes:

[0018] Use the sliding window statistical method to remove electromagnetic interference abnormal pulses with an amplitude range of ±15 g from the bearing vibration signal, bearing material property data, and bearing infrared thermal image data, and perform timestamp synchronization processing to obtain bearing multi-modal joint data;

[0019] Perform wavelet packet decomposition with a sub-band frequency of 0 - 3.9 kHz on the bearing multi-modal joint data to obtain bearing multi-modal decomposition data;

[0020] Perform cleaning on the distributed edge data set with a single-node storage capacity of ≥500 MB / h according to the bearing multi-modal decomposition data to obtain the bearing distributed edge data set.

[0021] The present invention effectively solves the common electromagnetic interference problem in high-frequency acquisition by adopting a sliding window statistical method and an electromagnetic interference abnormal pulse removal technique with an amplitude range of ±15g. Electromagnetic interference is one of the main factors affecting the accuracy of bearing vibration signals. Especially during high-frequency signal acquisition, electromagnetic pulses will introduce noise, affecting the authenticity and reliability of the signals. By using the sliding window statistical method to clean the vibration signals, material property data, and infrared thermal image data, abnormal pulses are removed, ensuring the purity of the data, thereby improving the accuracy of subsequent analysis. In addition, the timestamp synchronization processing step further ensures the time consistency of multi-modal data, avoiding the problem of timing misalignment between data from different sensors, enabling subsequent joint analysis and model training to be carried out within an accurate time frame. Through these processes, the system obtains high-quality bearing multi-modal joint data, which provides a stable and reliable data basis for bearing health monitoring. Based on the bearing multi-modal joint data, the system adopts wavelet packet decomposition technology to decompose the data into sub-band frequencies of 0 - 3.9 kHz, which can effectively extract the frequency characteristics in the bearing vibration signals and combine these frequency components with material properties and thermal image data, thereby enhancing the multi-dimensional characteristics of fault diagnosis. Wavelet packet decomposition can provide detailed signal characteristics in the time-frequency domain, helping to distinguish different types of fault modes, especially having significant advantages in early diagnosis and the detection of minor faults. Then, through the distributed edge data set cleaning step with a single-node storage capacity ≥500MB / h, the effectiveness and consistency of data during large-scale data storage and processing are ensured, avoiding problems of storage redundancy and data loss, enabling the system to remain efficient and stable when processing and transmitting large amounts of data. Finally, after this series of data processing steps, the obtained bearing distributed edge data set provides high-quality basic data for subsequent fault diagnosis and health assessment, helping to improve the accuracy of fault detection and the real-time response ability of the system.

[0022] Preferably, the calculation of the stiffness change curve in step S2 includes:

[0023] Identify elastic pulses based on the bearing distributed edge data set to obtain bearing elastic pulse data;

[0024] Extract the ball diameter based on the bearing distributed edge data set and calculate the theoretical contact frequency to obtain bearing stress distribution data;

[0025] Segment the rotational speed into intervals of 100 rpm based on the bearing elastic pulse data and the bearing stress distribution data to obtain the bearing rotational speed intervals; calculate the impact amplitude stiffness for the bearing rotational speed intervals to obtain bearing stiffness calculation data;

[0026] Fill in the missing values of the curve based on the bearing stiffness calculation data to obtain the bearing stiffness change curve.

[0027] The present invention identifies elastic pulses through a bearing distributed edge data set, extracts the elastic pulse data of the bearing, and provides important information about the dynamic response of the bearing for subsequent analysis. Elastic pulses are important signals reflecting the contact mechanical characteristics of the bearing, and can reflect the minute changes and early fault signs in the working state of the bearing. Next, by extracting the ball diameter and calculating the theoretical contact frequency, the system can calculate the stress distribution data of the bearing during operation, which provides theoretical support for further understanding the load-bearing capacity and fatigue characteristics of the bearing under different loads. The stress distribution data provides a key physical basis for the fault analysis of the bearing, especially important when identifying problems such as early fatigue cracks or poor contact. Subsequently, the system divides the rotational speed range according to the bearing elastic pulse data and stress distribution data, and particularly conducts a detailed analysis on the rotational speed range of 100 rpm, and then obtains the stiffness calculation data of the bearing through the calculation of the impact amplitude stiffness. The division of the rotational speed range can reveal the changes in the mechanical characteristics of the bearing at different rotational speeds, and the calculation of the impact amplitude stiffness can directly reflect the change trend of the stiffness of the bearing during operation, further providing a basis for fault characteristics. Finally, the system fills the missing values of the curve for the stiffness calculation data to ensure the integrity and continuity of the data, and obtains an accurate bearing stiffness change curve. The bearing stiffness change curve is one of the key indicators for monitoring the health status of the bearing, and its change can directly reflect whether there is overload, wear or other potential faults in the bearing. Therefore, through a series of refined data processing methods, the system comprehensively evaluates the operating state of the bearing from multiple dimensions, can improve the accuracy of early fault diagnosis, and provides a reliable basis for subsequent fault prediction and health assessment.

[0028] Preferably, the random parameter modeling described in step S2 includes:

[0029] Performing slope analysis based on the bearing stiffness change curve to obtain the bearing stiffness curve slope;

[0030] Performing Bayesian optimization physical theory interval analysis according to the bearing stiffness curve slope, the bearing rotational speed range and the bearing infrared thermal image data to obtain the bearing physical layer theory interval data; constructing a physical constraint branch layer through the bearing physical layer theory interval data to obtain the bearing physical feature layer;

[0031] Performing cross-time-domain data analysis according to the bearing vibration signal, the bearing elastic pulse data and the bearing stress distribution data to obtain the bearing data layer time-domain data; constructing a data-driven branch layer through the bearing data layer time-domain data to obtain the bearing data feature layer;

[0032] Performing random weight allocation according to the bearing physical feature layer and the bearing data feature layer to obtain the bearing random data model.

[0033] The present invention obtains the slope of the bearing stiffness curve by performing slope analysis on the bearing stiffness change curve. This process reveals the rate at which the bearing stiffness changes over time, reflecting the damage evolution process of the bearing and its health status. The change in bearing stiffness is one of the important indicators for judging the degree of bearing fatigue and the impending failure. The slope analysis can effectively capture the signal of sudden or abnormal change in stiffness, providing key information for early fault diagnosis. Next, the system combines the slope of the bearing stiffness curve, the bearing speed range and the infrared thermal imaging data to perform Bayesian optimization physical theory interval analysis, and obtains the theoretical interval data of the bearing physical level. This process clarifies the theoretical behavior range of the bearing under different working conditions by comprehensively considering the physical constraints, so that subsequent analysis can be carried out within the theoretical physics framework, avoiding the deviation and inaccuracy caused by the pure data-driven method. Based on the physical level data, the system further constructs a physical constraint branch layer to obtain a bearing physical feature layer, which provides reliable theoretical support for the physical property analysis and fault prediction of the bearing. At the same time, the system also performs cross-time domain data analysis based on the bearing vibration signal, elastic pulse data and stress distribution data to obtain the bearing data level time domain data. Cross-time domain data analysis can extract information from multiple moments and capture the dynamic changes of bearing status, overcoming the problem of ignoring time correlation in traditional methods and providing more comprehensive bearing health assessment data. Through these time domain data, the system constructs a data-driven branch layer and generates a bearing data feature layer, further improving the modeling capability from the data level and enhancing the accuracy of the system in fault diagnosis and prediction. Finally, combining the physical feature layer and the data feature layer, the system obtains a random data model of the bearing through random weight allocation. This model closely combines physical behavior with data features, has strong robustness and adaptability, and can effectively cope with the complex and changeable working conditions and failure modes that appear in practical applications.

[0034] Preferably, the physical modeling benchmark reference in step S2 includes:

[0035] Based on the bearing stiffness change curve, the peak value of the curve is abnormally located to obtain the abnormal point of the bearing stiffness;

[0036] Perform pitting and crack detection according to the abnormal points of bearing stiffness, mark them, and obtain bearing stiffness abnormality data;

[0037] Perform physical stress analysis based on the abnormal bearing stiffness data to obtain the bearing physical stress analysis data;

[0038] The bearing physical stress analysis data is used as a physical modeling benchmark reference for the bearing random data model to obtain a bearing fault physical-data hybrid model.

[0039] The present invention locates peak anomalies in the bearing stiffness change curve to identify stiffness anomaly points. Stiffness is an important parameter reflecting the working state of the bearing, and its change is directly related to fault modes such as bearing fatigue and wear. By locating peak anomalies in the stiffness change curve, potential fault signals can be detected in a timely manner, providing accurate data support for early warning. Next, the system detects pitting and cracks based on the stiffness anomaly points, marks them, and generates bearing stiffness anomaly data. Pitting and cracks are common fault types of bearings, and marking anomaly points can provide important basic data for in-depth analysis of subsequent faults. Through further analysis of the bearing stiffness anomaly data, the system can infer the physical stress state of the bearing under specific working conditions, and then obtain bearing physical stress analysis data. Physical stress analysis can accurately quantify the stress on each part of the bearing, thereby revealing potential problems such as overload or fatigue, and helping to diagnose whether there is a risk of excessive wear or damage during the bearing operation. Finally, these physical stress analysis data are used as a reference for physical modeling of the bearing random data model, and a bearing fault physical-data hybrid model is successfully constructed. This model integrates physical stress data and a random data model, has strong adaptability and robustness, can accurately simulate the real working state of the bearing, and predict the occurrence of potential faults. Through this series of data processing and analysis, the system realizes a comprehensive diagnosis from the data level to the physical level, significantly improves the accuracy of fault detection and prediction, and can provide a more accurate and reliable basis for bearing health monitoring.

[0040] Preferably, step S3 includes the following steps:

[0041] Step S31: Obtain bearing operation data; output the bearing fault model result of the bearing fault physical-data hybrid model;

[0042] Step S32: Conduct time-domain comparative analysis on the bearing fault model result and the bearing operation data, and perform characteristic fault assessment to obtain bearing fault prediction assessment data;

[0043] Step S33: Generate hyperparameters for the bearing fault prediction assessment data according to the learning rate index of the bearing fault physical-data hybrid model to obtain bearing fault optimized hyperparameters; perform hyperparameter backpropagation optimization on the bearing fault physical-data hybrid model based on the bearing fault optimized hyperparameters to obtain a bearing fault physical-data optimized model.

[0044] The present invention obtains the operation data of the bearing and outputs the fault model result of the physical-data hybrid model of the bearing fault based on these data. This stage ensures the timeliness and accuracy of model prediction and fault diagnosis by collecting real-time operation data, providing rich dynamic data support for subsequent analysis. The time-domain comparative analysis is carried out on the bearing fault model result and the bearing operation data, and the characteristic fault evaluation is carried out. This process can intuitively reveal the change law of the bearing under different working states through time-domain analysis, and at the same time, through the characteristic fault evaluation, the potential fault characteristics and abnormal data can be identified. This kind of characteristic evaluation data can provide detailed information about the health state of the bearing, help to detect the fault signs in advance and provide data support for subsequent decision-making. Based on the learning rate index of the physical-data hybrid model of the bearing, the hyperparameters are generated for the bearing fault prediction evaluation data, and then the optimized hyperparameters of the bearing fault are obtained. This process can improve the adaptability and accuracy of the model for fault prediction by dynamically adjusting the hyperparameters in the model, ensuring high prediction accuracy under different operating conditions and working conditions. Finally, based on the obtained optimized hyperparameters, the system performs hyperparameter backpropagation optimization on the physical-data hybrid model of the bearing fault, so as to generate a more accurate and optimized physical-data optimized model of the bearing fault. Through this series of steps, the system can not only monitor and predict the bearing fault in real time, but also ensure the continuous accuracy and effectiveness of the fault prediction result through the model adaptive optimization method.

[0045] Preferably, step S32 includes the following steps:

[0046] Step S321: Perform time-domain comparative analysis on the bearing fault model result data and the bearing operation data to obtain bearing fault type comparison data;

[0047] Step S322: Extract the envelope waveform from the bearing fault type comparison data to obtain the shaft envelope waveform data; perform waveform ridge line tracking according to the shaft envelope waveform data to obtain the bearing ridge line tracking data;

[0048] Step S323: Perform characteristic fidelity evaluation according to the bearing ridge line tracking data, and the evaluation criterion is that the envelope spectrum correlation coefficient > 0.9 to obtain the bearing fault prediction evaluation data, where the characteristic fidelity evaluation includes fault frequency amplitude evaluation and harmonic amplitude deviation evaluation.

[0049] The present invention conducts time-domain comparative analysis on the result data of the bearing fault model and the actual bearing operation data. This process can reveal the deviation between the model prediction and the actual data, thereby providing strong data support for the subsequent identification and diagnosis of fault types. Through comparative analysis, the system can clearly identify the specific types of bearing faults and form bearing fault type comparison data. Envelope waveform extraction is performed on the bearing fault type comparison data to obtain shaft envelope waveform data. The envelope waveform is an important feature carrying fault information in the bearing vibration signal. By extracting the envelope waveform, the system can more clearly distinguish the details of bearing faults, especially the signal characteristics related to internal defects of the bearing (such as pitting, cracks, etc.). Based on these envelope waveform data, the system further performs waveform ridge line tracking to obtain bearing ridge line tracking data. Ridge line tracking can effectively capture the key feature points in the signal, and then reveal the specific location and characteristics of the fault source, laying a foundation for subsequent fault analysis. The system conducts feature fidelity evaluation according to the bearing ridge line tracking data, and the evaluation criterion is that the envelope spectrum correlation coefficient is greater than 0.9. Through this high-precision evaluation criterion, the system can ensure that the extracted features have high fidelity in the diagnosis process, that is, the extracted features can truly reflect the fault state of the bearing. This evaluation also includes fault frequency amplitude evaluation and harmonic amplitude deviation evaluation, and these two evaluations provide detailed and accurate data bases for determining the bearing fault type and development trend. Finally, through these multi-dimensional data analyses, the system generates bearing fault prediction evaluation data, providing accurate data support for bearing fault prediction and subsequent maintenance.

[0050] Preferably, step S4 includes the following steps:

[0051] Step S41: Trace the source of the fault cause according to the bearing fault prediction evaluation data to obtain bearing fault cause data;

[0052] Step S42: Perform fixed-point detection on the bearing based on the bearing fault cause data to obtain bearing fault detection data;

[0053] Step S43: Construct a bearing fault report by combining the bearing fault detection data and the bearing fault cause data to obtain a bearing fault simulation report.

[0054] The present invention traces the root cause of the fault based on the bearing fault prediction and evaluation data. Through in-depth analysis of the fault prediction and evaluation data, this process can accurately trace the root cause of the bearing fault, providing an important basis for further fault diagnosis and treatment. Through fault cause tracing, the system can identify the key factors affecting the bearing performance, such as overload, fatigue, wear, etc., generate bearing fault cause data, and provide clear data support for subsequent analysis. Based on the bearing fault cause data, fixed-point detection of the bearing is carried out. Through fixed-point detection technology, this process comprehensively detects the bearing at specific positions, and then obtains bearing fault detection data. Fixed-point detection can detect various parts of the bearing in a relatively accurate range, identify local potential faults, such as cracks, corrosion, pitting, etc., ensuring high-precision and high-reliability of fault detection. By combining the bearing fault detection data with the fault cause data, a bearing fault report is constructed to generate a bearing fault simulation report. By integrating multi-dimensional data, the generated fault report not only provides the specific location and type of the fault, but also reveals the cause and development trend of the fault, thus providing intuitive and detailed fault information for maintenance personnel and effectively guiding subsequent repair and maintenance work.

[0055] In this specification, a bearing fault simulation system is provided for performing the above-mentioned bearing fault simulation method. The bearing fault simulation system includes:

[0056] A data acquisition and transmission module for acquiring multi-modal bearing data to construct a bearing distributed edge data set;

[0057] A data processing and modeling module for extracting the contact dynamic data of the bearing distributed edge data set, and calculating the bearing stiffness change curve according to the contact dynamic data; performing random parameter modeling based on the bearing distributed edge data set, and using the bearing stiffness change curve as a physical modeling reference to obtain a bearing fault physical-data hybrid model;

[0058] A model verification and optimization module for outputting bearing fault model result data from the bearing fault physical-data hybrid model; performing time-domain comparative analysis on the bearing fault model result data, and conducting characteristic fault evaluation to obtain bearing fault prediction and evaluation data; generating hyperparameters for the bearing fault prediction and evaluation data according to the learning rate index of the bearing fault physical-data hybrid model, so as to dynamically adjust the bearing fault physical-data hybrid model;

[0059] A fault tracing and report generation module for tracing the root cause of the fault according to the bearing fault prediction and evaluation data and generating a bearing simulation report.

[0060] The beneficial effects of the present invention are as follows. Through the deployment of a multi-modal sensor array and the construction of a distributed edge data set, the contact dynamic data of the bearing is comprehensively collected, providing accurate data support for subsequent fault diagnosis. After data acquisition, the system can calculate the stiffness change curve by extracting the contact dynamic data, which can reflect the stiffness change of the bearing under different working conditions in real time, thereby providing key physical characteristic information for fault diagnosis. The combination of physical modeling based on the stiffness change curve and stochastic parameter modeling can effectively capture the randomness and complexity during the operation of the bearing. The generated physical-data hybrid model of bearing faults fully integrates the characteristics of physical theory and actual operation data, greatly improving the accuracy and reliability of fault diagnosis. Through time-domain comparative analysis of the results of the fault model and characteristic fault assessment, not only can the fault type of the bearing be accurately identified, but also the degree and occurrence time of the fault can be predicted, providing an accurate basis for the maintenance and replacement of the bearing. In addition, by dynamically adjusting the hyperparameters of the fault prediction evaluation data using the learning rate index, the performance of the model is effectively optimized, enabling it to continuously improve the prediction accuracy according to the actual situation. Finally, through the construction of fault cause tracing and simulation reports, a detailed fault analysis report can be provided for engineers, helping to quickly locate the root cause of the problem and optimize the maintenance decision. Overall, this system can not only provide real-time bearing fault prediction and assessment, but also improve the long-term stability and adaptability of the system through dynamic adjustment and optimization. Therefore, the present invention solves the deficiencies of traditional bearing fault diagnosis methods in terms of accuracy and real-time performance by integrating multi-modal data acquisition, physical-data hybrid modeling, and dynamic hyperparameter adjustment mechanisms, improving the accuracy and response speed of fault prediction and early warning. Brief Description of the Drawings

[0061] Figure 1 It is a schematic flow chart of the steps of a bearing fault simulation method;

[0062] Figure 2 is Figure 1 a detailed implementation step flow chart of step S3 in

[0063] The realization, functional characteristics, and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed Embodiments

[0064] The technical method of the present invention for patents will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art within the scope of the present invention without creative work fall within the protection scope of the present invention.

[0065] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0066] It should be understood that although terms such as "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0067] To achieve the above object, please refer to Figures 1 to 2 , a bearing fault simulation method, the method comprising the following steps:

[0068] Step S1: Collect multi-modal bearing data to construct a bearing distributed edge data set;

[0069] Step S2: Extract the contact dynamic data of the bearing distributed edge data set, and calculate the bearing stiffness change curve according to the contact dynamic data; perform random parameter modeling based on the bearing distributed edge data set, and use the bearing stiffness change curve as a physical modeling reference to obtain a bearing fault physical-data hybrid model;

[0070] Step S3: Output the bearing fault model result data from the bearing fault physical-data hybrid model; perform time-domain comparative analysis on the bearing fault model result data, and perform characteristic fault evaluation to obtain bearing fault prediction evaluation data; generate hyperparameters for the bearing fault prediction evaluation data according to the learning rate index of the bearing fault physical-data hybrid model, so as to dynamically adjust the bearing fault physical-data hybrid model;

[0071] Step S4: Trace the cause of the fault according to the bearing fault prediction evaluation data and generate a bearing simulation report.

[0072] The beneficial effects of the present invention are as follows. By deploying a multi-modal sensor array and constructing a distributed edge data set, the contact dynamic data of the bearing is comprehensively collected, providing accurate data support for subsequent fault diagnosis. After data acquisition, the system can calculate the stiffness change curve by extracting the contact dynamic data, which can reflect the stiffness change of the bearing under different working conditions in real time, thereby providing key physical characteristic information for fault diagnosis. Combining the physical modeling based on the stiffness change curve with the stochastic parameter modeling can effectively capture the randomness and complexity in the operation of the bearing. The generated physical-data hybrid model of bearing faults fully integrates the characteristics of physical theory and actual operation data, greatly improving the accuracy and reliability of fault diagnosis. Through the time-domain comparative analysis of the results of the fault model and the characteristic fault assessment, not only can the fault type of the bearing be accurately identified, but also the degree and occurrence time of the fault can be predicted, providing an accurate basis for the maintenance and replacement of the bearing. In addition, by dynamically adjusting the hyperparameters of the fault prediction evaluation data through the learning rate index, the performance of the model is effectively optimized, enabling it to continuously improve the prediction accuracy according to the actual situation. Finally, by tracing the fault cause and constructing a simulation report, a detailed fault analysis report can be provided for engineers, helping to quickly locate the root cause of the problem and optimize the maintenance decision.

[0073] In the embodiment of the present invention, referring to Figure 1 as shown, in this example, the bearing fault simulation method includes the following steps:

[0074] Step S1: Collect multi-modal bearing data to construct a distributed edge data set of the bearing;

[0075] In the embodiments of the present invention, the multi-modal data acquisition of bearings usually includes multiple sensor data such as vibration signals, temperature changes, noise, and torque. These data can comprehensively reflect the operating state of the bearings under different working conditions. The vibration signal acquisition usually uses sensors with high-frequency response capabilities, such as MEMS accelerometers or vibration sensors with a bandwidth of 0-20 kHz, which can capture the minute vibration changes during the operation of the bearings and provide real-time monitoring of the bearing health status. To supplement and enrich the bearing data, the temperature information is usually collected through an infrared thermal imager with an accuracy of up to ±0.5°C and a high sampling rate of 30 Hz, which enables the accurate capture of the temperature data changes under different operating conditions. In addition, the use of torque sensors provides data on the bearing force changes, which are crucial for analyzing the performance of the bearings under different load conditions. Through the data acquisition of these different modalities, the system can provide multi-dimensional information for the health monitoring of the bearings. Then, the collected data will undergo certain preprocessing to ensure the data quality, including removing noise, correcting outliers, and synchronizing the timestamps of different sensor data. This process transforms the raw data into a structured format that can be used for subsequent analysis, and finally constructs a distributed edge dataset of bearings containing multi-modal data. These data are further processed and stored through distributed edge computing technology, which can efficiently support the operation of real-time monitoring and fault warning systems.

[0076] Step S2: Extract the contact dynamic data of the distributed edge dataset of the bearings, and calculate the bearing stiffness change curve based on the contact dynamic data; perform random parameter modeling based on the distributed edge dataset of the bearings, and use the bearing stiffness change curve as the physical modeling benchmark reference to obtain the bearing fault physical-data hybrid model;

[0077] In the embodiments of the present invention, contact dynamic data is extracted from the bearing distributed edge dataset. This process involves extracting contact-related dynamic characteristics from data such as vibration, pressure, and temperature collected by multiple sensors. The contact dynamic data usually includes information such as load changes, force frequencies, and dynamic responses during the operation of the bearing. By analyzing these data, the contact state and mechanical behavior of the bearing under different working conditions can be revealed. In the data extraction stage, signal processing techniques, such as time-frequency analysis and wavelet transform, are used to effectively separate signals related to bearing contact dynamics from complex raw data. These contact dynamic data provide the basis for the subsequent calculation of the bearing stiffness change curve. According to the extracted contact dynamic data, fitting and calculation methods are used to deduce the bearing stiffness change curve. This curve reflects the stiffness change of the bearing under different load conditions, which is usually calculated through frequency response functions or mechanical models. Combining the working load and dynamic characteristics of the bearing, quantitative analysis of the stiffness change is carried out. This process requires accurate modeling and analysis of the contact dynamic data to ensure that the calculated stiffness curve can accurately reflect the actual working state of the bearing. Based on the bearing distributed edge dataset, stochastic parameter modeling is performed. By constructing parameter models suitable for different working conditions, the influence of different environmental conditions, operating loads, and working states on the bearing performance can be considered. This modeling process usually involves using stochastic process models or machine learning algorithms to train and learn the bearing operation data to establish a mathematical model that can cope with various uncertain factors. Finally, the bearing stiffness change curve will be used as a reference for physical modeling, and combined with the results of stochastic parameter modeling, a bearing fault physical-data hybrid model will be jointly constructed.

[0078] Step S3: Output the bearing fault model result data from the bearing fault physical-data hybrid model; perform time-domain comparative analysis on the bearing fault model result data and conduct characteristic fault assessment to obtain bearing fault prediction assessment data; generate hyperparameters for the bearing fault prediction assessment data according to the learning rate index of the bearing fault physical-data hybrid model, so as to dynamically adjust the bearing fault physical-data hybrid model.

[0079] In the embodiments of the present invention, the bearing fault physical-data hybrid model outputs the bearing fault model result data. This process involves applying the constructed hybrid model to the actual bearing operation data to extract relevant information on the fault state. The result data output by this model includes the characteristic manifestations of the bearing under different fault modes, such as multi-dimensional data like vibration signals, temperature changes, stress, and torque, which reflect the nature and degree of the bearing fault. These output data provide a quantitative basis for subsequent fault diagnosis and early warning. The bearing fault model result data is processed using time-domain contrast analysis. Time-domain analysis reveals the characteristic changes in the signal by comparing the time-series signals of the bearing in normal and fault states, and then identifies the time point and characteristics of the fault occurrence. This process usually requires a comparative analysis of vibration signals, temperature changes, and other sensor data, and uses differential analysis methods (such as differential operations, feature extraction, etc.) to identify the fault mode. Through time-domain contrast analysis, the signal changes caused by the fault can be accurately captured, providing data support for subsequent characteristic fault assessment. In the characteristic fault assessment stage, feature extraction and signal processing techniques are used to further analyze the fault model result data. Common technical means include envelope analysis, spectral analysis, and wavelet transform, etc., which can extract key characteristics such as fault frequency and amplitude in the vibration signal. The assessment of these characteristics not only helps to determine the fault type (such as crack, pitting, wear, etc.), but also provides a basis for predicting the remaining service life of the bearing. Through characteristic fault assessment, the fault risk of the bearing can be quantified, and bearing fault prediction assessment data can be generated. According to the learning rate index of the bearing fault physical-data hybrid model, hyperparameters are generated for the bearing fault prediction assessment data. This process adjusts the learning rate hyperparameter in the model to dynamically optimize the performance of the model. The adjustment of the learning rate can affect the convergence speed and accuracy of the model, thereby improving the accuracy of fault prediction. Hyperparameter generation uses algorithm optimization means, such as cross-validation, grid search, or Bayesian optimization, etc., to efficiently search the parameter space of the model to obtain the hyperparameter combination most suitable for the current data set.

[0080] Step S4: Trace the cause of the fault based on the bearing fault prediction assessment data and generate a bearing simulation report.

[0081] In the embodiments of the present invention, by analyzing multi-dimensional feature information in the fault prediction and evaluation data (such as vibration signals, temperature changes, load changes, etc.), key data features closely related to the occurrence of faults are identified. This process usually involves machine learning algorithms, such as decision trees, support vector machines (SVMs), deep learning, etc. By comparing historical fault data with real-time monitoring data, the key factors leading to the occurrence of faults are extracted. For example, analyzing changes in vibration frequency, drastic fluctuations in temperature, uneven load distribution, etc. all indicate potential causes of faults. Fault cause tracing not only depends on static data analysis but also needs to combine trend changes in time-series data to infer the root cause of faults. Through tracing analysis, factors leading to faults, such as material fatigue, design defects, environmental impacts, improper operations, etc., can be revealed, thus providing a basis for subsequent preventive measures. Based on the fault cause data, a bearing simulation report is constructed. The construction of the simulation report depends on in-depth analysis of the fault causes and combines these causes with the prediction results of the fault modes to form a comprehensive fault analysis report. When constructing the simulation report, the key feature data of bearing faults needs to be combined with the output of the prediction model, and data visualization techniques (such as fault tree analysis, heat maps, curve graphs, etc.) are used to present the process and development trend of bearing faults. In addition, the report also needs to describe in detail the performance changes under different fault scenarios, provide comparative analysis, and quantitatively evaluate the impacts of different fault modes. For example, the report can include the analysis results of multi-dimensional data such as vibration frequency changes, temperature distribution, load changes, etc. of the bearing at different fault stages, as well as the fault occurrence mechanism based on physical modeling and data analysis.

[0082] Preferably, the construction of the bearing distributed edge dataset includes:

[0083] Deploy a multi-modal sensor array on the bearing test bench, where the multi-modal sensor array includes vibration sensors, torque sensors, and infrared thermal imagers;

[0084] Use a MEMS accelerometer and a vibration sensor with a bandwidth of 0 - 20 kHz to collect bearing vibration signals;

[0085] Use a torque sensor with a parameter of 0.1 rpm resolution to collect bearing material property data;

[0086] Use an infrared thermal imager with a sampling rate of 30 Hz and an accuracy of ±0.5 °C to collect bearing infrared thermal image data;

[0087] Perform data preprocessing on the bearing vibration signals, bearing material property data, and bearing infrared thermal image data to obtain the bearing distributed edge dataset.

[0088] In the embodiments of the present invention, synchronous acquisition of multiple data sources is achieved by deploying a multi-modal sensor array on a bearing test bench, aiming to comprehensively monitor the operating state of the bearing. These sensors include vibration sensors, torque sensors, and infrared thermal imagers, which can obtain the operating data of the bearing from different physical dimensions respectively. The vibration sensor uses a MEMS accelerometer with a bandwidth of 0 - 20 kHz, which can effectively capture the high-frequency components in the bearing vibration signal and provide key data on the dynamic response of the bearing. These vibration signal data can provide accurate information for the assessment of the bearing health state, especially in scenarios where high-frequency noise or vibration changes are significant. The torque sensor is used to collect bearing material property data with a resolution of 0.1 rpm, which can monitor the dynamic changes of the bearing material under different operating conditions with high precision. Especially when affected by load changes, it can provide data reflecting the material performance and fatigue degree. The infrared thermal imager plays a role in temperature monitoring during this process, with a sampling rate of 30 Hz and an accuracy of ±0.5 °C, which can accurately capture the minute fluctuations in the bearing surface temperature, and this is of great significance for analyzing the thermal state of the bearing and predicting potential failures (such as overheating, increased friction, etc.). For the multi-modal data obtained by these sensors, the data preprocessing step is a key link to ensure the effectiveness of subsequent analysis. The data preprocessing process includes operations such as denoising, synchronization, calibration, and standardization to remove errors introduced due to environmental interference, sensor errors, or signal drift, etc., so as to ensure the accuracy and consistency of the data. This process usually involves steps such as signal filtering, timestamp alignment, and missing value filling, etc., in order to merge and analyze data from different sources on the same time scale. Finally, these preprocessed data will be constructed into a distributed edge data set of the bearing for subsequent health monitoring and fault diagnosis.

[0089] Preferably, the data preprocessing of the bearing vibration signal, bearing material property data, and bearing infrared thermal image data includes the following steps:

[0090] Adopt the sliding window statistical method to remove electromagnetic interference abnormal pulses with an amplitude range of ±15 g from the bearing vibration signal, bearing material property data, and bearing infrared thermal image data, and perform timestamp synchronization processing to obtain bearing multi-modal joint data;

[0091] Perform wavelet packet decomposition with a sub-band frequency of 0 - 3.9 kHz on the bearing multi-modal joint data to obtain bearing multi-modal decomposition data;

[0092] Perform cleaning on the distributed edge data set with a single-node storage capacity of ≥500 MB / h according to the bearing multi-modal decomposition data to obtain the bearing distributed edge data set.

[0093] In the embodiments of the present invention, the sliding window statistical method is used to remove electromagnetic interference abnormal pulses from the bearing vibration signal, bearing material attribute data, and bearing infrared thermal image data. In this process, by setting a threshold with an amplitude range of ±15g, the abnormal pulse data caused by electromagnetic interference is screened out, reducing the influence of environmental noise on data acquisition. The application of the sliding window statistical method ensures the smoothing of data changes within the time window, helps to eliminate sudden interferences, and makes the data more conform to the actual physical laws. Then, timestamp synchronization processing is carried out to ensure that the data of different sensors can be compared and analyzed at the same time point, which is crucial for subsequent multi-modal data fusion and correlation analysis. Especially for the simultaneously collected vibration, material attribute, and thermal image data, time synchronization can ensure the time consistency of the data to the greatest extent. Through these two technical means, the bearing multi-modal joint data is obtained, which can comprehensively reflect the working state of the bearing. Secondly, based on the bearing multi-modal joint data, wavelet packet decomposition of the sub-band frequency of 0 - 3.9 kHz is performed. This operation analyzes the signal in multiple levels and frequency bands through wavelet packet decomposition, decomposes the signal into multiple frequency bands, and thus can extract the characteristic information of different frequency bands, which helps to reveal the working characteristics of the bearing in different frequency ranges. Especially for the processing of vibration signals, it can reveal potential fault characteristics or abnormal fluctuations. The wavelet packet decomposition provides more detailed and accurate characteristic data for subsequent fault diagnosis and health monitoring. Finally, according to the bearing multi-modal decomposition data, the distributed edge data set cleaning with a single-node storage capacity ≥500 MB / h is carried out. In the data cleaning stage, the distributed edge data set is filtered and optimized, redundant data and irrelevant information are removed, ensuring the efficiency of storage and transmission and the data quality. This cleaning process not only improves the efficiency of data processing but also provides clean and accurate input data for the subsequent analysis stage.

[0094] Preferably, the calculation of the stiffness change curve described in step S2 includes:

[0095] Identifying elastic pulses of the bearing based on the bearing distributed edge data set to obtain bearing elastic pulse data;

[0096] Extracting the ball diameter according to the bearing distributed edge data set and calculating the theoretical contact frequency to obtain bearing stress distribution data;

[0097] Segmenting the rotational speed in the range of 100 rpm according to the bearing elastic pulse data and the bearing stress distribution data to obtain the bearing rotational speed range; calculating the impact amplitude stiffness for the bearing rotational speed range to obtain bearing stiffness calculation data;

[0098] Filling in the missing values of the curve according to the bearing stiffness calculation data to obtain the bearing stiffness change curve.

[0099] In the embodiments of the present invention, elastic pulse identification is performed on the bearing distributed edge data set to extract bearing elastic pulse data. This step uses signal processing techniques, such as pulse detection algorithms or threshold-based event detection methods, to identify elastic pulse signals generated by the interaction between the bearing and its load from the vibration data. These pulse signals reflect the dynamic characteristics and abnormal behaviors of the bearing during operation. Next, the ball diameter information is extracted based on the bearing distributed edge data set, and the theoretical contact frequency is calculated to obtain the stress distribution data of the bearing. This process combines the material mechanics model and geometric parameters to deduce the motion trajectory of the balls in the bearing and its contact mechanical characteristics, thereby accurately calculating the stress distribution. The contact frequency of the bearing is closely related to its working state, and the theoretical calculation results provide the key physical background and stress distribution data for subsequent analysis. Subsequently, the bearing elastic pulse data and stress distribution data are combined and analyzed in intervals segmented based on a rotational speed of 100 rpm to obtain the rotational speed interval data of the bearing. The division of the rotational speed interval can effectively reveal the performance changes of the bearing under different working states, especially for fault diagnosis and performance evaluation. Then, the impact amplitude stiffness of the bearing in each rotational speed interval is calculated to obtain the bearing stiffness calculation data. This technical means calculates the stiffness change by analyzing the impact amplitude in the bearing vibration signal and combining the physical characteristics of the bearing. As a key indicator reflecting the bearing's load-bearing capacity and operating state, stiffness can effectively indicate the damage progress and fault state of the bearing. Finally, based on the bearing stiffness calculation data, the missing values in its change curve are filled to obtain a complete bearing stiffness change curve. The missing value filling technique uses interpolation or regression methods to ensure the continuity and integrity of the data, thereby avoiding analysis errors caused by data missing and providing a more accurate stiffness change trend for subsequent fault analysis.

[0100] Preferably, the random parameter modeling described in step S2 includes:

[0101] Perform slope analysis on the bearing stiffness change curve to obtain the bearing stiffness curve slope;

[0102] Perform Bayesian optimization physical theory interval analysis based on the bearing stiffness curve slope, the bearing rotational speed interval, and the bearing infrared thermal image data to obtain the bearing physical layer theory interval data; construct the physical constraint branch layer through the bearing physical layer theory interval data to obtain the bearing physical feature layer;

[0103] Perform cross-time-domain data analysis based on the bearing vibration signal, the bearing elastic pulse data, and the bearing stress distribution data to obtain the bearing data layer time-domain data; construct the data-driven branch layer through the bearing data layer time-domain data to obtain the bearing data feature layer;

[0104] Random weight assignment is performed according to the bearing physical feature layer and the bearing data feature layer to obtain a bearing random data model.

[0105] In the embodiment of the present invention, slope analysis is performed based on the bearing stiffness change curve. This step uses the numerical differentiation method to calculate the change rate of the bearing stiffness curve at different points, thereby obtaining the slope of the bearing stiffness. As an important index, the slope can reveal the sensitivity of the bearing stiffness to changes in time or working conditions, and is the basis for analyzing key issues such as bearing fatigue damage and abnormal wear. Then, Bayesian optimization physical theory interval analysis is performed in combination with the bearing stiffness curve slope, the bearing speed range, and the bearing infrared thermal image data. This step deeply analyzes the correlation between physical parameters such as speed and temperature and the bearing stiffness change to obtain the theoretical interval data of the bearing at the physical level. These data provide theoretical support for further understanding the physical behavior of the bearing and can provide a physical background for fault diagnosis. Based on these physical level data, a physical constraint branch layer is constructed to form a bearing physical feature layer. This step uses constraint optimization and model construction methods to integrate physical theory and actual data, establishing a multi-dimensional feature description system, which helps to improve the accuracy and interpretability of the model. Cross-time domain data analysis is performed in combination with the bearing vibration signal, the bearing elastic pulse data, and the bearing stress distribution data to obtain the time domain data at the bearing data level. This process uses time domain analysis methods to systematically model and analyze the dynamic behavior of the bearing, and can capture the state characteristics of the bearing at different time points. Especially through cross-time domain analysis, the health state evolution of the bearing under different time periods or operating conditions is revealed. The data-driven branch layer is constructed in combination with the time domain data to obtain the bearing data feature layer. This step uses machine learning and deep learning algorithms to extract higher-level features from the time domain data, automatically mining the potential laws in the data, and providing data support for subsequent fault prediction and health assessment. Random weight assignment is performed according to the bearing physical feature layer and the data feature layer to obtain a bearing random data model. This process uses a multi-dimensional fusion technology based on a weighted model, and assigns different weight values to the data of different feature layers, thereby comprehensively considering the advantages of physical and data-driven models to generate a random data model.

[0106] Preferably, the physical modeling benchmark reference described in step S2 includes:

[0107] Curve peak anomaly positioning is performed based on the bearing stiffness change curve to obtain the bearing stiffness anomaly points;

[0108] Pitting and crack detection are performed according to the bearing stiffness anomaly points and marked to obtain bearing stiffness anomaly data;

[0109] Physical stress force analysis is performed based on the bearing stiffness anomaly data to obtain bearing physical stress analysis data;

[0110] Using the bearing physical stress analysis data as a physical modeling reference for the bearing random data model, a bearing fault physical-data hybrid model is obtained.

[0111] In the embodiments of the present invention, the technical means for abnormal peak positioning based on the bearing stiffness change curve is to detect peaks and identify outliers in the bearing stiffness curve by applying signal processing methods (such as Fourier transform, wavelet transform, etc.), and extract abnormal fluctuations or peaks in the curve. These abnormal peaks usually correspond to the fault signs of the bearing, such as the initial stage of excessive wear or damage. After locating these peaks, potential bearing fault points can be further identified. Next, according to the identified abnormal bearing stiffness points, pitting and crack detection are carried out and marked. This process involves image processing technology and signal analysis methods. Usually, through high-frequency vibration signal analysis, combined with machine learning algorithms to classify different fault modes, the positions of pitting and cracks on the bearing surface are automatically marked. These marks will constitute the bearing stiffness abnormal data, providing input for subsequent fault analysis. Based on the bearing stiffness abnormal data, physical stress analysis is carried out, which is to evaluate the stress and load on the bearing under abnormal conditions through physical models and simulation calculations. This process usually uses numerical simulation technologies such as finite element analysis (FEA) to model the geometric shape, material properties, and load conditions of the bearing, and obtain the stress distribution map of the bearing under abnormal conditions. These physical stress analysis data provide a basis for further fault diagnosis, reveal the impact of faults on the bearing performance, and provide a deeper physical background for the specific location where the fault occurs. The bearing physical stress analysis data and the bearing random data model are used as a physical modeling reference to generate a bearing fault physical-data hybrid model. This process utilizes the combination of data fusion and physical modeling. First, the physical stress analysis results are used as a reference, combined with the actual operating data of the bearing (such as vibration signals, temperature data, etc.), and through the construction of a random data model, the influence of the physical state of the bearing and the actual operating environment is comprehensively considered.

[0112] As an example of the present invention, refer to Figure 2 As shown, in this example, step S3 includes:

[0113] Step S31: Obtain bearing operation data; output the bearing fault model result of the bearing fault physical-data hybrid model;

[0114] Step S32: Conduct time-domain comparative analysis on the bearing fault model result and the bearing operation data, and conduct characteristic fault evaluation to obtain bearing fault prediction evaluation data;

[0115] Step S33: Generate hyperparameters for the bearing fault prediction and evaluation data according to the learning rate index of the bearing fault physical-data hybrid model to obtain optimized hyperparameters for bearing faults; perform hyperparameter backpropagation optimization on the bearing fault physical-data hybrid model based on the optimized hyperparameters for bearing faults to obtain an optimized bearing fault physical-data model.

[0116] In the embodiment of the present invention, the bearing operation data is acquired and the fault model result of the bearing fault physical-data hybrid model is output. The specific technical means is to collect the real-time operation data of the bearing through a sensor array (such as vibration sensors, temperature sensors, etc.). The data includes vibration signals, temperature changes, pressure, or rotational speed, etc. These data are analyzed by an edge computing or central processing unit to generate a preliminary bearing fault physical-data hybrid model result. This model combines physical theory and data-driven methods, and can fuse the actual operation data with theoretical analysis, so as to more accurately depict the health state of the bearing. Perform time-domain comparative analysis on the bearing fault model result and the bearing operation data, and conduct characteristic fault evaluation to finally obtain the bearing fault prediction and evaluation data. The technical means of this step includes time-domain comparative analysis of signals. Usually, by comparing the data generated by the fault model and the actually collected operation data, their trends, amplitude changes, and frequency responses and other characteristics are compared in the time domain. This can be achieved through statistical methods such as correlation analysis and covariance analysis. On this basis, perform characteristic fault evaluation, and use methods such as time-frequency analysis and transient response analysis to identify whether there is a fault in the bearing, as well as the type and severity of the fault, so as to generate the bearing fault prediction and evaluation data. Generate hyperparameters based on the learning rate index of the bearing fault physical-data hybrid model and optimize the model. This technical means mainly involves hyperparameter tuning techniques in machine learning methods, such as grid search, random search, or Bayesian optimization, etc. By analyzing the bearing fault prediction and evaluation data, hyperparameters such as the learning rate and regularization coefficient in the model are adjusted. This process can be optimized through an automated machine learning framework (AutoML). During the optimization process, the parameters of the fault prediction model are adjusted through a feedback mechanism to improve its recognition ability and prediction accuracy for different fault modes.

[0117] Preferably, step S32 includes the following steps:

[0118] Step S321: Perform time-domain comparative analysis on the bearing fault model result data and the bearing operation data to obtain bearing fault type comparison data;

[0119] Step S322: Extract the envelope waveform of the bearing fault type comparison data to obtain shaft envelope waveform data; perform waveform ridge line tracking on the shaft envelope waveform data to obtain bearing ridge line tracking data;

[0120] Step S323: Conduct feature fidelity evaluation based on the bearing ridge line tracking data. The evaluation criterion is that the envelope spectrum correlation coefficient > 0.9, and bearing fault prediction evaluation data is obtained. The feature fidelity evaluation includes fault frequency amplitude evaluation and harmonic amplitude deviation evaluation.

[0121] In the embodiment of the present invention, time-domain comparative analysis is performed on the bearing fault model result data and the bearing operation data to obtain bearing fault type comparison data. Technically, time-domain comparative analysis analyzes the similarity and difference between the data generated by the fault model and the actually collected bearing operation data in the time domain by comparing the time characteristics of the signals, such as features like amplitude, waveform, frequency, etc., so as to accurately distinguish different types of faults. The core of this step is to identify and distinguish different bearing fault types by analyzing the change pattern of the data in the time domain. By extracting the envelope waveform of the bearing fault type comparison data, shaft envelope waveform data is obtained, and waveform ridge line tracking is performed to obtain bearing ridge line tracking data. The envelope waveform extraction technology extracts envelope information related to faults based on the characteristic frequencies of vibration signals, which helps to reveal the fault mode of the bearing, especially the high-frequency components hidden in the low-frequency band signals. Ridge line tracking is to identify the health status of the bearing by extracting key features in the envelope signal and using the change trend of the high-frequency components. Usually, methods such as filters and Hilbert transforms are used to process the signals to obtain the envelope waveform and ridge line tracking data, further providing effective information for fault evaluation. Feature fidelity evaluation is performed using the bearing ridge line tracking data. The evaluation criterion is that the envelope spectrum correlation coefficient > 0.9, and bearing fault prediction evaluation data is obtained. Feature fidelity evaluation mainly ensures that the extracted features accurately reflect the bearing fault information by comparing the correlation between the actually collected data and the model data. The envelope spectrum correlation coefficient is one of the criteria for measuring the similarity of two groups of signals in the frequency domain or time domain and can effectively detect the integrity and consistency of the signals. If the correlation coefficient is greater than 0.9, it indicates that the extracted features are highly consistent with the actual fault signals, thus providing highly reliable evaluation data for fault prediction.

[0122] Preferably, step S4 includes the following steps:

[0123] Step S41: Trace the source of the fault cause based on the bearing fault prediction evaluation data to obtain bearing fault cause data;

[0124] Step S42: Conduct fixed-point detection of the bearing based on the bearing fault cause data to obtain bearing fault detection data;

[0125] Step S43: Construct a bearing fault report by combining the bearing fault detection data and the bearing fault cause data to obtain a bearing fault simulation report.

[0126] In the embodiments of the present invention, by tracing the fault causes based on the bearing fault prediction and evaluation data, bearing fault cause data is obtained. Tracing the fault causes usually relies on in-depth analysis and correlation of fault modes, combining machine learning algorithms, a fault mode library, and previously known fault cases to infer the fault sources. This step generally uses a data-driven pattern recognition method to establish a mapping relationship model by comparing the fault prediction and evaluation data (such as fault frequency, amplitude, pulse, etc.) with historical data, and can infer the specific fault causes through the collected multi-dimensional data. Based on the bearing fault cause data, fixed-point detection of the bearing is carried out to obtain bearing fault detection data. This process uses fixed-point monitoring technology to select specific monitoring points or moments according to the bearing fault cause data to conduct detailed detection of the bearing. Technical means at the data level include collecting the local states of the bearing surface, structure, or interior, using sensors (such as accelerometers, temperature sensors, strain gauges, etc.) to monitor the operating state of the bearing in real time, and analyzing whether there are abnormal behaviors related to the fault causes. In this process, the collected data includes vibration signals, temperature changes, stress changes, etc., and the state of the bearing is accurately diagnosed through multi-modal sensor fusion technology to detect the exact location and type of the fault. The bearing fault detection data and the bearing fault cause data are combined to construct a bearing fault report. The technical means of this step include correlating the data analysis results with the fault cause tracing and detection results, and generating the corresponding fault report. This report is usually processed by an automated system to integrate the fault mode, fault location, cause, and the results of the detection data of the bearing to form a detailed diagnostic report.

[0127] In this specification, a bearing fault simulation system is provided for performing the above-mentioned bearing fault simulation method. The bearing fault simulation system includes:

[0128] A data acquisition and transmission module for acquiring multi-modal bearing data to construct a bearing distributed edge data set;

[0129] A data processing and modeling module for extracting the contact dynamic data of the bearing distributed edge data set, and calculating the bearing stiffness change curve according to the contact dynamic data; performing random parameter modeling based on the bearing distributed edge data set, and using the bearing stiffness change curve as a physical modeling reference to obtain a bearing fault physical-data hybrid model;

[0130] A model verification and optimization module for outputting bearing fault model result data from the bearing fault physical-data hybrid model; performing time-domain comparative analysis on the bearing fault model result data, and conducting characteristic fault evaluation to obtain bearing fault prediction and evaluation data; generating hyperparameters for the bearing fault prediction and evaluation data according to the learning rate index of the bearing fault physical-data hybrid model, so as to dynamically adjust the bearing fault physical-data hybrid model;

[0131] A fault tracing and report generation module, which is used to trace the cause of the fault according to the bearing fault prediction and evaluation data and generate a bearing simulation report.

[0132] The beneficial effects of the present invention are as follows: Through the deployment of a multi-modal sensor array and the construction of a distributed edge data set, the contact dynamic data of the bearing are comprehensively collected, providing accurate data support for subsequent fault diagnosis. After data acquisition, the system can reflect the stiffness change of the bearing under different working conditions in real time by extracting the contact dynamic data and calculating the stiffness change curve, thereby providing key physical characteristic information for fault diagnosis. The combination of physical modeling based on the stiffness change curve and stochastic parameter modeling can effectively capture the randomness and complexity in the bearing operation. The generated physical-data hybrid model of bearing faults fully integrates the characteristics of physical theory and actual operation data, greatly improving the accuracy and reliability of fault diagnosis. Through the time-domain comparative analysis of the results of the fault model and the characteristic fault evaluation, not only can the fault type of the bearing be accurately identified, but also the degree and occurrence time of the fault can be predicted, providing an accurate basis for the maintenance and replacement of the bearing. In addition, the hyperparameters are dynamically adjusted according to the learning rate index for the fault prediction and evaluation data, effectively optimizing the performance of the model and enabling it to continuously improve the prediction accuracy according to the actual situation. Finally, through the tracing of the cause of the fault and the construction of a simulation report, a detailed fault analysis report can be provided for engineers, helping to quickly locate the root cause of the problem and optimize the maintenance decision. Generally speaking, this system can not only provide real-time bearing fault prediction and evaluation, but also improve the long-term stability and adaptability of the system through dynamic adjustment and optimization. Therefore, the present invention solves the deficiencies of traditional bearing fault diagnosis methods in terms of accuracy and real-time performance by integrating multi-modal data acquisition, physical-data hybrid modeling and dynamic hyperparameter adjustment mechanism, and improves the accuracy and response speed of fault prediction and early warning.

[0133] Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be included in the present invention.

[0134] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A bearing fault simulation method, characterized in that: The following steps are involved: Step S1: Collect multimodal bearing data to construct a bearing distributed edge dataset; Step S2: extracting contact dynamic data of the bearing distributed edge data set, and calculating the bearing stiffness change curve according to the contact dynamic data; performing random parameter modeling based on the bearing distributed edge data set, and taking the bearing stiffness change curve as a physical modeling benchmark reference to obtain a bearing fault physical-data hybrid model; Step S3: outputting the bearing fault model result data from the bearing fault physics-data hybrid model; Perform time domain comparative analysis on the result data of the bearing fault model and perform characteristic fault evaluation to obtain bearing fault prediction evaluation data; generate hyperparameters for the bearing fault prediction evaluation data based on the learning rate index of the bearing fault physical-data hybrid model, thereby dynamically adjusting the bearing fault physical-data hybrid model; Step S4: trace the cause of the fault according to the bearing fault prediction and evaluation data, and generate a bearing simulation report.

2. The bearing fault simulation method according to claim 1, characterized in that: The step S1 of constructing a bearing distributed edge dataset includes: Deploy a multimodal sensor array on the bearing test bench, where the multimodal sensor array includes a vibration sensor, a torque sensor, and an infrared thermal imager; Use MEMS accelerometers and vibration sensors with a bandwidth of 0-20kHz to collect bearing vibration signals; The torque sensor parameters are used to collect bearing material property data with a resolution of 0.1rpm; The infrared thermal image data of the bearing is collected using an infrared thermal imager with a sampling rate of 30Hz and an accuracy of ±0.5℃; The bearing vibration signal, bearing material property data and bearing infrared thermal image data are preprocessed to obtain the bearing distributed edge data set.

3. The bearing fault simulation method according to claim 2, characterized in that: The data preprocessing of the bearing vibration signal, the bearing material property data and the bearing infrared thermal image data comprises: The sliding window statistical method is used to remove abnormal electromagnetic interference pulses with an amplitude range of ±15g from the bearing vibration signal, bearing material property data and bearing infrared thermal image data, and timestamp synchronization is performed to obtain bearing multi-modal joint data. According to the bearing multi-modal joint data, the sub-band frequency 0-3.9kHz wavelet packet decomposition is performed to obtain the bearing multi-modal decomposition data; According to the multimodal decomposition data of the bearing, the distributed edge data set with a single node storage capacity ≥ 500MB / h is cleaned to obtain the bearing distributed edge data set.

4. The bearing fault simulation method according to claim 1, characterized in that: The stiffness change curve calculation in step S2 includes: Perform elastic pulse identification based on the distributed edge data set of the bearing to obtain the elastic pulse data of the bearing; The ball diameter is extracted based on the bearing distributed edge data set, and the theoretical contact frequency is calculated to obtain the bearing stress distribution data; According to the bearing elastic pulse data and the bearing stress distribution data, the speed is segmented into 100 rpm intervals to obtain the bearing speed interval; the impact amplitude stiffness of the bearing speed interval is calculated to obtain the bearing stiffness calculation data; The missing values ​​of the curve are filled according to the bearing stiffness calculation data to obtain the bearing stiffness change curve.

5. The bearing fault simulation method according to claim 1, characterized in that: The random parameter modeling in step S2 includes: Perform slope analysis based on the bearing stiffness change curve to obtain the slope of the bearing stiffness curve; According to the slope of the bearing stiffness curve, the bearing speed range and the bearing infrared thermal image data, a Bayesian optimization physical theory interval analysis is performed to obtain the theoretical interval data of the bearing physical level; the physical constraint branch layer is constructed through the theoretical interval data of the bearing physical level to obtain the bearing physical characteristic layer; Perform cross-time domain data analysis based on bearing vibration signals, bearing elastic pulse data, and bearing stress distribution data to obtain time domain data at the bearing data level; construct a data-driven branch layer through the time domain data at the bearing data level to obtain a bearing data feature layer; Random weights are allocated according to the bearing physical feature layer and the bearing data feature layer to obtain a bearing random data model.

6. The bearing fault simulation method according to claim 1, characterized in that: The physical modeling benchmark reference in step S2 includes: Based on the bearing stiffness change curve, the peak value of the curve is abnormally located to obtain the abnormal point of the bearing stiffness; Perform pitting and crack detection according to the abnormal points of bearing stiffness, mark them, and obtain bearing stiffness abnormality data; Perform physical stress analysis based on the abnormal bearing stiffness data to obtain the bearing physical stress analysis data; The bearing physical stress analysis data is used as a physical modeling benchmark reference for the bearing random data model to obtain a bearing fault physical-data hybrid model.

7. The bearing fault simulation method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: Acquire bearing operation data; output a bearing fault model result of a bearing fault physical-data hybrid model; Step S32: performing time domain comparative analysis on the bearing fault model result and the bearing operation data, and performing characteristic fault evaluation to obtain bearing fault prediction evaluation data; Step S33: Generate hyperparameters for the bearing fault prediction evaluation data according to the learning rate index of the bearing fault physics-data hybrid model to obtain the bearing fault optimization hyperparameters; perform hyperparameter backpropagation optimization on the bearing fault physics-data hybrid model based on the bearing fault optimization hyperparameters to obtain the bearing fault physics-data optimization model.

8. The bearing fault simulation method according to claim 7, characterized in that: Step S32 includes the following steps: Step S321: performing time domain comparative analysis on the bearing fault model result data and the bearing operation data to obtain bearing fault type comparative data; Step S322: extracting the envelope waveform of the bearing fault type comparison data to obtain bearing envelope waveform data; performing waveform ridge tracking according to the bearing envelope waveform data to obtain bearing ridge tracking data; Step S323: Perform feature fidelity evaluation based on the bearing ridge tracking data, with the evaluation standard being envelope spectrum correlation coefficient>0.9, to obtain bearing fault prediction evaluation data, wherein the feature fidelity evaluation includes fault frequency amplitude evaluation and harmonic amplitude deviation evaluation.

9. The bearing fault simulation method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: tracing the cause of the failure according to the bearing failure prediction and evaluation data to obtain bearing failure cause data; Step S42: performing bearing fixed-point detection based on the bearing fault cause data to obtain bearing fault detection data; Step S43: construct a bearing fault report based on the bearing fault detection data and the bearing fault cause data to obtain a bearing fault simulation report.

10. A bearing fault simulation system, characterized in that: Used to execute the bearing fault simulation method according to claim 1, the bearing fault simulation system comprises: Data collection and transmission module, used to collect multi-modal bearing data to build a distributed edge data set of bearings; The data processing and modeling module is used to extract the contact dynamics data of the bearing distributed edge data set and calculate the bearing stiffness change curve based on the contact dynamics data; perform random parameter modeling based on the bearing distributed edge data set, and use the bearing stiffness change curve as a physical modeling benchmark reference to obtain a bearing fault physical-data hybrid model; The model verification and optimization module is used to output the bearing fault model result data from the bearing fault physical-data hybrid model; perform time domain comparative analysis on the bearing fault model result data, and perform characteristic fault evaluation to obtain bearing fault prediction evaluation data; generate hyperparameters for the bearing fault prediction evaluation data according to the learning rate index of the bearing fault physical-data hybrid model, so as to dynamically adjust the bearing fault physical-data hybrid model; The fault tracing and report generation module is used to trace the cause of the fault based on the bearing fault prediction and evaluation data and generate a bearing simulation report.

Images