Automatic anomaly detection method for high-speed rotating transmission device

Abstract

The invention relates to the technical field of fault diagnosis, in particular to an automatic anomaly detection method for a high-speed rotating transmission device, which comprises the following steps of: acquiring angular acceleration, rotating inertia and strain data of the high-speed rotating transmission device, calculating a transient strain gradient and matching a rotating speed change rate, adjusting a strain deviation range according to the change of the rotating inertia, and detecting the anomaly of the high-speed rotating transmission device. And screening strain abrupt change points to obtain a rotation strain matching result. According to the method, the dynamic strain response interval is set, the strain deviation range is adjusted, the strain abrupt change points are screened, the frequency response of the equipment is monitored in real time and compared with the normal frequency, so that continuous monitoring of the frequency deviation rate is more accurate, the frequency deviation rate jointly acts on identification and clustering of abnormal characteristics, the accuracy and the real-time performance of fault diagnosis are improved, and the fault diagnosis efficiency is improved. By calculating the deviation degree of the abnormal characteristics and the normal track, the abnormal detection threshold can be adjusted, potential faults can be identified and processed more effectively, the maintenance cost of the high-speed rotating transmission device is reduced, and the operation efficiency is improved.

Description

Technical Field

[0001] The invention relates to the technical field of fault diagnosis, and in particular to an automatic abnormality detection method for a high-speed rotating transmission device. Background Art

[0002] The field of fault diagnosis technology includes methods and technologies for identifying, analyzing and processing abnormalities or faults that occur in mechanical equipment and systems during operation. This field focuses on capturing equipment status data through various sensors and data acquisition technologies, and using analytical models to predict and diagnose potential mechanical failures. The core content includes data collection, feature recognition, condition monitoring, fault prediction and health management. By systematically integrating technologies, fault diagnosis can help improve equipment reliability and maintenance efficiency, reduce unplanned downtime, and optimize maintenance plans.

[0003] Among them, the automatic abnormality detection method of high-speed rotating transmission device refers to the use of automation technology to detect abnormal behaviors of high-speed rotating transmission device during operation. The technical matters targeted by this technical subject cover the establishment of data acquisition system, analysis and processing of multi-dimensional parameters, and automatic adjustment of process parameters. Specifically, the operation data is collected by sensors installed on the equipment, and the data is processed by graph neural network to identify abnormal patterns. At the same time, reinforcement learning is used to optimize and adjust the process parameters in the production process. The means work together to achieve real-time monitoring and automatic adjustment of the state of the transmission device.

[0004] Existing fault diagnosis technology relies on traditional data collection and processing methods, which shows its limitations when dealing with high-speed rotating equipment. Existing technology cannot provide sufficient sensitivity and response speed when capturing complex dynamic changes and transient abnormal behaviors. This leads to the failure of the fault diagnosis system to respond and adjust in a timely and accurate manner when the equipment encounters atypical failures or sudden abnormalities in actual operations, resulting in premature wear or failure of mechanical equipment, increasing the company's operating costs and the risk of production interruptions. Existing technologies lack adaptability and flexibility to failure modes in dynamic environments, making it difficult to achieve health management of the equipment throughout its life cycle, which is particularly evident in high-demand industrial applications. Summary of the invention

[0005] In order to solve the limitations of the existing technology when dealing with high-speed rotating equipment, that is, the existing technology cannot provide sufficient sensitivity and response speed when capturing complex dynamic changes and transient abnormal behaviors, resulting in that in actual operation, when the equipment encounters atypical failures or sudden abnormalities, the fault diagnosis system cannot respond and adjust in a timely and accurate manner, resulting in premature wear or failure of mechanical equipment, increasing the company's operating costs and the risk of production interruptions; and the existing technology lacks adaptability and flexibility to failure modes in a dynamic environment, making it difficult to achieve health management of the equipment throughout its life cycle. The present invention provides an automatic abnormality detection method for a high-speed rotating transmission device. The technical solution is as follows:

[0006] In one aspect, a method for automatic abnormality detection of a high-speed rotating transmission device is provided, the method comprising:

[0007] S1: Obtain the angular acceleration, rotational inertia and strain data of the high-speed rotating transmission device, calculate the transient strain gradient and match the speed change rate, adjust the strain deviation range according to the change of rotational inertia, screen the strain mutation point, and obtain the rotation strain matching result;

[0008] S2: Using the rotation strain matching result, record the real-time frequency response of the high-speed rotation transmission device, perform continuous monitoring to obtain the frequency deviation rate sequence, refer to the deviation of the abnormal feature points identified by the vibration acceleration, impact signal and transient stress, and obtain the rotation frequency deviation identification result;

[0009] S3: using the rotation frequency offset identification result, combined with vibration acceleration, transient stress and impact signal, analyzing the deviation degree between abnormal feature points and normal frequency trajectory, identifying local density distribution, screening continuous offset anomalies and sudden anomalies, and obtaining abnormal feature clustering results;

[0010] S4: calling the abnormal feature clustering result, calculating the fitting residual change rate of the continuous offset anomaly, setting the abnormal level according to the change rate, screening the abnormal feature points that meet the set range, obtaining the abnormal pattern matching result, automatically monitoring the abnormal signal of the high-speed rotating transmission device, and issuing an early warning.

[0011] As a further solution of the present invention, the rotational strain matching result includes a transient strain gradient threshold, a strain response interval, a strain deviation range, and a strain mutation; the rotational frequency offset identification result includes a frequency response ratio, a frequency offset rate threshold, and a feature point deviation amplitude; the abnormal feature clustering result includes a trajectory deviation degree, a local density distribution feature, and a continuous offset duration threshold.

[0012] As a further solution of the present invention, the step of obtaining the rotation strain matching result is specifically:

[0013] S101: acquiring angular acceleration, rotational inertia and strain data of the high-speed rotating transmission device, calculating the change of the rotational inertia, calculating the instantaneous increment of angular acceleration according to the change of the rotational inertia, matching the instantaneous increment of angular acceleration with the strain data, analyzing the degree of change of transient strain, and acquiring transient strain matching data;

[0014] S102: Based on the transient strain matching data, identifying the time variation trend of the rotational inertia, analyzing the distribution of the rotational inertia at the differentiated time points, setting a reference range of the rotational inertia variation, screening the transient strain data that meets the conditions, and obtaining the dynamic strain interval;

[0015] S103: calling the dynamic strain interval, analyzing the offset degree of the strain data, screening the mutation point in combination with the change trend of the rotational inertia, and performing comparison in combination with the instantaneous increment of the angular acceleration to obtain the rotational strain matching result.

[0016] As a further solution of the present invention, the step of obtaining the rotation frequency offset identification result is specifically:

[0017] S201: calling the rotation strain matching result, recording the real-time frequency response of the high-speed rotation transmission device, comparing the real-time operating frequency with the normal frequency, and obtaining a frequency deviation identification result;

[0018] S202: According to the frequency deviation identification result, continuously monitor the comparison value data, calculate the frequency change amplitude according to the time series, obtain the frequency offset rate sequence, call the frequency offset rate sequence, identify the dynamic range according to the frequency change trend, match and analyze the dynamic range data with the frequency offset rate, evaluate the corresponding relationship between the frequency change range and time, select the data points that meet the change range, and obtain the frequency offset dynamic range;

[0019] S203: Utilizing the frequency offset dynamic range, referring to vibration acceleration, impact signal and transient stress, analyzing the deviation of abnormal feature points, screening frequency offset data points, calculating frequency offset feature values, and comparing the corresponding transient stress fluctuations to obtain a rotation frequency offset identification result.

[0020] As a further solution of the present invention, the formula for calculating the frequency offset characteristic value is as follows:

[0021] ;

[0022] in, represents the frequency shift characteristic value, Representative The frequency value of the data point, represents the average frequency value of the data points, Representative The vibration acceleration amplitude corresponding to the data point is Represents the total number of data points.

[0023] As a further solution of the present invention, the step of obtaining the abnormal feature clustering result is specifically as follows:

[0024] S301: using the rotation frequency offset identification result, combined with vibration acceleration, transient stress and impact signal, identifying the distribution of abnormal feature points, comparing the abnormal feature points with the normal frequency trajectory, identifying the degree of deviation between the two, and obtaining frequency trajectory deviation data;

[0025] S302: calling the frequency trajectory deviation data, identifying the local density distribution according to the degree of deviation, calculating the number of abnormal feature points, identifying the density change in the local area, comparing the density fluctuation amplitude under the time series, and obtaining the local density distribution characteristics;

[0026] S303: Analyze the temporal variation trend of local density by using the local density distribution characteristics, screen out continuous deviation anomalies and sudden anomalies, classify abnormal feature points according to the amplitude and stability of density variation, and classify anomaly categories to obtain abnormal feature clustering results.

[0027] As a further solution of the present invention, the formula for calculating the number of abnormal feature points is as follows:

[0028] ;

[0029] in, Represents the number of abnormal feature points, Representative The deviation of feature points, Represents the average value of the deviation of feature points, Represents the standard deviation of the feature point deviation, Represents the total number of feature points in the differentiated area, Representative The feature point and The associated weight of the reference point, Representative The feature point and The influence factor of the reference point, Represents the total number of reference points.

[0030] As a further solution of the present invention, the step of obtaining the abnormal pattern matching result is specifically:

[0031] S401: extracting the data sequence of the continuous shift anomaly based on the abnormal feature clustering result, calculating the fitting residuals of the differentiated time points, comparing the residual difference values ​​of adjacent time points, and obtaining the continuous shift residual change rate;

[0032] S402: calling the continuous offset residual change rate, analyzing the abnormality level according to the change rate distribution, classifying the change rate according to the numerical range, setting the level standard, marking the corresponding level of the abnormal point, and obtaining the abnormality level classification list;

[0033] S403: using the abnormal level classification list, for sudden abnormalities, calculating the instantaneous signal change amplitude of the mutation factor, comparing the change amplitude with the set range, screening abnormal feature points that meet the range, and obtaining abnormal pattern matching results.

[0034] As a further solution of the present invention, the method further comprises step S5:

[0035] S5: Based on the abnormal pattern matching result, the difference between the real-time frequency deviation rate and the normal trajectory is calculated, the difference is called to adjust the abnormal detection threshold, the risk feature point ratio is calculated according to the abnormal feature point distribution, the rotation state abnormality assessment result is obtained, and the signal acquisition weight and data sampling frequency are adjusted;

[0036] The rotation state abnormality assessment result includes an abnormality detection threshold, a risk feature point ratio, and a frequency offset risk index.

[0037] As a further solution of the present invention, the step of obtaining the abnormal rotation state evaluation result is specifically:

[0038] S501: calling the abnormal pattern matching result, calculating the real-time frequency offset rate, extracting the frequency data of the normal trajectory, comparing the values ​​of the two, calculating the difference between the real-time frequency offset rate and the normal trajectory, and obtaining the frequency offset difference data;

[0039] S502: Using the frequency offset difference data, adjusting the anomaly detection threshold according to the numerical range, comparing the frequency offsets at differentiated time points, calculating the proportion of abnormal feature points, identifying the distribution of abnormal feature points, obtaining the rotation state anomaly assessment result, and adjusting the signal acquisition weight.

[0040] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0041] By carefully acquiring and analyzing the angular acceleration, rotational inertia and strain data of high-speed rotating transmission devices, the transient strain gradient is accurately calculated and matched with the speed change rate. The data-driven method optimizes the setting of the dynamic strain response interval and the adjustment of the strain deviation range, and effectively screens the strain mutation points. The frequency response of the equipment is monitored in real time and compared with the normal frequency, making the continuous monitoring of the frequency deviation rate more accurate, which works together to identify and cluster abnormal features, further improving the accuracy and real-time performance of fault diagnosis. By calculating the degree of deviation of abnormal features from the normal trajectory, the abnormal detection threshold can be adjusted to more effectively identify and handle potential faults, reduce the maintenance cost of high-speed rotating transmission devices and improve operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the workflow of the present invention; DETAILED DESCRIPTION

[0043] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0044] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0045] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0046] See also Figure 1 The embodiment of the present invention provides an automatic abnormality detection method for a high-speed rotating transmission device. The processing flow of the method may include the following steps:

[0047] S1: Obtain the angular acceleration, rotational inertia and strain data of the high-speed rotating transmission device, calculate the transient strain gradient and match the speed change rate, set the dynamic strain response range, adjust the strain deviation range according to the change of rotational inertia, screen the strain mutation point, and obtain the rotation strain matching result;

[0048] S2: Using the rotation strain matching results, record the real-time frequency response of the high-speed rotation transmission device, calculate the ratio of the real-time working frequency to the normal frequency, call the ratio for continuous monitoring to obtain the frequency offset rate sequence, call the dynamic range of the frequency offset rate, refer to the deviation of the abnormal feature points identified by the vibration acceleration, impact signal and transient stress, and obtain the rotation frequency offset identification result;

[0049] S3: Using the rotation frequency offset recognition results, combined with vibration acceleration, transient stress and impact signals, analyze the deviation degree between abnormal feature points and normal frequency trajectory, identify local density distribution, call the change characteristics of local density to screen continuous offset anomalies and sudden anomalies, and obtain abnormal feature clustering results;

[0050] S4: Call the abnormal feature clustering results, calculate the fitting residual change rate of the continuous offset anomaly, set the abnormal level according to the change rate, call the mutation factor of the sudden anomaly to calculate the instantaneous signal change amplitude, screen the abnormal feature points that meet the set range, and obtain the abnormal pattern matching results;

[0051] S5: Based on the abnormal pattern matching result, the difference between the real-time frequency deviation rate and the normal trajectory is calculated, the difference is called to adjust the abnormal detection threshold, the risk feature point ratio is calculated according to the abnormal feature point distribution, the rotation state abnormality assessment result is obtained, and the signal acquisition weight and data sampling frequency are adjusted;

[0052] The rotational strain matching results include transient strain gradient threshold, strain response interval, strain deviation range, and strain mutation. The rotational frequency offset identification results include frequency response ratio, frequency offset rate threshold, and feature point deviation amplitude. The abnormal feature clustering results include trajectory deviation, local density distribution characteristics, and continuous offset duration threshold. The abnormal pattern matching results include continuous offset anomaly level, sudden anomaly amplitude, and abnormal feature point credibility.

[0053] The specific steps for obtaining the rotation strain matching results are as follows:

[0054] S101: acquiring angular acceleration, rotational inertia and strain data of the high-speed rotating transmission device, calculating the change of the rotational inertia, calculating the instantaneous increment of angular acceleration according to the change of the rotational inertia, matching the instantaneous increment of angular acceleration with the strain data, analyzing the degree of change of transient strain, and acquiring transient strain matching data;

[0055] The angular acceleration, rotational inertia and strain data of the high-speed rotating transmission device are obtained. The angular acceleration is obtained by using a gyroscope sensor, which is installed on the rotating component and sets a suitable sampling frequency (for example, 10kHz) to ensure that the high-frequency signal is not lost. The data is transmitted to the computing unit in real time through the data acquisition card for storage and processing. The rotational inertia can be obtained by a dynamic test method. The torque on the drive shaft is measured by a high-precision torque sensor, and the rotational inertia is calculated in combination with the angular acceleration. The strain data is generally obtained by using strain gauges. The strain gauges are pasted on key parts, such as bearing seats, gear meshing areas or stress-prone parts. The resistance change is measured by a Wheatstone bridge and converted into a strain value. The change in rotational inertia is calculated by the adjacent The inertia difference at the time point is calculated, and the sliding window smoothing is used to reduce noise interference. The instantaneous increment of angular acceleration is calculated based on the time difference method. When matching the angular acceleration increment and strain data, the data needs to be processed, including detrending, filtering and normalization, to ensure the comparability between different physical quantities. The matching algorithm can use the dynamic time warping (DTW) method, which calculates the optimal matching path between the instantaneous increment sequence of angular acceleration and the strain sequence, and calculates the similarity score based on the matching path. The matching similarity threshold is set to 0.8. The setting of this value is based on batch experimental data statistics. The specific calculation method is to take the historical matching score distribution of a certain device under different working conditions, calculate its mean and standard deviation, and set is the matching reference lower limit of the device, that is: ;

[0056] Set the matching score data set, mean , standard deviation , then the matching similarity threshold is set to 0.8, and only the angular acceleration increment and strain data pairs with a matching degree exceeding 0.8 are retained to obtain the transient strain matching data.

[0057] S102: Based on the transient strain matching data, identify the time variation trend of the rotational inertia, analyze the distribution of the rotational inertia at the differentiated time points, set a reference range of the rotational inertia variation, screen the transient strain data that meets the conditions, and obtain the dynamic strain range;

[0058] To calculate the time variation trend of rotational inertia, it is necessary to analyze the distribution of rotational inertia over time. This can be achieved by counting the inertia values ​​at different time points. Regression analysis is used to process inertia data, and polynomial fitting or moving average is set to eliminate high-frequency fluctuations in the data. Time window technology can be used in the analysis process, that is, the mean and standard deviation of inertia change are calculated within a fixed time interval (such as 100ms), and the inertia data is classified, for example, 0-1000rpm is a low-speed interval, 1000-5000rpm is a medium-speed interval, and above 5000rpm is a high-speed interval. The inertia change rate is calculated in each interval to identify abnormal values. The benchmark range of inertia change is set as:

[0059] ;

[0060] in, is the mean inertia of the equipment in the corresponding speed range, is the standard deviation of inertia in this interval;

[0061] Set a certain device in the medium speed range of 1000-5000rpm, and its average inertia , standard deviation , then the inertia change reference range is calculated as follows:

[0062] ;

[0063] Data exceeding this range will be regarded as outliers. When screening transient strain data that meets the conditions, the sliding window method is used for local statistics. The window size can be set according to the speed characteristics of the equipment. For systems with speeds exceeding 5000rpm, the sliding window size is set to 50ms, and for systems below 1000rpm, it is set to 200ms. The basis for this setting is to ensure that each window contains at least 10 data points to ensure statistical stability and screen out the dynamic strain range.

[0064] S103: calling the dynamic strain interval, analyzing the offset degree of the strain data, screening the mutation point in combination with the change trend of the rotational inertia, and comparing with the instantaneous increment of the angular acceleration to obtain the rotational strain matching result;

[0065] It is necessary to analyze the degree of deviation of strain data. The deviation of strain data is measured by the relative rate of change. The calculation formula is as follows:

[0066] ;

[0067] in, is the strain value at the current time point, is the reference strain value, taking the mean value of the previous time window;

[0068] If the mean value of the strain value measured in a certain time window is set to 200 and the strain value at the current time point is 230, the degree of deviation is calculated as follows:

[0069] ;

[0070] The offset threshold is set to 10%. The value is set based on the offset fluctuation range statistics of the device under stable working conditions. The calculation method is:

[0071] ;

[0072] in, is the number of data points under normal operating conditions;

[0073] Set in 10000 normal working condition data points, calculate , then the threshold is:

[0074] ;

[0075] Combined with the change trend of rotational inertia, the mutation point can be selected. The judgment index of the mutation point can be set as the inertia change rate:

[0076] ;

[0077] If the inertia change rate at a certain time point exceeds the set threshold , then the point can be judged as a mutation point. The calculation of this value is based on the 95% confidence interval of the inertia change trend of the transmission device, combined with the instantaneous increment of angular acceleration for comparison. The angular acceleration increments before and after the mutation point are set to 200 and 240 respectively, and the incremental change rate is calculated as follows:

[0078] ;

[0079] The instantaneous incremental change threshold is set to 15%. The calculation method of this value is based on the distribution of angular acceleration change rate under normal working conditions and takes As the threshold, set , ,but for:

[0080] ;

[0081] If the calculated result exceeds 15%, it is determined that significant deformation occurs at that time point, and the rotational strain matching result is obtained.

[0082] The steps for obtaining the rotation frequency offset identification result are as follows:

[0083] S201: calling the rotation strain matching result, recording the real-time frequency response of the high-speed rotation transmission device, comparing the real-time working frequency with the normal frequency, and obtaining the frequency deviation identification result;

[0084] Record the real-time frequency response of high-speed rotating transmission devices. The real-time operating frequency is obtained by using a high-precision speed sensor, such as a photoelectric encoder or a laser tachometer. The sensor is generally installed at the shaft end or the gear meshing position to ensure measurement accuracy. The data acquisition system records the frequency changes at a sampling rate of 1kHz or higher. When comparing the real-time operating frequency with the normal frequency, it is necessary to set a normal operating frequency reference value. The setting of this value is based on the design rated speed of the equipment and empirical statistical data. The specific method is to collect the frequency data of the equipment under normal operating conditions and take the average value As a reference value, set For the normal working range, the rated speed of a certain device is set to 3000rpm, and the historical operation data statistics are obtained rpm, rpm, the normal frequency range is set to rpm. If the real-time operating frequency exceeds this range, it is considered that there is a frequency deviation. The frequency deviation is calculated using the relative deviation formula:

[0085] ;

[0086] in, is the real-time operating frequency, is the normal working frequency;

[0087] If the real-time measured speed is set to 3050rpm, the deviation is calculated as follows:

[0088] ;

[0089] If the frequency deviation threshold is set to 2%, the data point is marked as abnormal and the frequency deviation identification result is obtained.

[0090] S202: According to the frequency deviation identification result, continuously monitor the comparison value data, calculate the frequency change amplitude according to the time series, obtain the frequency offset rate sequence, call the frequency offset rate sequence, identify the dynamic range according to the frequency change trend, match and analyze the dynamic range data with the frequency offset rate, evaluate the corresponding relationship between the frequency change range and time, select the data points that meet the change range, and obtain the frequency offset dynamic range;

[0091] The comparison value data is continuously monitored, and the frequency change amplitude is calculated by time series analysis method. The calculation method adopts sliding window method, and the size of each window is set to 500ms. The difference between the maximum and minimum frequencies in the window is calculated as the change amplitude, and the frequency offset rate sequence is obtained. The frequency offset rate is calculated by normalization method, and the upper and lower limits of the normal frequency range are used as the benchmark to calculate the frequency change rate at the current moment. The frequency offset rate sequence is called, and the dynamic range is identified according to the frequency change trend. The dynamic range is set based on the 95% confidence interval of the offset rate. The calculation formula is:

[0092] ;

[0093] in, and are the mean and standard deviation of the frequency shift rate, respectively;

[0094] Set the frequency deviation rate statistics of a certain device to obtain , , the dynamic range is calculated as follows:

[0095] ;

[0096] The dynamic range data is matched and analyzed with the frequency offset rate. The sliding mean method is used to calculate the changing trend of the frequency offset rate in the time series. The corresponding relationship between the frequency change range and time is evaluated. The change range threshold is set to 2.5%. This value is set based on the maximum fluctuation rate statistics of the equipment under different load conditions. The data points that meet the change range are screened to obtain the dynamic range of the frequency offset.

[0097] S203: using the frequency offset dynamic range, referring to the vibration acceleration, impact signal and transient stress, analyzing the deviation of the abnormal feature points, screening the frequency offset data points, calculating the frequency offset feature values, and comparing the corresponding transient stress fluctuations to obtain the rotation frequency offset identification result;

[0098] The formula for calculating the frequency offset characteristic value is as follows:

[0099] ;

[0100] in, represents the frequency shift characteristic value, Representative The frequency value of the data point, represents the average frequency value of the data points, Representative The vibration acceleration amplitude corresponding to the data point is represents the total number of data points;

[0101] Parameter meaning and formula calculation derivation process:

[0102] Frequency value The vibration signal of the rotating machinery is measured and acquired by a high-precision spectrum analyzer, which performs Fourier transform on the vibration signal of the rotating machinery, extracts the main frequency components and detects the offset components therein;

[0103] Frequency Mean Calculated from all collected frequency data points, the calculation method is:

[0104] ;

[0105] Vibration acceleration amplitude Measured by the acceleration sensor, installed on the rotating part, real-time monitoring of vibration signals, the unit is ; Total number of data points Depends on the acquisition time and sampling rate, use a high sampling rate of more than 10kHz to ensure measurement accuracy;

[0106] The calculation steps are as follows:

[0107] The measured frequency offset data points and the corresponding vibration acceleration amplitudes are assumed to have collected 5 sets of data ( ), the data are as follows:

[0108] Hz, ;

[0109] Hz, ;

[0110] Hz, ;

[0111] Hz, ;

[0112] Hz, ;

[0113] Compute the frequency mean:

[0114] ;

[0115] Calculate the absolute value of the offset for each data point and multiply it by the vibration acceleration amplitude:

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] Sum operation:

[0122] ;

[0123] ;

[0124] ;

[0125] The results show that in the current measurement data, the frequency offset characteristic value based on the weighted vibration acceleration amplitude is 1.08 Hz. This value reflects the overall characteristics of the rotational frequency offset and can be used to analyze the frequency fluctuation under abnormal working conditions and further analyzed in combination with transient stress data.

[0126] The specific steps for obtaining abnormal feature clustering results are as follows:

[0127] S301: using the rotation frequency offset identification result, combined with the vibration acceleration, transient stress and impact signal, identifying the distribution of abnormal feature points, comparing the abnormal feature points with the normal frequency trajectory, identifying the degree of deviation between the two, and obtaining frequency trajectory deviation data;

[0128] Combined with vibration acceleration, transient stress and impact signals, analyze the abnormal feature points during the operation of the equipment, collect the rotation frequency, vibration acceleration, transient stress and impact signal data under normal operation of the equipment, establish a normal operation benchmark data set, and monitor the rotation frequency in real time when the equipment is running, and compare it with the benchmark data. If the offset exceeds the set threshold, it indicates that the equipment is abnormal. Record the vibration signal during the operation of the equipment, filter the vibration acceleration data and extract features, calculate the vibration amplitude, average value and peak value. When the average value or peak value of the vibration acceleration is significantly higher than the normal level, there is an abnormality. In transient stress analysis, Measure the stress of the equipment through strain gauges or pressure sensors, record the change curve of transient stress, and determine whether the stress value exceeds the allowable range of the material. If a sudden change or excessive stress occurs, the equipment faces structural problems. Obtain the impact signal through an acceleration sensor or impact detector, calculate the impact amplitude and impact frequency, and compare them with the normal impact mode. If the impact amplitude increases abnormally or the impact interval shortens significantly, it indicates that the equipment is subjected to abnormal external force. Mark all detected abnormal feature points, compare the feature points with the frequency trajectory under normal operating conditions, calculate the degree of deviation, and obtain the frequency trajectory deviation data if the deviation exceeds the set safety threshold.

[0129] S302: calling the frequency trajectory deviation data, identifying the local density distribution according to the degree of deviation, calculating the number of abnormal feature points, identifying the density change in the local area, comparing the density fluctuation amplitude under the time series, and obtaining the local density distribution characteristics;

[0130] The formula for calculating the number of abnormal feature points is as follows:

[0131] ;

[0132] in, Represents the number of abnormal feature points, Representative The deviation of feature points, Represents the average value of the deviation of feature points, Represents the standard deviation of the feature point deviation, Represents the total number of feature points in the differentiated area, Representative The feature point and The associated weight of the reference point, Representative The feature point and The influence factor of the reference point, represents the total number of reference points;

[0133] Parameter acquisition and calculation method:

[0134] Representative The deviation of a feature point is obtained by calculating the frequency trajectory deviation of the feature point in the current time window. The calculation method is:

[0135] ;

[0136] in, Representative The current frequency value of each feature point is collected by high-precision frequency monitoring equipment. The reference frequency value representing the feature point is obtained by long-term data statistics;

[0137] Set a certain area to collect feature point data:

[0138] ;

[0139] Reference frequency value:

[0140] ;

[0141] but:

[0142] ;

[0143] ;

[0144] ;

[0145] Represents the average value of the deviation of feature points:

[0146] ;

[0147] in, (Number of feature points);

[0148] ;

[0149] Represents the standard deviation of the feature point deviation:

[0150] ;

[0151] calculate:

[0152] ;

[0153] Representative The feature point and The association weight of each reference point is obtained by calculating the correlation coefficient between the frequency change of the feature point and the reference point data, and the data analysis is set to calculate:

[0154] ;

[0155] Representative The feature point and The influence factor of each reference point is calculated by the distance attenuation function from the reference point to the feature point:

[0156] ;

[0157] in, Represents the spatial distance from the feature point to the reference point, measured by the GIS system, set:

[0158] ;

[0159] but:

[0160] ;

[0161] ;

[0162] ;

[0163] Calculate the local density adjustment factor:

[0164] ;

[0165] Calculate each feature point:

[0166] ;

[0167] ;

[0168] ;

[0169] calculate :

[0170] ;

[0171] ;

[0172] The results show that the number of abnormal feature points is 2.138, which reflects the degree of density deviation of feature points within the statistical differentiation area and can be used for further density change analysis.

[0173] S303: Analyze the temporal variation trend of local density by using local density distribution characteristics, screen out continuous deviation anomalies and sudden anomalies, classify abnormal feature points according to the amplitude and stability of density variation, and classify anomaly categories to obtain abnormal feature clustering results;

[0174] Abnormal feature points are classified, and the judgment criteria for continuous offset anomalies and sudden anomalies are set. Under normal circumstances, the density of abnormal points in each grid unit of a local area of ​​a certain device is stable at around 5%. When the density exceeds 10% and does not decrease for multiple time windows, it is defined as a continuous offset anomaly. If the density surges in a short period of time, such as a sudden increase from 5% to 20%, it is defined as a sudden anomaly. In industrial equipment fault monitoring, if the density of abnormal points remains at 12% for five consecutive time windows during the operation of a certain device, it means that the abnormality in this area is a stable offset type. If the density of abnormal points suddenly increases from 6% to 25% in a certain time window, it indicates that a sudden failure occurs in this area. The stability of density fluctuation is calculated, and the fluctuation threshold is set, such as calculating the density mean and standard deviation. If the standard deviation is small, it indicates that the density is stable, and if the standard deviation is large, it indicates that the density changes violently. Based on the abnormal category collection criteria, the abnormal feature points are classified. If the density offset of a certain area exceeds 5%, it is classified as category 1, and if the density growth rate exceeds 10%, it is classified as category 2, forming a clustering result of abnormal features.

[0175] The specific steps for obtaining the abnormal pattern matching results are as follows:

[0176] S401: based on the abnormal feature clustering results, extract the data sequence of the continuous shift anomaly, calculate the fitting residuals of the differentiated time points, compare the residual differences of adjacent time points, and obtain the continuous shift residual change rate;

[0177] Extract the data series of continuous deviation anomalies, filter the classified anomaly data, arrange them in chronological order, form time series data, ensure the consistency of time intervals, and avoid calculation errors caused by missing data. In this process, if missing data points appear, use linear interpolation or historical mean filling to ensure data integrity. Select a suitable fitting method to perform trend fitting on the time series, such as using the least squares method for linear fitting or polynomial fitting to improve the trend matching accuracy. After obtaining the fitting function, calculate the fitting residual at each time point. The fitting residual is defined as the difference between the actual measured value and the fitting value. Record the residual data at each time point, and calculate the residual difference of adjacent time points to determine the trend of residual change. When the residual difference is small, it indicates that the abnormal change is relatively stable, and when the residual difference is large, It indicates that the abnormal change is more drastic. The threshold of the change rate of the residual difference is set. The threshold is set according to the residual change range during the normal operation period. The specific method is to select the residual change rate during the stable operation period of the equipment, calculate the mean and standard deviation, and set the threshold to the mean plus twice the standard deviation to ensure the sensitivity of anomaly detection and avoid false alarms. It is set that the residual change rate of a certain equipment is 0.05 and the standard deviation is 0.02 in the stable operation stage, then the threshold is set to 0.05+2×0.02=0.09. When the residual change rate exceeds 0.09, it is determined that the equipment has a continuous offset abnormality. If the equipment type or operating conditions are different, the standard deviation multiple can be adjusted according to the actual measurement data. By comparing the change rates in multiple time windows and analyzing the continuous changes of the abnormal points, the continuous offset residual change rate is obtained.

[0178] S402: calling the continuous offset residual change rate, analyzing the abnormality level according to the change rate distribution, classifying the change rate according to the numerical range, setting the level standard, marking the corresponding level of the abnormal point, and obtaining the abnormality level classification list;

[0179] According to the change rate distribution analysis, the abnormal level is analyzed, and the calculated change rate data is statistically analyzed to obtain the maximum, minimum, mean and standard deviation of the change rate. The numerical range of the change rate is set to divide it into multiple levels, such as setting low, medium and high levels. The division standard can be based on the operation requirements and safety range of the equipment. The specific method is to calculate the distribution of the change rate in the abnormal data set, set the low level range to be less than the change rate mean minus one standard deviation, the medium level range to be the mean minus one standard deviation to the mean plus one standard deviation, and the high level range to be greater than the mean plus one standard deviation. The change rate data distribution of a mechanical equipment is the mean. 0.06, standard deviation 0.015, then the low level range is change rate <0.045, the medium level range is 0.045≤change rate≤0.075, and the high level range is change rate>0.075. After the classification is completed, the abnormal points are marked, and the number of abnormal points of each level is counted to form an abnormal level classification list. If the equipment operating conditions are different, such as the vibration signal change rate of fans and pumps is low, the standard deviation multiple needs to be adjusted to between 1.5 and 2 to ensure the accuracy of the abnormal level classification. The distribution of abnormal points of different levels can be intuitively displayed to provide basic data support for further abnormal analysis and obtain the abnormal level classification list.

[0180] S403: using the abnormal level classification list, for sudden abnormalities, calculating the instantaneous signal change amplitude of the mutation factor, comparing the change amplitude with the set range, screening abnormal feature points that meet the range, and obtaining abnormal pattern matching results;

[0181] For sudden anomalies, calculate the instantaneous signal change amplitude of the mutation factor, select the data of the sudden abnormal point, obtain the instantaneous signal, such as vibration signal, current signal or temperature signal, calculate the signal change amplitude, and define the change amplitude as the difference between the signal value at the current moment and the signal value at the previous moment. The vibration amplitude at the moment is mm / s, The vibration amplitude at the moment is mm / s, the change range is calculated as:

[0182] ;

[0183] The instantaneous signal change amplitude is calculated for all time points, and the change amplitude is compared with the set range. The setting of this range is based on the normal fluctuation range of the equipment. The specific calculation method is to obtain the signal change amplitude data during the normal operation of the equipment, calculate the mean and standard deviation, and set the mutation judgment threshold to the mean plus 2-3 times the standard deviation. When a certain mechanical equipment is operating normally, the mean value of its vibration signal change amplitude is 1.2mm / s, and the standard deviation is 0.5mm / s. The threshold is set to 1.2+2×0.5=2.2mm / s. When the change amplitude exceeds 2.2mm / s, it is judged as a sudden abnormality. If the operating condition of a certain equipment fluctuates greatly, the threshold range can be adjusted to 2.5-3 times the standard deviation to reduce the false alarm rate. If the above calculation result 3.3mm / s is greater than the threshold 2.2mm / s, then the point is judged as a sudden abnormality. All abnormal feature points that meet the range are screened to obtain the abnormal pattern matching result.

[0184] The specific steps for obtaining the rotation state abnormality assessment results are as follows:

[0185] S501: calling the abnormal pattern matching result, calculating the real-time frequency offset rate, extracting the frequency data of the normal trajectory, comparing the values ​​of the two, calculating the difference between the real-time frequency offset rate and the normal trajectory, and obtaining the frequency offset difference data;

[0186] Calculate the real-time frequency offset rate, obtain the current operating data of the equipment, extract the real-time frequency data, and calculate the degree of offset relative to the reference frequency. In this process, ensure that the time interval of data collection is consistent to avoid data imbalance affecting the calculation results. At the same time, extract the frequency data of the normal operation trajectory of the equipment, establish a reference frequency curve, compare the real-time frequency offset data with the reference trajectory, and analyze the offset trend at different time points. If the frequency offset is found to be continuous or sudden, further mark the abnormal moment and calculate the change of the real-time offset rate. In the comparison process, use a sliding time window to smooth the data to reduce the impact of sudden interference on the calculation results and ensure the continuity of the data trend. After obtaining the offset rate, calculate its numerical difference with the normal trajectory to ensure the rationality of the offset rate and form the frequency offset difference data.

[0187] S502: using the frequency offset difference data, adjusting the anomaly detection threshold according to the value range, comparing the frequency offsets at the differentiated time points, calculating the proportion of abnormal feature points, identifying the distribution of abnormal feature points, obtaining the rotation state abnormality assessment result, and adjusting the signal acquisition weight;

[0188] The anomaly detection threshold is adjusted according to the numerical range, and the frequency offset difference data is statistically analyzed to extract the maximum value, minimum value and distribution range. The strategy for adjusting the threshold is set according to different working conditions. If the offset difference is within a fixed range, the anomaly detection threshold can be appropriately increased or decreased to avoid misjudgment or missed detection. The frequency offset at different time points is compared, the distribution law of anomalies in the time series is analyzed, and the proportion of abnormal feature points in all data points is calculated to determine the severity of the abnormal phenomenon. The distribution characteristics of abnormal feature points in the equipment operation cycle are identified. If the anomalies are concentrated in a specific time period, they are related to certain operating states. Through further data screening and adjustment, the rotation state anomaly evaluation results are formed, and the acquisition weight is adjusted for the signal acquisition process to improve the data acquisition accuracy of the key frequency band and optimize the accuracy of subsequent anomaly detection.

[0189] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for automatically detecting abnormalities in a high-speed rotating transmission device, characterized in that: The following steps are involved: S1: Obtain the angular acceleration, rotational inertia and strain data of the high-speed rotating transmission device, calculate the transient strain gradient and match the speed change rate, adjust the strain deviation range according to the change of rotational inertia, screen the strain mutation point, and obtain the rotation strain matching result; S2: Using the rotation strain matching result, record the real-time frequency response of the high-speed rotation transmission device, perform continuous monitoring to obtain the frequency deviation rate sequence, refer to the deviation of the abnormal feature points identified by the vibration acceleration, impact signal and transient stress, and obtain the rotation frequency deviation identification result; S3: using the rotation frequency offset identification result, combined with vibration acceleration, transient stress and impact signal, analyzing the deviation degree between abnormal feature points and normal frequency trajectory, identifying local density distribution, screening continuous offset anomalies and sudden anomalies, and obtaining abnormal feature clustering results; S4: calling the abnormal feature clustering result, calculating the fitting residual change rate of the continuous offset anomaly, setting the abnormal level according to the change rate, screening the abnormal feature points that meet the set range, obtaining the abnormal pattern matching result, automatically monitoring the abnormal signal of the high-speed rotating transmission device, and issuing an early warning.

2. The automatic abnormality detection method for a high-speed rotating transmission device according to claim 1, characterized in that: The rotation strain matching result includes transient strain gradient threshold, strain response interval, strain deviation range, and strain mutation. The rotation frequency offset identification result includes frequency response ratio, frequency offset rate threshold, and feature point deviation amplitude. The abnormal feature clustering result includes trajectory deviation, local density distribution characteristics, and continuous offset duration threshold.

3. The automatic abnormality detection method for a high-speed rotating transmission device according to claim 1, characterized in that: The steps for obtaining the rotation strain matching result are specifically as follows: S101: acquiring angular acceleration, rotational inertia and strain data of the high-speed rotating transmission device, calculating the change of the rotational inertia, calculating the instantaneous increment of angular acceleration according to the change of the rotational inertia, matching the instantaneous increment of angular acceleration with the strain data, analyzing the degree of change of transient strain, and acquiring transient strain matching data; S102: Based on the transient strain matching data, identifying the time variation trend of the rotational inertia, analyzing the distribution of the rotational inertia at the differentiated time points, setting a reference range of the rotational inertia variation, screening the transient strain data that meets the conditions, and obtaining the dynamic strain interval; S103: calling the dynamic strain interval, analyzing the offset degree of the strain data, screening the mutation point in combination with the change trend of the rotational inertia, and performing comparison in combination with the instantaneous increment of the angular acceleration to obtain the rotational strain matching result.

4. The automatic abnormality detection method of a high-speed rotating transmission device according to claim 1, characterized in that: The steps for obtaining the rotation frequency offset identification result are specifically as follows: S201: calling the rotation strain matching result, recording the real-time frequency response of the high-speed rotation transmission device, comparing the real-time operating frequency with the normal frequency, and obtaining a frequency deviation identification result; S202: According to the frequency deviation identification result, continuously monitor the comparison value data, calculate the frequency change amplitude according to the time series, obtain the frequency offset rate sequence, call the frequency offset rate sequence, identify the dynamic range according to the frequency change trend, match and analyze the dynamic range data with the frequency offset rate, evaluate the corresponding relationship between the frequency change range and time, select the data points that meet the change range, and obtain the frequency offset dynamic range; S203: Utilizing the frequency offset dynamic range, referring to vibration acceleration, impact signal and transient stress, analyzing the deviation of abnormal feature points, screening frequency offset data points, calculating frequency offset feature values, and comparing the corresponding transient stress fluctuations to obtain a rotation frequency offset identification result.

5. The automatic abnormality detection method of a high-speed rotating transmission device according to claim 4, characterized in that: The formula for calculating the frequency offset characteristic value is as follows: ； in, represents the frequency shift characteristic value, Representative The frequency value of the data point, represents the average frequency value of the data points, Representative The vibration acceleration amplitude corresponding to the data point is Represents the total number of data points.

6. The automatic abnormality detection method of a high-speed rotating transmission device according to claim 1, characterized in that: The steps for obtaining the abnormal feature clustering result are specifically as follows: S301: using the rotation frequency offset identification result, combined with vibration acceleration, transient stress and impact signal, identifying the distribution of abnormal feature points, comparing the abnormal feature points with the normal frequency trajectory, identifying the degree of deviation between the two, and obtaining frequency trajectory deviation data; S302: calling the frequency trajectory deviation data, identifying the local density distribution according to the degree of deviation, calculating the number of abnormal feature points, identifying the density change in the local area, comparing the density fluctuation amplitude under the time series, and obtaining the local density distribution characteristics; S303: Analyze the temporal variation trend of local density by using the local density distribution characteristics, screen out continuous deviation anomalies and sudden anomalies, classify abnormal feature points according to the amplitude and stability of density variation, and classify anomaly categories to obtain abnormal feature clustering results.

7. The automatic abnormality detection method of a high-speed rotating transmission device according to claim 6, characterized in that: The formula for calculating the number of abnormal feature points is as follows: ； in, Represents the number of abnormal feature points, Representative The deviation of feature points, Represents the average value of the deviation of feature points, Represents the standard deviation of the feature point deviation, Represents the total number of feature points in the differentiated area, Representative The feature point and The associated weight of the reference point, Representative The feature point and The influence factor of the reference point, Represents the total number of reference points.

8. The automatic abnormality detection method for a high-speed rotating transmission device according to claim 1, characterized in that: The steps for obtaining the abnormal pattern matching result are specifically as follows: S401: extracting the data sequence of the continuous shift anomaly based on the abnormal feature clustering result, calculating the fitting residuals of the differentiated time points, comparing the residual difference values ​​of adjacent time points, and obtaining the continuous shift residual change rate; S402: calling the continuous offset residual change rate, analyzing the abnormality level according to the change rate distribution, classifying the change rate according to the numerical range, setting the level standard, marking the corresponding level of the abnormal point, and obtaining the abnormality level classification list; S403: using the abnormal level classification list, for sudden abnormalities, calculating the instantaneous signal change amplitude of the mutation factor, comparing the change amplitude with the set range, screening abnormal feature points that meet the range, and obtaining abnormal pattern matching results.

9. The automatic abnormality detection method of a high-speed rotating transmission device according to claim 1, characterized in that: The method further comprises step S5: S5: Based on the abnormal pattern matching result, the difference between the real-time frequency deviation rate and the normal trajectory is calculated, the difference is called to adjust the abnormal detection threshold, the risk feature point ratio is calculated according to the abnormal feature point distribution, the rotation state abnormality assessment result is obtained, and the signal acquisition weight and data sampling frequency are adjusted; The rotation state abnormality assessment result includes an abnormality detection threshold, a risk feature point ratio, and a frequency offset risk index.

10. The automatic abnormality detection method of a high-speed rotating transmission device according to claim 9, characterized in that: The steps for obtaining the rotation state abnormality evaluation result are specifically as follows: S501: calling the abnormal pattern matching result, calculating the real-time frequency offset rate, extracting the frequency data of the normal trajectory, comparing the values ​​of the two, calculating the difference between the real-time frequency offset rate and the normal trajectory, and obtaining the frequency offset difference data; S502: Using the frequency offset difference data, adjusting the anomaly detection threshold according to the numerical range, comparing the frequency offsets at differentiated time points, calculating the proportion of abnormal feature points, identifying the distribution of abnormal feature points, obtaining the rotation state anomaly assessment result, and adjusting the signal acquisition weight.

Images