A cross-stage, multi-condition early fault diagnosis method based on permutation spectrum difference measurement

Abstract

An early fault diagnosis method for cross-stage multi-condition based on permutation spectrum difference measurement, including: collecting vibration signals of mechanical equipment under different states in the manufacturing test stage and the actual use stage, dividing the vibration signals into multiple subsequences through a combination of characteristic parameters, and reconstructing the subsequences into the original phase space; obtaining all possible permutation patterns according to the embedding dimension of the vibration signals, and converting the original phase space into a symbolic phase space; counting the number of occurrences of various permutation patterns in the symbolic phase space to obtain permutation spectrum characteristics; obtaining the difference degree between the vibration signals collected under different states; establishing a parameter selection model to select the best parameter combination; calculating the permutation spectrum characteristics of each time series using the best parameter combination, inputting the permutation spectrum characteristics of the data in the manufacturing test stage into a multi-class SVM classifier to obtain a training model, and inputting the permutation spectrum of the data in the actual use stage into the training model to identify the fault type and operating condition corresponding to the time series.

Description

Technical Field

[0001] This invention relates to a method for early fault diagnosis across multiple stages and operating conditions, and more specifically to a method for determining whether early faults occur in mechanical equipment under different operating conditions during actual use by using data from the manufacturing and testing stages, combined with permutation spectrum and difference measurement techniques. Background Technology

[0002] With the development of modern technology, mechanical equipment is evolving towards intelligence and data-driven approaches. From the manufacturing and testing phase to the actual use phase, massive amounts of data are generated. This data contains rich information about the equipment's status, reflecting potential underlying dynamic fluctuations. However, the actual use phase often only provides a large amount of normal operation data, making it difficult to obtain various fault data. In contrast, the product testing process can obtain normal and fault data under various operating conditions. Therefore, how to integrate and analyze multi-source data from the product manufacturing and testing phases and the actual use phases to establish a cross-phase mechanical equipment quality assurance and fault diagnosis method is a key research focus.

[0003] Before production, mechanical equipment undergoes a series of break-in and performance tests. Test data can be understood as usage data of the equipment under specific operating conditions. Therefore, without secondary disassembly and reassembly, the fluctuations in test data and actual usage data show certain similarities. By analyzing the underlying fluctuation characteristics of the output data from the manufacturing and testing phase of the mechanical equipment system under different states, and applying these characteristics to the output data from the actual usage phase, it is possible to promptly detect and accurately identify whether the mechanical equipment is malfunctioning and its current operating condition.

[0004] Localized damage to mechanical equipment manifests as repetitive pulse signals. Currently, relatively mature technologies exist for diagnosing late-stage faults in equipment. However, in the early stages of a fault, the pulse signals are easily drowned out by environmental noise, making early fault identification and fault type diagnosis difficult. Furthermore, the operating conditions of equipment are not static, and existing commonly used features are insufficient to identify operating condition information in signals; signal characteristics of various fault types can easily produce the same feature values ​​under different operating conditions. Therefore, finding sensitive features that can accurately identify early, weak pulses of different fault types under multiple operating conditions is the primary focus of fault diagnosis. Summary of the Invention

[0005] This invention provides a cross-stage, multi-condition early fault diagnosis method based on permutation spectrum dissimilarity measurement. Permutation spectrum (PS) features can be directly applied to the original time series without data transformation, and the calculation method is simple and quick. Dissimilarity measurement is a feature evaluation index based on the relative frequency of permutation patterns (PP), used for selecting the optimal parameter combination of PS.

[0006] Therefore, based on the vibration signals generated by mechanical equipment during the manufacturing and testing phases and the actual use phases, the arrangement patterns and arrangement spectrum characteristics of time series at different stages are obtained through phase space reconstruction. On this basis, selecting the optimal combination of feature parameters using a method based on difference measurement, and implementing early fault diagnosis of mechanical equipment under multiple operating conditions during the use phase using a multi-class Support Vector Machine (MSVM) is of great significance. Based on this, this invention provides a cross-stage, multi-operating condition early fault diagnosis method based on arrangement spectrum difference measurement. According to the embodiment of this invention, the cross-stage, multi-operating condition early fault diagnosis method based on arrangement spectrum difference measurement divides the original vibration signal into time series of a certain length. Based on this, phase space reconstruction is performed on the time series using different parameter combinations, the arrangement patterns of each phase space are calculated, and the frequency of each arrangement pattern is counted. The relative frequency distribution of the arrangement patterns is called the arrangement spectrum, which can be directly used as the input feature of the fault diagnosis model. A time series difference measurement method based on relative probability difference and pattern contribution is proposed, and the optimal parameter combination is selected using the difference measurement and pattern standard deviation. A MSVM classification model is trained using vibration signals from the manufacturing and testing phase. Based on this model, early fault diagnosis of mechanical equipment under multiple working conditions across different stages is achieved using vibration signals from the actual use phase, providing a technical basis for subsequent health management and other related tasks.

[0007] According to one embodiment of the present invention, a cross-stage multi-condition early fault diagnosis method based on permutation spectrum difference measurement is proposed, comprising the following steps:

[0008] Step 1: Signal Acquisition and Phase Space Reconstruction: The original vibration signals of the mechanical equipment under different conditions during the manufacturing and testing phases and the actual use phases are acquired using vibration sensors. For a given length of vibration signal, the vibration signal is divided into multiple subsequences by combining characteristic parameters, and the subsequences are reconstructed into the original phase space form.

[0009] Step 2: Constructing the arrangement pattern and symbol phase space: Obtain all possible arrangement patterns based on the embedding dimension of the vibration signal, and convert the original phase space of the vibration signal into a symbol phase space;

[0010] Step 3: Obtain the permutation spectrum features: Statistically count the number of occurrences of various permutation patterns in the symbol phase space, and obtain the corresponding relative frequency distribution, which is the permutation spectrum feature;

[0011] Step 4: Differentiation Measurement: Establish a difference measurement function to calculate the degree of difference between one-dimensional vibration signals collected by mechanical equipment under different conditions;

[0012] Step 5: Feature Parameter Selection: Establish a parameter selection model using the standard deviation of the arrangement spectrum features of multiple one-dimensional vibration signals and the degree of difference between the one-dimensional vibration signals, and select the optimal parameter combination;

[0013] Step Six: Cross-Stage Fault Diagnosis: Calculate the permutation spectrum characteristics of each time series, i.e., the one-dimensional vibration signal, using the optimal parameter combination. Input the permutation spectrum characteristics of the manufacturing and testing stage data into a multi-class SVM classifier to obtain a training model. Input the permutation spectrum characteristics of the actual use stage data into the obtained training model to identify the fault type and operating condition corresponding to the time series.

[0014] Optionally, the manufacturing and testing phase described in step one mainly refers to a series of break-in and performance tests conducted in the factory after all assembly work is completed before the mechanical equipment is put into production and use. The vibration signals collected during this phase are denoted as dataset T.

[0015] Optionally, the actual use stage described in step one mainly refers to the process of putting the mechanical equipment into actual use after production, and the vibration signals collected in this stage are denoted as dataset A.

[0016] Optionally, the combination of characteristic parameters mentioned in step one mainly refers to the parameters used in the phase space reconstruction of the original signal, including the following three parameters [m, τ, N], the definitions of which are explained in detail below.

[0017] The embedding dimension *m* represents the size of the moving window during phase space reconstruction, determining the structure and dimensionality of the phase space's spectral features. A value that is too large affects the model's computation speed and amplifies the influence of noise in the signal; a value that is too small fails to reflect the signal's potential fluctuations. In optional implementations, a range of *m* = 3 to 6 is chosen for the parameter discussion to obtain more accurate and appropriate results.

[0018] The time delay τ is a downsampling method that determines the shape of the permutation spectrum; different τ values ​​can yield different wave patterns for the same sequence. In optional implementations, τ = 1 to 20 can be selected as the range for parameter discussion.

[0019] The data length N is the length of data required to compute features, and its value is affected by the parameter embedding dimension m and the time delay τ. To ensure at least m! permutations are generated, the minimum length should satisfy N. min =m! +(m-1)τ.

[0020] Optionally, the reconstruction of the subsequence into the original phase space described in step one mainly refers to the reconstruction of the time series, which may further include the following steps:

[0021] For the vibration signal, a time series X = {x1, x2, ..., xn} with a data length of N is selected according to the parameter combination. N Reconstruct the phase space. Divide the time series X into N-(m-1)τ subsequences using the parameter embedding dimension m and the time delay τ. These subsequences are constructed into the original phase space S as shown below. m :

[0022]

[0023]

[0024] In the formula, Let i represent the subsequences, i = 1, 2, ..., N-(m-1)τ, and the length of each subsequence is the parameter embedding dimension, which is m.

[0025] Optionally, the arrangement pattern mentioned in step two mainly refers to the data fluctuation pattern determined according to the parameter embedding dimension m, and the arrangement pattern can be represented as follows:

[0026]

[0027] In the formula, r j Let j represent positive integers from 1 to m, where j = 1, 2, ..., m, r1, r2, ..., r j ,…,r m This represents a way to combine positive integers from 1 to m, where m is the parameter embedding dimension. Represents a subsequence with parameter embedding dimension m and time delay τ. The corresponding arrangement pattern.

[0028] Symbolic phase space Π m Each row in the dataset represents a permutation pattern. Different rows may have the same permutation pattern. Since each row has m elements and the elements are unique, the time series contains at most m! permutation patterns. Each permutation pattern is denoted as Π. m (k), where k represents the sequence number of the permutation pattern, k = 1, 2, ..., m!. For example, when m = 3, there are m! = 6 permutation patterns, namely Π 3(1) = [1,2,3], Π 3 (2) = [1,3,2], Π 3 (3) = [2, 1, 3], Π 3 (4) = [2,3,1], Π 3 (5) = [3, 1, 2], Π 3 (6) = [3,2,1].

[0029] Optionally, the symbolic phase space mentioned in step two refers to the subsequences in the original phase space. Converted into various permutation patterns The method, the symbolic phase space is denoted as Π m The steps are as follows:

[0030]

[0031]

[0032] Each subsequence Generated permutation pattern The following conditions must be met:

[0033] 1)

[0034] 2)

[0035] In the formula, l represents the index of an element in the subsequence, and m is the parameter embedding dimension. Condition 1) means arranging the elements in the subsequence in ascending order and representing them with positive integers from 1 to m, i.e., r1, r2, ..., rm. j ,…r m This represents positive integers from 1 to m; condition 2) indicates that when two elements have the same value, they are arranged according to their index. For example, for the sequence... The elements in the sequence satisfy x2 ≤ x3 ≤ x1, therefore its permutation pattern is as follows:

[0036]

[0037] By converting the original phase space into a symbolic phase space through step two above, it becomes easier to count the number of permutation patterns in the next step.

[0038] Optionally, the permutation spectrum features mentioned in step three mainly refer to feature vectors based on the number of permutation patterns. The calculation steps are as follows:

[0039] Statistical symbol phase space Π m Each permutation pattern Π m The number of occurrences of (k), denoted as k represents the sequence number of the permutation pattern, k = 1, 2, ..., m!. Then Π m The relative frequency corresponding to (k) can be expressed as:

[0040]

[0041] The symbolic phase space Π will be utilized m The relative frequency distributions of various modes obtained This is called the permutation spectrum feature, where m is the parameter embedding dimension, τ is the time delay, N is the data length, and k represents the permutation pattern number, k = 1, 2, ..., m!.

[0042] Step three above is the process of constructing the spectral features, which extracts key features from the perspective of the time series arrangement order. Its advantage is that it utilizes more effective information to obtain more accurate diagnostic results.

[0043] Optionally, the difference measurement function mentioned in step four mainly refers to a method for measuring the distinguishability of two vibration signals with different fault types, which can be expressed as:

[0044]

[0045] In the formula, X1 and X2 represent two vibration signals, and These represent the arrangement patterns Π in the two vibration signals, respectively. m The relative probability of (k), also known as the relative frequency. F(Π) represents the probability difference between permutation patterns; a larger value indicates a smaller similarity between signals. m (k) represents the permutation pattern Π m The weighting factor of (k). d m,τ (X1, X2) represents the difference between two vibration signals X1 and X2; the larger the value, the easier it is to distinguish between the two types of signals. The contribution of different arrangement patterns to the arrangement spectrum can be expressed as:

[0046]

[0047] In the formula, and The Π represents the arrangement pattern in the two vibration signals X1 and X2, respectively. m The Shannon entropy of (k). F(Π) m (k) represents the degree of contribution of different permutation patterns to the permutation spectrum. The more frequently a pattern appears, the greater its corresponding weight.

[0048] By utilizing the differences between vibration signals collected from mechanical equipment under different conditions during the manufacturing and testing phase in step four, a difference measurement function is proposed. Its advantage is that it can maximize the differences between the spectral characteristics of different signal arrangements, thereby making it easier to distinguish different fault types.

[0049] Optionally, the parameter selection model mentioned in step five mainly refers to the model for selecting the optimal combination of parameters, and the steps are as follows:

[0050] a) Collect different types of vibration signals B during the manufacturing and testing phase. For each type of vibration signal, obtain z1 time series of length N and set them as training groups. Collect different types of vibration signals during the actual use phase. For each type of vibration signal, obtain z2 time series of length N and set them as verification groups. Here, z1 + z2 = z, and z is the total number of groups for each type of vibration signal.

[0051] b) Calculate the permutation spectrum P of each of the z1 time series of the vibration signals of type B in training group B. Π,b (a), where B represents a total of B types of vibration signals, b represents the b-th vibration signal among the B types of vibration signals, b = 1, 2, ..., B, a represents the a-th time series in the training group and a = 1, 2, ..., z1, P ∏,b (a) represents the permutation spectrum of the a-th time series of the b-th vibration signal in the training group;

[0052] c) Calculate the standard deviation of the z1 time series permutations of each vibration signal in the training group, denoted as Std. b The larger the value, the more unstable the data. The mean of the difference measure between any two vibration signals is calculated.

[0053]

[0054] In the formula, b1 = 1, 2, ..., B, b2 = 1, 2, ..., B, and b1≠b2, X b1,a and X b2,a These represent two vibration signals X. b1 and X b2 The a-th time series, where a represents the a-th time series in the training group and a = 1, 2, ..., Z1, and b1 and b2 represent the b1-th and b2-th vibration signals in the B-type vibration signals, respectively.

[0055] d) Calculate the parameter selection index for any two vibration signals:

[0056]

[0057] In the formula, Ob1,b2 Represents two vibration signals X b1 and X b2 The parameter selection index, Std b1 and Std b2 These represent two vibration signals X. b1 and X b2 The standard deviation of the spectrum of the first z1 time series (training group).

[0058] e) When more than two vibration signals exist, it is necessary to compare multiple fault types. Calculate the parametric selection exponent for any two vibration signals and sum them. The final parametric selection exponent for all types of vibration signals under the parameters [m,τ] is denoted as... The above example uses a z1:z2 ratio to set the training and validation groups, with a total sample size of z. For instance, when z = 100 and z1:z2 = 20:80, the training group consists of 20 samples (20 time series), and the validation group consists of 80 samples, for a total of 100 samples. It should be understood that other suitable sample sizes and time series data and parameters can be selected based on different situations and needs; simply make the corresponding adjustments in the above steps.

[0059] The above step five proposes a parameter selection method. Its advantage is that by using the selected optimal parameter combination, the extracted permutation spectrum features can maximize the differentiation of signals in different states, thereby helping to distinguish different fault types.

[0060] Optionally, the "multi-class SVM classifier" mentioned in step six mainly refers to a multi-class data classification method. The model is trained using the training set data and its effectiveness is verified using the validation set data.

[0061] By utilizing the permutation spectrum features obtained through the optimal parameter combination in step six, a cross-stage diagnostic model is constructed. Under the condition that there are a small number of vibration signal samples from the actual use stage, the state of the mechanical equipment at this time can be accurately identified, thus solving the problem of few samples.

[0062] The method provided by the above embodiments of the present invention utilizes vibration signals from a portion of the manufacturing and testing phase to establish a cross-stage early fault diagnosis model for mechanical equipment, spanning from the manufacturing and testing phase to the actual use phase. The proposed permutation spectrum features and their parameter optimization method are computationally simple, capable of reflecting potential underlying data fluctuations in the time series, and timely identifying impact fluctuations caused by faults in different parts of the signal, as well as data changes due to variations in operating conditions. This enables multi-condition early fault diagnosis of mechanical equipment, providing reasonable guidance for on-site personnel and gaining valuable maintenance and diagnostic time.

[0063] The implementation method of this invention has at least the following advantages. This invention employs a cross-stage, multi-condition early fault identification method based on permutation spectrum difference, which can identify early faults in equipment under small sample conditions based on the underlying fluctuations of different fault types and operating conditions. Through a difference measurement function and parameter selection model, the optimal parameter combination suitable for different fault types can be selected. The permutation pattern and permutation spectrum characteristics of the signal calculated based on the optimal parameter combination can be directly applied to the original time series without data transformation, making the calculation method simple and time-efficient. Compared with traditional entropy-based feature methods, the cross-stage, multi-condition early fault diagnosis method based on permutation spectrum proposed in this invention can retain more underlying fluctuation information of the time series, being sensitive not only to the impact information caused by faults but also to signal fluctuations caused by changes in operating conditions. Since the permutation pattern focuses more on the sequential relationship between adjacent data values ​​rather than the numerical values ​​themselves, it avoids dependence on amplitude thresholds and has strong noise resistance. This method is a guiding technology for cross-stage, multi-condition identification and early fault type diagnosis of mechanical equipment, characterized by good noise resistance, simple calculation, and high diagnostic rate. The method has a certain degree of openness in practical applications and is suitable for real-time status detection of mechanical equipment during operation and post-operation fault detection. Attached Figure Description

[0064] The foregoing features of the invention will be more readily understood in conjunction with the accompanying drawings and the following detailed description, wherein:

[0065] Figure 1 A flowchart of a cross-stage multi-condition early fault diagnosis method based on permutation spectrum difference degree according to an embodiment of the present invention is shown.

[0066] Figure 2 A flowchart of a feature extraction and parameter selection method based on permutation pattern and permutation spectrum according to an embodiment of the present invention is shown;

[0067] Figure 3 The time-domain plots of 16 types of vibration signals during a test phase of a certain type of oil pump are shown in an example of an application according to an embodiment of the present invention.

[0068] Figure 4 The diagram shows the arrangement spectrum of 16 types of vibration signals during a test phase of a certain type of oil pump, as illustrated in an example of an application according to an embodiment of the present invention.

[0069] Figure 5 A three-dimensional plot showing the variation of variability with parameters is shown in an example of an application according to an embodiment of the present invention;

[0070] Figure 6The diagram illustrates a confusion matrix of diagnostic results for an early-stage fault in the actual use of a certain type of oil pump, taking into account both operating condition variations and fault types, in an example of an application according to an embodiment of the present invention.

[0071] Figure 7 The diagram illustrates a confusion matrix of diagnostic results for a certain type of oil pump during actual use, considering only the fault type and not changes in operating conditions, in an example of an application according to an embodiment of the present invention.

[0072] The serial numbers, symbols, and codes in the diagram are explained as follows:

[0073] Con 0: As an example, this describes the operating condition of a certain type of oil pump at a speed of 2000 rpm.

[0074] Con 1: As an example, the operating condition of a certain type of oil pump is 1975 rpm.

[0075] Con 2: As an example, the operating condition of a certain type of oil pump is 1950 rpm.

[0076] Con 3: As an example, the operating condition of a certain type of oil pump is 1925 rpm.

[0077] Normal: This is an example of the operating state of a certain type of oil pump, indicating a normal state.

[0078] Fault A: This is an example of the operating status of a certain type of oil pump, representing a Type I fault.

[0079] Fault B: As an example, this describes the operating status of a certain type of oil pump, representing a type 2 fault.

[0080] Fault C: As an example, this represents the operating status of a certain type of oil pump, indicating a Type III fault.

[0081] m: A parameter representing the embedding dimension of the permutation spectral features.

[0082] τ: A parameter representing the permutation spectrum characteristic, indicating the time delay.

[0083] N: A parameter representing the permutation spectrum characteristic, indicating the data length.

[0084] PS1: Arrangement spectrum characteristics of vibration signal X1

[0085] PS2: Arrangement spectrum characteristics of vibration signal X2

[0086] Std1: Standard deviation of PS1 permutation spectrum characteristics

[0087] Std2: Standard deviation of the PS2 permutation spectrum characteristics Detailed Implementation

[0088] With reference to the accompanying drawings, a specific implementation method is described in detail in one embodiment of the present invention, but the present invention is not limited to the specific implementation method.

[0089] The following describes in detail, with reference to the accompanying drawings, an early fault diagnosis method for multiple operating conditions across stages based on permutation spectrum difference measurement according to an embodiment of the present invention. Figure 1 A flowchart of a cross-stage, multi-condition early fault diagnosis method based on permutation spectrum difference degree according to an embodiment of the present invention is shown. Figure 2 A flowchart illustrating a feature extraction and parameter selection method based on permutation patterns and permutation spectra according to an embodiment of the present invention is shown. (Refer to...) Figure 1-2 This invention proposes a cross-stage, multi-condition early fault diagnosis method based on permutation spectrum difference measurement, comprising the following steps: Step 1: Signal acquisition and phase space reconstruction: Using vibration sensors, original vibration signals of mechanical equipment under different states during manufacturing testing and actual use are acquired. For a given length of vibration signal, the vibration signal is divided into multiple subsequences through feature parameter combinations, and these subsequences are reconstructed into the original phase space form; Step 2: Construction of permutation patterns and symbolic phase space: All possible permutation patterns are obtained based on the embedding dimension of the vibration signal, and the original phase space of the vibration signal is converted into a symbolic phase space; Step 3: Obtaining permutation spectrum features: The frequency of occurrence of various permutation patterns in the symbolic phase space is analyzed. The statistical analysis yields the corresponding relative frequency distribution, which is the permutation spectrum feature. Step four: Differentiation measurement: Establish a difference measurement function to calculate the difference between vibration signals collected from mechanical equipment under different conditions. Step five: Feature parameter selection: Utilize the standard deviation of the permutation spectrum features of multiple vibration signals and the difference between the vibration signals to establish a parameter selection model and select the optimal parameter combination. Step six: Cross-stage fault diagnosis: Calculate the permutation spectrum features of each time series, i.e., the vibration signal, using the optimal parameter combination. Input the permutation spectrum features of the manufacturing and testing stage data into a multi-class SVM classifier to obtain a training model, and input the permutation spectrum of the actual use stage data into the obtained training model to identify the fault type and operating condition corresponding to the time series.

[0090] Optionally, the manufacturing and testing phase described in step one mainly refers to a series of break-in and performance tests conducted in the factory after all assembly work is completed before the mechanical equipment is put into production and use. The vibration signals collected during this phase are denoted as dataset T. Optionally, the vibration signal is a one-dimensional vibration signal when a single sensor is used for signal acquisition. Alternatively, in other embodiments, when multiple sensors are required for signal acquisition, it can also be a multi-dimensional vibration signal.

[0091] Optionally, the actual use stage described in step one mainly refers to the process of putting the mechanical equipment into actual use after production, and the vibration signals collected in this stage are denoted as dataset A.

[0092] Optionally, the combination of characteristic parameters mentioned in step one mainly refers to the parameters used in the phase space reconstruction of the original signal, including the following three parameters [m, τ, N], the definitions of which are explained in detail below.

[0093] The embedding dimension *m* represents the size of the moving window during phase space reconstruction, determining the structure and dimensionality of the phase space's spectral features. A value that is too large affects the model's computation speed and amplifies the influence of noise in the signal; a value that is too small fails to reflect the signal's potential fluctuations. In optional implementations, a range of *m* = 3 to 6 is chosen for the parameter discussion to obtain more accurate and appropriate results.

[0094] The time delay τ is a downsampling method that determines the shape of the permutation spectrum; different τ values ​​can yield different fluctuation patterns for the same sequence. In optional implementations, τ can be selected as a range of 1 to 20 for discussion. It should be understood that in other implementations, this number can be adjusted according to the specific application requirements.

[0095] The data length N is the length of data required to compute features, and its value is affected by the parameter embedding dimension m and the time delay τ. To ensure at least m! permutations are generated, the minimum length should satisfy N. min =m! +(m-1)τ.

[0096] Optionally, the reconstruction of the subsequence into the original phase space described in step one mainly refers to the reconstruction of the time series, which may further include the following steps:

[0097] For the vibration signal, a time series X = {x1, x2, ..., xn} with a data length of N is selected according to the parameter combination. N Reconstruct the phase space. Divide the time series X into N-(m-1)τ subsequences using the parameter embedding dimension m and the time delay τ. These subsequences are constructed into the original phase space S as shown below. m :

[0098]

[0099]

[0100] In the formula, Let i represent the subsequences, i = 1, 2, ..., N-(m-1)τ, and the length of each subsequence is the parameter embedding dimension, which is m.

[0101] Optionally, the arrangement pattern mentioned in step two mainly refers to the data fluctuation pattern determined according to the parameter embedding dimension m, and the arrangement pattern can be represented as follows:

[0102]

[0103] In the formula, r j Let r1, r2, ..., r be positive integers from 1 to m, where j = 1, 2, ..., m. j ,…,r m This represents a way to combine positive integers from 1 to m, where m is the parameter embedding dimension. Represents a subsequence with parameter embedding dimension m and time delay τ. The corresponding arrangement pattern.

[0104] Symbolic phase space Π m Each row in the dataset represents a permutation pattern. Different rows may have the same permutation pattern. Since each row has m elements and the elements are unique, the time series contains at most m! permutation patterns. Each permutation pattern is denoted as Π. m (k), where k represents the sequence number of the permutation pattern, k = 1, 2, ..., m!. For example, when m = 3, there are m! = 6 permutation patterns, namely Π 3 (1) = [1,2,3], Π 3 (2) = [1,3,2], Π 3 (3) = [2, 1, 3], Π 3 (4) = [2,3,1], Π 3 (5) = [3, 1, 2], Π 3 (6) = [3,2,1].

[0105] Optionally, the symbolic phase space mentioned in step two refers to the subsequences in the original phase space. Converted into various permutation patterns The method, the symbolic phase space is denoted as Π m The steps to obtain it are as follows:

[0106]

[0107]

[0108] Each subsequence Generated permutation pattern The following conditions must be met:

[0109] 3)

[0110] 4)

[0111] In the formula, l represents the index of an element in the subsequence. Condition 1) indicates that the elements in the subsequence are arranged in ascending order and represented by positive integers from 1 to m, where m is the parameter embedding dimension; Condition 2) indicates that when two elements have the same value, they are arranged according to their index. For example, for the sequence... The elements in the sequence satisfy x2 ≤ x3 ≤ x1, therefore its permutation pattern is as follows:

[0112]

[0113] By converting the original phase space into a symbolic phase space through step two above, it becomes easier to count the number of permutation patterns in the next step.

[0114] Optionally, the permutation spectrum features mentioned in step three mainly refer to feature vectors based on the number of permutation patterns. The calculation steps are as follows:

[0115] Statistical symbol phase space Π m Each permutation pattern Π m The number of occurrences of (k), denoted as k represents the sequence number of the permutation pattern, k = 1, 2, ..., m!. Then Π m The relative frequency corresponding to (k) can be expressed as:

[0116]

[0117] The symbolic phase space Π will be utilized m The relative frequency distributions of various modes obtained This is called the permutation spectrum feature, where m is the parameter embedding dimension, τ is the time delay, N is the data length, and k represents the permutation pattern number, k = 1, 2, ..., m!.

[0118] Step three above is the process of constructing the spectral features, which extracts key features from the perspective of the time series arrangement order. Its advantage is that it utilizes more effective information to obtain more accurate diagnostic results.

[0119] Optionally, the difference measurement function mentioned in step four mainly refers to a method for measuring the distinguishability of two vibration signals with different fault types, which can be expressed as:

[0120]

[0121] In the formula, X1 and X2 represent two vibration signals, and These represent the arrangement patterns Π in the two vibration signals, respectively. m The relative probability of (k), also known as the relative frequency. F(Π) represents the probability difference between permutation patterns; a larger value indicates a smaller similarity between signals.m (k) represents the permutation pattern Π m The weighting factor of (k). d m,τ (X1, X2) represents the difference between two vibration signals X1 and X2; the larger the value, the easier it is to distinguish between the two types of signals. The contribution of different arrangement patterns to the arrangement spectrum can be expressed as:

[0122]

[0123] In the formula, and The Π represents the arrangement pattern in the two vibration signals X1 and X2, respectively. m The Shannon entropy of (k). F(Π) m (k) represents the degree of contribution of different permutation patterns to the permutation spectrum. The more frequently a pattern appears, the greater its corresponding weight.

[0124] By utilizing the differences between vibration signals collected from mechanical equipment under different conditions during the manufacturing and testing phase in step four, a difference measurement function is proposed. Its advantage is that it can maximize the differences between the spectral characteristics of different signal arrangements, thereby making it easier to distinguish different fault types.

[0125] Optionally, the parameter selection model mentioned in step five mainly refers to the model for selecting the optimal combination of parameters, and the steps are as follows:

[0126] a) Collect different types of vibration signals B during the manufacturing and testing phase. For each type of vibration signal, obtain z1 time series of length N and set them as training groups. Collect different types of vibration signals during the actual use phase. For each type of vibration signal, obtain z2 time series of length N and set them as verification groups. Here, z1 + z2 = z, and z is the total number of groups for each type of vibration signal.

[0127] b) Calculate the permutation spectrum P of each of the z1 time series of the vibration signals of type B in training group B. Π,b (a), where B represents a total of B types of vibration signals, b represents the b-th vibration signal among the B types of vibration signals, b = 1, 2, ..., B, a represents the a-th time series in the training group and a = 1, 2, ..., z1, P ∏,b (a) represents the permutation spectrum of the a-th time series of the b-th vibration signal in the training group;

[0128] c) Calculate the standard deviation of the z1 time series permutations of each vibration signal in the training group, denoted as Std. b The larger the value, the more unstable the data. The mean of the difference measure between any two vibration signals is calculated.

[0129]

[0130] In the formula, b1 = 1, 2, ..., B, b2 = 1, 2, ..., B, and b1≠b2, X b1,a and X b2,a These represent two vibration signals X. b1 and X b2 The a-th time series, where a represents the a-th time series in the training group and a = 1, 2, ..., z1, and b1 and b2 represent the b1-th and b2-th vibration signals in the B-type vibration signals, respectively.

[0131] d) Calculate the parameter selection index for any two vibration signals:

[0132]

[0133] In the formula, O b1,b2 Represents two vibration signals X b1 and X b2 The parameter selection index, Std b1 and Std b2 These represent two vibration signals X. b1 and X b2 The standard deviation of the spectrum of the first z1 time series (training group).

[0134] e) When more than two vibration signals exist, it is necessary to compare multiple fault types. Calculate the parametric selection exponent for any two vibration signals and sum them. The final parametric selection exponent for all types of vibration signals under the parameters [m,τ] is denoted as... The above example sets the z1:z2 ratio for the training and validation groups, with a total sample size of z. For instance, if z = 100 and z1:z2 = 20:80, the training group consists of 20 samples (20 time series), and the validation group consists of 80 samples, for a total of 100 samples. It should be understood that other suitable sample sizes and time series data and parameters can be selected based on different situations and needs; simply make the corresponding adjustments in the above steps.

[0135] The above step five proposes a parameter selection method. Its advantage is that by using the selected optimal parameter combination, the extracted permutation spectrum features can maximize the differentiation of signals in different states, thereby helping to distinguish different fault types.

[0136] Optionally, the "multi-class SVM classifier" mentioned in step six mainly refers to a multi-class data classification method. The model is trained using the training set data and its effectiveness is verified using the validation set data.

[0137] By utilizing the permutation spectrum features obtained through the optimal parameter combination in step six, a cross-stage diagnostic model is constructed. Under the condition that there are a small number of vibration signal samples from the actual use stage, the state of the mechanical equipment at this time can be accurately identified, thus solving the problem of few samples.

[0138] The following describes in detail, with reference to the accompanying drawings, an exemplary embodiment of an early fault diagnosis method for multiple operating conditions across stages based on permutation spectrum difference measurement according to an embodiment of the present invention.

[0139] Figure 3 The diagram shows a time-domain plot of 16 types of vibration signals during a test phase of a certain type of oil pump, as illustrated in an example of an application according to an embodiment of the present invention. Figure 4 The diagram shows the arrangement spectrum of 16 types of vibration signals during a test phase of a certain type of oil pump, as illustrated in an example of an application according to an embodiment of the present invention. Figure 5 A three-dimensional plot showing the variation of variability with parameters in an example application according to an embodiment of the present invention is shown. Figure 6 The diagram illustrates a confusion matrix of diagnostic results for an early stage of actual use of a certain type of oil pump, which simultaneously considers changes in operating conditions and fault types, in an example of an application according to an embodiment of the present invention. Figure 7 The diagram illustrates a confusion matrix of diagnostic results for a certain type of oil pump during actual use, considering only the fault type and not changes in operating conditions, in an example of an application according to an embodiment of the present invention.

[0140] Reference Figure 3-7 In an exemplary embodiment of the present invention, vibration signals from a servo oil pump during the manufacturing and testing phases and during actual use are used as experimental data for analysis. Besides the normal state, the oil pump experiences three critical friction pairs with faults classified as Fault A, Fault B, and Fault C. A three-phase acceleration vibration sensor is used to collect the vibration signals of the oil pump at a sampling frequency of 25600 Hz.

[0141] Each fault type operates under four conditions: 2000 rpm (Con0), 1975 rpm (Con1), 1950 rpm (Con2), and 1925 rpm (Con3). Therefore, including normal data, 16 data points are obtained, as shown in Table 1.

[0142] Table 1. Description of Vibration Signal Set

[0143]

[0144] Figure 1A flowchart of a cross-stage, multi-condition early fault diagnosis method based on permutation spectrum difference degree according to an embodiment of the present invention is shown. Figure 2 A flowchart of a feature extraction and evaluation method based on permutation patterns and permutation spectra according to an embodiment of the present invention is shown. The following references... Figure 1 and Figure 2 This paper describes a cross-stage multi-condition early fault diagnosis method based on permutation spectrum difference, according to one embodiment of the present invention. The cross-stage multi-condition early fault diagnosis method based on permutation spectrum difference according to one embodiment of the present invention includes the following steps:

[0145] Step 1: Signal Acquisition and Phase Space Reconstruction. Based on the experiment, 16 types of vibration signals were collected from the oil pump during both the manufacturing and testing phases and the actual usage phase. Figure 3 The original vibration signals of 16 types of mechanical equipment during the manufacturing and testing phase are shown in time-domain waveforms. The original vibration signals are converted into their original phase-space form based on the combination of characteristic parameters.

[0146] Step 2: Construct the permutation patterns and the symbolic phase space. Taking an embedding dimension of m=4 as an example, we can obtain m!=24 permutation patterns. Based on the permutation patterns, we convert the original phase space obtained in Step 1 into a symbolic phase space.

[0147] Step 3: Obtain Permutation Spectrum Characteristics: Statistically analyze the frequency and relative occurrence of various permutation patterns in the symbol phase space for each sample data point. The permutation spectra of various fault types are displayed in... Figure 4 .

[0148] Step 4: Dissimilarity Measurement: Based on the established dissimilarity measurement function, calculate the degree of dissimilarity between different types of data. Taking Normal0 and Fault A-0 data as examples, the degree of dissimilarity changes with parameter combinations as follows: Figure 5 As shown.

[0149] Step 5: Feature Parameter Selection: For each type of fault data, 20 time series with length N = 4096 were selected from the manufacturing and testing phase as the training group, and 80 time series with length N = 4096 were selected from the actual use phase as the validation group. Based on the established parameter selection model, the optimal parameter combination obtained in this paper is [m, τ, N] = [4, 11, 4096].

[0150] Step Six: Diagnose the oil pump fault data from two perspectives: considering operating conditions and not considering operating conditions. When considering operating conditions, the corresponding label is "label-ALL" in Table 1; when not considering operating conditions, i.e., only diagnosing the faulty part of the oil pump, the corresponding label is "label-4type" in Table 1. The diagnostic results obtained from the two perspectives are 97.16% and 100%, respectively. The confusion matrix diagrams of the diagnostic results are shown below. Figure 6 and Figure 7 As shown in the figure, the proposed permutation spectrum difference method can not only accurately identify data of different fault types, but is also highly sensitive to the operating condition information in the data. Figure 6 In this invention, the method only misidentifies a few Type I and Type II faults, but can guarantee a diagnosis rate of over 90% for each type.

[0151] Based on the above diagnostic results, relevant personnel can promptly identify anomalies and accurately determine the current fault type and operating condition during the early stages of mechanical equipment operation, avoiding misjudgments caused by amplitude changes due to variations in operating conditions. The diagnostic results can guide on-site personnel to repair or replace the mechanical equipment, ensuring the safe and effective operation of the entire mechanical system.

[0152] This invention proposes a cross-stage, multi-condition early fault diagnosis method based on permutation spectrum difference. It utilizes vibration signals from the product during the manufacturing and testing phase to promptly detect different data fluctuation patterns caused by faults and changes in operating conditions at the initial stage of actual use. Permutation spectrum characteristics are then derived based on these patterns. These characteristics are highly sensitive to data changes and are computationally simple, meeting the real-time requirements of mechanical equipment condition monitoring. The proposed model is applicable not only to the early fault diagnosis of the oil pump in the example but also to other mechanical equipment and their key components. This invention has good scalability and provides valuable reference for other skilled personnel in this field.

[0153] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0154] Furthermore, the foregoing only describes some embodiments, and changes, modifications, additions, and / or variations can be made without departing from the scope and spirit of the disclosed embodiments. These embodiments are illustrative and not restrictive. Moreover, the described embodiments relate to those currently considered most practical and preferred, and should be understood as not being limited to the disclosed embodiments, but rather intended to cover different modifications and equivalent arrangements included within the spirit and scope of those embodiments. Furthermore, the various embodiments described above can be used in conjunction with other embodiments; for example, an aspect of one embodiment can be combined with an aspect of another embodiment to implement yet another embodiment. Additionally, individual features or components of any given component can constitute another embodiment.

[0155] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A cross-stage, multi-condition early fault diagnosis method based on permutation spectrum difference measurement, characterized in that... Includes the following steps: Step 1: Signal Acquisition and Phase Space Reconstruction: The original vibration signals of the mechanical equipment under different conditions during the manufacturing and testing phases and the actual use phases are acquired using vibration sensors. For a given length of vibration signal, the vibration signal is divided into multiple subsequences by combining characteristic parameters, and the subsequences are reconstructed into the original phase space form. Step 2: Constructing the arrangement pattern and symbol phase space: Obtain all possible arrangement patterns based on the embedding dimension of the vibration signal, and convert the original phase space of the vibration signal into a symbol phase space; Step 3: Obtain the permutation spectrum features: Statistically count the number of occurrences of various permutation patterns in the symbol phase space, and obtain the corresponding relative frequency distribution, which is the permutation spectrum feature; Step 4: Differentiation Measurement: Establish a difference measurement function to obtain the degree of difference between vibration signals collected by mechanical equipment under different conditions; Step 5: Feature Parameter Selection: Establish a parameter selection model using the standard deviation of the arrangement spectrum features of multiple vibration signals and the degree of difference between the vibration signals, and select the optimal parameter combination; Step Six: Cross-Stage Fault Diagnosis: Calculate the arrangement spectrum characteristics of each time series, i.e. the vibration signal, using the optimal parameter combination. Input the arrangement spectrum characteristics of the manufacturing and testing stage data into a multi-class SVM classifier to obtain a training model. Input the arrangement spectrum characteristics of the actual use stage data into the obtained training model to identify the fault type and operating condition corresponding to the time series. Step one further includes the following steps: For the vibration signal, a time series with a data length of N is selected based on the parameter combination. Reconstructing phase space by embedding dimensions through parameters With time delay Time series Divided into Subsequences These subsequences are constructed into the original phase space shown below. : In the formula, This represents the subsequence that has been divided. , The length of each subsequence is also known as the parameter embedding dimension. The time delay is N, and the data length is N. Step two further includes the following steps: The arrangement pattern mentioned refers to the arrangement pattern based on the parameter embedding dimension. The defined data fluctuation pattern is represented by the following arrangement pattern: In the formula, Represents positive integers from 1 to m. , m is the parameter embedding dimension For time delay, Represents a way to combine positive integers from 1 to m. Indicates the parameter embedding dimension and time delay The following subsequence The corresponding arrangement pattern; Step two further includes the following steps: The symbolic phase space refers to the subsequences in the original phase space. Converted into various permutation patterns The method, the symbolic phase space is denoted as The steps are as follows: Symbolic phase space Each row in the array represents a permutation pattern, and different rows may have the same permutation pattern, because each row contains... There are a maximum of 100 elements and no two elements are repeated, so the time series contains at most 100 elements. There are several permutation patterns, each denoted as . , Indicates the sequence number of the arrangement pattern. ; Each subsequence Generated permutation pattern The following conditions must be met: 1) ; 2) In the formula, This represents the index of an element in the subsequence, where m is the parameter embedding dimension. For time delay, N is the data length. Condition 1) means to arrange the elements in the subsequence from smallest to largest and use positive integers from 1 to m. Condition 2) means to arrange the elements according to their index when two elements have the same value.

2. The cross-stage multi-condition early fault diagnosis method based on permutation spectrum difference measurement as described in claim 1, characterized in that... Step three further includes the following steps: The permutation spectrum features mentioned in step three refer to feature vectors based on the number of permutation patterns, which are obtained through the following steps: Statistical symbol phase space Various permutation patterns The number of occurrences is denoted as ,but The corresponding relative frequency is expressed as: The symbolic phase space will be utilized The relative frequency distributions of various modes obtained This is called the permutation spectral feature, where m is the parameter embedding dimension. The time delay is N, and the data length is N. Indicates the sequence number of the arrangement pattern. .

3. The cross-stage multi-condition early fault diagnosis method based on permutation spectrum difference measurement as described in claim 2, characterized in that... Step four further includes the following steps: The difference measurement function mentioned in step four refers to a method for measuring the distinguishability of two vibration signals with different fault types, which is expressed as: In the formula, and These represent two vibration signals. and These represent the arrangement patterns in the two vibration signals, respectively. The relative probability, or relative frequency, This represents the probability difference between permutation patterns; a larger value indicates a lower similarity between signals. Indicates the arrangement pattern Weighting factors Indicates two vibration signals and The greater the difference measure, the easier it is to distinguish between the two types of signals. The contribution of different arrangement patterns to the arrangement spectrum is expressed as follows: In the formula, and These represent two vibration signals respectively. and Middle arrangement pattern Shannon entropy, This represents the contribution of different permutation patterns to the permutation spectrum, where m is the parameter embedding dimension. For time delay, Indicates the sequence number of the arrangement pattern. .

4. The cross-stage multi-condition early fault diagnosis method based on permutation spectrum difference measurement as described in claim 3, characterized in that... The parameter selection model mentioned in step five refers to selecting a model for calculating the optimal parameter combination, and further includes the following steps: a) Collect different types of vibration signals B during the manufacturing and testing phase. For each vibration signal, obtain z1 time series of length N and set them as training groups. Collect different types of vibration signals during the actual use phase. For each vibration signal, obtain z2 time series of length N and set them as verification groups. Here, z1 + z2 = z, where z is the total number of groups for each type of vibration signal. b) Calculate the training groups separately The permutation spectra of the z1 time series of the vibration signals Where B represents a total of B types of vibration signals, and b represents the b-th vibration signal among the B types of vibration signals. , where 'a' represents the a-th time series in the training group and , This represents the permutation spectrum of the a-th time series of the b-th vibration signal in the training group; c) Calculate the standard deviation of the z1 time series permutations of each vibration signal in the training group, denoted as . The larger the value, the more unstable the data. The mean of the difference measure between any two vibration signals is calculated. : In the formula, , , and and These represent two vibration signals respectively. and The There are a time series, where 'a' represents the a-th time series in the training group and... b1 and b2 represent the b1-th and b2-th vibration signals in type B vibration signals, respectively, and m is the parameter embedding dimension. For time delay; d) Calculate the parameter selection index for any two vibration signals: In the formula, Representing two vibration signals and The parameter selection index, and These represent the two vibration signals of the training group, respectively. and forward Standard deviation of the time series permutation spectrum; e) When there are two or more vibration signals, it is necessary to compare multiple fault types, calculate the parameter selection index for any two vibration signals and sum them, and finally determine the parameter selection index for all types of vibration signals. The parameter selection index is denoted as .

Images