Fault diagnosis method and system

Abstract

The present invention discloses a fault diagnosis method and system, which belongs to the field of laser non-contact vibration detection and diagnosis technology. The present invention synergistically enhances the ability to extract fault features through multi-dimensional feature fusion of time domain framing, frequency domain envelope spectrum, frequency multiplication matching and energy entropy features, and realizes the collaborative identification of weak faults and multiple faults; reduces the influence of external interference on the diagnosis results through frequency conversion and frequency multiplication error control and peak interval analysis; dynamically judges the significance of fault features through energy entropy, replacing the traditional amplitude threshold method, thereby improving the robustness of the system; and solves the technical problem of limited performance of machine learning models due to the scarcity of fault samples through artificial feedback optimization algorithm, achieving the technical effect of continuous iterative optimization of the diagnosis system.

Description

Technical Field

[0001] The present invention relates to the technical field of laser non-contact vibration detection and diagnosis, in particular to a fault diagnosis method and system. Background Art

[0002] With the increasing integration and comprehension of industrial production, manual inspection and verification of the health of many large-scale equipment is no longer feasible. Furthermore, the increasing complexity of current production machinery and equipment has led to an increase in the coupling and subtle nature of faults. Analysis methods that rely on vibration signals often suffer from subtle fault signals and a high level of interference. This complicates fault identification for vibration analysis methods. Therefore, effectively identifying weak fault signals and multi-source fault signals is a pressing issue.

[0003] When it comes to motor bearing fault diagnosis, most current technologies focus on improving the strength of fault features in the original signal and reducing interference from noise and other industrial vibrations, laying the foundation for subsequent fault feature identification. However, based on these technologies, accurately and effectively identifying fault features remains a major challenge. Data-driven approaches, such as machine learning and deep learning, rely on large-scale data. However, the scarcity of fault samples in actual industrial scenarios limits the performance of big data modeling methods. Traditional signal processing methods primarily use vibration meters to collect equipment vibration signals, extract the signal's time, frequency, and time-frequency domain features, and directly match them to a library of mechanical equipment fault types. This approach leaves room for improvement in automatically and accurately identifying fault types.

[0004] In actual industrial applications, bearing vibration signals collected by vibration meters are subject to significant external interference, making it difficult to identify the fault type. In most cases, equipment failures do not occur suddenly, and they may not be a single fault. Early, subtle fault signatures are easily obscured by the main frequency signal, and multiple fault types may be mixed together. Fault impacts occur within a very short period of time, generating a periodic response along with the equipment's rotation. This makes accurate bearing fault diagnosis difficult. Summary of the Invention

[0005] In response to the above problems, a fault diagnosis method and system are provided. Based on traditional signal processing technology, the present invention constructs an improved signal processing method and artificial feedback system. On the basis of effectively identifying weak faults and multiple faults, combined with the artificial feedback of the platform module, the algorithm is regularly optimized to improve the accuracy of fault diagnosis. The present invention synergistically enhances the ability to extract fault features through multi-dimensional feature fusion of time domain framing, frequency domain envelope spectrum, frequency multiplication matching and energy entropy features, solves the problem that weak signals are easily masked, and thus realizes the collaborative identification of weak faults and multiple faults; reduces the impact of external interference on the diagnosis results through frequency conversion and multiplication error control and peak interval analysis; dynamically determines the significance of fault features through energy entropy, replacing the traditional amplitude threshold method, thereby improving the robustness of the system; solves the technical problem of limited performance of machine learning models due to scarcity of fault samples through artificial feedback optimization algorithm, and achieves the technical effect of continuous iterative optimization of the diagnosis system.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is.

[0007] A fault diagnosis method comprises the following steps:

[0008] Step 1: Collect vibration signals Perform frame processing to obtain multi-frame time domain signals ;

[0009] Step 2: Transform each frame of time domain signal into Convert to frequency domain signal ;

[0010] Step 3: Calculate the frequency domain signal through Hilbert transform The envelope spectrum of

[0011] Step 4: Based on the frequency Fz After calculating and filtering out the corresponding frequency octaves, before extracting the envelope spectrum amplitude n Group Peak Num 1 and the corresponding frequency F 1, and sorted in ascending order of frequency;

[0012] Step 5: Calculation F Forward difference value of adjacent frequency intervals in 1 ,like If the error ∈ [-2, 2], the step-by-step frequency multiplication fault judgment process is performed to calculate the corresponding fault, otherwise the calculation Num The maximum peak value in 1 Pm 0 and the corresponding frequency fm 0;

[0013] Step 6: Determine the significant peak by determining the amplitude prominence through wavelet energy entropy characteristics fx ,like fxand fm Output when it is a multiple of 0 fm 0 corresponds to a fault, otherwise a cyclic fault determination process is performed to calculate the corresponding fault.

[0014] Preferably, the frame processing formula in step 1 is as follows:

[0015]

[0016] in, t For time; l is the current frame; L is the total number of frames; is the rectangular window function of the current frame; inc Indicates frame shift; Len Indicates the frame length.

[0017] Preferably, the Fourier transform formula is as follows:

[0018]

[0019] in, T For the entire time domain; t For time; j is an imaginary number; is the frequency; is the time domain signal of each frame.

[0020] Preferably, the Hilbert transform formula is as follows:

[0021]

[0022] in, is the frequency; is the frequency domain signal; is a frequency domain signal The envelope spectrum.

[0023] Preferably, the frequency multiplication error ez1∈[-5,5], the front n In the group n Take 10.

[0024] Preferably, the F 1=[ f 1, f 2,…, fn ]; Num 1=[ N 1, N 2,…, Nn ], the forward difference value The calculation formula is: .

[0025] Preferably, the steps of the step-by-step frequency multiplication fault determination process are as follows:

[0026] S1: Calculation F 1 and frequency conversion Fz Ratio G , and extract the ratio G The frequency corresponding to the minimum value Fg ;

[0027] S2: If the error ez2= Fg - Fz In [-5, 5], calculate Fg 2 times the frequency Fg 2, 3 times the frequency Fg 3. Otherwise, the device frequency is determined to be normal;

[0028] S3: If there is any frequency doubling Fg 2 and or Fg 3, then the frequency at this time is calculated as Fg The corresponding fault type is displayed when the frequency is normal.

[0029] Preferably, the steps of determining the amplitude prominence degree by using the wavelet energy entropy feature are as follows:

[0030] S1: Envelope spectrum of frequency domain signal Perform continuous wavelet decomposition to obtain different scales s The wavelet coefficients under ;

[0031] S2: Calculate each scale s Energy under E ( s ):

[0032] ;

[0033] S3: Energy E ( s ) for standardization, and wait until the standardized energy :

[0034] ;

[0035] S4: Calculate wavelet energy entropy Entropy :

[0036] ;

[0037] S5: If the wavelet energy entropy Entropy Exceeding the threshold thr , it is judged as a prominent feature, otherwise it is not prominent; among them, the threshold thr The wavelet energy entropy of the normal operation state of the equipmentEntropy Calibration.

[0038] Preferably, the steps of the cycle fault determination process are as follows:

[0039] S1: Set the number of cycles i =1, floating threshold m = ±5;

[0040] S2: Calculate the maximum peak value of the current envelope spectrum respectively Pm 1 ( i ) and its frequency fm 1 ( i );[2 fm 1 ( i )- m ,2 fm 1 ( i )+ m ]The maximum peak value within the range Pm 2 ( i ) and its frequency fm 2 ( i );[3 fm 1 ( i )- m ,3 fm 1 ( i )+ m ]The maximum peak value within the range Pm 3 ( i ) and its frequency fm 3 ( i );

[0041] S3: Calculate the peak value of each group Pm ( i )=( Pm 1 ( i ), Pm 2 ( i ), Pm 3 ( i )) and the amplitude mean of its left and right neighbors, and take the smaller ratio of the left and right amplitude means as Ka ( i )=( Ka 1 ( i ), Ka 2 ( i ), Ka 3 ( i )); where the left neighborhood range is [ fm ( i )- Fz / 2, fm ( i )-1], the right neighbor range is [ fm ( i )+1, fm( i )+ Fz / 2];

[0042] S4: Remove the currently calculated peak from the envelope spectrum Pm ( i ) and corresponding frequency fm ( i )=( fm 1 ( i ), fm 2 ( i ), fm 3 ( i )),like i <3, then execute i = i +1 and return to step S2;

[0043] S5: Otherwise, the three groups of loops from the calculation Ka ( i ) select the maximum value, according to the selected group fm 1 ( i ) Finally determine the fault type.

[0044] A fault diagnosis system includes a data acquisition module for signal acquisition and a fault diagnosis platform for fault identification and early warning, wherein the fault diagnosis platform includes a signal processing module and a manual feedback module; the data acquisition module collects motor bearing vibration data and outputs it to the signal processing module in the fault diagnosis platform, the signal processing module identifies and warns of different fault types according to the fault diagnosis method, and outputs the fault type to the manual feedback module, the actual fault type is fed back by the manual feedback module and fed back to the signal processing module for updating the error parameters of the fault diagnosis method.

[0045] Due to the adoption of the above technical solution, the present invention has the following beneficial effects.

[0046] (1) The present invention extracts the envelope spectrum information of the vibration signal, combines the frequency-harmonic relationship analysis and the envelope spectrum amplitude prominence judgment, and solves the problem of weak fault characteristics being masked by the main frequency signal and the difficulty in identifying multiple fault types caused by mixing, thereby achieving the technical effect of accurately distinguishing weak fault types from multiple fault types.

[0047] (2) The present invention generates envelope spectrum by frame processing, Fourier transform and Hilbert transform, thereby solving the problem that traditional signal processing methods are insufficient in processing short-term fault impacts and periodic response signals, and realizing efficient extraction and feature enhancement of fault frequencies; through peak sorting and frequency multiplication relationship analysis technical means, it solves the technical problem of feature confusion when multiple fault types are mixed, and achieves the technical effect of accurately distinguishing multiple fault types.

[0048] (3) The present invention distinguishes between weak fault types and multiple fault types through the relationship between signal frequency and frequency doubling and the prominence of envelope spectrum amplitude, and provides actual feedback on the fault diagnosis platform to provide an effective optimization path for signal processing methods; the prominence of envelope spectrum amplitude is judged by the energy entropy characteristics after wavelet decomposition, and the prominence of signal in a short time is calculated by analyzing the overall energy fluctuation. Combined with peak sorting and frequency doubling matching, the limitations of traditional methods that rely on manual experience or large-scale data are solved, and the technical effect of automatically identifying the main / secondary fault types is achieved.

[0049] (4) The present invention sorts the first 10 groups of abnormally prominent peaks on the envelope spectrum, considers the relationship between the peak interval and the rotation frequency, and analyzes whether the frequency doublings of the frequencies corresponding to the first 10 groups of abnormally prominent peaks exist and the prominence of the peaks. By comparing the three groups of identified peaks, the group with the highest amplitude prominence among the 1st, 2nd and 3rd times of the frequencies corresponding to the peaks is selected as the main fault type identified, and the others are secondary fault types. The signal processing algorithm is optimized by feeding back the artificial feedback module to solve the problem of limited model performance caused by the scarcity of fault samples in actual industrial scenarios, and achieve the technical effect of dynamically improving the accuracy of fault diagnosis.

[0050] (5) The present invention enhances the fault feature extraction capability through the fusion of multi-dimensional features such as time domain framing, frequency domain envelope spectrum, frequency multiplication matching and energy entropy features, solves the problem of weak signals being easily masked, and thus realizes the collaborative identification of weak faults and multiple faults; reduces the influence of external interference on the diagnosis results through frequency conversion and multiplication error control and peak interval analysis; dynamically judges the significance of fault features through energy entropy, replacing the traditional amplitude threshold method, thereby improving the robustness of the system; solves the technical problem of limited performance of machine learning models due to scarcity of fault samples through artificial feedback optimization algorithm, and achieves the technical effect of continuous iterative optimization of the diagnosis system.

[0051] (6) The present invention solves the technical problems of masking of weak fault features and mixed identification of multiple faults in industrial scenarios through the technical means of envelope spectrum feature extraction, frequency conversion and multiplication matching, and wavelet energy entropy quantification, combined with an artificial feedback closed-loop optimization system, and achieves high-precision, automated, and dynamically optimized fault diagnosis effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The following detailed discussion discusses the making and use of preferred embodiments of the present invention. However, it should be understood that the present invention provides numerous applicable inventive concepts that can be embodied in a variety of specific contexts. The specific embodiments discussed are intended only to illustrate specific ways to make and use the present invention and are not intended to limit the scope of the invention. It would be readily apparent to one of ordinary skill in the art that other embodiments could be derived from these drawings without inventive effort.

[0053] Figure 1 Schematic diagram of the structure of the system of the present invention.

[0054] Figure 2 Flowchart of the fault diagnosis method of the present invention.

[0055] Figure 3 This is the cycle fault determination processing flow of the present invention.

[0056] Figure 4 The frequency spectrum and envelope spectrum corresponding to different bearing fault types. DETAILED DESCRIPTION

[0057] The following detailed discussion discusses the making and use of preferred embodiments of the present invention. However, it should be understood that the present invention provides many applicable inventive concepts that can be embodied in a variety of specific circumstances. The specific embodiments discussed are intended only to illustrate specific ways to make and use the present invention and are not intended to limit the scope of the present invention.

[0058] The present invention extracts the envelope spectrum information of the vibration signal, combines frequency-harmonic relationship analysis with envelope spectrum amplitude prominence determination, and solves the problem of weak fault characteristics being masked by the main frequency signal and the difficulty in identifying multiple fault types due to mixing, thereby achieving the technical effect of accurately distinguishing between weak fault types and multiple fault types. By generating the envelope spectrum through frame processing, Fourier transform, and Hilbert transform, the problem of traditional signal processing methods' insufficient processing of short-term fault impulses and periodic response signals is solved, achieving efficient extraction of fault frequencies and feature enhancement. By using peak sorting and frequency-harmonic relationship analysis techniques, the technical problem of feature confusion when multiple fault types are mixed is solved, achieving the technical effect of accurately distinguishing multiple fault types.

[0059] like Figure 2 The fault diagnosis method shown in the figure includes the following steps: Step 1: collecting the vibration signal Perform frame processing to obtain multi-frame time domain signals The framing processing formula is as follows:

[0060]

[0061] in, t For time; l is the current frame; L is the total number of frames; is the rectangular window function of the current frame; inc Indicates frame shift; Len Indicates the frame length.

[0062] Step 2: Transform each frame of time domain signal into Convert to frequency domain signal ; The Fourier transform formula is as follows:

[0063]

[0064] in, T For the entire time domain; t For time; j is an imaginary number; is the frequency; is the time domain signal of each frame.

[0065] Step 3: Calculate the frequency domain signal through Hilbert transform The envelope spectrum of . The Hilbert change formula is as follows:

[0066]

[0067] in, is the frequency; is the frequency domain signal; is a frequency domain signal The envelope spectrum.

[0068] Step 4: Based on the frequency Fz After calculating and filtering out the corresponding frequency octaves, before extracting the envelope spectrum amplitude n Group Peak Num 1 and the corresponding frequency F 1, and arranged in ascending order of frequency; the frequency multiplication error ez1∈[-5,5], in this embodiment, n Take 10.

[0069] Step 5: Calculation F Forward difference value of adjacent frequency intervals in 1 ,like If the error ∈ [-2, 2], the step-by-step frequency multiplication fault judgment process is performed to calculate the corresponding fault, otherwise the calculation Num The maximum peak value in 1 Pm 0 and the corresponding frequency fm 0; described F 1=[ f 1, f 2,…, fn ]; Num 1=[ N 1, N 2,…, Nn ], the forward difference value The calculation formula is: .

[0070] The steps of the step-by-step frequency multiplication fault determination process are as follows: S1: Calculate F 1 and frequency conversion Fz Ratio G , and extract the ratio G The frequency corresponding to the minimum valueFg .

[0071] S2: If the error ez2= Fg - Fz In [-5, 5], calculate Fg 2 times the frequency Fg 2, 3 times the frequency Fg 3. Otherwise, the device frequency is determined to be normal.

[0072] S3: If there is any frequency doubling Fg 2 and or Fg 3, then the frequency at this time is calculated as Fg The corresponding fault type is displayed when the frequency is normal.

[0073] Step 6: Determine the significant peak by determining the amplitude prominence through wavelet energy entropy characteristics fx The steps of determining the amplitude prominence degree by wavelet energy entropy feature are as follows: S1: Envelope spectrum of frequency domain signal Perform continuous wavelet decomposition to obtain different scales s The wavelet coefficients under .

[0074] S2: Calculate each scale s Energy under E ( s ):

[0075] ;

[0076] S3: Energy E ( s ) for standardization, and wait until the standardized energy :

[0077] ;

[0078] S4: Calculate wavelet energy entropy Entropy :

[0079] ;

[0080] S5: If the wavelet energy entropy Entropy Exceeding the threshold thr , it is judged as a prominent feature, otherwise it is not prominent; among them, the threshold thr The wavelet energy entropy of the normal operation state of the equipment Entropy Calibration.

[0081] like fx and fm Output when it is a multiple of 0 fm 0 corresponds to a fault, otherwise a cyclic fault determination process is performed to calculate the corresponding fault.Figure 3 As shown, the steps of the cycle fault determination process are as follows: S1: Set the number of cycles i =1, floating threshold m =±5.

[0082] S2: Calculate the maximum peak value of the current envelope spectrum respectively Pm 1 ( i ) and its frequency fm 1 ( i );[2 fm 1 ( i )- m ,2 fm 1 ( i )+ m ]The maximum peak value within the range Pm 2 ( i ) and its frequency fm 2 ( i );[3 fm 1 ( i )- m ,3 fm 1 ( i )+ m ]The maximum peak value within the range Pm 3 ( i ) and its frequency fm 3 ( i ).

[0083] S3: Calculate the peak value of each group Pm ( i )=( Pm 1 ( i ), Pm 2 ( i ), Pm 3 ( i )) and the amplitude mean of its left and right neighbors, and take the smaller ratio of the left and right amplitude means as Ka ( i )=( Ka 1 ( i ), Ka 2 ( i ), Ka 3 ( i )); where the left neighborhood range is [ fm ( i )- Fz / 2, [[ID=A115]]fm ( i )-1], the right neighbor range is [ fm ( i )+1, fm ( i )+ Fz / 2].

[0084] S4: Remove the currently calculated peak from the envelope spectrum Pm ( i ) and corresponding frequency fm ( i )=( fm 1 ( i ), fm 2 ( i ), fm 3 ( i )),like i <3, then execute i = i +1 and return to step S2.

[0085] S5: Otherwise, the three groups of loops from the calculation Ka ( i ) select the maximum value, according to the selected group fm 1 ( i ) Finally determine the fault type.

[0086] In addition, if Figure 1 As shown, this embodiment also provides a fault diagnosis system, including a data acquisition module for signal acquisition and a fault diagnosis platform for fault identification and early warning, the fault diagnosis platform including a signal processing module and a manual feedback module; the data acquisition module collects motor bearing vibration data and outputs it to the signal processing module in the fault diagnosis platform, the signal processing module identifies and warns of different fault types according to the fault diagnosis method, and outputs the fault type to the manual feedback module, the actual fault type is fed back by the manual feedback module and fed back to the signal processing module for updating the error parameters of the fault diagnosis method.

[0087] The present invention solves the technical problems of masking weak fault features and mixed identification of multiple faults in industrial scenarios through the technical means of envelope spectrum feature extraction, frequency conversion and multiplication matching, and wavelet energy entropy quantification, combined with an artificial feedback closed-loop optimization system, to achieve high-precision, automated, and dynamically optimized fault diagnosis effects.

[0088] The following is combined with Figure 1 - 4 Further elaborate.

[0089] like Figure 2 The fault diagnosis method shown in the figure includes the following steps: Assuming that the vibration signal currently collected is , the signal is divided into frames and the signal of each frame is obtained as , expressed as follows:

[0090]

[0091] in, t For time; l is the current frame;L is the total number of frames; is the rectangular window function of the current frame; inc Indicates frame shift; Len Indicates the frame length.

[0092] Convert the time domain signal into the frequency domain signal. Perform Fourier transform to convert the time domain signal into the frequency domain signal :

[0093]

[0094] in, T For the entire time domain; t For time; j is an imaginary number; is the frequency; is the time domain signal of each frame.

[0095] Calculate frequency domain signal through Hilbert transform The envelope spectrum is obtained by performing Hilbert transform on the spectrum signal and obtaining the absolute value to obtain the envelope spectrum amplitude. :

[0096]

[0097] in, is the frequency; is the frequency domain signal; is a frequency domain signal The envelope spectrum.

[0098] Known frequency Fz , ask for frequency Fz The frequency multiplication is removed, and the frequency multiplication is within the error range of ez1=±5. Then all the data are sorted from large to small, and then the first 10 groups of data values ​​are sorted. Num 1=[ N 1, N 2,… N 10] The corresponding frequency value F1 =[ f 1, f 2,…, f 10] Sort from small to large. Then calculate F 1=[ f 1, f 2,…, f 10], and obtain the interval difference between each two numbers, that is, .like If the error ∈ [-2, 2], the step-by-step frequency multiplication fault determination process is performed. The steps of the step-by-step frequency multiplication fault determination process are as follows: S1: Calculate F 1=[ f 1,f 2,…, f 10] Convert from Fz Ratio G , and extract the ratio G The frequency corresponding to the minimum value Fg .

[0099] S2: If the error ez2= Fg - Fz In [-5, 5], calculate Fg 2 times the frequency Fg 2, 3 times the frequency Fg 3. Otherwise, the device frequency is determined to be normal, that is, the surface device is in a normal state in the fault corresponding to the frequency.

[0100] S3: If there is any frequency doubling Fg 2=2 Fg Sum or frequency doubling Fg 3=3 Fg , then the frequency at this time is calculated as Fg Otherwise, the device frequency is normal, indicating that the device is in a normal state in the fault corresponding to the frequency.

[0101] like If the error exceeds [-2, 2], then calculate Num 1=[ N 1, N 2,… N 10] the maximum peak Pm 0 and the corresponding frequency fm 0, and the significant peak is determined by judging the amplitude prominence through the wavelet energy entropy feature fx The steps of determining the amplitude prominence degree by wavelet energy entropy feature are as follows: S1: Envelope spectrum of frequency domain signal Perform continuous wavelet decomposition to obtain different scales s The wavelet coefficients under .

[0102] S2: Calculate each scale s Energy under E ( s ):

[0103] ;

[0104] S3: Energy E ( s ) for standardization, and wait until the standardized energy :

[0105] ;

[0106] S4: Calculate wavelet energy entropyEntropy :

[0107] ;

[0108] S5: If the wavelet energy entropy Entropy Exceeding the threshold thr , it is judged as a prominent feature, otherwise it is not prominent; among them, the threshold thr The wavelet energy entropy of the normal operation state of the equipment Entropy Calibration.

[0109] like fm 0 and frequency conversion Fz Is there a frequency between fx and fm 0 is a multiple of fx The corresponding amplitude is a larger peak value, which satisfies both the peak condition and the peak prominence degree. The prominence degree can be judged by referring to the prominence degree judgment method. fx and fm Output when it is a multiple of 0 fm 0 corresponds to a fault, otherwise a cyclic fault determination process is performed to calculate the corresponding fault. The steps of the cyclic fault determination process are as follows: Figure 3 Shown: S1: Set the number of cycles i =1, floating threshold m =±5.

[0110] S2: Calculate the maximum peak value of the current envelope spectrum respectively Pm 1 ( i ) and its frequency fm 1 ( i );[2 fm 1 ( i )- m ,2 fm 1 ( i )+ m ]The maximum peak value within the range Pm 2 ( i ) and its frequency fm 2 ( i );[3 fm 1 ( i )- m ,3 fm 1 ( i )+ m ]The maximum peak value within the range Pm 3 ( i ) and its frequency fm 3 ( i ).

[0111] S3: Calculate the peak value of each group Pm ( i )=(Pm 1 ( i ), Pm 2 ( i ), Pm 3 ( i )) and the amplitude mean of its left and right neighbors, and take the smaller ratio of the left and right amplitude means as Ka ( i )=( Ka 1 ( i ), Ka 2 ( i ), Ka 3 ( i )); where the left neighborhood range is [ fm ( i )- Fz / 2, fm ( i )-1], the right neighbor range is [ fm ( i )+1, fm ( i )+ Fz / 2].

[0112] S4: Remove the currently calculated peak from the envelope spectrum Pm ( i ) and corresponding frequency fm ( i )= ( fm 1 ( i ), fm 2 ( i ), fm 3 ( i )), that is, remove the previous stage calculated data from the original envelope spectrum data Pm ( i )=( Pm 1 ( i ), Pm 2 ( i ), Pm 3 ( i ) and its corresponding frequency fm ( i )=( fm 1 ( i ), fm 2 ( i ), fm 3 ( i )). If the number of cycles i <3, then execute i = i +1 and return to step S2.

[0113] S5: Otherwise, select from the three sets of data in the calculation cycle Ka ( i)=( Ka 1 ( i ), Ka 2 ( i ), Ka 3 ( i )), where as long as Ka 1 ( i ), Ka 2 ( i ), Ka 3 ( i ) have two groups that meet the larger requirement, then the current i Group Ka ( i ) is larger. Then calculate the larger Ka ( i ) i The corresponding group data fm ( i )=( fm 1 ( i ), fm 2 ( i ), fm 3 ( i ))Medium frequency fm 1 ( i ), and according to the frequency fm 1 ( i ) The corresponding fault type is used to finally determine the bearing fault type.

[0114] This embodiment uses the real data collected by the vibration meter to identify the fault type of the motor bearing. The collected fault data is divided into bearing outer ring fault data, bearing inner ring fault data, and bearing ball fault data. Figure 1 The fault diagnosis system includes a data acquisition module for signal acquisition and a fault diagnosis platform for fault identification and early warning. The fault diagnosis platform includes a signal processing module and a manual feedback module. The data acquisition module primarily collects motor bearing vibration data from a vibration meter in real time and automatically uploads it to the fault diagnosis platform via a network. The data acquisition module collects motor bearing vibration data and outputs it to a signal processing module in the fault diagnosis platform. The signal processing module identifies and warns of different fault types according to the fault diagnosis method and outputs the fault type to the manual feedback module. The manual feedback module then provides feedback on the actual fault type and feeds it back to the signal processing module for updating the error parameters of the fault diagnosis method.

[0115] Figure 4 The frequency domain and envelope spectrum of bearing outer ring fault data, bearing inner ring fault data and bearing ball fault data are given. Figure 4It can be seen that under different fault types, the fault frequency of the signal is different. According to the fault diagnosis method of the present invention, different categories of fault types can be clearly distinguished, and then the actual fault type is fed back into the signal processing module in combination with the manual feedback module to update the error parameters of the fault diagnosis method, thereby realizing the optimization of the fault diagnosis method of the present invention. Through the manual feedback optimization algorithm, the technical problem of limited performance of the machine learning model due to scarcity of fault samples is solved, and the technical effect of continuous iterative optimization of the diagnosis system is achieved.

[0116] The present invention synergistically enhances the fault feature extraction capability through multi-dimensional feature fusion of time domain framing, frequency domain envelope spectrum, frequency multiplication matching and energy entropy features, solves the problem of weak signals being easily masked, and thus realizes the collaborative identification of weak faults and multiple faults; reduces the impact of external interference on the diagnosis results through frequency conversion and multiplication error control and peak interval analysis; and dynamically determines the significance of fault features through energy entropy, replacing the traditional amplitude threshold method, thereby improving the robustness of the system.

[0117] Although the specification has been described in detail, it should be understood that various changes, substitutions, and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims. In addition, the specific embodiments described are not intended to limit the scope of the invention, and a person of ordinary skill in the art can readily understand based on the present invention that currently existing or later developed processes, machines, manufactures, material compositions, means, methods, or steps may perform substantially the same functions or obtain substantially the same results as the embodiments of the present invention. Therefore, the appended claims are intended to include such processes, machines, manufactures, material compositions, means, methods, or steps within their scope.

Claims

1. A fault diagnosis method, characterized in that: The following steps are involved: Step 1: Frame the collected vibration signal x(t) to obtain a multi-frame time domain signal x l (t); Step 2: Transform each frame of time domain signal x into l (t) is converted into a frequency domain signal X l (ω); Step 3: Calculate the frequency domain signal X through Hilbert transform l (t) envelope spectrum; Step 4: After calculating and filtering out the corresponding frequency multiplication based on the rotation frequency Fz, extract the first n peaks Num1 and the corresponding frequencies F1 of the envelope spectrum amplitude and arrange them in ascending order of frequency; Step 5: Calculate the forward difference value Δf of the adjacent frequency intervals in F1. If the error of Δf is ∈[-2, 2], perform the step-by-step frequency multiplication fault determination process to calculate the corresponding fault. Otherwise, calculate the maximum peak value Pm0 and the corresponding frequency fm0 in Num1. The steps of the step-by-step frequency multiplication fault determination process are as follows: S1: Calculate the ratio G of F1 and the rotation frequency Fz, and extract the frequency Fg corresponding to the minimum value of the ratio G; S2: If the error ez2 = Fg - Fz is within [-5, 5], then calculate the double frequency Fg2 and triple frequency Fg3 of Fg; otherwise, determine that the device frequency is normal; S3: If any of the multiple frequencies Fg2 and / or Fg3 exists, calculate the corresponding fault type when the frequency is Fg; otherwise, determine that the device frequency is normal. Step 6: Determine the significant peak value fx by judging the amplitude prominence through the wavelet energy entropy feature. If fx is a multiple of fm0, output fm0 corresponding to the fault; otherwise, perform cyclic fault determination processing to calculate the corresponding fault.

2. A fault diagnosis method according to claim 1, characterized in that: The framing processing formula in step 1 is as follows: x l (t)=x(t)·w l (t),l=1,...,L Where t is time; l is the current frame; L is the total number of frames; w l (t) is the rectangular window function of the current frame; inc represents the frame shift; Len represents the frame length.

3. A fault diagnosis method according to claim 1, characterized in that: The Fourier transform formula in step 2 is as follows: Where T is the entire time domain; t is time; j is an imaginary number; ω is frequency; x l (t) is the time domain signal of each frame.

4. A fault diagnosis method according to claim 1, characterized in that: The Hilbert transformation formula in step 3 is as follows: Where ω is the frequency; X l (ω) is the frequency domain signal; X hl (ω) is the frequency domain signal X l (ω) envelope spectrum.

5. A fault diagnosis method according to claim 1, characterized in that: The frequency multiplication error ez1∈[-5, 5] in step 4, and n in the first n groups is 10.

6. A fault diagnosis method according to claim 1, characterized in that: In step 5, F1 = [f1, f2, ..., fn]; Num1 = [N1, N2, ..., Nn], and the calculation formula of the forward differential value Δf is Δf = f(k+1)-f(k), k = 1, ..., (n-1).

7. A fault diagnosis method according to claim 1, characterized in that: The steps of determining the amplitude prominence degree by the wavelet energy entropy feature in step 6 are as follows: S1: Envelope spectrum X of frequency domain signal hl (ω) Perform continuous wavelet decomposition to obtain wavelet coefficients at different scales s S2: Calculate the energy E(s) at each scale s: S3: Normalize the energy E(s) until the normalized energy S4: Calculate the wavelet energy entropy Entropy: S5: If the wavelet energy entropy Entropy exceeds the threshold thr, it is determined to be a prominent feature, otherwise it is not prominent; wherein the threshold thr is calibrated by the wavelet energy entropy Entropy of the normal operating state of the equipment.

8. A fault diagnosis method according to claim 1, characterized in that: The steps of the cycle fault determination process are as follows: S1: Set the number of cycles i = 1, and the floating threshold m = ±5; S2: Calculate the maximum peak value Pm1(i) of the current envelope spectrum and its frequency fm1(i); the maximum peak value Pm2(i) in the range [2fm1(i)-m, 2fm1(i)+m] and its frequency fm2(i); the maximum peak value Pm3(i) in the range [3fm1(i)-m, 3fm1(i)+m] and its frequency fm3(i); S3: Calculate the amplitude mean of each peak group Pm(i) = (Pm1(i), Pm2(i), Pm3(i)) and its left and right neighboring values, and take the smaller ratio of the left and right amplitude means as Ka(i) = (Ka1(i), Ka2(i), Ka3(i)); where the left neighboring range is [fm(i)-Fz / 2, fm(i)-1], and the right neighboring range is [fm(i)+1, fm(i)+Fz / 2]; S4: Remove the currently calculated peak value Pm(i) and the corresponding frequency fm(i)=(fm1(i), fm2(i), fm3(i)) from the envelope spectrum. If i<3, execute i=i+1 and return to step S2; S5: Otherwise, the maximum value is selected from the three calculated cyclic groups Ka(i), and the fault type is finally determined based on fm1(i) of the selected group.

9. A fault diagnosis system, characterized in that: It includes a data acquisition module for signal acquisition and a fault diagnosis platform for fault identification and early warning, and the fault diagnosis platform includes a signal processing module and a manual feedback module; the data acquisition module collects motor bearing vibration data and outputs it to the signal processing module in the fault diagnosis platform, and the signal processing module identifies and warns of different fault types according to the fault diagnosis method described in any one of claims 1 to 8, and outputs the fault type to the manual feedback module, and the actual fault type is fed back by the manual feedback module and fed back to the signal processing module for updating the error parameters of the fault diagnosis method.

Images