Rolling bearing fault diagnosis method based on adversarial auto-encoder

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a rolling bearing fault diagnosis method based on an adversarial auto-encoder, and the method comprises the steps: calculating a historical fault diagnosis threshold value of a to-be-monitored bearing based on an analysis result; acquiring real-time operation data of the to-be-monitored bearing, and processing the real-time operation data by adopting a confrontation auto-encoder to obtain real-time feature data; calculating a fault early warning index of the to-be-monitored bearing based on the real-time feature data and the historical feature data; selecting a corresponding adjustment coefficient according to the fault early warning index, and adjusting the fault early warning index based on the adjustment coefficient to obtain a final fault early warning index; and judging whether the to-be-monitored bearing has a fault according to the final fault early warning index, and if so, performing fault early warning. The rolling bearing fault diagnosis method based on the anti-auto-encoder has the advantages of high precision, high reliability, real-time performance and the like, can effectively solve the problems in rolling bearing fault diagnosis, and provides powerful guarantee for stable operation of industrial equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a rolling bearing fault diagnosis method based on an adversarial autoencoder. Background Art

[0002] Adversarial autoencoders are deep learning models that introduce adversarial training mechanisms based on traditional autoencoders. By cleverly combining adversarial training of the generator and the discriminator, this model can effectively learn the potential feature representation of the data while effectively removing unimportant noise information, thereby making the generated data closer to the distribution of real data.

[0003] In the current field of fault diagnosis, many methods still rely on manual feature extraction and the accumulation of expert experience. Although this method can achieve certain results in some cases, its dependence limits the accuracy and efficiency of fault diagnosis. This is because manual feature extraction often requires a lot of time and professional knowledge, and the subjectivity of expert experience may also lead to instability in the diagnosis results.

[0004] Therefore, it is necessary to design a rolling bearing fault diagnosis method based on adversarial autoencoder to solve the problems existing in current technology. Summary of the invention

[0005] In view of this, the present invention proposes a rolling bearing fault diagnosis method based on adversarial autoencoder, aiming to solve the problem that the fault diagnosis methods in the current technology often rely on manual feature extraction and expert experience, which limits the accuracy and efficiency of diagnosis.

[0006] The present invention proposes a rolling bearing fault diagnosis method based on an adversarial autoencoder, comprising the following steps:

[0007] S100: determining a bearing to be monitored, collecting historical operation records of the bearing to be monitored, parsing the historical operation records, and calculating a historical fault diagnosis threshold of the bearing to be monitored based on the parsing result;

[0008] S200: Acquire real-time operating data of the bearing to be monitored, and process the real-time operating data using an adversarial autoencoder to obtain real-time feature data;

[0009] S300: Acquire historical characteristic data corresponding to the real-time characteristic data in the previous normal operation record, and calculate the fault warning index of the bearing to be monitored based on the real-time characteristic data and the historical characteristic data;

[0010] S400: selecting a corresponding adjustment coefficient according to the fault warning index, and adjusting the fault warning index based on the adjustment coefficient to obtain a final fault warning index;

[0011] S500: judging whether the bearing to be monitored has a fault according to the final fault warning index, and if so, issuing a fault warning.

[0012] Furthermore, the historical operation records are parsed, and the historical fault diagnosis threshold of the bearing to be monitored is calculated based on the parsing result, including:

[0013] Analyze the historical operation records to obtain normal historical operation records and abnormal historical operation records;

[0014] respectively counting the number of first records of the abnormal historical operation records and the number of second records of the normal historical operation records;

[0015] Determine an abnormal operation characteristic fitting curve and a normal operation characteristic fitting curve according to the abnormal historical operation record and the normal historical operation record;

[0016] Determine an abnormal offset value corresponding to each abnormal historical operation record based on the abnormal operation characteristic fitting curve and the normal operation characteristic fitting curve;

[0017] The historical fault diagnosis threshold of the bearing to be monitored is calculated according to the abnormal offset value corresponding to each abnormal historical operation record.

[0018] Further, when determining the abnormal offset value corresponding to each abnormal historical operation record based on the abnormal operation characteristic fitting curve and the normal operation characteristic fitting curve, it includes:

[0019] Determine the normal operating characteristic value corresponding to each normal historical operating record;

[0020] Perform smooth fitting on all normal operation characteristic values ​​based on the acquisition time series to determine a normal operation characteristic fitting curve; wherein the horizontal axis of the normal operation characteristic fitting curve is the acquisition time, and the vertical axis is the normal operation characteristic value;

[0021] Determine the abnormal operation characteristic value corresponding to each abnormal historical operation record;

[0022] Performing smooth fitting on all abnormal operation characteristic values ​​based on the acquisition time series to determine an abnormal operation characteristic fitting curve, wherein the abscissa of the abnormal operation characteristic fitting curve is the acquisition time, and the ordinate is the abnormal operation characteristic value;

[0023] Determining whether there is an intersection between the normal operation characteristic fitting curve and the abnormal operation characteristic fitting curve;

[0024] If so, the abnormal operation characteristic value at the intersection is extracted, the shortest distance from the remaining abnormal operation characteristic values ​​on the abnormal operation characteristic fitting curve to the normal operation characteristic fitting curve is calculated, and the shortest distance is used as the corresponding abnormal offset value; the minimum shortest distance is extracted from all the shortest distances, and used as the abnormal offset value of the abnormal operation characteristic value at the intersection;

[0025] If not, the shortest distance from each abnormal operation characteristic value on the abnormal operation characteristic fitting curve to the normal operation characteristic fitting curve is calculated, and the shortest distance is used as the corresponding abnormal offset value.

[0026] Further, when calculating the historical fault diagnosis threshold of the bearing to be monitored according to the abnormal offset value corresponding to each abnormal historical operation record, it includes:

[0027] The historical fault warning threshold is obtained by the following formula:

[0028]

[0029] Among them, T h Indicates the historical fault warning threshold of the bearing to be monitored; N 1 Indicates the number of the first record; N 2 Indicates the second record quantity; D a,i represents the abnormal offset value corresponding to the i-th abnormal historical operation record; max(Da) represents the maximum abnormal offset value corresponding to the abnormal historical record; ω i represents the weight coefficient corresponding to the i-th abnormal offset value; γ represents the adjustment coefficient.

[0030] Furthermore, obtaining the real-time operation data of the bearing to be monitored and processing the real-time operation data using an adversarial autoencoder to obtain real-time feature data includes:

[0031] Preprocessing the real-time operation data to remove noise and interference information;

[0032] The preprocessed real-time operation data is input into the adversarial autoencoder, and the real-time feature data of the bearing to be monitored is extracted through adversarial training of the encoder and the decoder.

[0033] Further, when calculating the fault warning index of the bearing to be monitored based on the real-time characteristic data and the historical characteristic data, it includes:

[0034] Acquire a standard data value corresponding to the real-time characteristic data;

[0035] Determining a controllable data range according to the standard data value and the historical characteristic data, and dividing the real-time characteristic data into long-term abnormal data, short-term abnormal data and normal characteristic data according to the controllable data range;

[0036] Normalizing the long-term abnormal data to obtain a normalized data sequence;

[0037] Performing standardization processing on the long-term abnormal data to obtain a standardized data sequence;

[0038] Determine the maximum normalized value and the minimum normalized value in the normalized data sequence, and calculate the normalized sum and the normalized difference of the maximum normalized value and the minimum normalized value respectively;

[0039] Determine the maximum standardized value and the minimum standardized value in the standardized processed data sequence, and calculate the standardized sum and the standardized difference of the maximum standardized value and the minimum standardized value respectively;

[0040] Determining a first processed value of the normalized processed data sequence according to the normalized sum value and the normalized difference value;

[0041] Determine a second processed value of the standardized processed data sequence according to the standardized sum value and the standardized difference value;

[0042] Extracting all long-term abnormal data greater than the first processing value in the normalized processing data sequence to construct a first data sequence; extracting all long-term abnormal data greater than the second processing value in the standardized processing data sequence to construct a second data sequence;

[0043] Extracting long-term abnormal data from the intersection of the first data sequence and the second data sequence;

[0044] Analyze the short-term abnormal data and the normal characteristic data respectively to determine the corresponding intersection short-term abnormal data and intersection normal characteristic data;

[0045] The fault warning index of the bearing to be monitored is calculated according to the intersection long-term abnormal data, the intersection short-term abnormal data and the intersection normal characteristic data.

[0046] Further, when the real-time characteristic data is divided into long-term abnormal data, short-term abnormal data and normal characteristic data according to the controllable data range, it includes:

[0047] Determine the numerical value relationship between the standard data value and the historical characteristic data, and determine the left boundary value and the right boundary value based on the numerical value relationship;

[0048] Constructing the controllable data range according to the left boundary value and the right boundary value;

[0049] If the real-time feature data is less than or equal to the right boundary value, the corresponding real-time feature data is used as the normal feature data;

[0050] If the real-time feature data is greater than the right boundary value, an abnormal mark is generated for the corresponding real-time feature data;

[0051] Analyze all abnormal historical operation behaviors and determine the abnormal characteristic data corresponding to each abnormal historical operation behavior;

[0052] Pair the real-time feature data that generates the abnormal mark with all the abnormal feature data, and count the number of pairs;

[0053] When the number of pairs is less than the preset number of pairs, the corresponding real-time feature data is used as the short-term abnormal data;

[0054] When the pairing quantity is greater than or equal to the preset pairing quantity, the corresponding real-time feature data is used as the long-term abnormal data.

[0055] Further, when calculating the fault warning index of the bearing to be monitored according to the intersection long-term abnormal data, the intersection short-term abnormal data and the intersection normal characteristic data, it includes:

[0056] Respectively counting the first abnormal number of the intersection long-term abnormal data, the second abnormal number of the intersection short-term abnormal data, and the third abnormal number of the intersection normal feature data;

[0057] The fault warning index is obtained by the following formula:

[0058]

[0059] Among them, FPI represents the fault warning index; M 1 Indicates the first abnormal number; M 2 Indicates the second abnormal number; M 3 Indicates the third abnormal number; L j represents the characteristic value of the j-th long-term abnormal data; S k represents the characteristic value of the kth short-term abnormal data; N l represents the characteristic value of the lth normal characteristic data; p j represents the weight coefficient of the j-th long-term abnormal data; α represents the weight adjustment coefficient of the first abnormal number; β represents the weight adjustment coefficient of the second abnormal number; η represents the weight adjustment coefficient of the third abnormal number.

[0060] Further, selecting a corresponding adjustment coefficient according to the fault warning index, and adjusting the fault warning index based on the adjustment coefficient to obtain a final fault warning index includes:

[0061] Presetting a first preset fault warning index and a second preset fault warning index;

[0062] Presetting a first adjustment coefficient, a second adjustment coefficient, and a third adjustment coefficient;

[0063] Select the adjustment coefficient according to the fault warning index:

[0064] When the fault warning index is less than the first preset fault warning index, calculating a first product value of the first adjustment coefficient and the fault warning index, and using the first product value as the final fault warning index;

[0065] When the fault warning index is greater than or equal to the first preset fault warning index and less than the second preset fault warning index, calculating a second product value of the second adjustment coefficient and the fault warning index, and using the second product value as the final fault warning index;

[0066] When the fault warning index is greater than or equal to the second preset fault warning index, a third product value of the third adjustment coefficient and the fault warning index is calculated, and the third product value is used as the final fault warning index.

[0067] Further, judging whether the bearing to be monitored has a fault according to the final fault warning index, and if so, performing a fault warning includes:

[0068] Determining whether the bearing to be monitored has a fault according to the relationship between the final fault warning index and a preset fault warning threshold;

[0069] When the final fault warning index is less than the preset fault warning threshold, it is judged that the bearing to be monitored is not in a fault state; the current operating state is recorded as normal, and no further action is required; when the final fault warning index is greater than or equal to the preset fault warning threshold, it is judged that the bearing to be monitored has a fault, triggering a fault warning mechanism and generating fault alarm information; an alarm signal or prompt information is output to notify maintenance personnel to inspect or replace the bearing.

[0070] Compared with the prior art, the beneficial effect of the present invention is that the rolling bearing fault diagnosis method based on the adversarial autoencoder provided by the present invention has the advantages of high precision, high reliability and real-time performance, can effectively solve the difficulties in rolling bearing fault diagnosis, and provide strong guarantee for the stable operation of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0072] Figure 1 A flowchart of a rolling bearing fault diagnosis method based on an adversarial autoencoder provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0074] See also Figure 1 As shown, in some embodiments of the present application, this embodiment provides a rolling bearing fault diagnosis method based on an adversarial autoencoder, comprising the following steps:

[0075] S100: determining a bearing to be monitored, collecting historical operation records of the bearing to be monitored, parsing the historical operation records, and calculating a historical fault diagnosis threshold of the bearing to be monitored based on the parsing result;

[0076] S200: Acquire real-time operating data of the bearing to be monitored, and process the real-time operating data using an adversarial autoencoder to obtain real-time feature data;

[0077] S300: Acquire historical characteristic data corresponding to the real-time characteristic data in the previous normal operation record, and calculate the fault warning index of the bearing to be monitored based on the real-time characteristic data and the historical characteristic data;

[0078] S400: selecting a corresponding adjustment coefficient according to the fault warning index, and adjusting the fault warning index based on the adjustment coefficient to obtain a final fault warning index;

[0079] S500: judging whether the bearing to be monitored has a fault according to the final fault warning index, and if so, issuing a fault warning.

[0080] It can be understood that the rolling bearing fault diagnosis method based on the adversarial autoencoder provided in this embodiment can effectively extract key features from complex rolling bearing operation data through the powerful feature extraction capability of the adversarial autoencoder, thereby realizing high-precision fault diagnosis. In the specific implementation process, firstly, the bearing to be monitored is determined and its historical operation record is collected through step S100, which is the basis for subsequent fault diagnosis. Through the in-depth analysis of the historical operation record, the historical fault diagnosis threshold of the bearing to be monitored can be calculated, and this step is crucial for the subsequent judgment of whether the bearing fails. In step S200, the operation data of the bearing to be monitored is obtained in real time, and these data are processed by the adversarial autoencoder to obtain real-time feature data. This step makes full use of the advantages of the adversarial autoencoder, and can accurately extract the key features in the rolling bearing operation data, providing strong support for subsequent fault diagnosis. Then, in step S300, the historical feature data corresponding to the real-time feature data in the previous normal operation record is obtained, and the fault warning index of the bearing to be monitored is calculated based on these data. By comparing the real-time feature data with the historical feature data, the abnormal changes in the operation state of the bearing can be discovered in time, thereby early warning of potential fault risks. In step S400, the corresponding adjustment coefficient is selected according to the fault warning index, and the fault warning index is adjusted to obtain the final fault warning index. This step can further improve the accuracy and reliability of fault diagnosis by fine-tuning the fault warning index. Finally, in step S500, it is determined whether the bearing to be monitored has a fault based on the final fault warning index, and a fault warning is issued when necessary. When the final fault warning index exceeds the preset threshold, the system will trigger the fault warning mechanism and promptly notify the maintenance personnel to repair or replace the bearing to avoid equipment damage or production accidents caused by bearing failure.

[0081] It can be understood that the rolling bearing fault diagnosis method based on the adversarial autoencoder provided in this embodiment has the advantages of high precision, high reliability and real-time performance, and can effectively solve the difficulties in rolling bearing fault diagnosis and provide strong guarantee for the stable operation of industrial equipment.

[0082] Specifically, the historical operation records are parsed, and the historical fault diagnosis threshold of the bearing to be monitored is calculated based on the parsing result, including:

[0083] Analyze the historical operation records to obtain normal historical operation records and abnormal historical operation records;

[0084] respectively counting the number of first records of the abnormal historical operation records and the number of second records of the normal historical operation records;

[0085] Determine an abnormal operation characteristic fitting curve and a normal operation characteristic fitting curve according to the abnormal historical operation record and the normal historical operation record;

[0086] Determine an abnormal offset value corresponding to each abnormal historical operation record based on the abnormal operation characteristic fitting curve and the normal operation characteristic fitting curve;

[0087] The historical fault diagnosis threshold of the bearing to be monitored is calculated according to the abnormal offset value corresponding to each abnormal historical operation record.

[0088] In this embodiment, the normal historical operation records refer to the operation records of the bearing in normal operation, which reflect the various parameters and characteristics of the bearing in normal operation. The abnormal historical operation records refer to the operation records of the bearing in a faulty or abnormal state, which contain key information when the fault or abnormality occurs. The types of bearing faults and abnormal states are bearing wear, bearing overheating, bearing cracks, or poor bearing lubrication.

[0089] Specifically, when determining the abnormal offset value corresponding to each abnormal historical operation record based on the abnormal operation characteristic fitting curve and the normal operation characteristic fitting curve, it includes:

[0090] Determine the normal operating characteristic value corresponding to each normal historical operating record;

[0091] Perform smooth fitting on all normal operation characteristic values ​​based on the acquisition time series to determine a normal operation characteristic fitting curve; wherein the horizontal axis of the normal operation characteristic fitting curve is the acquisition time, and the vertical axis is the normal operation characteristic value;

[0092] Determine the abnormal operation characteristic value corresponding to each abnormal historical operation record;

[0093] Performing smooth fitting on all abnormal operation characteristic values ​​based on the acquisition time series to determine an abnormal operation characteristic fitting curve, wherein the abscissa of the abnormal operation characteristic fitting curve is the acquisition time, and the ordinate is the abnormal operation characteristic value;

[0094] Determining whether there is an intersection between the normal operation characteristic fitting curve and the abnormal operation characteristic fitting curve;

[0095] If so, the abnormal operation characteristic value at the intersection is extracted, the shortest distance from the remaining abnormal operation characteristic values ​​on the abnormal operation characteristic fitting curve to the normal operation characteristic fitting curve is calculated, and the shortest distance is used as the corresponding abnormal offset value; the minimum shortest distance is extracted from all the shortest distances, and used as the abnormal offset value of the abnormal operation characteristic value at the intersection;

[0096] If not, the shortest distance from each abnormal operation characteristic value on the abnormal operation characteristic fitting curve to the normal operation characteristic fitting curve is calculated, and the shortest distance is used as the corresponding abnormal offset value.

[0097] It is understandable that we can accurately quantify the abnormal conditions that occur during the operation of the bearing, thereby providing a more accurate basis for its fault diagnosis. After determining the abnormal offset values, the historical fault diagnosis threshold can be set based on these offset values.

[0098] Specifically, when calculating the historical fault diagnosis threshold of the bearing to be monitored according to the abnormal offset value corresponding to each abnormal historical operation record, it includes:

[0099] The historical fault warning threshold is obtained by the following formula:

[0100]

[0101] Among them, T h Indicates the historical fault warning threshold of the bearing to be monitored; N 1 Indicates the number of the first record; N 2 Indicates the second record quantity; D a,i represents the abnormal offset value corresponding to the i-th abnormal historical operation record; max(Da) represents the maximum abnormal offset value corresponding to the abnormal historical record; ω i represents the weight coefficient corresponding to the i-th abnormal offset value; γ represents the adjustment coefficient.

[0102] It can be understood that the calculation method of the historical fault diagnosis threshold comprehensively considers the number of abnormal historical operation records, the size of the abnormal offset value and the corresponding weight coefficient, so that the calculated threshold is more in line with the actual situation and the accuracy of fault diagnosis is improved. The determination of the weight coefficient ωi can be adjusted according to the specific situation of the abnormal historical operation record. For example, for abnormal types that occur frequently and have a greater impact on the operating status of the bearing, a larger weight coefficient can be assigned. The introduction of the adjustment coefficient γ further increases the flexibility of the calculation process, so that the historical fault diagnosis threshold can be adaptively adjusted according to different application scenarios and requirements.

[0103] Specifically, obtaining the real-time operating data of the bearing to be monitored, and processing the real-time operating data using an adversarial autoencoder to obtain real-time feature data includes:

[0104] Preprocessing the real-time operation data to remove noise and interference information;

[0105] The preprocessed real-time operation data is input into the adversarial autoencoder, and the real-time feature data of the bearing to be monitored is extracted through adversarial training of the encoder and the decoder.

[0106] It is understandable that real-time operation data often contains a lot of noise and interference information. If this information is not preprocessed, it may affect the accuracy and reliability of subsequent feature extraction. Therefore, before the real-time operation data is input into the adversarial autoencoder, it needs to be preprocessed. The main purpose of preprocessing is to remove noise and interference information in the data and retain key information useful for fault diagnosis. This can not only improve the accuracy of feature extraction, but also reduce the amount of calculation and improve the efficiency of the entire fault diagnosis method. After the preprocessed real-time operation data is input into the adversarial autoencoder, the encoder and decoder will perform adversarial training. The encoder is responsible for mapping the input data to the latent feature space, while the decoder is responsible for reconstructing the data in the latent feature space back to the original data space. Through adversarial training, the encoder and decoder will continuously optimize their own parameters to improve the ability and accuracy of feature extraction. Finally, through the processing of the adversarial autoencoder, we can extract key features, namely real-time feature data, from the real-time operation data. These real-time feature data can accurately reflect the operating status of the bearing to be monitored and provide strong support for subsequent fault diagnosis.

[0107] Specifically, when calculating the fault warning index of the bearing to be monitored based on the real-time characteristic data and the historical characteristic data, it includes:

[0108] Acquire a standard data value corresponding to the real-time characteristic data;

[0109] Determining a controllable data range according to the standard data value and the historical characteristic data, and dividing the real-time characteristic data into long-term abnormal data, short-term abnormal data and normal characteristic data according to the controllable data range;

[0110] Normalizing the long-term abnormal data to obtain a normalized data sequence;

[0111] Performing standardization processing on the long-term abnormal data to obtain a standardized data sequence;

[0112] Determine the maximum normalized value and the minimum normalized value in the normalized data sequence, and calculate the normalized sum and the normalized difference of the maximum normalized value and the minimum normalized value respectively;

[0113] Determine the maximum standardized value and the minimum standardized value in the standardized processed data sequence, and calculate the standardized sum and the standardized difference of the maximum standardized value and the minimum standardized value respectively;

[0114] Determining a first processed value of the normalized processed data sequence according to the normalized sum value and the normalized difference value;

[0115] Determine a second processed value of the standardized processed data sequence according to the standardized sum value and the standardized difference value;

[0116] Extracting all long-term abnormal data greater than the first processing value in the normalized processing data sequence to construct a first data sequence; extracting all long-term abnormal data greater than the second processing value in the standardized processing data sequence to construct a second data sequence;

[0117] Extracting long-term abnormal data from the intersection of the first data sequence and the second data sequence;

[0118] Analyze the short-term abnormal data and the normal characteristic data respectively to determine the corresponding intersection short-term abnormal data and intersection normal characteristic data;

[0119] The fault warning index of the bearing to be monitored is calculated according to the intersection long-term abnormal data, the intersection short-term abnormal data and the intersection normal characteristic data.

[0120] It is understandable that when calculating the fault warning index, the correlation and difference between real-time feature data and historical feature data are fully considered. Through careful analysis and processing of real-time feature data, we can accurately identify abnormal changes in the operating status of the bearing and calculate the fault warning index accordingly. The level of this index directly reflects the risk of bearing failure, providing an important basis for subsequent fault warning and maintenance decisions.

[0121] Specifically, when the real-time characteristic data is divided into long-term abnormal data, short-term abnormal data and normal characteristic data according to the controllable data range, it includes:

[0122] Determine the numerical value relationship between the standard data value and the historical characteristic data, and determine the left boundary value and the right boundary value based on the numerical value relationship;

[0123] Constructing the controllable data range according to the left boundary value and the right boundary value;

[0124] If the real-time feature data is less than or equal to the right boundary value, the corresponding real-time feature data is used as the normal feature data;

[0125] If the real-time feature data is greater than the right boundary value, an abnormal mark is generated for the corresponding real-time feature data;

[0126] Analyze all abnormal historical operation behaviors and determine the abnormal characteristic data corresponding to each abnormal historical operation behavior;

[0127] Pair the real-time feature data that generates the abnormal mark with all the abnormal feature data, and count the number of pairs;

[0128] When the number of pairs is less than the preset number of pairs, the corresponding real-time feature data is used as the short-term abnormal data;

[0129] When the pairing quantity is greater than or equal to the preset pairing quantity, the corresponding real-time feature data is used as the long-term abnormal data.

[0130] It is understandable that through the detailed division and processing of real-time feature data, we can more accurately identify abnormal changes in the operating status of the bearing, providing a more accurate basis for subsequent fault warning and maintenance decisions. When determining the controllable data range, the numerical relationship between the standard data value and the historical feature data is fully considered to ensure that the determined range can truly reflect the normal operating status of the bearing. On this basis, the real-time feature data is compared with the controllable data range, and it is divided into long-term abnormal data, short-term abnormal data and normal feature data according to different data conditions. This division process not only helps us to have a deeper understanding of the operating status of the bearing, but also provides an important basis for the subsequent calculation of the fault warning index. In the calculation process of the fault warning index, we fully consider the correlation and difference between long-term abnormal data, short-term abnormal data and normal feature data. Through in-depth analysis and processing of these data, we can accurately calculate the fault warning index, so as to timely discover abnormal changes in the operating status of the bearing, and provide important support for subsequent fault warning and maintenance decisions.

[0131] Specifically, when calculating the fault warning index of the bearing to be monitored according to the intersection long-term abnormal data, the intersection short-term abnormal data and the intersection normal characteristic data, it includes:

[0132] Respectively counting the first abnormal number of the intersection long-term abnormal data, the second abnormal number of the intersection short-term abnormal data, and the third abnormal number of the intersection normal feature data;

[0133] The fault warning index is obtained by the following formula:

[0134]

[0135] Among them, FPI represents the fault warning index; M 1 Indicates the first abnormal number; M 2 Indicates the second abnormal number; M 3 Indicates the third abnormal number; L j represents the characteristic value of the j-th long-term abnormal data; S k represents the characteristic value of the kth short-term abnormal data; N l represents the characteristic value of the lth normal characteristic data; p jrepresents the weight coefficient of the j-th long-term abnormal data; α represents the weight adjustment coefficient of the first abnormal number; β represents the weight adjustment coefficient of the second abnormal number; η represents the weight adjustment coefficient of the third abnormal number.

[0136] It can be understood that the calculation method of the fault warning index comprehensively considers the number and characteristic values ​​of the intersection long-term abnormal data, the intersection short-term abnormal data and the intersection normal characteristic data, as well as the corresponding weight coefficients and weight adjustment coefficients, so that the calculated fault warning index is more accurate and reliable. The characteristic values ​​Lj, Sk and Nl represent the specific characteristic performances of long-term abnormal data, short-term abnormal data and normal characteristic data respectively, and their sizes directly reflect the abnormal degree of the bearing operation status. The introduction of weight coefficients pj, α, β and η further considers the differences in the importance of different types of data in fault warning, making the calculation process more in line with the actual situation.

[0137] Specifically, selecting a corresponding adjustment coefficient according to the fault warning index, and adjusting the fault warning index based on the adjustment coefficient to obtain a final fault warning index includes:

[0138] Presetting a first preset fault warning index and a second preset fault warning index;

[0139] Presetting a first adjustment coefficient, a second adjustment coefficient, and a third adjustment coefficient;

[0140] Select the adjustment coefficient according to the fault warning index:

[0141] When the fault warning index is less than the first preset fault warning index, calculating a first product value of the first adjustment coefficient and the fault warning index, and using the first product value as the final fault warning index;

[0142] When the fault warning index is greater than or equal to the first preset fault warning index and less than the second preset fault warning index, calculating a second product value of the second adjustment coefficient and the fault warning index, and using the second product value as the final fault warning index;

[0143] When the fault warning index is greater than or equal to the second preset fault warning index, a third product value of the third adjustment coefficient and the fault warning index is calculated, and the third product value is used as the final fault warning index.

[0144] It is understandable that by further adjusting the fault warning index, we can obtain a final fault warning index that is more accurate and in line with the actual situation. This adjustment process fully considers the size of the fault warning index and the corresponding adjustment coefficient, so that the final warning index can more accurately reflect the risk of bearing failure. The setting of the first preset fault warning index and the second preset fault warning index provides us with an important reference for judging the size of the fault warning index. The introduction of the first adjustment coefficient, the second adjustment coefficient and the third adjustment coefficient further increases the flexibility and accuracy of the adjustment process. According to the different sizes of the fault warning index, we can select the appropriate adjustment coefficient for calculation to obtain the final fault warning index.

[0145] Specifically, judging whether the bearing to be monitored has a fault according to the final fault warning index, and if so, performing a fault warning includes:

[0146] Determining whether the bearing to be monitored has a fault according to the relationship between the final fault warning index and a preset fault warning threshold;

[0147] When the final fault warning index is less than the preset fault warning threshold, it is determined that the bearing to be monitored is not in a fault state; the current operating state is recorded as normal, and no further action is required;

[0148] When the final fault warning index is greater than or equal to the preset fault warning threshold, it is determined that the monitored bearing has a fault, the fault warning mechanism is triggered, and fault alarm information is generated; an alarm signal or prompt information is output to notify maintenance personnel to inspect or replace the bearing.

[0149] It is understandable that by comparing the final fault warning index with the preset fault warning threshold, it is possible to timely and accurately determine whether the bearing to be monitored has a fault. When the final fault warning index is lower than the preset fault warning threshold, it means that the current operating state of the bearing is normal and no additional maintenance measures are required. In this case, we can continue to monitor the operating state of the bearing to ensure its stable operation. When the final fault warning index reaches or exceeds the preset fault warning threshold, it means that the bearing may have a fault, and the fault warning mechanism needs to be triggered immediately. The triggering of the fault warning mechanism will automatically generate fault alarm information and notify the maintenance personnel in time by outputting an alarm signal or prompt information. After receiving the alarm information, the maintenance personnel can respond quickly and inspect the bearing to determine the specific cause and extent of the fault. According to the inspection results, the maintenance personnel can take corresponding repair or replacement measures to avoid further development of the fault and ensure the reliability and safety of the equipment. This process not only improves the accuracy and efficiency of fault diagnosis, but also provides strong support for preventive maintenance of equipment, which helps to extend the service life of the equipment and reduce maintenance costs.

[0150] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may 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 codes.

[0151] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0152] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0153] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A rolling bearing fault diagnosis method based on an adversarial autoencoder, characterized in that: include: Determine a bearing to be monitored, collect historical operation records of the bearing to be monitored, analyze the historical operation records, and calculate a historical fault diagnosis threshold of the bearing to be monitored based on the analysis results; Acquire real-time operating data of the bearing to be monitored, and process the real-time operating data using an adversarial autoencoder to obtain real-time feature data; Acquire historical characteristic data corresponding to the real-time characteristic data in the previous normal operation record, and calculate the fault warning index of the bearing to be monitored based on the real-time characteristic data and the historical characteristic data; Selecting a corresponding adjustment coefficient according to the fault warning index, and adjusting the fault warning index based on the adjustment coefficient to obtain a final fault warning index; It is determined whether the bearing to be monitored has a fault according to the final fault warning index, and if so, a fault warning is issued.

2. The rolling bearing fault diagnosis method based on the adversarial autoencoder according to claim 1 is characterized in that: The historical operation records are parsed, and the historical fault diagnosis threshold of the bearing to be monitored is calculated based on the parsing result, including: Analyze the historical operation records to obtain normal historical operation records and abnormal historical operation records; respectively counting the number of first records of the abnormal historical operation records and the number of second records of the normal historical operation records; Determine an abnormal operation characteristic fitting curve and a normal operation characteristic fitting curve according to the abnormal historical operation record and the normal historical operation record; Determine an abnormal offset value corresponding to each abnormal historical operation record based on the abnormal operation characteristic fitting curve and the normal operation characteristic fitting curve; The historical fault diagnosis threshold of the bearing to be monitored is calculated according to the abnormal offset value corresponding to each abnormal historical operation record.

3. The rolling bearing fault diagnosis method based on the adversarial autoencoder according to claim 2 is characterized in that: When determining the abnormal offset value corresponding to each abnormal historical operation record based on the abnormal operation characteristic fitting curve and the normal operation characteristic fitting curve, it includes: Determine the normal operating characteristic value corresponding to each normal historical operating record; Perform smooth fitting on all normal operation characteristic values ​​based on the acquisition time series to determine a normal operation characteristic fitting curve; wherein the horizontal axis of the normal operation characteristic fitting curve is the acquisition time, and the vertical axis is the normal operation characteristic value; Determine the abnormal operation characteristic value corresponding to each abnormal historical operation record; Performing smooth fitting on all abnormal operation characteristic values ​​based on the acquisition time series to determine an abnormal operation characteristic fitting curve, wherein the abscissa of the abnormal operation characteristic fitting curve is the acquisition time, and the ordinate is the abnormal operation characteristic value; Determining whether there is an intersection between the normal operation characteristic fitting curve and the abnormal operation characteristic fitting curve; If so, the abnormal operation characteristic value at the intersection is extracted, the shortest distance from the remaining abnormal operation characteristic values ​​on the abnormal operation characteristic fitting curve to the normal operation characteristic fitting curve is calculated, and the shortest distance is used as the corresponding abnormal offset value; the minimum shortest distance is extracted from all the shortest distances, and used as the abnormal offset value of the abnormal operation characteristic value at the intersection; If not, the shortest distance from each abnormal operation characteristic value on the abnormal operation characteristic fitting curve to the normal operation characteristic fitting curve is calculated, and the shortest distance is used as the corresponding abnormal offset value.

4. The rolling bearing fault diagnosis method based on the adversarial autoencoder according to claim 3 is characterized in that: When calculating the historical fault diagnosis threshold of the bearing to be monitored according to the abnormal offset value corresponding to each abnormal historical operation record, it includes: The historical fault warning threshold is obtained by the following formula: Among them, T h represents the historical fault warning threshold of the bearing to be monitored; N1 represents the number of first records; N2 represents the number of second records; D a,i represents the abnormal offset value corresponding to the i-th abnormal historical operation record; max(Da) represents the maximum abnormal offset value corresponding to the abnormal historical record; ω i represents the weight coefficient corresponding to the i-th abnormal offset value; γ represents the adjustment coefficient.

5. The rolling bearing fault diagnosis method based on adversarial autoencoder according to claim 1, characterized in that: Acquiring the real-time operating data of the bearing to be monitored, and processing the real-time operating data using an adversarial autoencoder to obtain real-time feature data includes: Preprocessing the real-time operation data to remove noise and interference information; The preprocessed real-time operation data is input into the adversarial autoencoder, and the real-time feature data of the bearing to be monitored is extracted through adversarial training of the encoder and the decoder.

6. The rolling bearing fault diagnosis method based on adversarial autoencoder according to claim 2 is characterized in that: When calculating the fault warning index of the bearing to be monitored based on the real-time characteristic data and the historical characteristic data, it includes: Acquire a standard data value corresponding to the real-time characteristic data; Determining a controllable data range according to the standard data value and the historical characteristic data, and dividing the real-time characteristic data into long-term abnormal data, short-term abnormal data and normal characteristic data according to the controllable data range; Normalizing the long-term abnormal data to obtain a normalized data sequence; Performing standardization processing on the long-term abnormal data to obtain a standardized data sequence; Determine the maximum normalized value and the minimum normalized value in the normalized data sequence, and calculate the normalized sum and the normalized difference of the maximum normalized value and the minimum normalized value respectively; Determine the maximum standardized value and the minimum standardized value in the standardized processed data sequence, and calculate the standardized sum and the standardized difference of the maximum standardized value and the minimum standardized value respectively; Determining a first processed value of the normalized processed data sequence according to the normalized sum value and the normalized difference value; Determine a second processed value of the standardized processed data sequence according to the standardized sum value and the standardized difference value; Extracting all long-term abnormal data greater than the first processing value in the normalized processing data sequence to construct a first data sequence; extracting all long-term abnormal data greater than the second processing value in the standardized processing data sequence to construct a second data sequence; Extracting long-term abnormal data from the intersection of the first data sequence and the second data sequence; Analyze the short-term abnormal data and the normal characteristic data respectively to determine the corresponding intersection short-term abnormal data and intersection normal characteristic data; The fault warning index of the bearing to be monitored is calculated according to the intersection long-term abnormal data, the intersection short-term abnormal data and the intersection normal characteristic data.

7. The rolling bearing fault diagnosis method based on adversarial autoencoder according to claim 6 is characterized in that: When the real-time characteristic data is divided into long-term abnormal data, short-term abnormal data and normal characteristic data according to the controllable data range, it includes: Determine the numerical value relationship between the standard data value and the historical characteristic data, and determine the left boundary value and the right boundary value based on the numerical value relationship; Constructing the controllable data range according to the left boundary value and the right boundary value; If the real-time feature data is less than or equal to the right boundary value, the corresponding real-time feature data is used as the normal feature data; If the real-time feature data is greater than the right boundary value, an abnormal mark is generated for the corresponding real-time feature data; Analyze all abnormal historical operation behaviors and determine the abnormal characteristic data corresponding to each abnormal historical operation behavior; Pair the real-time feature data that generates the abnormal mark with all the abnormal feature data, and count the number of pairs; When the number of pairs is less than the preset number of pairs, the corresponding real-time feature data is used as the short-term abnormal data; When the pairing quantity is greater than or equal to the preset pairing quantity, the corresponding real-time feature data is used as the long-term abnormal data.

8. The rolling bearing fault diagnosis method based on adversarial autoencoder according to claim 7 is characterized in that: When calculating the fault warning index of the bearing to be monitored according to the intersection long-term abnormal data, the intersection short-term abnormal data and the intersection normal characteristic data, it includes: Respectively counting the first abnormal number of the intersection long-term abnormal data, the second abnormal number of the intersection short-term abnormal data, and the third abnormal number of the intersection normal feature data; The fault warning index is obtained by the following formula: Among them, FPI represents the fault warning index; M1 represents the number of first abnormalities; M2 represents the number of second abnormalities; M3 represents the number of third abnormalities; L j represents the characteristic value of the j-th long-term abnormal data; S k represents the characteristic value of the kth short-term abnormal data; N l represents the characteristic value of the lth normal characteristic data; p j represents the weight coefficient of the j-th long-term abnormal data; α represents the weight adjustment coefficient of the first abnormal number; β represents the weight adjustment coefficient of the second abnormal number; η represents the weight adjustment coefficient of the third abnormal number.

9. The rolling bearing fault diagnosis method based on adversarial autoencoder according to claim 1, characterized in that: Selecting a corresponding adjustment coefficient according to the fault warning index, and adjusting the fault warning index based on the adjustment coefficient to obtain a final fault warning index includes: Presetting a first preset fault warning index and a second preset fault warning index; Presetting a first adjustment coefficient, a second adjustment coefficient, and a third adjustment coefficient; Select the adjustment coefficient according to the fault warning index: When the fault warning index is less than the first preset fault warning index, calculating a first product value of the first adjustment coefficient and the fault warning index, and using the first product value as the final fault warning index; When the fault warning index is greater than or equal to the first preset fault warning index and less than the second preset fault warning index, calculating a second product value of the second adjustment coefficient and the fault warning index, and using the second product value as the final fault warning index; When the fault warning index is greater than or equal to the second preset fault warning index, a third product value of the third adjustment coefficient and the fault warning index is calculated, and the third product value is used as the final fault warning index.

10. The rolling bearing fault diagnosis method based on adversarial autoencoder according to claim 1, characterized in that: Judging whether the bearing to be monitored has a fault according to the final fault warning index, and if so, performing a fault warning includes: Determining whether the bearing to be monitored has a fault according to the relationship between the final fault warning index and a preset fault warning threshold; When the final fault warning index is less than the preset fault warning threshold, it is determined that the bearing to be monitored is not in a fault state; the current operating state is recorded as normal, and no further action is required; When the final fault warning index is greater than or equal to the preset fault warning threshold, it is determined that the monitored bearing has a fault, the fault warning mechanism is triggered, and fault alarm information is generated; an alarm signal or prompt information is output to notify maintenance personnel to inspect or replace the bearing.

Images