Variable working condition motor fault diagnosis method based on vibration signals

Abstract

The invention discloses a variable-working-condition motor fault diagnosis method based on a vibration signal, and the method comprises the steps: collecting the vibration acceleration data of a housing of a motor in various states, carrying out the data preprocessing and feature extraction, constructing a motor fault diagnosis model through employing the extracted features, carrying out the training and parameter optimization of the model, and carrying out the fault diagnosis of the motor. Therefore, the fault diagnosis model capable of quickly and accurately diagnosing the motor fault is obtained and is further used for real-time monitoring and fault diagnosis of the motor, so that the monitoring cost and the time cost are reduced, the accuracy of motor fault diagnosis is improved, and the safety and the stability of a motor system are improved.

Description

Technical Field

[0001] The present invention relates to the fields of motor fault diagnosis, digital signal processing, and machine learning, and specifically to a variable-condition motor fault diagnosis method based on vibration signals. Background Art

[0002] Motors are widely used in industrial production. As a power source, motors require more stable performance. However, during actual operation, they are easily affected by various factors. Once a fault occurs, the resulting losses are incalculable. Therefore, it is of great significance to perform real-time condition monitoring and fault diagnosis on motors to ensure the safe and stable operation of equipment.

[0003] During industrial production, motors often operate in complex and harsh environments. It is simple, convenient, and easy to implement motor fault diagnosis based on the vibration signals collected on the motor housing. In addition, when various faults occur in the motor, the vibration signals often change differently. Therefore, the accuracy of motor fault diagnosis based on vibration signals is relatively high.

[0004] Motors often operate under variable-condition conditions. At this time, the collected vibration signals are non-stationary signals. As the motor speed and load change, the time-domain characteristics and frequency-domain characteristics of the vibration signals will change, which causes difficulties in fault diagnosis based on vibration signals. Using digital signal processing technology to perform stationary processing on the signals from the perspectives of the time domain and the transform domain respectively can remove the influence of working condition changes on the signals, making similar faults exhibit similar signal characteristics under different working conditions, thereby improving the accuracy of fault diagnosis. Specifically, when the motor runs at variable speed, the change in the motor speed is mainly manifested as a slow translation of the vibration signal spectrum, and the change in the vibration signal spectrum is closely related to the change in the motor speed. The right shift of the spectrum corresponds to an increase in the motor speed, and the left shift of the spectrum corresponds to a decrease in the motor speed. When the motor runs under variable load, the change in the motor load is mainly manifested as the mutation of the vibration signal envelope and the short-term fluctuation after the mutation.

[0005] Based on digital signal processing technology to process the vibration signals of motors under variable conditions, starting from the perspectives of fault mechanisms and signal characteristics, it is possible to accurately extract the motor state characteristics, thereby accurately identifying and diagnosing various types of motor faults. Machine learning algorithms, as intelligent algorithms, can learn sample characteristics from samples and classify the samples. Weighted machine learning algorithms such as random forest can solve the problem of sample imbalance. Using weighted machine learning algorithms for motor fault diagnosis can largely solve the problem of few fault samples. Combining digital signal processing technology and machine learning algorithms for variable-condition motor fault diagnosis based on vibration signals is simple, efficient, has high accuracy, and good interpretability. Summary of the Invention

[0006] The object of the present invention is to provide a variable-condition motor fault diagnosis method based on vibration signals, so as to overcome the problems of difficult acquisition of motor state information and inaccurate fault diagnosis caused by the fact that motors often operate under variable speed and variable load conditions. The present invention designs a processing method for vibration signals under variable conditions of motors, combines a fault diagnosis model with strong generalization ability, and realizes efficient and accurate fault diagnosis of motors under variable conditions based on vibration signals.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A variable-condition motor fault diagnosis method based on vibration signals includes the following steps:

[0009] 1. Data acquisition

[0010] Collect vibration acceleration data on the outer shell of the motor in different healthy states, including normal state and various fault states.

[0011] 2. Data preprocessing

[0012] 2.1 Process missing values and outliers in the data.

[0013] 2.2 Signal stationary processing: Extract the constant load signal according to the vibration signal envelope, estimate the rotation frequency according to the frequency spectrum of the vibration signal, and then perform equiangular resampling on the signal according to the rotation frequency.

[0014] 3. Feature extraction

[0015] Perform wavelet packet decomposition on the vibration signal after preprocessing. The formula is as follows:

[0016] w i+1，2j (n) = ∑ k h(k)w i，j (2n - k)

[0017] w i+1，2j+1 (n) = ∑ k g(k)w i，j (2n - k)

[0018] Among them, w i，j (n) is the wavelet packet decomposition coefficient of the i-th level and j-th node, w 0，0 (n) is the original vibration signal x(n), and h(k) and g(k) are the unit impulse responses of the wavelet low-pass filter and the wavelet high-pass filter respectively.

[0019] Calculate the total energy of the signal. The formula is as follows:

[0020]

[0021] Among them, E is the total energy of the signal, and x(n) is the vibration signal.

[0022] Calculate the energy of each node at the last level of wavelet packet decomposition. The formula is as follows:

[0023] E i，j = ∑ n |w i，j (n)| 2

[0024] Where E i，j is the energy of node j at the i-th level, and w i，j (n) is the wavelet packet decomposition coefficient of node j at the i-th level.

[0025] Calculate the ratio of the energy of each node at the last level of wavelet packet decomposition to the total energy of the signal. The formula is as follows:

[0026] R i，j = E i，j / E

[0027] Where R i，j is the ratio of the energy of node j at the i-th level to the total energy of the signal, E i，j is the wavelet packet energy of node j at the i-th level, and E is the total energy of the signal.

[0028] Use the energy distribution of each node at the last level of wavelet packet decomposition to construct the feature vector of the sample X = [R i,0 , R i,1 , R i,2 ,...].

[0029] 4. Construct the motor state feature dataset

[0030] Construct the motor state feature dataset according to the feature vector and sample label of the sample, and divide it into a training set and a test set in a ratio of 8:2 and perform normalization.

[0031] 5. Build a fault diagnosis model

[0032] Create a random forest classifier, and use the motor state feature dataset to train and optimize the parameters.

[0033] 6. Actual fault diagnosis

[0034] Perform preprocessing and feature extraction on the vibration acceleration signal collected in real time in sequence, and then put it into the constructed fault diagnosis model for fault diagnosis.

[0035] Beneficial effects

[0036] The motor fault diagnosis method of the present invention starts from the correlation between vibration signals and motor states and the changes in motor operating conditions. By segmenting and resampling the vibration signals of the motor during variable-speed and variable-load operation at equal angles, the vibration signals are stabilized from the perspectives of the time domain and the transform domain, greatly removing the influence of variable operating conditions of the motor on the signals, and making the vibration signals of the motor in the same state under different operating conditions exhibit similar time-frequency characteristics. Then, the vibration signals are decomposed by wavelet packet and the energy distribution of different order bands is calculated to construct the feature vectors of the samples, making the feature vectors of different states of the motor have large differences while further ensuring that the feature vectors of the same state of the motor under different operating conditions are similar. Finally, a random forest classifier is used to classify the samples based on the wavelet packet energy features, realizing the stabilization processing and feature extraction of the motor vibration signals under variable operating conditions, and further realizing the fault diagnosis of the motor under variable operating conditions, overcoming the defect of low diagnostic accuracy of traditional diagnostic models during variable operating conditions of the motor, expanding the engineering application scope of the diagnostic model, and improving the generalization ability and fault diagnosis accuracy of the diagnostic model.

[0037] Meanwhile, the present invention combines digital signal processing technology and machine learning theory to propose a fault diagnosis model based on digital signal processing technology and random forest algorithm. By using digital signal processing technologies such as envelope analysis, wavelet analysis, and order analysis to accurately extract fault features, and using a random forest classifier with weights for fault diagnosis, it can overcome the problem of insufficient fault samples faced in the process of motor fault diagnosis to a certain extent while maintaining a high diagnostic accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flowchart of a variable-condition motor fault diagnosis method based on vibration signals according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0039] The following describes the specific embodiments of the present invention with reference to the drawings, so that those skilled in the art can better understand the present invention. It should be particularly noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0040] As Figure 1 shown, a variable-condition motor fault diagnosis method based on vibration signals according to the present invention includes the following steps:

[0041] S1. Data acquisition

[0042] Collect the vibration acceleration signals on the outer shell of the motor in different healthy states, including the normal state and six fault states of bearing fault, gear fault, imbalance fault, stator fault, rotor fault, and shaft misalignment fault, and record them as states 0-6 respectively.

[0043] S2. Data preprocessing

[0044] S2.1 Outlier processing: Use a cosine curve to perform interpolation replacement on the outliers in the data collected in step S1.

[0045] S2.2 Stationarization processing: Perform stationarization processing on the vibration signal after being processed in step S2.1. Specifically:

[0046] (1) Calculate the envelope of the vibration signal x(n) after being processed in step S2.1 to obtain Env(n).

[0047] (2) Use a moving average filter to smooth the envelope to obtain env(n).

[0048] (3) Set a start threshold according to env(n) when the motor is not started. Only take the continuous signal segments with the envelope greater than this threshold as valid signals, and the signals less than this threshold are the noise when the motor is not started.

[0049] (4) Calculate the change rate of env(n) and take the absolute value to obtain absder env (n). Set a change rate threshold according to the actual operating parameters of the motor, and extract the continuous signal segments corresponding to the original signal x(n) where the change rate of absder env (n) is less than this threshold as constant load signals.

[0050] (5) Perform short-time Fourier transform on the constant load signals, and set the frequency resolution of each frame of the signal to 1 Hz.

[0051] (6) Calculate the maximum rotational frequency rf of the motor according to the maximum rotational speed of the motor max , and take out the spectrum with a frequency less than rf max in each frame of the signal.

[0052] (7) Use peak detection technology to estimate the fundamental frequency corresponding to each frame of the signal from the spectrum with a frequency less than rf max in each frame of the signal by setting an appropriate threshold, and use it as the rotational frequency of the motor in each time frame. The fundamental frequencies of all frames are connected together in time sequence to form the rotational frequency of the motor.

[0053] (8) Perform equal-angle resampling on the vibration signal according to the rotational frequency of the motor.

[0054] S3. Feature extraction

[0055] Extract the wavelet packet energy distribution characteristics of the signal. The specific steps are as follows:

[0056] (1) Set the wavelet basis and the wavelet decomposition level L, and perform L-level wavelet packet decomposition on the vibration signal after preprocessing. The formula is as follows:

[0057] w i+1，2j (n) = ∑ k h(k)w i，j (2n - k)

[0058] w i+1，2j+1 (n) = ∑ k g(k)w i，j (2n - k)

[0059] where i = 0, 1, 2, ... L - 1, j = 0, 1, ... i, w i，j (n) is the wavelet packet decomposition coefficient of the j-th node at the i-th level, w 0，0 (n) is the original vibration signal x(n), and h(k) and g(k) are the unit impulse responses of the wavelet low-pass filter and the wavelet high-pass filter, respectively.

[0060] (2) Calculate the total energy of the signal, and the formula is as follows:

[0061]

[0062] where E is the total energy of the signal, x(n) is the vibration signal, and N is the signal length.

[0063] (3) Calculate the energy of each node at the last level of the wavelet packet decomposition, and the formula is as follows:

[0064] E j = ∑ n |w L，j (n)| 2

[0065] where j = 0, 1, ... 2 L - 1, E j is the energy of the j-th node at the L-th level, and w L，j (n) is the wavelet packet decomposition coefficient of the j-th node at the L-th level.

[0066] (4) Calculate the ratio of the energy of the wavelet packet coefficients of each node at the last level of the wavelet packet decomposition to the total energy of the signal, and the formula is as follows:

[0067] R j = E j / E

[0068] where R j is the ratio of the energy of the j-th node at the L-th level to the total energy of the signal, E j is the energy of the j-th node at the L-th level, and E is the total energy of the signal.

[0069] (5) Use the energy distribution of each node at the last level of the wavelet packet decomposition to construct the feature vector of the sample

[0070] S4. Construct the motor state feature dataset

[0071] Use the feature vectors and state numbers of the vibration signals in the seven states of normal motor, bearing fault, gear fault, unbalance fault, stator fault, rotor fault, and shaft misalignment fault to form a feature matrix and a label matrix respectively, and construct the motor state feature dataset.

[0072] S5. Build a motor fault diagnosis model

[0073] Create a random forest classifier as the fault diagnosis model, and use the constructed motor state feature dataset for training and parameter optimization. The parameters to be optimized in this invention can be divided into two categories: feature extraction parameters and classification model parameters. This invention uses a parameter optimization method based on the final classification accuracy to optimize the relevant parameters. The specific steps are as follows:

[0074] (1) Optimize the feature extraction parameters: First, fix the type of wavelet basis, and set the wavelet packet decomposition level to where n is the sample length, represents rounding down. Then perform feature extraction. At each level of the wavelet packet decomposition level, randomly divide the dataset into a training set and a test set in a ratio of 8:2, and use the random forest classifier (default parameters) for training and testing. Repeat the experiment 1000 times, and determine the optimal wavelet packet decomposition level according to the classification accuracy on the test set of the 1000 experiments. Specifically: The one with the highest average value and the smallest variance of the classification accuracy is the optimal. If the accuracy of the optimal wavelet packet decomposition level of this wavelet basis cannot meet the requirements, replace the wavelet basis and repeat the above operations until the optimal wavelet basis and the optimal wavelet packet decomposition level are found.

[0075] (2) Optimize the hyperparameters of the random forest classifier: Fix the feature extraction parameters as the optimal parameters found in step (1), and use the Bayesian search method to find the optimal hyperparameters of the random forest classifier.

[0076] S6. Actual fault diagnosis

[0077] During the actual production process, use a vibration acceleration sensor to collect the vibration acceleration signals on the motor cover at equal time intervals for a period of time. After preprocessing the collected signals, multiple vibration signals under different loads may be obtained. At this time, the signals with too short time should be discarded first. Then, perform feature extraction on the obtained multiple signals respectively and conduct fault diagnosis through the fault diagnosis model. If the diagnosis results are different, the diagnosis result of the signal with the longest time shall prevail. If the diagnosis results are different and the time lengths are the same, perform feature extraction on the vibration signals with the same time length respectively, calculate the average value of the feature vectors, and then put the average feature vector into the fault diagnosis model for fault diagnosis, and take the diagnosis result of the average feature vector as the standard.

Claims

1. A variable working condition motor fault diagnosis method based on vibration signals, characterized in that, It includes the following steps: (1) Data acquisition: Collect the vibration acceleration data on the outer shell of the motor under different health states, including the normal state and multiple fault states; (2) Data preprocessing: Perform outlier processing and stationary processing on the data collected in step (1); (3) Feature extraction: Extract the wavelet packet energy distribution features from the data after step (2); (4) Construct the motor state feature dataset: Construct the motor state feature dataset according to the feature vectors and sample labels extracted in step (3); (5) Build a motor fault diagnosis model: Create a random forest classifier as the fault diagnosis model, and use the motor state feature dataset constructed in step (4) for training and parameter optimization; (6) Actual fault diagnosis: After preprocessing and feature extraction on the real-time collected vibration data in sequence, put it into the constructed fault diagnosis model for fault diagnosis.

2. The variable operating condition motor fault diagnosis method based on vibration signals according to claim 1, wherein The data acquisition in step (1) is specifically as follows: Collect the vibration acceleration signals on the outer shell of the motor under different health states, including the normal state and six fault states of bearing fault, gear fault, imbalance fault, stator fault, rotor fault, and shaft misalignment fault, and record them as states 0 to 6 respectively.

3. The variable working condition motor fault diagnosis method based on vibration signals according to claim 1, wherein, The data preprocessing in step (2) is specifically as follows: (2.1) Outlier processing: Use a cosine curve to perform interpolation replacement processing on the outliers in the data collected in step (1); (2.2) Stationary processing: Perform stationary processing on the vibration signal after step (2.1) processing.

4. The method for diagnosing variable-condition motor faults based on vibration signals according to claim 3, characterized in that The signal stationary processing in step (2.2) is specifically as follows: (2.2.1) Obtain Env(n) by finding the envelope of the vibration signal x(n) after step (2.1) processing; (2.2.2) Use a moving average filter to smooth the envelope to obtain env(n); (2.2.3) Set a start threshold according to env(n) when the motor is not started, and only take the continuous signal segments with the envelope greater than this threshold as valid signals, and the signals less than this threshold are the noise when the motor is not started; (2.2.4) Obtain the change rate of env(n) and take the absolute value to get absder env (n). Set a change rate threshold according to the actual operating parameters of the motor, and extract the continuous signal segments corresponding to the original signal x(n) whose absder env (n) change rate is less than this threshold as constant load signals; (2.2.5) Perform short-time Fourier transform on the constant load signal, and set the frequency resolution of each frame of the signal to 1 Hz; (2.2.6) Calculate the maximum rotational frequency rf of the motor based on the maximum rotational speed of the motor max , and extract the spectrum with a frequency less than rf in each frame of the signal max ; (2.2.7) Using peak detection technology, estimate the fundamental frequency corresponding to each frame of signal from the spectrum where the frequency of each frame of signal is less than rf by setting an appropriate threshold, and use it as the rotational frequency of the motor within each time frame. The fundamental frequencies of all frames are connected together in chronological order to form the rotational frequency of the motor; max ​ (2.2.8) Perform equal-angle resampling on the vibration signal according to the rotation frequency of the motor.

5. The method for diagnosing variable working condition motor faults based on vibration signals according to claim 1, wherein The feature extraction in step (3) is specifically as follows: (3.1) Set the wavelet basis and the wavelet decomposition level L, and perform L-level wavelet packet decomposition on the vibration signal after step (2) processing. The formula is as follows: w i+1，2j (n) = ∑ k h(k)w i，j (2n - k) w i+1，2j+1 (n) = Σ k g(k)w i，j (2n - k) where \(i = 0, 1, 2, \cdots, L - 1\), \(j = 0, 1, \cdots, i\), \(w\) i，j \((n)\) is the wavelet packet decomposition coefficient of the \(j\) node at the \(i\)th level, \(w\) 0，0 \((n)\) is the original vibration signal \(x(n)\), and \(h(k)\) and \(g(k)\) are the unit impulse responses of the wavelet low-pass filter and the wavelet high-pass filter respectively; (3.2) Calculate the total energy of the signal. The formula is as follows: where E is the total energy of the signal, x(n) is the vibration signal, and N is the signal length; (3.3) Calculate the energy of each node at the last level of wavelet packet decomposition. The formula is as follows: E j = ∑ n |w L，j (n)| 2 where j = 0, 1,... 2 L -1, E j is the energy of the j-th node at the L-th level, and w L，j (n) is the wavelet packet decomposition coefficient of the j-th node at the L-th level; (3.4) Calculate the ratio of the energy of the wavelet packet coefficients of each node at the last level of wavelet packet decomposition to the total energy of the signal. The formula is as follows: R j = E j / E (3.5)Construct the eigenvector of the sample using the energy distribution of each node at the last level of wavelet packet decomposition 6. The variable working condition motor fault diagnosis method based on vibration signals according to claim 1, characterized in that The construction of the motor fault feature dataset in step (4) is specifically as follows: Use the feature vectors and state numbers of the vibration signals in the seven states of normal motor, bearing fault, gear fault, imbalance fault, stator fault, rotor fault, and shaft misalignment fault to form a feature matrix and a label matrix respectively, and construct the motor fault feature dataset.

7. The method for diagnosing variable-condition motor faults based on vibration signals according to claim 1, wherein The step (5) constructs a motor fault diagnosis model, specifically: Create a random forest classifier as the fault diagnosis model, and use the constructed motor fault feature dataset for training and parameter optimization.

8. The method for diagnosing variable working condition motor faults based on vibration signals according to claim 7, wherein Parameter optimization is specifically as follows: (5.1) Optimize feature extraction parameters: First, fix the type of wavelet basis, and set the wavelet packet decomposition level to where n is the sample length, denotes rounding down. Then, perform feature extraction. At each wavelet packet decomposition level, randomly divide the dataset into a training set and a test set at a ratio of 8:

2. Use a random forest classifier (default parameters) for training and testing, and repeat the experiment 1000 times. Determine the optimal wavelet packet decomposition level based on the classification accuracy on the test set in the 1000 experiments. Specifically, the one with the highest average classification accuracy and the smallest variance is considered the optimal. If the accuracy of the optimal wavelet packet decomposition level for this type of wavelet basis cannot meet the requirements, replace the wavelet basis and repeat the above operations until the optimal wavelet basis and the optimal wavelet packet decomposition level are found; (5.2) Optimize the hyperparameters of the random forest classifier: Fix the feature extraction parameters as the optimal parameters found in step (5.1), and use the Bayesian search method to find the optimal hyperparameters of the random forest classifier.

9. The method for diagnosing variable working condition motor faults based on vibration signals according to claim 1, wherein, The step (6) for actual fault diagnosis is specifically: During the actual production process, use a vibration acceleration sensor to collect the vibration acceleration signals on the motor cover at equal time intervals for a period of time. After preprocessing the collected signals, multiple vibration signals under different loads may be obtained. At this time, the signals with too short time should be discarded first. Then, feature extraction should be performed on the obtained multiple signals respectively, and fault diagnosis should be carried out through the fault diagnosis model. If the diagnosis results are different, the diagnosis result of the signal with the longest time shall prevail. If the diagnosis results are different and the time lengths are the same, the vibration signals with the same time length should be respectively subjected to feature extraction, and then the average value of the feature vectors should be calculated. Then, the average feature vector should be put into the fault diagnosis model for fault diagnosis, and the diagnosis result of the average feature vector shall prevail.

Images