Fault diagnosis system and diagnosis method for large variable speed units

By using techniques such as order resampling and nearest neighbor interpolation in large variable speed equipment, data stability problems in variable speed scenarios are solved, efficient fault diagnosis is achieved, and diagnostic accuracy and efficiency are improved.

CN118520402BActive Publication Date: 2024-11-08LEVI INTELLIGENT (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410971162.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-08
Estimated Expiration
2044-07-19

AI Technical Summary

Technical Problem

In the variable speed scenario, the stability of the collected data is damaged, resulting in low diagnostic accuracy of large variable speed equipment.

Method used

Using data acquisition unit, data analysis unit and data alarm unit, the error data exclusion, steady-state data screening and signal steady-stateization are carried out through order resampling method, nearest neighbor interpolation method, isolated forest algorithm and sliding window inspection method, to achieve The non-steady state data is converted into steady state data and fault diagnosis is performed based on the alarm triggering rules.

Benefits of technology

It improves the diagnostic accuracy and efficiency of large-scale variable speed equipment, enhances the depth and efficiency of vibration signal analysis, processes data flow in real time, and promptly reflects the latest status.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118520402B_ABST
    Figure CN118520402B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of analysis and diagnosis of non-steady-state data, and specifically, to a large variable speed unit fault diagnosis system and a diagnosis method thereof. It includes a data acquisition unit for collecting vibration signals and speed data during the operation of large variable speed equipment; a data analysis unit analyzes the non-steady-state vibration signals and speed data based on an order resampling method, and converts the non-steady-state data into steady-state data; a data alarm unit alarms steady-state data exceeding a threshold based on an alarm trigger rule; and a data acquisition unit eliminates erroneous data based on an isolation forest algorithm. The design principle of the algorithm of the present invention is intuitive, applicable to various data models, and screens steady-state data based on a sliding window test method, capable of processing data streams in real time, continuously updating the data in the window as the data arrives, and timely reflecting the latest status; the data analysis unit analyzes the data based on the order resampling method, which greatly enhances the depth and efficiency of vibration signal analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of analysis and diagnosis of non-steady-state data, and in particular to a fault diagnosis system and a diagnosis method for a large variable speed machine set. Background Art

[0002] For the monitoring scenario of stable speed, since the speed of the motor equipment is fixed, the collected data is mainly steady-state. At this time, the system will be configured with analysis and diagnosis methods related to steady-state data; but in the scenario of variable speed, since the stability of the collected data is destroyed, the system's steady-state analysis method loses its meaning; especially for the scenario of large-scale variable speed, it is necessary to develop a set of non-steady-state data analysis and diagnosis system, which can effectively improve the accuracy and efficiency of diagnosis in large-scale variable speed equipment. Therefore, a large-scale variable speed unit fault diagnosis system and its diagnosis method are provided. Summary of the invention

[0003] The purpose of the present invention is to provide a large variable speed unit fault diagnosis system and a diagnosis method thereof, so as to solve the problem of low diagnostic accuracy of large variable speed equipment in variable speed scenarios due to the destruction of the stability of collected data raised in the above background technology.

[0004] To achieve the above object, on the one hand, the present invention aims to provide a large variable speed unit fault diagnosis system, comprising:

[0005] A data acquisition unit, the data acquisition unit is used to collect vibration signals and speed data of large variable speed equipment during operation;

[0006] A data analysis unit, wherein the data analysis unit adopts an order resampling method to analyze the non-steady-state vibration signal and the speed data through a nearest neighbor interpolation method, and introduces neighbor point weights to optimize the interpolation process of the nearest neighbor interpolation method on the signal, so as to improve the accuracy of converting the non-steady-state data into steady-state data;

[0007] Among them, the optimized nearest neighbor interpolation method is:

[0008]

[0009] in, represents the value after introducing neighbor weights, represents the neighboring point, is the value of the neighboring point, Representative The weight of the neighboring points, Represents the normalized weight;

[0010] A data alarm unit is used to generate an alarm for steady-state data exceeding a threshold based on an alarm triggering rule.

[0011] As a further improvement of the technical solution, the data acquisition unit includes an erroneous data elimination module and a steady-state data screening module;

[0012] Among them, the error data elimination module pre-processes the data, and eliminates the error data based on the isolation forest algorithm to obtain the vibration signal and speed data without noise, smooth speed, and synchronous alignment;

[0013] The steady-state data screening module performs steady-state data screening on the preprocessed vibration signal and speed data through a sliding window test method to obtain non-steady-state vibration signal and speed data.

[0014] As a further improvement of the technical solution, the erroneous data elimination module is used to identify abnormal data in the data based on the isolation forest algorithm and eliminate it, including the following steps:

[0015] S3.1. Collect the data set to be tested, where the data set includes normal and abnormal vibration signals and speed data;

[0016] S3.2, select the speed feature, and select the middle value between the maximum and minimum values ​​of the feature as the split point, divide the data set into two left and right subsets according to the split point, and use the two subsets as the input of the left and right subtrees respectively to form an isolated tree; construct the isolated tree to form an isolated forest;

[0017] S3.3. For each data point in the data set, calculate its path length H in each isolated tree and take the average value as the anomaly score E of the data point;

[0018] S3.4. Set a threshold K. Data points with anomaly scores exceeding the threshold are considered outliers and excluded.

[0019] As a further improvement of the technical solution, the steady-state data screening module processes the time series data based on the sliding window inspection method to identify the steady-state data, including the following steps:

[0020] S4.1. Set the sliding window size S;

[0021] S4.2, setting an initial window at the beginning of the time series containing the vibration signal and the speed data, and the number of data points in the window is equal to the window size;

[0022] S4.3, using polynomial regression filtering technology to smooth the window data;

[0023] S4.4, calculating the coefficient of the first-order term by the least square method according to the filtered data, and introducing the regression parameter to optimize the calculation of the least square method;

[0024] S4.5, move the window forward along the time series by one step, the new window excludes the oldest data point, and includes the next new data point;

[0025] S4.6, repeat the data smoothing and linear term coefficient calculation operations for each new window, and then compare the calculated linear term coefficient with the result of the previous window. If the index change does not exceed the threshold value B, the data in the window is considered to be in a steady state;

[0026] S4.7. Mark the data as steady-state segments for screening to obtain steady-state data.

[0027] As a further improvement of the technical solution, in S4.3, the window data is smoothed by using a polynomial regression filtering technique, which includes the following steps:

[0028] S4.31. Determine the order of the polynomial regression to be second order;

[0029] S4.32, selecting a sliding window and determining a set of data points contained in the window;

[0030] S4.33. For each time point in the window , construct a design matrix , where each row corresponds to a data point and the columns represent the polynomial terms corresponding to the data point time;

[0031] S4.34. Use the least squares method to solve the coefficients of the polynomial regression and minimize the residual sum of squares, that is, find the coefficient vector Make Minimum, among which is the number of data points in the window, Is the first The intensity value of the vibration signal in the group observation, Indicates The row vector associated with each data point, Including constant terms, speed values ​​and their higher powers;

[0032] S4.35, use the obtained coefficient and the design matrix Predict the smoothed value at each time point, for each point in the window ,calculate ,in It is the smoothed value of the point after polynomial regression filtering.

[0033] In S4.4, the least squares method is optimized as follows:

[0034]

[0035] in, is the coefficient of the first-order term, b is the intercept, represents the parameter vector, parameter , is the regression parameter, Represents the parameter vector The dimension of is the total number of parameters in the model, Represents the number of observation samples in the data set, that is, the number of data points in the window, Represents the speed values ​​measured at different time points in the data set represents the vibration intensity measurement at a given time point in the data set, Represents the parameter vector The elements.

[0036] As a further improvement of the technical solution, the data analysis unit converts the non-steady-state data into steady-state data based on the order resampling method, comprising the following steps:

[0037] S7.1. Obtain accurate speed information through encoder equipment and perform constant angle incremental sampling on the data;

[0038] S7.2. First, perform fast Fourier transform on the steady-state signal in the angle domain to obtain a frequency domain signal; divide the spectrum by order to distinguish the frequency shift signal caused by the speed change, and each order corresponds to a stable frequency component;

[0039] S7.3, taking the vibration as a reference order, interpolating the signal by the nearest neighbor interpolation method based on the reference order, and introducing the weight of the neighboring points to optimize the interpolation process of the signal by the nearest neighbor interpolation method, so that it is evenly distributed in each order;

[0040] S7.4, synchronously average the signals of the same order, eliminate random noise, and smooth the speed information, especially for non-steady-state signals with periodic changes;

[0041] S7.5. Obtain steady-state vibration signals and speed data, and use a minimum-maximum normalization algorithm to quantify the steady-state vibration signals and speed data.

[0042] As a further improvement of the technical solution, in S7.5, the minimum-maximum normalization algorithm is specifically:

[0043]

[0044] in, represents the vibration signal and speed data in the data set, Represents the value obtained after the minimum-maximum normalization algorithm, Represents the maximum value in the data set. Represents the minimum value in the data set.

[0045] As a further improvement of the technical solution, the data alarm unit determines whether to alarm by setting a threshold based on the alarm triggering rule, including the following steps:

[0046] S9.1. preprocessing the quantified steady-state data collected by the data analysis unit;

[0047] S9.2, set the alarm threshold P;

[0048] S9.3. When it is detected that the quantified steady-state data exceeds the preset threshold value P, an alarm is triggered and an alarm message is sent.

[0049] On the other hand, the present invention provides a large variable speed unit fault diagnosis method, which is used in the above large variable speed unit fault diagnosis system, and includes the following steps:

[0050] S10.1. Eliminate erroneous data based on the isolation forest algorithm and screen steady-state data based on the sliding window test method;

[0051] S10.2, analyzing the data based on the order resampling method to convert the non-steady-state data into steady-state data;

[0052] S10.3. Based on the alarm triggering rules, an alarm is issued for steady-state data exceeding the threshold.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. In the fault diagnosis system and its diagnosis method for large variable speed units, the data acquisition unit eliminates erroneous data based on the isolation forest algorithm. The algorithm principle is intuitive. It measures the "isolation" of data points by building random decision trees to determine whether they are abnormal. It is applicable to various data models and performs steady-state data screening based on the sliding window test method. It can process data streams in real time, continuously update the data in the window as the data arrives, and reflect the latest status in a timely manner.

[0055] 2. In the fault diagnosis system and diagnostic method of large variable speed units, the data analysis unit analyzes the data based on the order resampling method and converts the non-steady-state data into steady-state data. The order resampling method has the advantages of signal stabilization, enhanced fault characteristics, convenient order spectrum analysis, and simplified data processing flow, which greatly enhances the depth and efficiency of vibration signal analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flowchart of the overall process of the present invention.

[0057] The meaning of each number in the figure is:

[0058] 1. Data acquisition unit; 2. Data analysis unit; 3. Data alarm unit. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Embodiment 1:

[0061] See also Figure 1 As shown, a large variable speed unit fault diagnosis system is provided, including:

[0062] The data acquisition unit 1 is used to collect vibration signals and speed data of large variable speed equipment during operation;

[0063] In this embodiment, the data acquisition unit 1 includes an erroneous data elimination module and a steady-state data screening module;

[0064] Among them, the error data elimination module pre-processes the data, and eliminates the error data based on the isolation forest algorithm to obtain the vibration signal and speed data without noise, smooth speed, and synchronous alignment; the vibration signal includes frequency, amplitude, and phase, and the speed data includes speed value, speed range, and speed fluctuation;

[0065] The steady-state data screening module performs steady-state data screening on the pre-processed vibration signal and speed data through a sliding window inspection method to obtain non-steady-state vibration signal and speed data;

[0066] Among them, the wrong data elimination module is based on the isolation forest algorithm to identify outliers in the data and eliminate them. The isolation forest algorithm is based on the idea of ​​random forest and identifies outliers in the data set by constructing multiple decision trees. It assumes that outliers are easier to be isolated than normal points. Therefore, the anomaly detection process is converted into a problem of finding the shortest path, including the following steps:

[0067] S3.1. Collect the data set to be tested, where the data set includes normal and abnormal vibration signals and speed data;

[0068] S3.2, select the speed feature, and select the middle value between the maximum and minimum values ​​of the feature as the split point, divide the data set into two left and right subsets according to the split point, and use the two subsets as the input of the left and right subtrees respectively to form an isolated tree; construct the isolated tree to form an isolated forest;

[0069] S3.3. For each data point in the data set, calculate its path length H in each isolated tree and take the average value as the anomaly score E of the data point;

[0070] S3.4. Set a threshold K. Data points with anomaly scores exceeding the threshold are considered outliers and excluded.

[0071] Among them, the steady-state data screening module processes time series data based on the sliding window test method to identify steady-state data. The core idea is to move a fixed-size window on the data sequence, and analyze or process the data in the window each time it moves, including the following steps:

[0072] S4.1. Set the sliding window size S. The window size should be large enough to include meaningful data features, but not too large to ignore details or too small to cause noise interference;

[0073] S4.2, setting an initial window at the beginning of the time series containing the vibration signal and the speed data, and the number of data points in the window is equal to the window size;

[0074] S4.3, using polynomial regression filtering technology to smooth the window data;

[0075] Among them, the polynomial regression of the polynomial regression filtering technology can fit nonlinear relationships, which is particularly useful for data that cannot be accurately described by linear models. By increasing the order of the polynomial, it can approximate complex signal forms and adapt to a wider range of signal types, improving the model's ability to fit data. It can remove high-frequency noise while maintaining the shape and peak position of low-frequency signals, and is suitable for signal smoothing and noise reduction processing. The specific algorithm steps are:

[0076] S4.31. Determine the order of the polynomial regression to be second order;

[0077] S4.32, selecting a sliding window and determining a set of data points contained in the window;

[0078] S4.33. For each time point in the window , construct a design matrix , where each row corresponds to a data point and the columns represent the polynomial terms corresponding to the data point time;

[0079] S4.34. Use the least squares method to solve the coefficients of the polynomial regression and minimize the residual sum of squares, that is, find the coefficient vector Make Minimum, among which is the number of data points in the window, Is the first The intensity value of the vibration signal in the group observation, Indicates The row vector associated with each data point, Including constant terms, speed values ​​and their higher powers;

[0080] S4.35, use the obtained coefficient and the design matrix Predict the smoothed value at each time point, for each point in the window ,calculate ,in It is the smoothed value of the point after polynomial regression filtering.

[0081] S4.4, calculating the coefficient of the first-order term by the least square method according to the filtered data, and introducing the regression parameter to optimize the calculation of the least square method;

[0082] Among them, the regression parameters are introduced into the least squares method to calculate the coefficients of the first-order terms. The regression parameters are introduced to limit the size of the coefficients to prevent the high-order polynomial from overfitting the noise and the low-order polynomial from being unable to capture complex trends. It can help the model ignore random fluctuations in the data and focus on capturing more stable and common patterns. The regression parameters The relationship between fitting error and model complexity is balanced, which helps to improve the stability of the model, reduce the drastic changes in coefficients caused by small changes or noise in the data, and make the model more robust to small changes in the data. After least squares optimization, it is:

[0083]

[0084] in, is the coefficient of the first-order term, b is the intercept, represents the parameter vector, parameter , is the regression parameter, Represents the parameter vector The dimension of is the total number of parameters in the model, Represents the number of observation samples in the data set, that is, the number of data points in the window, represents the speed values ​​measured at different time points in the data set, represents the vibration intensity measurement at a given time point in the data set, Represents the parameter vector The elements;

[0085] S4.5, move the window forward along the time series by one step, the new window excludes the oldest data point, and includes the next new data point;

[0086] S4.6, repeat the data smoothing and linear term coefficient calculation operations for each new window, and then compare the calculated linear term coefficient with the result of the previous window. If the index change does not exceed the threshold value B, the data in the window is considered to be in a steady state;

[0087] S4.7. Mark the data as steady-state segments for screening to obtain steady-state data.

[0088] A data analysis unit 2, wherein the data analysis unit 2 analyzes the data based on an order resampling method to convert the non-steady-state data into steady-state data;

[0089] In this embodiment, the data analysis unit 2 converts the non-steady-state data into steady-state data that is easy to analyze based on the order resampling method. The order resampling method has the advantages of signal stabilization, enhanced fault characteristics, convenient order spectrum analysis, and simplified data processing flow, which greatly enhances the depth and efficiency of vibration signal analysis, and includes the following steps:

[0090] S7.1. Obtain accurate speed information through encoder equipment and perform constant angle increment sampling on the data, so that the relative stability of the data can be maintained even when the speed changes;

[0091] S7.2. Order is a frequency unit related to the rotation speed. The frequency of the nth order signal is n times the basic frequency. First, the steady-state signal in the angle domain is fast Fourier transformed to obtain the frequency domain signal. The spectrum is divided by order to distinguish the frequency shift signal caused by the rotation speed change. Each order corresponds to a stable frequency component.

[0092] Among them, the steady-state signal in the angle domain is fast Fourier transformed to obtain the frequency domain signal. The Fourier transform converts the time domain signal into the frequency domain signal, which enables us to intuitively understand the component frequency and amplitude of the signal, and can easily identify the specific frequency components in the signal. The complex time domain convolution operation is simplified to the multiplication operation in the frequency domain, which greatly reduces the computational complexity and simplifies the understanding and processing of complex signals. The following steps are included:

[0093] S7.21, ensure data collection and speed synchronization;

[0094] S7.22, the data points are sorted according to the angle of rotation of the rotor, forming a signal representation in the angle domain;

[0095] S7.23, perform fast Fourier transform on this angle domain signal. Since the signal is sampled according to the rotor rotation angle, the result of fast Fourier transform will directly reflect the frequency components related to the rotation speed and its order;

[0096] Furthermore, the fast Fourier transform algorithm is specifically as follows:

[0097]

[0098] in, represents a complex value in the frequency domain, represents the discrete frequency in the frequency domain, represents a complex value in the time domain, represents the discrete time in the time domain, Indicates the number of time domain sampling points, Is a long The complex roots of whose principal value is 1;

[0099] S7.3. Determine the vibration as a reference order, interpolate the signal based on the reference order by the nearest neighbor interpolation method, and introduce the weight of the neighboring points to optimize the interpolation process of the signal by the nearest neighbor interpolation method, so that it is evenly distributed on each order to ensure that the data remains continuous and smooth on the new time axis;

[0100] Furthermore, introducing the weights of neighboring points into the nearest neighbor interpolation algorithm to interpolate the signal can produce smoother transitions and more natural image quality, which reduces the jagged edges and mosaic effects commonly seen in nearest neighbor interpolation, can better preserve image details, and reduce information loss; introducing the weights of neighboring points can make the nearest interpolation algorithm more accurate and detailed, providing better visual effects and signal fidelity. After optimization, it helps to more accurately separate and identify different order components and reduce pseudo-steady-state identification caused by mutations between data points. The nearest neighbor interpolation algorithm is optimized as follows:

[0101]

[0102] in, represents the value after introducing neighbor weights, is the value of the neighboring point, Representative The weight of the neighboring points, The specific algorithm is:

[0103]

[0104] Represents the normalized weight. To ensure that the sum of the weights is 1, the specific algorithm is:

[0105]

[0106] in, is the distance between each neighboring point and the interpolation position, The standard deviation controls the speed at which the weight decreases. This means a wider weight distribution and a wider impact range; Indicates the lead point; Indicates The weight of each neighbor;

[0107] S7.4, synchronously average the signals of the same order, eliminate random noise, and smooth the speed information, especially for non-steady-state signals with periodic changes;

[0108] S7.5, obtaining a steady-state vibration signal and a rotation speed data, and quantizing the steady-state vibration signal and the rotation speed data by using a minimum-maximum normalization algorithm;

[0109] Among them, the minimum-maximum normalization algorithm formula is:

[0110]

[0111] in, represents the vibration signal and speed data in the data set, Represents the value obtained after the minimum-maximum normalization algorithm, Represents the maximum value in the data set. Represents the minimum value in the data set.

[0112] A data alarm unit 3, wherein the data alarm unit 3 generates an alarm for steady-state data exceeding a threshold value based on an alarm triggering rule;

[0113] In this embodiment, the data alarm unit 3 determines whether to alarm by setting a threshold based on the alarm triggering rule, including the following steps:

[0114] S9.1. Clean the quantified steady-state data collected by the data analysis unit 2 to remove invalid, erroneous or incomplete data;

[0115] Among them, the quantified steady-state data include the mean, standard deviation and kurtosis of vibration, the average value and fluctuation range of rotation speed;

[0116] S9.2, set the alarm threshold P;

[0117] S9.3. When it is detected that the quantified steady-state data exceeds the preset threshold value P, an alarm is triggered and an alarm message is sent.

[0118] Embodiment 2:

[0119] The difference between Example 2 of the present invention and Example 1 is that this example introduces the static mechanical performance collection and analysis method used in the above-mentioned large variable speed unit fault diagnosis system.

[0120] A large variable speed unit fault diagnosis method, used in the above large variable speed unit fault diagnosis system, comprises the following steps:

[0121] S10.1. Eliminate erroneous data based on the isolation forest algorithm and screen steady-state data based on the sliding window test method;

[0122] S10.2, analyzing the data based on the order resampling method to convert the non-steady-state data into steady-state data;

[0123] S10.3. Based on the alarm triggering rules, an alarm is issued for steady-state data exceeding the threshold.

[0124] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. Fault diagnosis system for large variable speed units, It is characterized in that include: A data acquisition unit (1), the data acquisition unit (1) being used to collect vibration signals and speed data of large variable speed equipment during operation; A data analysis unit (2), wherein the data analysis unit (2) adopts an order resampling method to analyze the non-steady-state vibration signal and speed data through a nearest neighbor interpolation method, introduces neighbor point weights to optimize the interpolation process of the nearest neighbor interpolation method on the signal, and is used to improve the accuracy of converting the non-steady-state data into steady-state data; Among them, the optimized nearest neighbor interpolation method is: ; in, represents the value after introducing neighbor weights, represents the neighboring point, is the value of the neighboring point, Representative The weight of the neighboring points, Represents the normalized weight; A data alarm unit (3), wherein the data alarm unit (3) generates an alarm for steady-state data exceeding a threshold value based on an alarm triggering rule; The data acquisition unit (1) comprises an erroneous data elimination module and a steady-state data screening module; Among them, the error data elimination module is used to pre-process the data. It eliminates the error data based on the isolation forest algorithm to obtain the vibration signal and speed data without noise, smooth speed, and synchronous alignment; The steady-state data screening module performs steady-state data screening on the pre-processed vibration signal and speed data through a sliding window inspection method to obtain steady-state vibration signal and speed data; The steady-state data screening module processes the time series data based on the sliding window test method to identify the steady-state data, including the following steps: S4.

1. Set the sliding window size S; S4.2, setting an initial window at the beginning of the time series containing the vibration signal and the speed data, and the number of data points in the window is equal to the window size; S4.3, using polynomial regression filtering technology to smooth the window data; S4.

4. Calculate the coefficient of the first-order term by the least square method based on the filtered data, and introduce regression parameters to optimize the calculation of the least square method to determine the change rate of the vibration signal and the speed data; S4.5, move the window forward along the time series by one step, the new window excludes the oldest data point, and includes the next new data point; S4.6, repeat the data smoothing and linear term coefficient calculation operations for each new window, and then compare the calculated linear term coefficient with the result of the previous window. If the comparison result does not exceed the threshold value B, it is considered that the data in the window is in a steady state; S4.7, marking the data as steady-state segments for screening, and obtaining steady-state vibration signals and speed data; In S4.3, the window data is smoothed by using a polynomial regression filtering technique, which includes the following steps: S4.

31. Determine the order of the polynomial regression to be first order; S4.32, selecting a sliding window, and determining a set of data points contained in the window, the set including vibration signals and rotation speed data; S4.

33. For each time point in the window , construct a design matrix , where each row corresponds to a data point and the columns represent the polynomial terms corresponding to the data point time; S4.

34. Solve the coefficients of polynomial regression using the least squares method and find the coefficient vector Make Minimum, among which is the number of data points in the window, Is the first The intensity value of the vibration signal in the group observation, Indicates The row vector associated with each data point, Including constant terms, speed values ​​and their higher powers; S4.35, use the obtained coefficient and the design matrix Predict the smoothed value at each time point, for each point in the window ,calculate ,in is the smoothed value of the point after polynomial regression filtering; In S4.4, the least squares method is optimized as follows: ; in, is the coefficient of the first-order term, is the intercept, represents the coefficient vector, the coefficient vector , is the regression parameter, Represents the coefficient vector The dimension of represents the number of observation samples in the dataset, represents the speed values ​​measured at different time points in the data set, represents the vibration intensity measurement at a given time point in the data set, Represents the coefficient vector The elements.

2. The large variable speed unit fault diagnosis system according to claim 1, Features: The erroneous data elimination module is used to identify abnormal data in the data based on the isolation forest algorithm and eliminate it, including the following steps: S3.

1. Collect the data set to be tested, where the data set includes normal and abnormal vibration signals and speed data; S3.2, select the speed feature, and select the middle value between the maximum and minimum values ​​of the feature as the split point, divide the data set into two left and right subsets according to the split point, and use the two subsets as the input of the left and right subtrees respectively to form an isolated tree; construct the isolated tree to form an isolated forest; S3.

3. For each data point in the data set, calculate its path length H in each isolated tree and take the average value as the anomaly score E of the data point; S3.

4. Set a threshold K. Data points with anomaly scores exceeding the threshold are considered outliers and excluded.

3. The large variable speed unit fault diagnosis system according to claim 1, Features: The data analysis unit (2) converts the non-steady-state data into steady-state data based on the order resampling method, comprising the following steps: S7.

1. Obtain accurate speed information through encoder equipment and perform constant angle incremental sampling on the data; S7.

2. First, perform fast Fourier transform on the steady-state signal in the angle domain to obtain a frequency domain signal; divide the spectrum by order to distinguish the frequency shift signal caused by the speed change, and each order corresponds to a stable frequency component; S7.3, taking the vibration as a reference order, interpolating the signal by the nearest neighbor interpolation method based on the reference order, and introducing the weight of the neighboring points to optimize the interpolation process of the signal by the nearest neighbor interpolation method, so that it is evenly distributed in each order; S7.4, synchronously average the signals of the same order, eliminate random noise, and smooth the speed information; S7.

5. Obtain steady-state vibration signals and speed data, and use a minimum-maximum normalization algorithm to quantify the steady-state vibration signals and speed data.

4. The large variable speed unit fault diagnosis system according to claim 3, Features: In S7.5, the minimum-maximum normalization algorithm is specifically: ; in, represents the vibration signal and speed data in the data set, Represents the value obtained after the minimum-maximum normalization algorithm, Represents the maximum value in the data set. Represents the minimum value in the data set.

5. The large variable speed unit fault diagnosis system according to claim 1, Features: The data alarm unit (3) determines whether to alarm by setting a threshold based on the alarm triggering rule, including the following steps: S9.

1. preprocessing the quantified steady-state data collected by the data analysis unit (2); S9.2, set the alarm threshold P; S9.

3. When it is detected that the quantified steady-state data exceeds the preset threshold value P, an alarm is triggered and an alarm message is sent.

6. A large variable speed unit fault diagnosis method, used in a large variable speed unit fault diagnosis system as claimed in any one of claims 1 to 5, It is characterized in that The steps include: S10.

1. Eliminate erroneous data based on the isolation forest algorithm and screen steady-state data based on the sliding window test method; S10.2, analyzing the data based on the order resampling method to convert the non-steady-state data into steady-state data; S10.

3. Based on the alarm triggering rules, an alarm is issued for steady-state data exceeding the threshold.