A device fundamental frequency data generation system and method based on device fundamental frequency prediction

Abstract

The present invention relates to a system and method for generating equipment fundamental frequency data based on equipment fundamental frequency prediction, wherein the system includes a data acquisition module, an abnormal screening module, a weight configuration module and a fundamental frequency prediction module. The data acquisition module is used to obtain vibration signals and fundamental frequency data under various operating conditions; the abnormal screening module is used to identify and remove abnormal values in the fundamental frequency data; the weight configuration module is used to configure training weights according to operating condition characteristics and abnormal data analysis results; the fundamental frequency prediction module is used to train a fundamental frequency predictor and generate fundamental frequency data of a target device. The present application solves the technical problems in the prior art that the fundamental frequency data acquisition process of rotating equipment has abnormal data interference, and the reliability of data under different operating conditions has large differences, resulting in low fundamental frequency prediction accuracy, and achieves the technical effect of improving the accuracy and robustness of equipment fundamental frequency prediction through abnormal data screening driven by operating condition characteristics and differentiated weight configuration.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and particularly to a device fundamental frequency data generation system and method based on device fundamental frequency prediction. Background Art

[0002] Rotating devices are widely used in industrial production, and the monitoring of their operating states is of great significance for ensuring the safe operation of the devices and production efficiency. The fundamental frequency of a rotating device, as an important parameter of the device operating state, its accurate acquisition has important value for device state monitoring, fault diagnosis, and predictive maintenance.

[0003] Currently, the acquisition of rotating device fundamental frequency data is mainly through direct collection by sensors or analysis and extraction from vibration signals. However, in the actual industrial environment, the device will have different degrees of wear and noise interference under different operating conditions, resulting in large errors in the collected fundamental frequency data. Especially when the device operates for a long time or works under high-load conditions, the collected fundamental frequency data may contain a large number of outliers, seriously affecting the accuracy and reliability of the fundamental frequency data. Traditional fundamental frequency prediction methods usually adopt a unified data processing standard, ignoring the differences in data reliability under different operating conditions, and unable to effectively screen out abnormal data. At the same time, the prior art also fails to perform differential processing according to the operating condition characteristics and the degree of data abnormality, resulting in abnormal data having a large interference on the prediction result during the model training process, ultimately affecting the accuracy and robustness of the fundamental frequency prediction. Summary of the Invention

[0004] Aiming at the technical problems in the prior art that there are abnormal data interferences in the process of collecting rotating device fundamental frequency data and the large differences in data reliability under different operating conditions lead to low fundamental frequency prediction accuracy, the present invention provides a device fundamental frequency data generation system and method based on device fundamental frequency prediction to solve the problems.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] In a first aspect, the present invention provides a device fundamental frequency data generation system based on device fundamental frequency prediction, comprising: a data acquisition module, configured to obtain a variety of operating conditions, collect vibration signals of a device under the variety of operating conditions within a historical time, and perform fundamental frequency processing and annotation to obtain a plurality of sample condition vibration signal sets and a plurality of sample condition fundamental frequency data sets; an abnormal screening module, configured to obtain condition characteristic information of a variety of operating conditions, configure a plurality of abnormal fundamental frequency screening ratios, respectively perform isolated abnormal fundamental frequency screening on the plurality of sample condition fundamental frequency data sets, obtain a plurality of abnormal fundamental frequency data sets and delete them, to obtain a plurality of normal condition vibration signal sets and a plurality of normal condition fundamental frequency data sets; a weight configuration module, configured to calculate the abnormality degrees of the plurality of abnormal fundamental frequency data sets, and in combination with a plurality of condition characteristic information, configure a plurality of initial training weights; a fundamental frequency prediction module, configured to, according to the plurality of initial training weights, use the plurality of normal condition vibration signal sets and the plurality of normal condition fundamental frequency data sets to train a device fundamental frequency predictor, collect vibration signals of a current target device, input them into the device fundamental frequency predictor, and predict and output to obtain device fundamental frequency data.

[0007] Optionally, the data acquisition module includes: a condition acquisition unit, configured to obtain a variety of operating conditions of a target device; a vibration signal acquisition unit, configured to collect vibration signals of a same-type device under the variety of operating conditions according to historical operation monitoring data of the same-type device, to obtain a plurality of sample condition vibration signal sets; a fundamental frequency processing and annotation unit, configured to process each sample condition vibration signal by Fourier transform and harmonic product spectrum to obtain fundamental frequency data, and perform annotation to obtain a plurality of sample condition fundamental frequency data sets.

[0008] Optionally, the abnormal screening module includes: a condition characteristic acquisition unit, configured to obtain the operating time and average operating load of a same-type device under a variety of operating conditions as a plurality of condition characteristic information; a coefficient calculation unit, configured to calculate, within the plurality of condition characteristic information, the ratio of each operating time to the mean of the plurality of operating times to obtain a plurality of time coefficients, and calculate the ratio of each average operating load to the mean of the plurality of average operating loads to obtain a plurality of load coefficients; an adjustment coefficient calculation unit, configured to calculate a plurality of screening ratio adjustment coefficients according to the plurality of time coefficients and the plurality of load coefficients; a preset ratio acquisition unit, configured to obtain a preset abnormal fundamental frequency screening ratio; an abnormal ratio configuration unit, configured to respectively perform configuration calculation on the preset abnormal fundamental frequency screening ratio by using the plurality of screening ratio adjustment coefficients to obtain a plurality of abnormal fundamental frequency screening ratios.

[0009] Optionally, the abnormal screening module further includes: a fundamental frequency interval obtaining unit for obtaining the fundamental frequency data interval of the device; an isolated data screening unit for randomly generating first fundamental frequency data within the fundamental frequency data interval, performing binary classification on the multiple sample working condition fundamental frequency data sets, and determining whether there is sample working condition fundamental frequency data classified as isolated fundamental frequency data; an abnormal data classification unit for, if so, classifying the isolated fundamental frequency data as abnormal fundamental frequency data, and if not, continuing to randomly generate second fundamental frequency data for isolated abnormal fundamental frequency screening; a screening ratio control unit for stopping the isolated abnormal fundamental frequency screening until the proportion of abnormal fundamental frequency data divided within each sample working condition fundamental frequency data set reaches the multiple abnormal fundamental frequency screening ratios, and obtaining multiple screened abnormal fundamental frequency data sets; a normal data obtaining unit for obtaining the multiple abnormal working condition vibration signal sets corresponding to the multiple abnormal fundamental frequency data sets, deleting them, and obtaining multiple normal working condition vibration signal sets and multiple normal working condition fundamental frequency data sets.

[0010] Optionally, the weight configuration module includes: an average value calculation unit for respectively calculating the average values of the multiple normal working condition fundamental frequency data sets to obtain multiple average normal fundamental frequency data; an abnormality calculation unit for respectively calculating the deviation amplitudes between the abnormal fundamental frequency data in the multiple abnormal fundamental frequency data sets and the multiple average normal fundamental frequency data to obtain multiple abnormal deviation amplitude sets, and calculating the average value to obtain multiple abnormality degrees.

[0011] Optionally, the weight configuration module further includes: a time weight calculation unit for calculating, within the multiple working condition characteristic information, the ratio of each working condition running time to the sum of the multiple working condition running times as multiple running time weights; a load weight calculation unit for calculating the ratio of each average running load to the sum of the multiple average running loads as multiple load weights; an abnormal weight allocation unit for allocating, according to the multiple abnormality degrees, multiple abnormal training weights for multiple operating conditions, where the magnitude of the abnormality degree is negatively correlated with the magnitude of the abnormal training weight; a weight fusion calculation unit for performing weight fusion calculation according to the multiple running time weights, multiple load weights, and multiple abnormal training weights to obtain multiple initial training weights.

[0012] Optionally, the fundamental frequency prediction module includes: a training sample combination unit configured to combine the plurality of normal operating condition vibration signal sets and the plurality of normal operating condition fundamental frequency data sets to obtain multiple sets of normal fundamental frequency training sample sets and obtain a plurality of normal sample data amounts; a sample weight allocation unit configured to respectively divide the plurality of initial training weights by the plurality of normal sample data amounts to perform data-level weight allocation to obtain a plurality of sample data weight sets; a predictor construction unit configured to construct an equipment fundamental frequency predictor based on machine learning; a predictor training unit configured to use the multiple sets of normal fundamental frequency training sample sets to allocate training resources according to the plurality of sample data weight sets to perform supervised training and testing on the equipment fundamental frequency predictor until convergence; and a fundamental frequency data generation unit configured to collect the vibration signal of the current target equipment, input same into the equipment fundamental frequency predictor, and predict and output to obtain the equipment fundamental frequency data.

[0013] In a second aspect, the present invention provides a method for generating equipment fundamental frequency data based on equipment fundamental frequency prediction, including: obtaining a plurality of operating conditions, collecting vibration signals of an equipment when in the plurality of operating conditions during a historical time and performing fundamental frequency processing and annotation to obtain a plurality of sample operating condition vibration signal sets and a plurality of sample operating condition fundamental frequency data sets; obtaining condition characteristic information of the plurality of operating conditions, configuring a plurality of abnormal fundamental frequency screening ratios, respectively performing isolated abnormal fundamental frequency screening on the plurality of sample operating condition fundamental frequency data sets to obtain a plurality of abnormal fundamental frequency data sets and deleting same to obtain a plurality of normal operating condition vibration signal sets and a plurality of normal operating condition fundamental frequency data sets; calculating the abnormality degrees of the plurality of abnormal fundamental frequency data sets, and combining the plurality of condition characteristic information to configure a plurality of initial training weights; training an equipment fundamental frequency predictor according to the plurality of initial training weights by using the plurality of normal operating condition vibration signal sets and the plurality of normal operating condition fundamental frequency data sets, collecting the vibration signal of the current target equipment, inputting same into the equipment fundamental frequency predictor, and predicting and outputting to obtain the equipment fundamental frequency data.

[0014] The beneficial effects of the present invention are:

[0015] The data acquisition module obtains various operating conditions, collects vibration signals of the equipment under various operating conditions in historical time and performs fundamental frequency processing and annotation, obtaining multiple sample condition vibration signal sets and multiple sample condition fundamental frequency data sets. By collecting sample data under different conditions, it provides basic data support for subsequent fundamental frequency prediction, ensuring the diversity and representativeness of the data source; the abnormal screening module obtains the condition characteristic information of various operating conditions, configures multiple abnormal fundamental frequency screening ratios, separately performs isolated abnormal fundamental frequency screening on multiple sample condition fundamental frequency data sets, obtains multiple abnormal fundamental frequency data sets and deletes them, obtaining multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets, thereby addressing the problem of sample fundamental frequency data errors caused by possible wear, noise, etc. under different conditions of the equipment. Through the method of isolating abnormal data screening, the quality of sample data is improved; the weight configuration module calculates the abnormality degrees of multiple abnormal fundamental frequency data sets, combines multiple condition characteristic information, and configures multiple initial training weights, such that the greater the abnormality degree of a condition, the smaller its corresponding weight; this differential weight configuration ensures the reasonable influence degree of data under different conditions in model training; the fundamental frequency prediction module trains the equipment fundamental frequency predictor according to multiple initial training weights, using multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets, collects the vibration signal of the current target equipment, inputs it into the equipment fundamental frequency predictor, and predicts and outputs to obtain the equipment fundamental frequency data, thereby accurately predicting the equipment fundamental frequency data.

[0016] Through the above technical solution, the present invention realizes abnormal data screening and differential weight configuration based on condition characteristics, effectively improving the accuracy and robustness of equipment fundamental frequency prediction. Description of the Drawings

[0017] Figure 1 It is a schematic structural diagram of a device fundamental frequency data generation system based on device fundamental frequency prediction provided by the present invention;

[0018] Figure 2 It is a schematic flow diagram of a device fundamental frequency data generation method based on device fundamental frequency prediction provided by the present invention.

[0019] In the drawings, the components represented by each reference numeral are as follows:

[0020] Data acquisition module 11, abnormal screening module 12, weight configuration module 13, fundamental frequency prediction module 14. Detailed Embodiment

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0022] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined.

[0023] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as more preferred or more advantageous than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid unnecessary details from obscuring the description of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown, but is in line with the broadest scope consistent with the principles and features disclosed in the present invention.

[0024] Embodiment 1:

[0025] As Figure 1 shown, the embodiment of the present invention provides a device fundamental frequency data generation system based on device fundamental frequency prediction, including a data acquisition module 11, an anomaly screening module 12, a weight configuration module 13, and a fundamental frequency prediction module 14.

[0026] Among them, the data acquisition module 11 is used to obtain a variety of operating conditions, collect vibration signals of the device under the variety of operating conditions in historical time and perform fundamental frequency processing and annotation, so as to obtain a plurality of sample condition vibration signal sets and a plurality of sample condition fundamental frequency data sets.

[0027] The anomaly screening module 12 is used to obtain the condition feature information of a variety of operating conditions, configure a plurality of anomaly fundamental frequency screening ratios, and respectively perform isolated anomaly fundamental frequency screening on the plurality of sample condition fundamental frequency data sets, obtain a plurality of anomaly fundamental frequency data sets and delete them, so as to obtain a plurality of normal condition vibration signal sets and a plurality of normal condition fundamental frequency data sets.

[0028] The weight configuration module 13 is used to calculate the abnormality degrees of the multiple abnormal fundamental frequency data sets, and configure multiple initial training weights in combination with multiple working condition characteristic information.

[0029] The fundamental frequency prediction module 14 is used to train the equipment fundamental frequency predictor according to the multiple initial training weights, using the multiple normal working condition vibration signal sets and multiple normal working condition fundamental frequency data sets, collect the vibration signals of the current target equipment, input them into the equipment fundamental frequency predictor, and predict and output to obtain the equipment fundamental frequency data.

[0030] Specifically, the data acquisition module 11 is responsible for acquiring the historical vibration data of the equipment under different working conditions and performing fundamental frequency extraction and annotation processing. Specifically, first, the data acquisition module 11 obtains various operating conditions of the target equipment during historical operation through the equipment monitoring system. These conditions may include different load conditions, different operating durations, and different environmental parameters, etc. For example, for a certain rotating equipment, it includes various operating states such as full-load operation condition, 75% load operation condition, 50% load operation condition, etc. Secondly, for each identified operating condition, the data acquisition module 11 acquires the vibration signals generated by the equipment under the corresponding condition to form multiple sample condition vibration signal sets. For example, under the full-load operation condition, a vibration signal sample set can be acquired within a certain period of time; under the 75% load operation condition, another set of vibration signal samples is acquired, and so on. These vibration signals are usually obtained through vibration sensors on the equipment and stored after appropriate digital processing. Then, the data acquisition module 11 performs fundamental frequency processing and annotation on the acquired vibration signals. For example, methods such as Fourier transform and harmonic product spectrum are applied to extract the fundamental frequency information from the vibration signals, and the extracted fundamental frequency data is associated and annotated with the corresponding vibration signals to form multiple sample condition fundamental frequency data sets. For example, after analyzing the vibration signals under the full-load operation condition, a data annotation result with a fundamental frequency of 49.8 Hz may be obtained. The multiple sample condition vibration signal sets and multiple sample condition fundamental frequency data sets obtained through the data acquisition module 11 provide the necessary raw data basis for subsequent abnormal data screening and prediction model training, ensuring that the system can perform subsequent predictions based on real historical operation data.

[0031] The abnormal screening module 12 is responsible for screening and removing abnormal data in the sample fundamental frequency dataset to ensure the accuracy and reliability of subsequent training data. Specifically, the abnormal screening module 12 first obtains the working condition characteristic information of the device under various operating conditions, such as key parameters like the running time and average running load of the device. These characteristic information reflect the running state and wear degree of the device under different working conditions and are important bases for determining the abnormal screening ratio. For example, for a certain rotating device, characteristic information such as its cumulative running time and average temperature under different loads is obtained. Secondly, the abnormal screening module 12 configures multiple abnormal fundamental frequency screening ratios according to the working condition characteristic information. The proportion of abnormal data under different working conditions is often different. For example, a working condition with a longer running time may accumulate more wear and noise interference, resulting in a higher proportion of abnormal data; under a high-load operating condition, the device is subjected to greater forces and may also generate more abnormal vibration signal data. Therefore, corresponding abnormal fundamental frequency screening ratios are configured for different working condition characteristics. Thirdly, the abnormal screening module 12 performs isolated abnormal fundamental frequency screening on each sample working condition fundamental frequency dataset to identify the abnormal fundamental frequency dataset. Subsequently, the abnormal screening module 12 deletes the identified abnormal fundamental frequency dataset and its corresponding vibration signal from the original dataset, thereby obtaining multiple normal working condition vibration signal sets and multiple normal working condition fundamental frequency datasets. The screened dataset can more accurately reflect the normal fundamental frequency characteristics of the device under each working condition and provide a more reliable data basis for subsequent weight configuration and prediction model training. Through the processing of the abnormal screening module 12, the system effectively removes abnormal data caused by factors such as device wear and environmental noise, significantly improving the accuracy and stability of fundamental frequency prediction.

[0032] The weight configuration module 13 is responsible for analyzing the characteristics of abnormal data and intelligently configuring the training weights in combination with the working condition information. Specifically, the weight configuration module 13 first calculates the abnormality degree for multiple abnormal fundamental frequency data sets screened by the abnormal screening module. The abnormality degree is a quantitative index measuring the degree of deviation of abnormal data from normal data, and is obtained by calculating the deviation amplitude between the abnormal fundamental frequency data and the mean value of the normal fundamental frequency data. For example, for an abnormal fundamental frequency data set under a certain working condition, if the normal fundamental frequency mean value is 50 Hz and the average value of the abnormal data is 45 Hz, the abnormality degree in this working condition is relatively large; if the average value of the abnormal data under another working condition is 49 Hz, its abnormality degree is relatively small. Secondly, the weight configuration module 13 combines various operating condition characteristic information obtained from the data acquisition link, such as operating time, average load, etc., to perform weight configuration calculation. The operating condition characteristic information reflects the operating state and data reliability of the equipment under different operating conditions, and is an important reference basis for weight configuration. For example, the data collected under the working conditions with a longer operating time and a relatively stable load may be more representative and should be given a higher weight. After that, the weight configuration module 13 configures multiple initial training weights by comprehensively considering the abnormality degree and the operating condition characteristic information. Among them, a larger abnormality degree in a working condition usually indicates that the data quality under this working condition is relatively low, and the corresponding training weight will be appropriately reduced; on the contrary, the data quality is higher under the working condition with a smaller abnormality degree, and the corresponding training weight will be increased accordingly. For example, a training weight of 0.8 is configured for the working condition with an abnormality degree of 10%, while only a training weight of 0.4 is configured for the working condition with an abnormality degree of 30%. Through the weight configuration module 13, the system can reasonably allocate training resources according to the data quality and working condition characteristics, focus on high-quality and high-reliability data samples, and effectively improve the training efficiency and prediction accuracy of the subsequent fundamental frequency prediction model.

[0033] The fundamental frequency prediction module 14 is responsible for constructing and training a device fundamental frequency predictor based on the data processed by the foregoing modules and the configured weights, so as to achieve accurate prediction of the fundamental frequency of the target device. Specifically, the fundamental frequency prediction module 14 first receives multiple initial training weights from the weight configuration module 13, as well as multiple sets of normal operating condition vibration signal sets and multiple sets of normal operating condition fundamental frequency data sets screened by the abnormal screening module 12, as the input data for training the device fundamental frequency predictor. For example, for a certain rotating device, there are multiple sets of training data, including the normal vibration signal set and its corresponding fundamental frequency data set under full load conditions, the normal vibration signal set and its corresponding fundamental frequency data set under 75% load conditions, etc. Secondly, this module assigns weights to the training data according to the initial training weights, refining the training weights to the data sample level, so that the training process pays more attention to high-quality and high-reliability data samples. For example, more training resources, such as more iteration times or a higher learning rate, may be allocated to the operating condition data with higher weights, while the training resource allocation for the operating condition data with lower weights is correspondingly reduced. Thirdly, the fundamental frequency prediction module 14 constructs a device fundamental frequency predictor based on machine learning algorithms. This predictor can establish a prediction model from vibration signals to fundamental frequency data by learning the mapping relationship between vibration signals and fundamental frequency data. Among them, algorithms such as deep neural networks, support vector machines or random forests can be used to construct the predictor, and it is supervised and trained according to the weight assignment until the prediction performance reaches the preset convergence standard. When the predictor training is completed, the fundamental frequency prediction module 14 collects the real-time vibration signals of the current target device and inputs them into the trained device fundamental frequency predictor to quickly obtain the corresponding device fundamental frequency data output. For example, for a certain rotating device in operation, the vibration sensor data is collected, preprocessed and then input into the predictor, and a prediction result such as "the current device fundamental frequency is 49.8 Hz" can be directly output. Through the fundamental frequency prediction module 14, the system realizes the whole process from historical data training to real-time fundamental frequency prediction, provides accurate and reliable fundamental frequency data support for equipment condition monitoring, fault warning and maintenance decision-making, and significantly improves the intelligent level and prediction accuracy of equipment monitoring.

[0034] Further, the data acquisition module 11 includes a working condition acquisition unit, a vibration signal acquisition unit and a fundamental frequency processing and annotation unit. Among them, the working condition acquisition unit is used to acquire various operating conditions of the target device; the vibration signal acquisition unit is used to collect vibration signals of the same type of equipment under the various operating conditions according to the historical operation monitoring data of the same type of equipment, and obtain multiple sample working condition vibration signal sets; the fundamental frequency processing and annotation unit is used to process each sample working condition vibration signal by using Fourier transform and harmonic product spectrum to obtain fundamental frequency data, and perform annotation to obtain multiple sample working condition fundamental frequency data sets.

[0035] In an alternative embodiment, the data acquisition module 11 includes three functional units: a working condition acquisition unit, a vibration signal acquisition unit, and a fundamental frequency processing and annotation unit. These units work together to form a complete data acquisition process.

[0036] The working condition acquisition unit is responsible for identifying and acquiring various operating condition information of the target device. This unit obtains the operating condition parameters of the device in different operating states through data interaction with the device control system, including but not limited to key information such as load rate, speed range, ambient temperature, operating duration, etc. For example, for the rotating equipment in a certain power plant, this unit identifies various typical operating conditions such as full load (100%), high load (80%-90%), medium load (50%-70%), and low load (30%-40%). Subsequently, based on the acquired working condition information, the vibration signal acquisition unit specifically acquires the vibration signals of similar devices under the above-mentioned various operating conditions. The vibration signal acquisition unit extracts the vibration signal records of similar devices under the corresponding working conditions by accessing the historical operation monitoring database, and performs preliminary denoising and standardization processing on these signals. For example, for the full load condition, the vibration signal acquisition unit acquires 200 groups of vibration signal samples; for the medium load condition, it may acquire 200 groups of vibration signal samples, thus forming multiple sample vibration signal sets corresponding to the working conditions. After that, the fundamental frequency processing and annotation unit receives the multiple sample working condition vibration signal sets output by the vibration signal acquisition unit, and conducts in-depth processing and analysis on each sample vibration signal. Specifically, the fundamental frequency processing and annotation unit first applies the Fourier transform to the vibration signal to convert the time-domain signal into a frequency-domain signal, showing the various frequency components contained in the signal; then, through the harmonic product spectrum analysis technique, it identifies and extracts the fundamental frequency component in the signal, filtering out the interference and noise effects; afterwards, it associates and annotates the extracted fundamental frequency data with the corresponding vibration signal and working condition information to form a structured sample working condition fundamental frequency data set. For example, after processing the vibration signal under a certain full load condition, it is annotated as "full load condition, vibration signal ID_123, fundamental frequency 49.8Hz".

[0037] Through the collaborative work of the working condition acquisition unit, the vibration signal acquisition unit, and the fundamental frequency processing and annotation unit, the data acquisition module 11 can efficiently acquire the vibration signals and fundamental frequency data under various working conditions, and perform processing and annotation, laying a data foundation for subsequent abnormal screening and prediction model training.

[0038] Further, the abnormal screening module 12 includes a working condition feature acquisition unit, a coefficient calculation unit, an adjustment coefficient calculation unit, a preset ratio acquisition unit, and an abnormal ratio configuration unit. Among them, the working condition feature acquisition unit is used to acquire the working condition operation time and average operation load of the same type of equipment under multiple operating conditions as multiple working condition feature information; the coefficient calculation unit is used to calculate the ratio of each working condition operation time to the average value of multiple working condition operation times within the multiple working condition feature information to obtain multiple time coefficients, and calculate the ratio of each average operation load to the average value of multiple average operation loads to obtain multiple load coefficients; the adjustment coefficient calculation unit is used to calculate and obtain multiple screening ratio adjustment coefficients according to the multiple time coefficients and multiple load coefficients; the preset ratio acquisition unit is used to acquire the preset abnormal fundamental frequency screening ratio; the abnormal ratio configuration unit is used to respectively use the multiple screening ratio adjustment coefficients to perform configuration calculation on the preset abnormal fundamental frequency screening ratio to obtain multiple abnormal fundamental frequency screening ratios.

[0039] In a preferred embodiment, the abnormal screening module 12 includes a working condition feature acquisition unit, a coefficient calculation unit, an adjustment coefficient calculation unit, a preset ratio acquisition unit, and an abnormal ratio configuration unit to realize the adaptive abnormal data screening based on the working condition features.

[0040] Specifically, the working condition feature acquisition unit is responsible for acquiring the key feature information of the same type of equipment under multiple operating conditions, mainly including the working condition operation time and the average operation load. For example, for a certain rotating equipment, the working condition feature acquisition unit obtains that the cumulative operation time under the full load condition is 5000 hours and the average operation load is 95%; the cumulative operation time under the medium load condition is 8000 hours and the average operation load is 65%, etc. These feature information directly reflect the usage intensity and wear degree of the equipment under each working condition and are an important basis for judging the abnormal data ratio. After receiving the working condition feature information acquired by the working condition feature acquisition unit, the coefficient calculation unit performs the standardization processing of the working condition features. The coefficient calculation unit calculates the ratio of each working condition operation time to the average value of multiple working condition operation times respectively to obtain multiple time coefficients; at the same time, it calculates the ratio of each working condition average operation load to the average value of multiple working condition average operation loads to obtain multiple load coefficients. For example, if the operation times of the three working conditions are 5000 hours, 8000 hours, and 2000 hours respectively, the average value is 5000 hours, and the corresponding time coefficients are 1.0, 1.6, and 0.4 respectively; if the average operation loads of the three working conditions are 95%, 65%, and 40% respectively, the average value is 66.7%, and the corresponding load coefficients are 1.42, 0.97, and 0.60 respectively.

[0041] Subsequently, the adjustment coefficient calculation unit calculates the final screening ratio adjustment coefficient based on the time coefficient and the load coefficient output by the coefficient calculation unit. This calculation is performed by multiplying the time coefficient by the load coefficient, reflecting the combined influence of the operating time and load on the generation of abnormal data. For example, for the above three operating conditions, the screening ratio adjustment coefficients are 1.0×1.42 = 1.42, 1.6×0.97 = 1.55, and 0.4×0.60 = 0.24 respectively. This means that a relatively high screening ratio of abnormal data may be required for the first and second operating conditions, while the proportion of abnormal data for the third operating condition may be relatively low. At the same time, the preset ratio acquisition unit obtains the benchmark value of the preset abnormal fundamental frequency screening ratio of the system. This ratio is determined based on historical statistical data and represents the average proportion of abnormal fundamental frequency data of the equipment under standard operating conditions. For example, according to the historical monitoring records of the equipment, the preset abnormal fundamental frequency screening ratio is set to 5%, indicating that under standard operating conditions, approximately 5% of the fundamental frequency data shows abnormalities due to various interference factors. After that, the abnormal ratio configuration unit applies the screening ratio adjustment coefficient output by the adjustment coefficient calculation unit to the preset abnormal fundamental frequency screening ratio, and through multiplication, obtains the actual abnormal fundamental frequency screening ratio for each operating condition. For example, if the preset abnormal fundamental frequency screening ratio is 5%, the abnormal fundamental frequency screening ratios for the above three operating conditions are 5%×1.42 = 7.1%, 5%×1.55 = 7.75%, and 5%×0.24 = 1.2% respectively. This means that when screening abnormal fundamental frequency data, 7.1% of the data will be screened out as abnormal data for the first operating condition, 7.75% of the data will be screened out as abnormal data for the second operating condition, and only 1.2% of the data will be screened out as abnormal data for the third operating condition.

[0042] Through the collaborative work of the above five functional units, the abnormal screening module 12 can adaptively configure the abnormal fundamental frequency screening ratio for each operating condition based on the operating condition characteristic information, realizing the precision and personalization of abnormal data screening, and providing a ratio basis for subsequent abnormal fundamental frequency screening operations.

[0043] Further, the abnormal screening module 12 further includes a fundamental frequency range acquisition unit, an isolated data screening unit, an abnormal data classification unit, a screening ratio control unit, and a normal data acquisition unit. Among them, the fundamental frequency range acquisition unit is used to acquire the fundamental frequency data range of the device; the isolated data screening unit is used to randomly generate first fundamental frequency data within the fundamental frequency data range, perform binary classification on the multiple sample working condition fundamental frequency data sets, and determine whether there is sample working condition fundamental frequency data classified as isolated fundamental frequency data; the abnormal data classification unit is used to, if so, classify the isolated fundamental frequency data as abnormal fundamental frequency data, and if not, continue to randomly generate second fundamental frequency data for isolated abnormal fundamental frequency screening; the screening ratio control unit is used to stop the isolated abnormal fundamental frequency screening until the proportion of abnormal fundamental frequency data divided in each sample working condition fundamental frequency data set reaches the multiple abnormal fundamental frequency screening ratios, and obtain multiple screened abnormal fundamental frequency data sets; the normal data acquisition unit is used to acquire the multiple abnormal working condition vibration signal sets corresponding to the multiple abnormal fundamental frequency data sets, delete them, and obtain multiple normal working condition vibration signal sets and multiple normal working condition fundamental frequency data sets.

[0044] Specifically, the abnormal screening module 12 further includes a fundamental frequency range acquisition unit, an isolated data screening unit, an abnormal data classification unit, a screening ratio control unit, and a normal data acquisition unit. These units together constitute a complete execution system for abnormal fundamental frequency screening to achieve accurate abnormal data screening functions.

[0045] The fundamental frequency interval acquisition unit is responsible for determining the effective interval range of the device's fundamental frequency data. By analyzing the device's specification parameters and historical operation data, it obtains the theoretical range and actual distribution interval of the device's fundamental frequency. For example, for a certain type of rotating device, the theoretical fundamental frequency value may be 50 Hz. Considering the fluctuations in actual operation, the fundamental frequency interval may be determined as 45 Hz - 55 Hz. The determination of this interval provides the data space range for subsequent screening of abnormal data. Subsequently, the isolated data screening unit uses a method combining random sampling and binary classification to identify isolated abnormal points in the fundamental frequency data set. The isolated data screening unit first randomly generates a fundamental frequency data point (referred to as the first fundamental frequency data) within the determined fundamental frequency data interval, and then uses this point as a boundary to perform binary classification on the fundamental frequency data sets of multiple sample working conditions, and determines whether there are isolated fundamental frequency data formed by this classification. For example, if the randomly generated fundamental frequency data is 47 Hz, and most of the fundamental frequency data under a certain working condition is distributed in the interval of 49 Hz - 51 Hz, and only a small amount of data is distributed in the interval of 46 Hz - 48 Hz, then these small amounts of data may be classified as isolated data. Subsequently, the abnormal data classification unit processes the judgment result of the isolated data screening unit. When the judgment result is "yes", that is, there are fundamental frequency data classified as isolated, the abnormal data classification unit classifies these isolated fundamental frequency data as abnormal fundamental frequency data; when the judgment result is "no", that is, no obvious isolated data is found, this unit will instruct the isolated data screening unit to continue randomly generating new fundamental frequency data points (referred to as the second fundamental frequency data) and re-execute the process of screening isolated abnormal fundamental frequencies. For example, if the first randomly generated 47 Hz fails to effectively divide isolated data, the system will generate a second random point of 52 Hz and re-perform the division judgment.

[0046] The screening ratio control unit is responsible for monitoring the progress and ratio of abnormal data screening, calculating in real time the proportion of the divided abnormal fundamental frequency data in the fundamental frequency data set of each sample working condition, and comparing it with the previously configured abnormal fundamental frequency screening ratio. When the proportion of abnormal data in a certain working condition reaches its corresponding screening ratio, the isolated abnormal fundamental frequency screening process for this working condition stops; when the screening of abnormal data for all working conditions reaches the predetermined ratio, the entire screening process ends, and the system obtains multiple screened abnormal fundamental frequency data sets. For example, if the abnormal fundamental frequency screening ratio for a certain working condition is 7.1%, then when the fundamental frequency data marked as abnormal in this working condition reaches 7.1% of the total data volume, the screening process for this working condition stops. After completing the screening of abnormal data, the normal data acquisition unit performs data separation operations. The normal data acquisition unit first obtains the vibration signals corresponding to multiple abnormal fundamental frequency data sets to form multiple abnormal working condition vibration signal sets; then deletes these abnormal data from the original sample set, and finally obtains multiple normal working condition vibration signal sets and multiple normal working condition fundamental frequency data sets without abnormal data. For example, if there are 100 groups of fundamental frequency data in a certain working condition, and 7 of them are marked as abnormal, then after deleting these 7 groups of data and their corresponding vibration signals, 93 groups of normal data in this working condition will be retained for subsequent processing.

[0047] Through the above processing, the abnormal screening module 12 can screen out the abnormal fundamental frequency data under each working condition according to a predetermined ratio and remove it from the training data set, effectively improving the data quality and prediction accuracy of subsequent model training.

[0048] Furthermore, the weight configuration module 13 includes a mean calculation unit and an abnormality calculation unit. Among them, the mean calculation unit is used to calculate the means of the multiple normal working condition fundamental frequency data sets respectively to obtain multiple average normal fundamental frequency data; the abnormality calculation unit is used to calculate the deviation amplitudes between the abnormal fundamental frequency data in the multiple abnormal fundamental frequency data sets and the multiple average normal fundamental frequency data respectively to obtain multiple abnormal deviation amplitude sets, and calculate the mean value to obtain multiple abnormality degrees.

[0049] In a preferred implementation manner, the weight configuration module 13 includes a mean calculation unit and an abnormality calculation unit, which can accurately analyze the characteristics of abnormal fundamental frequency data to quantify the deviation degree of abnormal data under each working condition.

[0050] The mean calculation unit is responsible for statistically analyzing the normal fundamental frequency data processed by the abnormal screening module. The mean calculation unit receives multiple normal operating condition fundamental frequency data sets as inputs, calculates the arithmetic mean for each normal fundamental frequency data set under each operating condition respectively, and obtains the average normal fundamental frequency data under each operating condition. For example, for the normal fundamental frequency data set under the full load operating condition, if it contains 93 groups of data with values of 49.8 Hz, 50.1 Hz, 49.9 Hz, etc., the calculated average normal fundamental frequency data is 50.0 Hz; for the medium load operating condition, the average value is 49.5 Hz, and so on. These average values represent the normal fundamental frequency levels of the equipment under each operating condition and are the reference values for subsequent abnormality calculation. Then, based on the output result of the mean calculation unit, the abnormality calculation unit quantitatively evaluates the deviation degree of the abnormal fundamental frequency data. The abnormality calculation unit first calculates the deviation amplitude (which can be in the form of absolute difference or relative difference) between each data point in the abnormal fundamental frequency data set under each operating condition and the average normal fundamental frequency data corresponding to that operating condition, forming multiple abnormal deviation amplitude sets. For example, if an abnormal fundamental frequency data under the full load operating condition is 45.0 Hz and the average normal fundamental frequency data for this operating condition is 50.0 Hz, the deviation amplitude of this abnormal data is 5.0 Hz or 10% (depending on whether absolute difference or relative difference is used for calculation). Subsequently, the abnormality calculation unit calculates the mean for each abnormal deviation amplitude set to obtain the comprehensive abnormality index under each operating condition. For example, if the mean deviation amplitude of 7 groups of abnormal data under the full load operating condition is 4.5 Hz, the abnormality for this operating condition is 4.5 Hz; if the abnormality under the medium load operating condition is 2.8 Hz, it indicates that the abnormal data under the full load operating condition has a greater deviation degree.

[0051] Through the mean calculation unit and the abnormality calculation unit, the weight configuration module 13 can quantify the severity of abnormal data under each operating condition, providing a scientific basis for subsequent weight configuration. The greater the abnormality of an operating condition, the lower the possible data quality, and a lower weight should be assigned during training; the smaller the abnormality of an operating condition, the relatively higher the data quality, and a higher training weight should be assigned. This adaptive weight configuration mechanism based on data quality can improve the pertinence and effectiveness of model training.

[0052] Further, the weight configuration module 13 further includes a time weight calculation unit, a load weight calculation unit, an anomaly weight allocation unit, and a weight fusion calculation unit. Among them, the time weight calculation unit is used to calculate, within the multiple working condition feature information, the ratio of the running time of each working condition to the sum of the running times of the multiple working conditions as multiple running time weights; the load weight calculation unit is used to calculate the ratio of each average running load to the sum of the multiple average running loads as multiple load weights; the anomaly weight allocation unit is used to allocate, according to the multiple degrees of anomaly, multiple anomaly training weights for multiple operating conditions, where the magnitude of the degree of anomaly is negatively correlated with the magnitude of the anomaly training weight; the weight fusion calculation unit is used to perform weight fusion calculation according to the multiple running time weights, the multiple load weights, and the multiple anomaly training weights to obtain multiple initial training weights.

[0053] In a feasible implementation manner, the weight configuration module 13 further includes a time weight calculation unit, a load weight calculation unit, an anomaly weight allocation unit, and a weight fusion calculation unit to implement weight configuration.

[0054] The time weight calculation unit is responsible for allocating corresponding weight coefficients based on the equipment running time. This unit extracts the running time data of each working condition from the working condition feature information, calculates the ratio of the running time of each working condition to the total running time of all working conditions, and uses this ratio as the running time weight of the corresponding working condition. For example, if the running times of three working conditions are 5000 hours, 8000 hours, and 2000 hours respectively, the total running time is 15000 hours, and the corresponding running time weights are 5000 / 15000 = 0.33, 8000 / 15000 = 0.53, and 2000 / 15000 = 0.13 respectively. This way of weight allocation reflects that the working condition with a longer running time may contain more valuable running information and should obtain a relatively higher weight in training. The load weight calculation unit adopts a similar proportional allocation method to configure weight coefficients based on the average running load. This unit calculates the ratio of the average running load of each working condition to the sum of the average running loads of all working conditions as the load weight of the corresponding working condition. For example, if the average running loads of three working conditions are 95%, 65%, and 40% respectively, the total load is 200%, and the corresponding load weights are 95 / 200 = 0.475, 65 / 200 = 0.325, and 40 / 200 = 0.2 respectively. This configuration method reflects that the data under the working condition with a higher load represents the operating characteristics of the equipment more fully and should obtain a relatively higher training weight.

[0055] The abnormal weight allocation unit allocates abnormal training weights in a reverse mapping manner based on the previously calculated abnormality metrics. This unit assigns weights according to the magnitude of the abnormality for each working condition, following the principle that the abnormality and the weight are negatively correlated. That is, for a working condition with a greater abnormality, its abnormal training weight is smaller; for a working condition with a smaller abnormality, its abnormal training weight is larger. For example, if the abnormalities of three working conditions are 4.5 Hz, 2.8 Hz, and 1.5 Hz respectively, a certain inverse function relationship can be used to map them to abnormal training weights of 0.15, 0.25, and 0.45 (the specific mapping function can be designed according to actual needs). This configuration method ensures that working conditions with less abnormal data and higher data quality have a greater influence in training. Subsequently, the weight fusion calculation unit is responsible for integrating the above three different-dimensional weights to form the final initial training weight. This unit first calculates the sum of the running time weight, load weight, and abnormal training weight for each working condition, and then calculates the ratio of this sum value to the total sum of the three weights of all working conditions as the initial training weight of this working condition. For example, if the sum of the three weights of the first working condition is 0.33 + 0.475 + 0.15 = 0.955, the second working condition is 0.53 + 0.325 + 0.25 = 1.105, and the third working condition is 0.13 + 0.2 + 0.45 = 0.78, then the total sum of the weights of the three working conditions is 2.84, and the corresponding initial training weights are 0.955 / 2.84 = 0.336, 1.105 / 2.84 = 0.389, and 0.78 / 2.84 = 0.275. This fusion method comprehensively considers three key factors: the running time, load level, and data quality of the working condition, and forms a weight allocation scheme.

[0056] Through the weight calculation unit, load weight calculation unit, abnormal weight allocation unit, and weight fusion calculation unit, the weight configuration module 13 can comprehensively evaluate the importance and reliability of the data of each working condition from multiple dimensions, provide an accurate weight configuration for the subsequent fundamental frequency prediction model training, and effectively improve the training efficiency and prediction accuracy.

[0057] Furthermore, the fundamental frequency prediction module 14 includes a training sample combination unit, a sample weight assignment unit, a predictor construction unit, a predictor training unit, and a fundamental frequency data generation unit. Among them, the training sample combination unit is used to combine the multiple normal operating condition vibration signal sets and the multiple normal operating condition fundamental frequency data sets to obtain multiple groups of normal fundamental frequency training sample sets and obtain multiple normal sample data volumes; the sample weight assignment unit is used to divide the multiple initial training weights by the multiple normal sample data volumes respectively for data-level weight assignment to obtain multiple sample data weight sets; the predictor construction unit is used to construct an equipment fundamental frequency predictor based on machine learning; the predictor training unit is used to use the multiple groups of normal fundamental frequency training sample sets, allocate training resources according to the multiple sample data weight sets, and perform supervised training and testing on the equipment fundamental frequency predictor until convergence; the fundamental frequency data generation unit is used to collect the vibration signal of the current target equipment, input it into the equipment fundamental frequency predictor, and predict and output to obtain the equipment fundamental frequency data.

[0058] In a preferred embodiment, the fundamental frequency prediction module 14 includes five functional units: a training sample combination unit, a sample weight assignment unit, a predictor construction unit, a predictor training unit, and a fundamental frequency data generation unit, forming a complete predictor training and application process to achieve an accurate and efficient equipment fundamental frequency prediction function.

[0059] The training sample combination unit is responsible for integrating the normal data processed by the previous module and constructing a standardized training sample set. The training sample combination unit receives multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets as inputs, pairs and combines the vibration signals under each condition with the corresponding fundamental frequency data to form multiple normal fundamental frequency training sample sets with clear "input-output" corresponding relationships. At the same time, the training sample combination unit also calculates the number of normal samples under each condition to obtain multiple normal sample data volumes. For example, the full load condition may contain 93 sets of normal sample data, the medium load condition may contain 88 sets of normal sample data, and the low load condition may contain 95 sets of normal sample data. These data will be organized into a structured training sample set to provide basic data support for subsequent model training. The sample weight allocation unit refines the initial training weights at the condition level output by the weight configuration module to the data sample level. The sample weight allocation unit divides multiple initial training weights by the corresponding normal sample data volumes respectively to achieve uniform dispersion of the weights and obtains multiple sample data weight sets. For example, if the initial training weight for the full load condition is 0.336 and the normal sample data volume is 93, then the weight of each sample under this condition is 0.336 / 93 = 0.0036; if the initial training weight for the medium load condition is 0.389 and the normal sample data volume is 88, then the weight of each sample under this condition is 0.389 / 88 = 0.0044. This refined weight allocation ensures that the influence of each condition in the training process is consistent with its preset weight regardless of the number of its samples.

[0060] The predictor construction unit is responsible for constructing the basic framework of the device fundamental frequency predictor based on machine learning techniques. The predictor construction unit can select appropriate machine learning algorithms, such as support vector regression, random forest, deep neural network, etc., according to the characteristics of the prediction task and the complexity of the data, and construct a prediction model structure that can map from vibration signal features to fundamental frequency data. For example, for a complex non-linear mapping relationship, a deep neural network containing multiple hidden layers is constructed, where the input layer receives vibration signal features and the output layer predicts fundamental frequency data; for a relatively simple mapping relationship, a support vector regression model is adopted to achieve non-linear mapping through a kernel function. The predictor training unit executes the core process of model training. The predictor training unit first divides multiple groups of normal fundamental frequency training sample sets into a training set and a test set, and then allocates training resources (such as the number of iterations, learning rate, etc.) according to the sample data weight set to perform supervised gradient descent training on the predictor. For example, a sample with a weight of 0.0044 may obtain more iterative training times or a higher learning rate, while a sample with a weight of 0.0036 will have a corresponding reduction in training resource allocation. During the training process, the predictor training unit continuously monitors the performance of the model on the test set, evaluates the prediction error and the convergence state, until the model performance reaches the preset convergence standard, and completes the parameter optimization of the predictor. The fundamental frequency data generation unit is responsible for applying the trained predictor to the actual prediction scenario. The fundamental frequency data generation unit collects the vibration signal of the current target device in real time through a vibration sensor, performs preprocessing on the signal that is consistent with the training data, and then inputs the processed vibration signal features into the trained device fundamental frequency predictor to quickly obtain the corresponding device fundamental frequency data output. For example, for a running rotating device, after collecting its vibration signal and inputting it into the predictor, a prediction result of "the current device fundamental frequency is 49.8 Hz" can be directly obtained, providing real-time data support for device status monitoring and fault diagnosis.

[0061] Through the collaborative work of the above functional units, the fundamental frequency prediction module 14 realizes the complete process from training sample organization to model training and then to actual prediction, enabling the system to intelligently learn the internal mapping relationship between vibration signals and fundamental frequency data based on historical data, and applying this relationship to the fundamental frequency prediction of new devices, improving the accuracy and robustness of device fundamental frequency prediction.

[0062] Embodiment 2:

[0063] As Figure 2 shown, based on the same inventive concept as the device fundamental frequency data generation system for device fundamental frequency prediction provided in Embodiment 1, the present invention embodiment also provides a method for generating device fundamental frequency data for device fundamental frequency prediction, including:

[0064] Obtain multiple operating conditions, collect the vibration signals of the device under the multiple operating conditions within historical time, and perform fundamental frequency processing and annotation to obtain multiple sample condition vibration signal sets and multiple sample condition fundamental frequency data sets;

[0065] Obtain the condition characteristic information of multiple operating conditions, configure multiple abnormal fundamental frequency screening ratios, respectively perform isolated abnormal fundamental frequency screening on the multiple sample condition fundamental frequency data sets, obtain multiple abnormal fundamental frequency data sets and delete them, and obtain multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets;

[0066] Calculate the abnormality degree of the multiple abnormal fundamental frequency data sets, and combine multiple condition characteristic information to configure multiple initial training weights;

[0067] According to the multiple initial training weights, use the multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets to train the device fundamental frequency predictor, collect the vibration signals of the current target device, input them into the device fundamental frequency predictor, and predict and output to obtain the device fundamental frequency data.

[0068] Further, obtaining multiple operating conditions, collecting the vibration signals of the device under the multiple operating conditions within historical time, and performing fundamental frequency processing and annotation to obtain multiple sample condition vibration signal sets and multiple sample condition fundamental frequency data sets includes:

[0069] Obtain multiple operating conditions of the target device;

[0070] According to the historical operation monitoring data of similar devices, collect the vibration signals of the similar devices under the multiple operating conditions to obtain multiple sample condition vibration signal sets;

[0071] Process each sample condition vibration signal using Fourier transform and harmonic product spectrum to obtain fundamental frequency data, and perform annotation to obtain multiple sample condition fundamental frequency data sets.

[0072] Further, obtaining the condition characteristic information of multiple operating conditions and configuring multiple abnormal fundamental frequency screening ratios includes:

[0073] Obtain the operating time and average operating load of similar devices under multiple operating conditions as multiple condition characteristic information;

[0074] Within the multiple condition characteristic information, calculate the ratio of each operating time to the average value of the multiple operating times to obtain multiple time coefficients, and calculate the ratio of each average operating load to the average value of the multiple average operating loads to obtain multiple load coefficients;

[0075] Calculate multiple screening ratio adjustment coefficients based on the multiple time coefficients and multiple load coefficients;

[0076] Obtain the preset abnormal fundamental frequency screening ratio;

[0077] Respectively use the multiple screening ratio adjustment coefficients to configure and calculate the preset abnormal fundamental frequency screening ratio to obtain multiple abnormal fundamental frequency screening ratios.

[0078] Furthermore, respectively perform isolated abnormal fundamental frequency screening on the multiple sample working condition fundamental frequency data sets, obtain multiple abnormal fundamental frequency data sets and delete them, and obtain multiple normal working condition vibration signal sets and multiple normal working condition fundamental frequency data sets, including:

[0079] Obtain the fundamental frequency data interval of the equipment;

[0080] Randomly generate the first fundamental frequency data within the fundamental frequency data interval, perform binary classification on the multiple sample working condition fundamental frequency data sets, and judge whether there is sample working condition fundamental frequency data classified as isolated fundamental frequency data;

[0081] If so, classify the isolated fundamental frequency data as abnormal fundamental frequency data. If not, continue to randomly generate the second fundamental frequency data for isolated abnormal fundamental frequency screening;

[0082] Until the proportion of the abnormal fundamental frequency data divided in each sample working condition fundamental frequency data set reaches the multiple abnormal fundamental frequency screening ratios, stop the isolated abnormal fundamental frequency screening, and obtain multiple screened abnormal fundamental frequency data sets;

[0083] Obtain the multiple abnormal working condition vibration signal sets corresponding to the multiple abnormal fundamental frequency data sets, delete them, and obtain multiple normal working condition vibration signal sets and multiple normal working condition fundamental frequency data sets.

[0084] Furthermore, it is characterized in that calculating the abnormality degree of the multiple abnormal fundamental frequency data sets includes:

[0085] Respectively calculate the means of the multiple normal working condition fundamental frequency data sets to obtain multiple average normal fundamental frequency data;

[0086] Respectively calculate the deviation amplitudes between the abnormal fundamental frequency data in the multiple abnormal fundamental frequency data sets and the multiple average normal fundamental frequency data to obtain multiple abnormal deviation amplitude sets, and calculate the means to obtain multiple abnormality degrees.

[0087] Furthermore, it is characterized in that combining multiple working condition characteristic information to configure multiple initial training weights includes:

[0088] Within the multiple working condition characteristic information, calculate the ratio of each working condition running time to the sum of the multiple working condition running times as multiple running time weights;

[0089] Calculate the ratio of each average running load to the sum of the multiple average running loads as multiple load weights;

[0090] According to the multiple abnormality degrees, multiple abnormal training weights for obtaining multiple operating conditions are allocated, wherein the magnitude of the abnormality degree is negatively correlated with the magnitude of the abnormal training weight;

[0091] According to multiple operating time weights, multiple load weights, and multiple abnormal training weights, weight fusion calculation is performed to obtain multiple initial training weights.

[0092] Furthermore, it is characterized in that, according to the multiple initial training weights, using the multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets, an equipment fundamental frequency predictor is trained, the vibration signal of the current target equipment is collected, input into the equipment fundamental frequency predictor, and the equipment fundamental frequency data is obtained by predictive output, including:

[0093] Combine the multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets to obtain multiple groups of normal fundamental frequency training sample sets, and obtain multiple normal sample data volumes;

[0094] Respectively divide the multiple initial training weights by the multiple normal sample data volumes to perform data-level weight allocation, and obtain multiple sample data weight sets;

[0095] Based on machine learning, an equipment fundamental frequency predictor is constructed;

[0096] Use the multiple groups of normal fundamental frequency training sample sets, allocate training resources according to the multiple sample data weight sets, and perform supervised training and testing on the equipment fundamental frequency predictor until convergence;

[0097] Collect the vibration signal of the current target equipment, input it into the equipment fundamental frequency predictor, and obtain the equipment fundamental frequency data by predictive output.

[0098] It should be noted that in the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailedly described in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0099] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can adopt the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a means for implementing the functions specified in multiple blocks.

[0101] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction means that implements the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a means for implementing the functions specified in multiple blocks.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a means for implementing the functions specified in multiple blocks.

[0103] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic inventive concept.

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

Claims

1. A device fundamental frequency data generation system based on device fundamental frequency prediction, characterized in that The system includes: A data acquisition module, which is used to obtain multiple operating conditions, collect vibration signals of the device under the multiple operating conditions in historical time and perform fundamental frequency processing and annotation, so as to obtain a plurality of sample condition vibration signal sets and a plurality of sample condition fundamental frequency data sets; An abnormal screening module, which is used to obtain the condition characteristic information of multiple operating conditions, configure a plurality of abnormal fundamental frequency screening ratios, respectively perform isolated abnormal fundamental frequency screening on the plurality of sample condition fundamental frequency data sets, obtain a plurality of abnormal fundamental frequency data sets and delete them, so as to obtain a plurality of normal condition vibration signal sets and a plurality of normal condition fundamental frequency data sets; A weight configuration module, which is used to calculate the abnormality degree of the plurality of abnormal fundamental frequency data sets, and combine a plurality of condition characteristic information to configure a plurality of initial training weights; A fundamental frequency prediction module, which is used to train a device fundamental frequency predictor according to the plurality of initial training weights, using the plurality of normal condition vibration signal sets and the plurality of normal condition fundamental frequency data sets, collect the vibration signal of the current target device, input it into the device fundamental frequency predictor, and predict and output to obtain the device fundamental frequency data; Among them, the weight configuration module includes: A mean calculation unit, which is used to calculate the mean values of the plurality of normal condition fundamental frequency data sets respectively to obtain a plurality of average normal fundamental frequency data; An abnormality degree calculation unit, which is used to calculate the deviation amplitudes between the abnormal fundamental frequency data in the plurality of abnormal fundamental frequency data sets and the plurality of average normal fundamental frequency data respectively, obtain a plurality of abnormal deviation amplitude sets, and calculate the mean value to obtain a plurality of abnormality degrees; A time weight calculation unit, which is used to calculate the ratio of the operating time of each condition to the sum of the operating times of multiple conditions in the plurality of condition characteristic information as a plurality of operating time weights; A load weight calculation unit, which is used to calculate the ratio of each average operating load to the sum of multiple average operating loads as a plurality of load weights; An abnormal weight allocation unit, which is used to allocate a plurality of abnormal training weights for multiple operating conditions according to the plurality of abnormality degrees, wherein the magnitude of the abnormality degree is negatively correlated with the magnitude of the abnormal training weight; A weight fusion calculation unit, which is used to perform weight fusion calculation according to a plurality of operating time weights, a plurality of load weights and a plurality of abnormal training weights to obtain a plurality of initial training weights.

2. The device fundamental frequency data generation system based on device fundamental frequency prediction according to claim 1, wherein The data acquisition module includes: A condition acquisition unit, which is used to obtain multiple operating conditions of the target device; A vibration signal acquisition unit, which is used to collect vibration signals of the same type of device under the multiple operating conditions according to the historical operation monitoring data of the same type of device, so as to obtain a plurality of sample condition vibration signal sets; A fundamental frequency processing and annotation unit, which is used to process each sample condition vibration signal by Fourier transform and harmonic product spectrum to obtain fundamental frequency data, and perform annotation to obtain a plurality of sample condition fundamental frequency data sets.

3. The device fundamental frequency data generation system based on device fundamental frequency prediction according to claim 1, wherein The abnormal screening module includes: A condition characteristic acquisition unit, which is used to obtain the operating time and average operating load of the same type of device under multiple operating conditions as a plurality of condition characteristic information; A coefficient calculation unit, configured to calculate, among multiple operating condition characteristic information, the ratio of the operating time of each operating condition to the average value of the operating times of multiple operating conditions, to obtain multiple time coefficients, and calculate the ratio of each average operating load to the average value of multiple average operating loads, to obtain multiple load coefficients; An adjustment coefficient calculation unit, configured to calculate and obtain multiple screening ratio adjustment coefficients according to the multiple time coefficients and the multiple load coefficients; A preset ratio acquisition unit, configured to acquire a preset abnormal fundamental frequency screening ratio; An abnormal ratio configuration unit, configured to respectively use the multiple screening ratio adjustment coefficients to perform configuration calculation on the preset abnormal fundamental frequency screening ratio, to obtain multiple abnormal fundamental frequency screening ratios.

4. The device fundamental frequency data generation system based on device fundamental frequency prediction according to claim 1, characterized in that, The abnormal screening module further includes: A fundamental frequency range acquisition unit, configured to acquire the fundamental frequency data range of the device; An isolated data screening unit, configured to randomly generate first fundamental frequency data within the fundamental frequency data range, perform binary classification on the multiple sample operating condition fundamental frequency data sets, and determine whether there is sample operating condition fundamental frequency data classified as isolated fundamental frequency data; An abnormal data classification unit, configured to, if so, classify the isolated fundamental frequency data as abnormal fundamental frequency data, and if not, continue to randomly generate second fundamental frequency data for isolated abnormal fundamental frequency screening; A screening ratio control unit, configured to stop the isolated abnormal fundamental frequency screening until the proportion of the abnormal fundamental frequency data divided in each sample operating condition fundamental frequency data set reaches the multiple abnormal fundamental frequency screening ratios, to obtain multiple abnormal fundamental frequency data sets with screening completed; A normal data acquisition unit, configured to acquire the multiple abnormal operating condition vibration signal sets corresponding to the multiple abnormal fundamental frequency data sets, perform deletion, to obtain multiple normal operating condition vibration signal sets and multiple normal operating condition fundamental frequency data sets.

5. The device fundamental frequency data generation system based on device fundamental frequency prediction according to claim 1, characterized in that The fundamental frequency prediction module includes: A training sample combination unit, configured to combine the multiple normal operating condition vibration signal sets and the multiple normal operating condition fundamental frequency data sets, to obtain multiple groups of normal fundamental frequency training sample sets, and obtain multiple normal sample data amounts; A sample weight distribution unit, configured to respectively divide the multiple initial training weights by the multiple normal sample data amounts to perform data-level weight distribution, to obtain multiple sample data weight sets; A predictor construction unit, configured to construct a device fundamental frequency predictor based on machine learning; A predictor training unit, configured to use the multiple groups of normal fundamental frequency training sample sets, allocate training resources according to the multiple sample data weight sets, and perform supervised training and testing on the device fundamental frequency predictor until convergence; A fundamental frequency data generation unit, configured to collect the vibration signal of the current target device, input it into the device fundamental frequency predictor, and predict and output to obtain the device fundamental frequency data.

6. A method for generating device fundamental frequency data based on device fundamental frequency prediction, characterized in that, The method includes: Obtain multiple operating conditions, collect the vibration signals of the device under the multiple operating conditions in historical time and perform fundamental frequency processing and annotation, to obtain multiple sample operating condition vibration signal sets and multiple sample operating condition fundamental frequency data sets; Obtain the condition characteristic information of multiple operating conditions, configure multiple abnormal fundamental frequency screening ratios, respectively screen the isolated abnormal fundamental frequencies for the multiple sample condition fundamental frequency data sets, obtain multiple abnormal fundamental frequency data sets and delete them, and obtain multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets; Calculate the abnormality degrees of the multiple abnormal fundamental frequency data sets, and configure multiple initial training weights in combination with multiple condition characteristic information; According to the multiple initial training weights, use the multiple normal condition vibration signal sets and multiple normal condition fundamental frequency data sets to train the equipment fundamental frequency predictor, collect the vibration signals of the current target equipment, input them into the equipment fundamental frequency predictor, and predict and output to obtain the equipment fundamental frequency data; Among them, calculating the abnormality degrees of the multiple abnormal fundamental frequency data sets and configuring multiple initial training weights in combination with multiple condition characteristic information includes: Calculate the means of the multiple normal condition fundamental frequency data sets respectively to obtain multiple average normal fundamental frequency data; Calculate the deviation amplitudes between the abnormal fundamental frequency data in the multiple abnormal fundamental frequency data sets and the multiple average normal fundamental frequency data respectively to obtain multiple abnormal deviation amplitude sets, and calculate the means to obtain multiple abnormality degrees; In the multiple condition characteristic information, calculate the ratio of each condition running time to the sum of multiple condition running times as multiple running time weights; Calculate the ratio of each average running load to the sum of multiple average running loads as multiple load weights; According to the multiple abnormality degrees, allocate multiple abnormal training weights for multiple operating conditions, where the magnitude of the abnormality degree is negatively correlated with the magnitude of the abnormal training weight; Perform weight fusion calculation according to multiple running time weights, multiple load weights and multiple abnormal training weights to obtain multiple initial training weights.

Images