A bearing fault prediction evaluation system based on data augmentation

Abstract

The application discloses a bearing fault prediction evaluation system based on data enhancement, and relates to the technical field of intelligent fault diagnosis and prediction.The system comprises a data acquisition module for extracting bearing operation feature vectors from real-time monitoring data, a data enhancement and training module for generating an extended sample set through a data enhancement technique and training a deep learning model using the sample set to obtain a fault prediction model, an online calibration module for online calibration of the fault prediction model according to real-time monitoring data and historical data and output of optimized abnormal detection indexes, and a threshold adjustment module for dynamic adjustment of abnormal detection thresholds according to the optimized abnormal detection indexes.The bearing fault prediction evaluation system based on data enhancement avoids equipment downtime and production efficiency reduction caused by failure to discover faults in time, helps to reduce maintenance costs and prolong the service life of equipment, and significantly improves the overall reliability and production efficiency of industrial equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent fault diagnosis and prediction, and particularly relates to a bearing fault prediction evaluation system based on data enhancement. BACKGROUND

[0002] With the continuous development of industrial equipment intelligence and automation, equipment fault diagnosis and prediction has become one of the key technologies to improve production efficiency and equipment reliability. In mechanical equipment, bearings as important bearing components, their fault detection and prediction technology have attracted widespread attention. Traditional bearing fault diagnosis methods usually rely on vibration signal analysis, combined with time domain, frequency domain and time-frequency domain feature extraction, for fault recognition and classification. However, with the increasing complexity of industrial equipment operating state, traditional fault prediction systems face a series of challenges, such as complex fault types, unbalanced data, and frequent false alarms and missed alarms in real-time monitoring process.

[0003] In order to solve the above problems, in recent years, bearing fault prediction systems based on machine learning and data enhancement have been widely studied. Data enhancement technology generates diverse samples to increase the size of the training data set and improve the generalization ability of the model. However, in the real-time monitoring process, the existing bearing fault detection system often has difficulty in adapting to changing working environment and operating state, resulting in frequent false alarms or missed alarms, affecting the accuracy and stability of the prediction system. In these systems, the setting of fault detection threshold is crucial to the performance of the system. The traditional fixed threshold method has limitations and cannot adapt to the dynamic changes of equipment state. The method of dynamically adjusting the anomaly detection threshold can improve the accuracy of prediction to a certain extent and reduce the occurrence of false alarms and missed alarms by monitoring the changes of real-time data and feedback of historical fault data. However, how to design a reasonable dynamic adjustment mechanism so that the adjustment of the threshold can match the real-time operating state of the equipment is still a difficulty in current technology. SUMMARY

[0004] The purpose of the present application is to provide a bearing fault prediction evaluation system based on data enhancement, which dynamically adjusts the anomaly detection threshold to solve the problem of frequent false alarms and missed alarms in real-time monitoring process.

[0005] To achieve the above purpose, the present application provides the following technical scheme: a bearing fault prediction evaluation system based on data enhancement, the system comprising:

[0006] a data acquisition module for extracting bearing operating feature vectors from real-time monitoring data, acquiring real-time monitoring data through sensors, the real-time monitoring data including vibration signals, temperature signals and noise signals;

[0007] The data enhancement and training module connected with the data acquisition module is used to generate an extended sample set through a data enhancement technique and train a deep learning model using the sample set to obtain a fault prediction model.

[0008] The online calibration module connected with the data enhancement and training module is used to calibrate the fault prediction model online according to real-time monitoring data and historical data, output an optimized abnormality detection index, and compare the real-time monitoring data and the historical data, adjust the change rate of bearing faults of the prediction model, compare the real-time monitoring data and the historical data, and adjust the decay rate constant of the prediction model to ensure that the prediction model can reflect the dynamic changes of the health status and fault development of the equipment.

[0009] The online calibration module further comprises an error correction unit for adjusting the parameters of the prediction model according to the difference value output by the data comparison unit to reduce errors and improve prediction accuracy.

[0010] The threshold adjustment module connected with the online calibration module is used to dynamically adjust the abnormality detection threshold according to the optimized abnormality detection index, including measuring the rate of equipment fault wear according to real-time monitoring data.

[0011] The detection threshold is adjusted according to the rate of bearing equipment fault wear to ensure that the equipment will not be false or missed due to environmental factors or equipment aging.

[0012] Preferably, the data acquisition module extracts the bearing operation feature vector from the real-time monitoring data, including evaluating the parameters of each bearing monitoring point, determining the fault degree value of each bearing monitoring point, and calculating the bearing fault evaluation index, and the specific formula is:

[0013] Wherein, E represents the bearing fault evaluation index, M represents the number of all bearing monitoring collection points, Q represents the fault degree value of the bearing monitoring point, C represents the bearing fault influence factor, and i represents the index of the bearing monitoring point.

[0014] Preferably, the data enhancement and training module generates an extended sample set through a data enhancement technique and trains a deep learning model using the sample set to obtain a fault prediction model, including generating enhanced samples by simulating different fault types to enhance the sample set, changing different equipment fault working conditions, and enhancing the model generalization ability, and the specific formula is: G = X / T.

[0015] Wherein, G represents the speed of sample generation in the data enhancement process, X represents the feature offset after data sample transformation, and T represents the time spent in the data enhancement process.

[0016] Preferably, the online calibration module calibrates the fault prediction model according to real-time monitoring data and historical data, and outputs an optimized abnormality detection index including a comparison between real-time monitoring data and historical data, adjusts a change rate of bearing fault of the prediction model, compares real-time monitoring data and historical data, and adjusts a decay rate constant of the prediction model, so as to ensure that the prediction model can reflect dynamic changes of the health status and fault development of the equipment, and the specific formula is: Z(t) = Z0e -αt ;

[0017] Wherein, Z(t) represents the change rate of bearing fault, t represents time, Z0 represents the initial fault state of the model, e represents the base number of natural logarithm, and a represents the decay rate constant of the fault prediction model.

[0018] Preferably, the data enhancement and training module generates an extended sample set through a data enhancement technique, and trains a deep learning model by using the sample set to obtain a fault prediction model. The data enhancement technique in the fault prediction model includes sample expansion, data disturbance and data interpolation operation on an existing bearing fault data set to generate a diversified data set and improve the generalization ability of the trained model.

[0019] Preferably, the threshold adjustment module dynamically adjusts the abnormality detection threshold according to the optimized abnormality detection index, and the calculation formula of the loss quantization value B of the bearing due to fault wear is: B = L / R.

[0020] Wherein, B represents the loss quantization value of the bearing due to fault wear, L represents the loss amount caused by the fault within a given time, and R represents the time period of the fault loss.

[0021] Preferably, the threshold adjustment module dynamically adjusts the abnormality detection threshold according to the optimized abnormality detection index, and the threshold adjustment further includes calculating a real-time abnormality score based on the optimized abnormality detection index, obtaining a maximum normal score S max and a minimum fault score S min in historical abnormality detection, and adjusting the threshold, and the specific formula is:

[0022] U = (S max +S min ) / 2 x v.

[0023] Wherein, U represents the threshold, S max represents the maximum normal score, S min represents the minimum fault score, and v represents a weight coefficient.

[0024] Preferably, the threshold adjustment module dynamically adjusts the abnormality detection threshold according to the optimized abnormality detection index, and the specific formula for measuring the rate of equipment fault wear according to real-time monitoring data is: r = B / (A x D).

[0025] Wherein, r represents the rate of bearing equipment failure wear, B represents the loss quantitative value of bearing due to failure wear, A represents the bearing surface area, and D represents the time period of bearing failure wear.

[0026] From the above technical solution, the present application has the following beneficial effects:

[0027] The bearing operation feature vector is extracted from the real-time monitoring data by the data acquisition module, the data enhancement and training module generates an expanded sample set through data enhancement technology, and the deep learning model is trained using the sample set to obtain a fault prediction model. The online calibration module performs online calibration on the fault prediction model according to the real-time monitoring data and historical data, and outputs the optimized abnormal detection index. The threshold adjustment module dynamically adjusts the abnormal detection threshold according to the optimized abnormal detection index, can generate diversified samples, significantly increase the size of the training data set, and thus improve the generalization ability of the fault diagnosis model. Especially in dealing with complex fault types and variable working environments, it can effectively improve the prediction accuracy and more accurately match the real-time running state of the equipment. Compared with the traditional fixed threshold method, the dynamic adjustment mechanism of the present application can significantly reduce the false positives and false negatives caused by environmental changes and equipment aging, can adapt to the running state changes of the equipment in real time, and improve the stability of the system. This mechanism ensures that the system can maintain high prediction performance when facing different fault modes and working environments, further improves the accuracy and reliability of real-time monitoring and fault prediction, avoids equipment downtime and production efficiency due to failure to discover in time, helps to reduce maintenance costs and prolong the service life of the equipment, and thus significantly improves the overall reliability and production efficiency of industrial equipment. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The system module connection diagram of the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0030] As shown in Figure 1 A technical solution is provided by the present application: a bearing fault prediction evaluation system based on data enhancement, which comprises:

[0031] A data acquisition module is used to extract bearing operation feature vectors from real-time monitoring data.

[0032] The data augmentation and training module connected with the data acquisition module is used to generate an extended sample set through a data augmentation technique and train a deep learning model using the sample set to obtain a fault prediction model.

[0033] The online calibration module connected with the data augmentation and training module is used to perform online calibration on the fault prediction model according to real-time monitoring data and historical data and output an optimized abnormality detection index.

[0034] The threshold adjustment module connected with the online calibration module is used to dynamically adjust an abnormality detection threshold according to the optimized abnormality detection index, including measuring a rate of equipment fault wear according to real-time monitoring data, and a specific formula is as follows: r = B / (A x D).

[0035] Wherein, r represents the rate of bearing equipment fault wear, B represents a loss quantification value of the bearing due to fault wear, A represents a bearing surface area, and D represents a time period of bearing fault wear.

[0036] The detection threshold is adjusted according to the rate of bearing equipment fault wear to ensure that the equipment will not be false reported or missed reported due to environmental factors or equipment aging.

[0037] In an embodiment of the present application, the system extracts the operating data of the bearing from the monitoring device in real time through the data acquisition module. This data contains the dynamic characteristics of the bearing under different working environments and is collected by sensors through multiple physical parameters such as temperature, vibration and sound. The data acquisition module transmits these real-time data to the data enhancement module. The data enhancement module uses specific algorithms to process the collected data to generate a diversified enhanced sample set to make up for the deficiencies in the original data. The enhanced data set is used to train the deep learning model, thereby constructing a fault prediction model capable of predicting bearing failures. In order to ensure the prediction accuracy and adaptability of the model, the system will calibrate the model online according to the historical data, and continuously optimize the output of the fault detection index. The results of these optimizations help the system to make fault predictions under different operating conditions and identify potential faults in the early stages of the bearing. By adjusting the fault detection threshold and optimizing the model parameters, the system can respond to abnormal conditions in the operation of the equipment in a timely manner, thereby ensuring the smooth operation of the equipment. The present application has significant beneficial effects. First, the data enhancement technology generates diversified training samples, which can effectively solve the problem of insufficient fault data and improve the accuracy of the fault prediction model. The enhanced sample set not only enriches the training data of the model, but also improves the model's ability to recognize different fault patterns, thereby ensuring the stability and reliability of the prediction results. Second, the online adjustment module of the system can dynamically optimize the fault prediction model according to the real-time data and historical data of the equipment, continuously adjust the detection parameters, and improve the sensitivity and accuracy of fault detection, so that the equipment can be monitored in real time and respond quickly to potential faults during operation. In addition, the system can identify and predict potential faults in the bearing in advance, helping maintenance personnel to take measures in advance to avoid sudden equipment failures, thereby significantly reducing equipment downtime and reducing maintenance and replacement costs. Overall, the present application not only improves the reliability of the equipment, but also prolongs the service life of the equipment and reduces unnecessary maintenance costs, providing an efficient and accurate fault prediction solution for industrial production.

[0038] The data acquisition module extracts the bearing operating feature vector from the real-time monitoring data, which includes evaluating the parameters of each bearing monitoring point, determining the fault degree value of each bearing monitoring point, and calculating the bearing fault evaluation index, with the specific formula being:

[0039] Wherein, E represents the bearing fault evaluation index, M represents the number of all bearing monitoring collection points, Q represents the fault degree value of the bearing monitoring point, C represents the bearing fault influence factor, and i represents the index of the bearing monitoring point.

[0040] In one possible implementation, the data acquisition module extracts the operating data of the bearings through real-time monitoring of the equipment, covering physical characteristics such as temperature, vibration, and pressure. These data provide important basis for evaluating bearing failures. Through this data, the system not only obtains the failure degree of each bearing monitoring point, but also calculates the failure degree value in the process of real-time data acquisition. Specifically, the system uses the formula: wherein E is the bearing failure evaluation index of the entire system, representing the comprehensive failure condition of the bearings; M is the number of all monitoring collection points of the bearings; Q i is the failure degree of the i-th monitoring point; and C i represents the influencing factors of the failure of the monitoring point. The bearing failure evaluation index E is obtained by weighted calculation of the failure degrees of all monitoring points, and the weight is determined by the influencing factors. Finally, through the calculation of the failure evaluation value, the system can monitor and optimize the state of the bearings in real time, and provide decision support for subsequent maintenance and optimization. The present application can comprehensively evaluate the failure of the equipment by calculating the bearing failure evaluation index E, and accurately reflect the operating condition of the equipment. By collecting and calculating the failure degree of each bearing monitoring point in real time, the system can establish a fine failure analysis model between different monitoring points, thereby identifying potential failure risks. The bearing failure evaluation index E integrates the failure degrees and influencing factors of multiple monitoring points, so that the system can not only accurately diagnose the failure position of the equipment, but also can give early warning and response according to the severity of the failure. In this way, the system greatly improves the accuracy and response speed of bearing failure detection, effectively prolongs the service life of the equipment, reduces the downtime of the equipment, and reduces the cost of maintenance and maintenance.

[0041] The data augmentation and training module generates an expanded sample set through data augmentation technology, and trains a deep learning model using the sample set to obtain a failure prediction model. The sample set is generated by simulating different fault types, changing the fault working conditions of different equipment, and enhancing the generalization ability of the model. The specific formula is: G = X / T;

[0042] wherein G represents the speed of sample generation in the data augmentation process, X represents the characteristic offset of the data sample after transformation, and T represents the time spent in the data augmentation process.

[0043] In one possible implementation, the data augmentation and training module generates an expanded sample set by simulating different fault types. First, the data augmentation technique processes the original data by adjusting and transforming different device fault conditions to generate new sample data. This process simulates data characteristics under different fault modes through specific algorithms, thereby generating representative augmented samples. These augmented samples are used for training the deep learning model to enhance the model's generalization ability, enabling it to identify more unknown fault types. Specifically, the sample generation speed G in the data augmentation process can be quantified by the formula G = X / T, where X is the feature offset of the data samples after transformation, reflecting the degree of sample feature change, and T is the time required to generate these augmented samples. By adjusting these parameters, the system can optimize the data augmentation process while maintaining efficiency, generating high-quality training samples and improving the accuracy and reliability of the fault prediction model. The expanded sample set generated by the data augmentation technique of the present invention improves the training effect of the deep learning model. By simulating different types of faults and generating augmented samples, the system can cover more fault modes, enabling the model to handle more diverse fault conditions and improving the model's generalization ability. The generation of augmented samples accelerates the model training process, ensuring that the model learns enough fault features in a short time and improving the sensitivity and accuracy of fault detection. In addition, the data augmentation technique avoids overfitting caused by insufficient training data by adjusting sample features and transforming conditions, enhancing the system's adaptability in actual applications, reducing equipment downtime and maintenance costs, and improving the overall performance of the system.

[0044] The online calibration module calibrates the fault prediction model based on real-time monitoring data and historical data, outputs the optimized abnormal detection index including the comparison of real-time monitoring data and historical data, adjusts the change rate of bearing fault of the prediction model, compares the real-time monitoring data and historical data, adjusts the decay rate constant of the prediction model, and ensures that the prediction model can reflect the dynamic changes of the health status and fault development of the equipment. The specific formula is: Z(t) = Z0e -αt ;

[0045] Wherein, Z(t) represents the change rate of bearing fault, t represents time, Z0 represents the initial fault state of the model, e represents the base of natural logarithm, and a represents the decay rate constant of the fault prediction model.

[0046] In this embodiment, the system combines real-time monitoring data with historical data through an online standard module to optimize the fault prediction model. Real-time monitoring data is continuously collected by sensors to capture the operating characteristics of the bearing and compared with historical fault data to adjust the output of the fault prediction model, ensuring that the model accurately reflects the dynamic changes in the health status and fault development of the equipment. Specifically, the system expresses the change rate of bearing failure through the formula Z(t) = Z0e -αt , where Z(t) is the failure rate at any time t, Z0 is the initial value of the failure state, and a is the change rate constant, representing the development rate of the failure. This formula dynamically adjusts the model parameter a by comparing real-time data with historical data, enabling accurate prediction of changes in the health status of the equipment and further optimizing the fault prediction capability of the bearing. The present invention dynamically adjusts the change rate of the fault prediction model in real time, enabling the system to respond to changes in the state of the equipment in real time and accurately assess the health status and fault development trend of the bearing. Through the formula Z(t) = Z0e -αt , the system can dynamically adjust the failure rate at different time points, ensuring that the fault prediction model works effectively throughout the life cycle of the equipment. In addition, the comparison of historical data and real-time monitoring data optimizes the accuracy of fault prediction, enabling early warning of potential failures and reducing production downtime and maintenance costs caused by equipment failure. Dynamic adjustment of the fault prediction model also improves the adaptability of the system, enabling it to better cope with changes in different working environments and equipment operating conditions, thereby enhancing the reliability and stability of equipment operation.

[0047] The data acquisition module extracts the bearing operating feature vector from the real-time monitoring data. Real-time monitoring data is obtained through sensors, including vibration signals, temperature signals, and noise signals.

[0048] In this embodiment, the system extracts the operating feature vector of the bearing from the real-time monitoring device through the data acquisition module. Real-time monitoring data is obtained through sensors installed on the device, including multiple signal types: vibration signals, temperature signals, and noise signals. Vibration signals reflect the mechanical vibration situation of the bearing, temperature signals indicate the thermal state of the bearing, and noise signals help identify abnormal sounds that may occur during the operation of the bearing. Through real-time collection and analysis of these signals, the data acquisition module converts them into feature vectors as input data for subsequent fault prediction models. These feature vectors, combined with other processing modules such as the data enhancement module, will be used to train deep learning models to improve the accuracy and reliability of fault detection. This invention combines vibration signals, temperature signals, and noise signals and other real-time monitoring data to fully utilize the rich information generated during the operation of the bearing, improving the accuracy of fault detection. By collecting different types of signals in real time, the running state of the bearing can be fully understood, and potential faults can be detected in a timely manner. For example, vibration signals can effectively reflect mechanical problems of the bearing, temperature signals can help determine whether the bearing is overheating, and noise signals can identify relatively small abnormal sounds. By converting these signals into feature vectors, the system can more accurately identify and predict bearing failures, improving the efficiency of equipment maintenance and early warning capabilities. In addition, using sensor-acquired data, the system can achieve real-time monitoring, reducing human factors that interfere with data collection, making fault detection more reliable and efficient.

[0049] The data augmentation and training module generates an expanded sample set through data augmentation techniques and trains a deep learning model using the sample set to obtain a fault prediction model. Data augmentation techniques include generating a diversified data set by expanding the sample set, data perturbation, and data interpolation operation on the existing bearing fault data set to improve the generalization ability of the trained model. In this embodiment, the data augmentation and training module generates an expanded sample set through various data augmentation techniques to improve the accuracy and generalization ability of the fault prediction model. First, by expanding the sample set of the existing bearing fault data set, new samples are generated using the existing data. These samples retain the basic characteristics of the original data and are appropriately transformed to enhance the diversity of the data. Second, the data perturbation technique is used to apply noise or make small changes to the original data to simulate various working conditions that may occur in the real world to enhance the robustness of the model. Finally, through data interpolation operation, intermediate samples are generated by interpolation algorithm combined with existing data points to fill the data gap and enrich the content of the data set. Through these data augmentation techniques, the diversified data set generated is used to train the deep learning model to improve the generalization ability of the fault prediction model and make it adapt to bearing fault conditions in different working environments. The present application effectively improves the diversity and richness of the data set through various data augmentation techniques such as sample expansion, data perturbation, and data interpolation. By increasing the sample size, data augmentation techniques enable the model to learn more potential fault features during training, thereby improving the model's generalization ability and prediction accuracy. In particular, data perturbation and interpolation operations can simulate the performance of bearings under different working conditions, making the model more robust when facing different types of faults, reducing problems caused by insufficient training data or overfitting. In addition, the expanded data set can help the model better handle rare fault conditions, thereby improving the system's fault prediction ability and ensuring the health management and early warning of equipment.

[0050] The threshold adjustment module dynamically adjusts the abnormality detection threshold according to the optimized abnormality detection index. The calculation formula of the loss quantification value B caused by bearing failure wear is B = L / R.

[0051] Wherein, B represents the loss quantification value caused by bearing failure wear, L represents the loss amount caused by failure within a given time, and R represents the time period of failure loss.

[0052] In this embodiment, the threshold adjustment module dynamically adjusts the anomaly detection threshold in the fault prediction model according to the optimized anomaly detection index. Specifically, the loss quantification value B represents the loss caused by the bearing failure due to wear and tear, which is calculated by the formula B = L / R, where L is the loss amount caused by the failure within a given time period, and R is the time period during which the loss occurs. In this way, the system can monitor the loss caused by bearing failure in real time and adjust the threshold for anomaly detection according to these loss quantification values, thereby achieving dynamic evaluation of the health status of the equipment. The threshold adjustment module adjusts the prediction range of the fault prediction model according to the changing loss quantification value, so that the system can respond and adapt to changes in different working conditions, ensuring the accuracy and reliability of fault prediction. By dynamically adjusting the anomaly detection threshold and combining the loss quantification value B, the present application can more accurately reflect the loss caused by the bearing failure due to wear and tear, improving the real-time performance and sensitivity of the fault detection system. According to the loss amount L caused by the failure and the time period R, the loss quantification value B calculated provides an effective index for measuring the health status of the equipment. In this way, the system can adjust the detection threshold in real time during equipment operation and provide accurate fault warning according to actual conditions, avoiding fault detection errors caused by overly lenient or overly strict threshold settings. Dynamic adjustment of the threshold not only improves the adaptability of the system, but also effectively reduces the downtime of the equipment, reduces maintenance costs, and prolongs the service life of the equipment.

[0053] The threshold adjustment module dynamically adjusts the anomaly detection threshold based on the optimized anomaly detection index, which further includes calculating a real-time anomaly score based on the optimized anomaly detection index, obtaining the maximum normal score S max and the minimum fault score S min in historical anomaly detection, and adjusting the threshold, with the specific formula being:

[0054] U = (S max +S min ) / 2 x v.

[0055] where U represents the threshold, S max represents the maximum normal score, S min represents the minimum fault score, and v represents the weight coefficient.

[0056] In this embodiment, the threshold adjustment module calculates a real-time anomaly score and dynamically adjusts the threshold based on the optimized anomaly detection index. Specifically, the system obtains the maximum normal score S max and the minimum fault score S min from historical anomaly detection data, which reflect the anomaly detection results of the equipment in normal operation and fault states, respectively. The threshold U is adjusted according to the formula U = (S max +S min) / 2 x v, where v is a weight coefficient used to adjust the influence of the maximum normal score and the minimum failure score. This threshold is used to determine the health status of the equipment during real-time monitoring. The real-time anomaly score is compared with the adjusted threshold, and if the score exceeds the threshold, the system will issue a warning signal to alert potential equipment failure risks. The present application can more accurately assess the health status of the equipment by dynamically adjusting the anomaly detection threshold. Through the calculation of the maximum normal score S max and the minimum failure score S min , the system can adjust the threshold based on the historical operation data of the equipment, making the anomaly detection more accurate under different working conditions. After introducing the weight coefficient v, the system can flexibly adjust the sensitivity of the threshold, thereby optimizing the failure warning mechanism according to different application scenarios. This method can reduce false positives and false negatives caused by excessively low or high threshold settings, ensuring the accuracy and timeliness of failure detection. Ultimately, this dynamic adjustment mechanism improves the reliability of failure warning, helping maintenance personnel to take effective measures in a timely manner when potential problems occur in the equipment, thereby avoiding serious damage to the equipment and reducing maintenance costs.

[0057] The online calibration module further comprises a data comparison unit for comparing the real-time monitoring data with the historical data and outputting the difference values required for model calibration. In this embodiment, the online calibration module can compare and analyze the real-time monitoring data and the historical data through the design of the data comparison unit. The data comparison unit calculates the difference values between the real-time collected bearing operation data and the historical operation data of the equipment. These difference values are crucial parameters for model calibration. Specifically, the real-time data reflects the current working state of the bearing, while the historical data is a standard reference based on the past equipment operation and failure mode. By calculating the difference between the real-time data and the historical data, the data comparison unit can extract potential abnormal signals and trends, thereby providing accurate calibration data for the failure prediction model. These calibration data further optimize the failure prediction model, improve its accuracy and adaptability, and enable more accurate failure prediction under different working environments and equipment states. The present application compares the real-time monitoring data with the historical data through the introduction of the data comparison unit, generates the difference values required for calibration, and enables the system to timely reflect the changes and potential failure risks of the equipment. Through comparative analysis, the system can capture the deviation and change between the real-time data and the historical data, thereby effectively calibrating and optimizing the model and enhancing the accuracy and stability of the model. This method is particularly suitable for dynamic working environments and can help the system automatically adjust and optimize the failure prediction model according to the running conditions and environmental conditions of different equipment. Real-time model calibration improves the early warning capability of the equipment, ensures timely response when the equipment is abnormal, reduces the risk of equipment failure, and improves the overall production efficiency and equipment reliability.

[0058] The online calibration module further includes an error correction unit for adjusting the parameters of the prediction model according to the difference value output by the data comparison unit, reducing errors and improving prediction accuracy. In this embodiment, the online calibration module includes an error correction unit in addition to the data comparison unit. The error correction unit further adjusts the parameters of the prediction model according to the difference value output by the data comparison unit. Specifically, after comparing the real-time monitoring data with the historical data, the data comparison unit outputs the difference value between the two, reflecting the deviation between the model prediction and the actual state. The error correction unit adjusts the parameters of the prediction model (such as weights, thresholds, etc.) based on these difference values, thereby reducing the prediction error of the model. This adjustment process enables the model to more accurately adapt to the current device operating state and environmental conditions, improving the accuracy and reliability of the prediction. Through repeated adjustment and correction, the system continuously optimizes the prediction model, enabling it to maintain high prediction accuracy during device operation, detect potential faults in a timely manner, and accurately warn. By introducing the error correction unit, the present application effectively reduces the prediction error caused by changes in device operating state, environmental factor interference, etc., further improving the accuracy of the fault prediction model. By calculating and adjusting the model parameters in real time, the error correction unit enables the prediction model to be more flexible and adaptable, thereby maintaining high prediction accuracy under different operating conditions. This dynamic error correction mechanism greatly improves the system's fault detection capability, ensuring that the device can discover potential problems and issue warnings in a timely manner during operation, reducing downtime and maintenance costs caused by device failures. In addition, the error correction unit can continuously adjust the parameters according to real-time data and historical data, ensuring that the system can adapt to the aging process of the device and changes in the environment, improving the long-term stability of the system.

[0059] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data augmentation based bearing failure prediction assessment system, characterized in that, The system comprises: a data acquisition module for extracting bearing operation feature vectors from real-time monitoring data; a data enhancement and training module connected to the data acquisition module, for generating an expanded sample set through data enhancement techniques and training a deep learning model using the sample set to obtain a fault prediction model; an online calibration module connected to the data enhancement and training module, for online calibration of the fault prediction model according to real-time monitoring data and historical data, outputting an optimized abnormality detection index, including comparison of real-time monitoring data and historical data, adjusting the change rate of bearing faults of the prediction model, comparing real-time monitoring data and historical data, adjusting the decay rate constant of the prediction model, and ensuring that the prediction model can reflect the dynamic changes of the health status and fault development of the equipment; a threshold adjustment module connected to the online calibration module, for dynamically adjusting the abnormality detection threshold according to the optimized abnormality detection index, including measuring the rate of equipment fault wear according to real-time monitoring data; adjusting the detection threshold according to the rate of bearing equipment fault wear to ensure that the equipment will not be false positive or false negative due to environmental factors or equipment aging; The online calibration module calibrates the fault prediction model according to real-time monitoring data and historical data, and outputs an optimized abnormality detection index, including comparison of real-time monitoring data and historical data, adjustment of a change rate of bearing fault of the prediction model, comparison of real-time monitoring data and historical data, and adjustment of a decay rate constant of the prediction model, so as to ensure that the prediction model can reflect dynamic changes of the health status and fault development of the equipment, and the specific formula is: Z(t) = Z0 x e -αt ; wherein Z(t) represents the change rate of bearing faults, t represents time, Z0 represents the initial fault state of the model, e represents the base of the natural logarithm, and a represents the decay rate constant of the fault prediction model; The threshold adjusting module dynamically adjusts the abnormality detection threshold according to the optimized abnormality detection index. A calculation formula of a loss quantification value B caused by bearing failure wear in the abnormality detection threshold is as follows: wherein B represents the loss quantification value of the bearing due to fault wear, L represents the loss caused by the fault within a given time, and R represents the time period of the fault loss. The threshold adjusting module dynamically adjusts the anomaly detection threshold according to the optimized anomaly detection index, and the method further comprises the following steps of: calculating a real-time anomaly score based on the optimized anomaly detection index, obtaining a maximum normal score S max and a minimum fault score S min adjusting the threshold, and the specific formula is as follows: where U represents a threshold value, S max represents a maximum normal score, S min represents a minimum failure score, and υ represents a weight coefficient.

2. The data-augmentation-based bearing failure prediction evaluation system of claim 1, wherein: The data acquisition module extracts bearing operation feature vectors from real-time monitoring data, including evaluating the parameters of each bearing monitoring point, determining the fault degree value of each bearing monitoring point, and calculating the bearing fault evaluation index, with the specific formula being: wherein E represents the bearing fault evaluation index, M represents the number of all bearing monitoring collection points, Q represents the fault degree value of the bearing monitoring point, C represents the bearing fault influence factor, and i represents the index of the bearing monitoring point.

3. The data-augmentation-based bearing failure prediction evaluation system of claim 1, wherein: The data enhancement and training module generates an expanded sample set through a data enhancement technique and trains a deep learning model using the sample set to obtain a fault prediction model, including generating enhanced samples by simulating different fault types, enhancing the model generalization capability by changing different equipment fault working conditions, and the specific formula is: wherein G represents the speed of sample generation in the data enhancement process, X represents the feature offset after data sample transformation, and T represents the time spent in the data enhancement process.

4. The data-augmentation-based bearing failure prediction evaluation system of claim 1, wherein: The data acquisition module extracts bearing operation feature vectors from real-time monitoring data, which includes vibration signals, temperature signals, and noise signals.

5. The data-augmentation-based bearing failure prediction evaluation system of claim 1, wherein: The data enhancement and training module generates an expanded sample set through data enhancement techniques and trains a deep learning model using the sample set to obtain a fault prediction model. The data enhancement techniques include sample expansion, data perturbation, and data interpolation operations on existing bearing fault data sets to generate diversified data sets and improve the generalization ability of the training model.

6. The data-augmentation-based bearing failure prediction evaluation system of claim 1, wherein: The online calibration module further includes a data comparison unit for comparing real-time monitoring data with historical data and outputting the difference value required for model calibration.

7. The data-augmentation-based bearing failure prediction evaluation system of claim 1, wherein: The online calibration module further includes an error correction unit for adjusting the parameters of the prediction model according to the difference value output by the data comparison unit to reduce errors and improve prediction accuracy.

Images