A method for evaluating bearing performance degradation of construction robots

Abstract

The present invention specifically discloses a construction robot bearing performance degradation assessment method, the method comprising the following steps: S1, collecting vibration signals of the construction robot bearing in a healthy state; S2, extracting the collected vibration signals in the time domain and time-frequency domain to obtain an initial feature set; S3, using the principal component analysis method to fuse and reduce the normalized initial feature set, and selecting the input feature vector of the self-organizing map network assessment model; S4, taking the feature vector obtained in step S3 and inputting it into the self-organizing map network for training; S5, collecting the vibration signal of the construction robot bearing to be evaluated and repeating steps S2 and S3, the obtained feature vector is used as a test set and input into the trained assessment model to calculate the performance degradation index, and the performance degradation degree of the construction robot bearing is determined according to the degradation starting threshold. The present invention has no label setting, accurately assesses the degradation degree of the construction robot bearing in the degradation process, and saves labor and time costs.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment failure prediction and health management, and in particular to a method for evaluating bearing performance degradation of a construction robot. Background Art

[0002] Construction robots can not only free construction workers from dangerous construction environments and protect their lives, but also improve the efficiency of construction, solving the problem of labor shortage to a certain extent. However, since the working environment of construction robots is very complex, and they are bound to fail after long-term, high-intensity operation, once a construction robot fails, the entire mechanical equipment will be unusable, which may cause downtime, affect production and cause economic losses to the company, or even cause casualties and serious accidents. Therefore, it is very important to ensure the reliable operation of construction robots.

[0003] All construction robot failures are caused by failures of its internal components, among which bearings have the highest failure frequency among all parts of construction robots. The main function of bearings is to support and fix the output shaft of the construction robot, thereby reducing the friction generated when the internal components are in operation. It has the characteristics of high speed, large load changes, and high failure rate. Therefore, it is easy to fail after long-term high-intensity operation. Bearing failure will also lead to a series of chain effects such as wear of mechanical parts, transmission failure, motor damage, etc., so once the bearing fails, it will directly affect the performance, quality and reliability of the construction robot.

[0004] Performance degradation assessment of construction robot bearings is an important means to achieve fault prediction and health management. It mainly maps the bearing vibration signal to the degradation state by constructing the corresponding assessment model. By evaluating the performance degradation of construction robot bearings, the operating status of the bearings can be monitored in real time, and the operating status of the construction robot bearings can be visualized, which is of great significance for timely detection of construction robot bearing faults and formulation of maintenance strategies. At present, most performance degradation assessment models are supervised training methods, which also means that samples must be labeled first. However, the selection of labels increases manual participation, resulting in a lack of certain universality, and it is impossible to train an accurate model in the absence of fault samples.

[0005] In view of this, designing a performance degradation assessment method for construction robot bearings that does not require label setting and can accurately detect early bearing failures is a technical problem that needs to be urgently solved in this technical field. Summary of the invention

[0006] The purpose of the present invention is to provide a method for evaluating the performance degradation of a construction robot bearing, which has no label setting and can accurately detect the early failure of the bearing. In the present invention, IMF is Intrinsic Mode Function.

[0007] In order to solve the above technical problems, the present invention provides a method for evaluating bearing performance degradation of a construction robot, comprising the following steps:

[0008] S1. Install a vibration sensor on the outer ring of the construction robot bearing, and use the vibration sensor to collect the vibration signal S(t) of the construction robot bearing in a healthy state;

[0009] S2, extracting the features of the collected vibration signal in the time domain and the time-frequency domain to obtain an initial feature set;

[0010] S3, using principal component analysis to fuse and reduce the dimension of the normalized initial feature set, and selecting the principal component with a cumulative contribution rate greater than 90% as the input feature vector of the self-organizing map network evaluation model;

[0011] S4, taking the feature vector obtained in step S3 as a training set and inputting it into the self-organizing map network for training, thereby obtaining a trained self-organizing map network evaluation model;

[0012] S5. Collect the vibration signal of the construction robot bearing to be evaluated and repeat steps S2 and S3. The obtained feature vector is used as a test set and input into the trained self-organizing map network evaluation model to calculate the performance degradation index DI, and the degree of performance degradation of the construction robot bearing is determined according to the degradation starting threshold.

[0013] Preferably, the specific implementation method of step S2 includes the following steps:

[0014] S21, extracting eight time domain features of the bearing vibration signal of the construction robot collected in step S1, including mean, root mean square, variance, peak-to-peak value, kurtosis, skewness, kurtosis, and standard deviation;

[0015] S22, performing lumped empirical mode decomposition on the bearing vibration signal of the construction robot collected in step S1, setting two parameters of lumped empirical mode decomposition, white noise amplitude a=0.2, lumped average times M=100, and obtaining N IMF (Intrinsic Mode Function) components based on this decomposition, and selecting R effective IMF components with higher correlation signals with the original signal x(t) from the N IMF components based on the correlation coefficient method, where R≤N;

[0016] S23, calculating the energy entropy of the R effective IMF components extracted in step S22, the calculation formula is:

[0017]

[0018] In the formula, H i is the energy entropy of the ith IMF component, E is the total energy value of IMF, and E i is the energy value of the i-th IMF component;

[0019] S24, the R effective IMF components extracted in step S22 form a matrix M = [imf 1 ,imf 2 ,...,imf R ] T , and then perform singular value decomposition on the matrix, the calculation formula is:

[0020] M=UAV T (4)

[0021] Where U is an m-order orthogonal matrix, V is an n-order orthogonal matrix, V T is the transposed matrix of V, A is a semi-positive definite m×n-order diagonal matrix, the elements on the diagonal of A are the singular values ​​of A, calculate the first S singular values ​​of the matrix M, where S≤R, and combine the R energy entropies of step S23 to form the time-frequency domain features;

[0022] S25, constructing an initial feature set by combining the 8 time domain features extracted in step S21, the R effective IMF component energy entropies extracted in step S23, and the S IMF matrix singular values ​​calculated in step S24.

[0023] Preferably, the time domain characteristics of the bearing vibration signal S(t) extracted in step S21 can be expressed by the formula:

[0024]

[0025] F 4 (t) = max{S i (t)}-min{S i (t)} (8)

[0026]

[0027]

[0028] In formulas (5) to (12), F 1 (t)~F 8 (t) are respectively expressed as mean, root mean square, variance, peak-to-peak value, kurtosis, skewness, kurtosis, and standard deviation, {S i (t)} i＝1,2,...,n Represents the i-th sampling sequence of the vibration signal S(t) at time t.

[0029] Preferably, the specific implementation method of the lumped empirical mode decomposition in step S22 comprises the following steps:

[0030] S221, setting two parameters of lumped empirical mode decomposition, the white noise amplitude a is set to 0.2 times the standard deviation of the original data, the lumped average number M is set to 100, and the original signal x(t) is input;

[0031] S222, add a white noise sequence n of a certain amplitude to the original signal x(t) m (t), we get the signal x m (t), x m (t) = x(t) + n m (t);

[0032] S223, the signal x after adding noise m (t) Perform empirical mode decomposition (EMD) to obtain N IMF components a i,m (i=1,2,...,N);

[0033] S224, repeat step S223 until white noise sequences with different amplitudes are added to x(t) M times;

[0034] S225. The IMF obtained from each decomposition is The lumped average calculation is performed, and the calculation result is the N IMF components obtained by the lumped empirical mode decomposition.

[0035] Preferably, the calculation formula of the effective IMF component selection method based on the correlation coefficient method in step S22 is:

[0036]

[0037] In the formula, r i is the correlation coefficient between the ith IMF component and the original vibration signal, x i represents the normalized value of the ith IMF component, y i is the normalized value of the original vibration signal. The standard deviation λ of the N correlation coefficient values ​​is calculated as the threshold to filter out the IMF components with higher correlation, that is, to select the correlation coefficient r i >λ, eliminating r i The IMF component of <λ, the calculation formula of λ is:

[0038]

[0039] Preferably, in step S3, the initial feature set obtained in S2 is firstly Normalization processing is performed to unify the feature scale, and then the principal component analysis method is used to fuse and reduce the dimension of the normalized initial feature set to reduce feature redundancy. The principal components are sorted according to the contribution rate, and n principal components with a cumulative contribution rate greater than 90% and the smallest dimension are selected as the input feature vector of the self-organizing map network evaluation model.

[0040] Preferably, the specific implementation method of step S4 of self-organizing map network training comprises the following steps:

[0041] S41, set the number of input neurons n, the number of output layer neurons m, and the maximum number of iterations l max , learning rate α;

[0042] S42, set the initial weights of the input layer and the output layer to random values ​​close to 0, and convert the input vector X = [x 1 ,x 2 ,...,x n ] and the weight vector corresponding to the output layer neurons Normalized to and

[0043] S43. Calculate input vector and the weight vector of each neuron Euclidean distance

[0044] S44. Select the neuron with the smallest Euclidean distance as the winning neuron and record it as W j * , update the weights of the winning neuron and its neighboring neurons:

[0045]

[0046] W j (t+1)=W j (t) (16)

[0047] In the formula, the learning rate α is (0, 1], t is the training time, and α decreases as the learning dimension increases, that is, as the learning proceeds, the weights will transition from coarse adjustment to fine adjustment;

[0048] S45, increase the number of iterations l = l + 1, when l > l max (maximum number of iterations) or α<α min (minimum learning rate), the network training ends; otherwise, the network training continues back to step S43.

[0049] Preferably, the specific implementation method of step S5 includes the following steps:

[0050] S51, collecting the bearing vibration signal of the construction robot to be evaluated and repeating steps S2 and S3;

[0051] S52, taking the feature vector obtained in step S51 as a test set and inputting it into the trained self-organizing map network evaluation model, calculating the Euclidean distance with the best matching unit weight vector as the performance degradation index DI, and the calculation formula is:

[0052] DI=||Yw BMU || (17)

[0053] Where Y represents the input feature vector of the SOM network, w BMU Represents the weight vector of the best matching unit corresponding to the input vector, and || || represents the Euclidean distance calculation. When the DI value is large, it means that the current state of the bearing is very different from the healthy state; conversely, the smaller the index value, the smaller the difference between the current state of the bearing and the healthy state;

[0054] S53. Set a degradation starting threshold, and evaluate the performance degradation status of the bearing through quantitative analysis of degradation indicators. If the DI value is less than the degradation starting threshold, it means that the bearing is in a healthy stage; if the DI value is greater than the degradation starting threshold, it means that the bearing is in a fault degradation stage; and the larger the DI value, the more serious the bearing fault.

[0055] Preferably, the degradation starting threshold in step S53 is determined by the commonly used 3σ criterion: assuming that the training data conforms to the Gaussian normal distribution N(μ,σ), the probability value of the real-time new data belonging to the interval [μ-3σ,μ+3σ] is 99.73%, where μ and σ are the mean and standard deviation of the statistical data, respectively. Once the performance index value DI at time t is not within this range, it means that the bearing has seriously deviated from the normal state. Therefore, time t can be used as the starting point of bearing degradation.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention proposes a construction robot bearing performance degradation evaluation method. First, a vibration sensor is used to collect the vibration signal of the construction robot bearing in a healthy state, and the time domain and time-frequency domain feature extraction of the vibration signal is performed to construct an initial feature set. Then, the principal component analysis method is used to fuse and reduce the dimension of the normalized initial feature set, and the principal component with a cumulative contribution rate greater than 90% is selected for the training of the self-organizing map network evaluation model. Finally, the vibration signal of the bearing to be evaluated is collected and input into the trained self-organizing map network evaluation model after feature extraction and dimension reduction. The Euclidean distance between the input vector and the corresponding optimal matching unit is calculated as a degradation index, and the bearing performance degradation degree is detected according to the degradation starting threshold. If the degradation index is greater than the degradation starting threshold, the bearing is in a fault degradation stage, and the larger the degradation index, the higher the fault degree. The method is more sensitive to identifying early faults of construction robot bearings, and improves the versatility and accuracy of the evaluation model through an unsupervised training network, can better reflect the performance change trend of the construction robot bearing, and ensure the reliable operation of the construction robot. In addition, no label is required, which reduces labor costs and time costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flow chart of a construction robot bearing performance degradation assessment method of the present invention,

[0059] Figure 2 is a flow chart of the method for constructing an initial feature set in step S2 of the present invention,

[0060] Figure 3 is a flow chart of the method for lumped empirical mode decomposition in step S22 of the present invention,

[0061] Figure 4 is the self-organizing map network structure diagram in step S4 of the present invention,

[0062] Figure 5 is a flow chart of the method for training the self-organizing map network in step S4 of the present invention,

[0063] Figure 6 is a flow chart of determining the bearing degradation state in step S5 of the present invention;

[0064] Figure 7 is a diagram of the original data signal used in a specific embodiment of the present invention;

[0065] Figure 8 is a performance degradation evaluation result diagram of the method according to a specific embodiment of the present invention;

[0066] Fig. 9 This is a result diagram of using the RMS value of the comparison method of an embodiment of the present invention as a performance degradation indicator. DETAILED DESCRIPTION

[0067] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0068] In order to make those skilled in the art better understand the technical solution of the present invention, the following will be combined with the attached embodiment of the present invention. Figure 1-9 , the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0070] See also Figure 1 To achieve the above-mentioned purpose, the first embodiment of the present invention provides a construction robot bearing performance degradation assessment method. The construction robot bearing performance degradation assessment method provided by the present invention is applied to construction robot bearing degradation state assessment and early fault detection. The method comprises the following steps:

[0071] S1. Install a vibration sensor on the outer ring of the construction robot bearing, and use the vibration sensor to collect the vibration signal S(t) of the construction robot bearing in a healthy state;

[0072] S2, extracting the features of the collected vibration signal in the time domain and the time-frequency domain to obtain an initial feature set;

[0073] S3, using principal component analysis to fuse and reduce the dimension of the normalized initial feature set, and selecting the principal component with a cumulative contribution rate greater than 90% as the input feature vector of the self-organizing map network evaluation model;

[0074] S4, taking the feature vector obtained in step S3 as a training set and inputting it into the self-organizing map network for training, thereby obtaining a trained self-organizing map network evaluation model;

[0075] S5. Collect the vibration signal of the construction robot bearing to be evaluated and repeat steps S2 and S3. The obtained feature vector is used as a test set and input into the trained self-organizing map network evaluation model to calculate the performance degradation index DI, and the degree of performance degradation of the construction robot bearing is determined according to the degradation starting threshold.

[0076] In this embodiment, firstly, a vibration sensor is used to collect the vibration signal of a construction robot bearing in a healthy state, and the time domain and time-frequency domain features of the vibration signal are extracted to construct an initial feature set. Then, the principal component analysis method is used to fuse and reduce the dimension of the normalized initial feature set. The principal component with a cumulative contribution rate greater than 90% is selected for training the self-organizing map network evaluation model. Finally, the vibration signal of the bearing to be evaluated is collected and input into the trained self-organizing map network evaluation model after feature extraction and dimension reduction. The Euclidean distance between the input vector and its corresponding optimal matching unit is calculated as a performance degradation index, and the degree of bearing performance degradation is detected according to the degradation starting threshold. If the performance degradation index value is greater than the degradation starting threshold, the bearing is in a fault degradation stage, and the larger the performance degradation index value, the higher the degree of fault.

[0077] See also Figure 2 The specific implementation of step S2 includes:

[0078] S21, extracting eight time domain features of the bearing vibration signal of the construction robot collected in step S1, including mean, root mean square, variance, peak-to-peak value, kurtosis, skewness, kurtosis, and standard deviation;

[0079] S22, performing lumped empirical mode decomposition on the bearing vibration signal of the construction robot collected in step S1, setting two parameters of the lumped empirical mode decomposition, white noise amplitude a=0.2, lumped average times M=100, and obtaining N IMF (Intrinsic Mode Function) components based on this decomposition, and selecting R (R≤N) effective IMF components with higher correlation with the original signal x(t) from the N IMF components based on the correlation coefficient method;

[0080] S23, calculating the energy entropy of the R effective IMF components extracted in step S22, the calculation formula is:

[0081]

[0082] In the formula, H i is the energy entropy of the ith IMF component, E is the total energy value of IMF, and E i is the energy value of the i-th IMF component;

[0083] S24, the R effective IMF components extracted in step S22 form a matrix M = [imf 1 ,imf 2 ,...,imf R ] T , and then perform singular value decomposition on the matrix, the calculation formula is:

[0084] M=UAV T (4)

[0085] Where U is an m-order orthogonal matrix, V is an n-order orthogonal matrix, V T is the transposed matrix of V, A is a semi-positive definite m×n order diagonal matrix, and the elements on the diagonal of A are the singular values ​​of A. The first S (S≤R) singular values ​​of the matrix M are calculated, and the R energy entropies in step S23 are combined to form the time-frequency domain features.

[0086] S25, constructing an initial feature set by combining the 8 time domain features extracted in step S21, the R effective IMF component energy entropies extracted in step S23, and the S IMF matrix singular values ​​calculated in step S24.

[0087] In this embodiment, the time domain characteristics of the bearing vibration signal S(t) extracted in step S21 can be expressed by the formula:

[0088]

[0089] F 4 (t) = max{S i (t)}-min{S i (t)} (8)

[0090]

[0091]

[0092] In formulas (5) to (12), F 1 (t)~F 8 (t) are respectively expressed as mean, root mean square, variance, peak-to-peak value, kurtosis, skewness, kurtosis, and standard deviation, {S i (t)} i＝1,2,...,n Represents the i-th sampling sequence of the vibration signal S(t) at time t.

[0093] See also Figure 3 The specific implementation steps of the lumped empirical mode decomposition method in step S22 in this embodiment include:

[0094] S221, setting the white noise amplitude a of two parameters of lumped empirical mode decomposition to 0.2 times the standard deviation of the original data, and the lumped average number M to 100;

[0095] S222, add a white noise sequence n of a certain amplitude to the original signal x(t) m (t), we get the signal x m (t), x m (t) = x(t) + n m (t);

[0096] S223, the signal x after adding noise m (t) Perform empirical mode decomposition to obtain N IMF components a i,m (i=1,2,...,N);

[0097] S224, repeat step S223 until white noise sequences with different amplitudes are added to x(t) M times;

[0098] S225. The IMF obtained from each decomposition is The lumped average calculation is performed, and the calculation result is the N IMF components obtained by the lumped empirical mode decomposition.

[0099] In this embodiment, the calculation formula of the method for selecting effective IMF components based on the correlation coefficient method from the N IMF components obtained by the lumped empirical mode decomposition in step S225 in step S222 is:

[0100]

[0101] In the formula, r i is the correlation coefficient between the ith IMF component and the original vibration signal, x i represents the normalized value of the ith IMF component, y i is the normalized value of the original vibration signal. The standard deviation λ of the N correlation coefficient values ​​is calculated as the threshold to filter out the IMF components with higher correlation, that is, to select the correlation coefficient r i >λ, eliminating r i The IMF component of <λ, the calculation formula of λ is:

[0102]

[0103] In this embodiment, in step S3, the initial feature set is first Normalization processing is performed to unify the feature scale, and then the principal component analysis method is used to fuse and reduce the dimension of the normalized initial feature set to reduce feature redundancy. The principal components are sorted according to the contribution rate, and n principal components with a cumulative contribution rate greater than 90% and the smallest dimension are selected as the input feature vector of the self-organizing map network evaluation model.

[0104] See also Figure 4 and Figure 5 The specific implementation method of step S4 in this embodiment includes:

[0105] S41, set the number of input neurons n, the number of output layer neurons m, and the maximum number of iterations l max , learning rate α;

[0106] S42, set the initial weights of the input layer and the output layer to random values ​​close to 0, and convert the input vector X = [x 1 ,x 2 ,...,x n ] and the weight vector corresponding to the output layer neurons Normalized to and

[0107] S43. Calculate input vector and the weight vector of each neuron Euclidean distance

[0108] S44. Select the neuron with the smallest Euclidean distance as the winning neuron and record it as W j * , update the weights of the winning neuron and its neighboring neurons:

[0109]

[0110] W j (t+1)=W j (t) (16)

[0111] In the formula, the learning rate α is (0, 1], and t is the training time. α decreases as the learning dimension increases, that is, as the learning proceeds, the weights will transition from coarse tuning to fine tuning.

[0112] S45, increase the number of iterations l = l + 1, when l > l max (maximum number of iterations) or α<α min (minimum learning rate), the network training is terminated, otherwise, the network training is continued back to step S43.

[0113] See also Figure 6 In this embodiment, the specific implementation of step S5 includes:

[0114] S51, collecting the bearing vibration signal of the construction robot to be evaluated and repeating steps S2 and S3;

[0115] S52, taking the feature vector obtained in step S51 as a test set and inputting it into the trained self-organizing map network evaluation model, calculating the Euclidean distance with the best matching unit weight vector as the degradation index DI, and the calculation formula is:

[0116] DI=||Yw BMU || (17)

[0117] Where Y represents the input feature vector of the SOM network, w BMURepresents the weight vector of the best matching unit corresponding to the input vector, and || || represents the Euclidean distance calculation. When the DI value is large, it means that the current state of the bearing is very different from the healthy state; conversely, the smaller the index value, the smaller the difference between the current state of the bearing and the healthy state;

[0118] S53. Set a degradation starting threshold, and evaluate the performance degradation status of the bearing through quantitative analysis of degradation indicators. If the DI value is less than the degradation starting threshold, it means that the bearing is in a healthy stage; if the DI value is greater than the degradation starting threshold, it means that the bearing is in a fault degradation stage; and the larger the DI value, the more serious the bearing fault.

[0119] In this embodiment, the commonly used 3σ criterion is used in step S53 to determine the degradation starting threshold: assuming that the training data conforms to the Gaussian normal distribution N(μ,σ), the probability value of the real-time new data belonging to the interval [μ-3σ,μ+3σ] is 99.73%, where μ and σ are the mean and standard deviation of the statistical data, respectively. Once the performance index value DI at time t is not within this range, it means that the bearing has seriously deviated from the normal state. Therefore, time t can be used as the starting point of bearing degradation.

[0120] The present invention uses bearing 2_1 in the IMS bearing data set to conduct experimental verification of the present invention. Bearing 2_1 has undergone an accelerated aging experiment for 163 hours and 50 minutes, and finally an outer ring failure occurred. During the experiment, the speed of bearing 2_1 was 2000r / min, the radial load was 6000N, the sensor sampling frequency was 20.48KHz, the sampling time was 1s, and the sampling was performed every 10 minutes. The data set of bearing 2_1 contains a total of 984 files, each file contains 20480 data, and records the full life cycle vibration data of the bearing from a healthy state to the final outer ring failure.

[0121] See also Figure 7 , which is a diagram of the original data signal used in a specific embodiment of the present invention, and the original signal is a vibration signal of the bearing 2_1 in the horizontal direction.

[0122] In this experiment, the first 100 groups of bearing data in normal state were selected as training samples to train the self-organizing map network evaluation model. The number of neurons in the input layer and output layer of the model was 4 and 48 respectively. The regular hexagon was used as the topological neighborhood shape, the initial learning rate was set to 0.6, and the maximum number of iterations was set to 100. 984 groups of bearing full life cycle data were input into the above-trained self-organizing map network evaluation model to obtain the performance degradation evaluation results of the bearing.

[0123] See also Figure 8, is a performance degradation evaluation result diagram based on principal component analysis and self-organizing map network according to an embodiment of the present invention. The degradation starting point is the 532nd sample using the 3σ criterion. It can be seen from the figure that the degradation value has been kept near 0 before the 532nd sample. When the 532nd sample comes, a large jump occurs in the degradation curve, and the subsequent sample curves maintain a continuous upward trend, indicating that the evaluation model has detected a fault signal at the 532nd sampling point, and the bearing has entered an early degradation state at this time; the DI value ( Figure 8 The degradation value in the bearing first rises and then falls. This is because the fault point of the bearing is worn flat, and the bearing begins to enter a state of deepening degradation. The DI value rises steadily in the 782-940 sample segment, and the bearing enters a state of severe degradation. After 940 samples, the DI value increases exponentially, and the bearing has already suffered a serious fault and is close to complete failure.

[0124] To further demonstrate the superiority of the present invention, the commonly used root mean square value RMS is selected as the performance degradation index to evaluate the bearing 2_1. Fig. 9 .

[0125] like Fig. 9 As shown in FIG. 1 , the RMS value of the comparative method of the present invention is used as the performance degradation index result diagram. It can be seen that the bearing 2_1 is basically stable in the 0-700 sample segment, and a large jump occurs at the 700th sample; the DI value ( Fig. 9 The characteristic value in the sample ( ) gradually decreases, at which time the fault point is worn flat, and the bearing enters a state of deepening degradation; the DI value rises steadily in the 793-912 sample segment, and the bearing enters a state of severe degradation; after 912 samples, the DI value increases exponentially, and the bearing has already suffered a serious fault and is close to a state of complete failure. Therefore, RMS is 168 samples later than the sample time of the present invention for detecting early fault points, indicating that the method of the present invention has certain advantages in detecting early bearing faults.

Claims

1. A construction robot bearing performance degradation assessment method, characterized in that: The following steps are involved: S1. Install a vibration sensor on the outer ring of the construction robot bearing, and use the vibration sensor to collect the vibration signal S(t) of the construction robot bearing in a healthy state; S2, extracting the features of the collected vibration signal in the time domain and the time-frequency domain to obtain an initial feature set; S21, extracting eight time domain features of the bearing vibration signal of the construction robot collected in step S1, including mean, root mean square, variance, peak-to-peak value, kurtosis, skewness, kurtosis, and standard deviation; S22, performing lumped empirical mode decomposition on the bearing vibration signal of the construction robot collected in step S1, setting two parameters of lumped empirical mode decomposition, white noise amplitude a=0.2, lumped average times M=100, decomposing to obtain N IMF components, and then selecting R effective IMF components with higher correlation with the original signal x(t) from the N IMF components based on the correlation coefficient method, where R≤N; S23, calculating the energy entropy of the R effective IMF components extracted in step S22, the calculation formula is: In the formula, H i is the energy entropy of the ith IMF component, E is the total energy value of IMF, and E i is the energy value of the i-th IMF component; S24, the R effective IMF components extracted in step S22 are used to form a matrix M = [imf1, imf2, ..., imf R ] T , and then perform singular value decomposition on the matrix, the calculation formula is: M = UAV T (4) Where U is an m-order orthogonal matrix, V is an n-order orthogonal matrix, V T is the transposed matrix of V, A is a semi-positive definite m×n-order diagonal matrix, the elements on the diagonal of A are the singular values ​​of A, calculate the first S singular values ​​of the matrix M, where S≤R, and combine the R energy entropies of step S23 to form the time-frequency domain features; S25, constructing an initial feature set by combining the 8 time domain features extracted in step S21, the R effective IMF component energy entropies extracted in step S23, and the S IMF matrix singular values ​​calculated in step S24; S3, using principal component analysis to fuse and reduce the dimension of the normalized initial feature set, and selecting the principal component with a cumulative contribution rate greater than 90% as the input feature vector of the self-organizing map network evaluation model; S4, taking the feature vector obtained in step S3 as a training set and inputting it into the self-organizing map network for training, thereby obtaining a trained self-organizing map network evaluation model; S5. Collect the vibration signal of the construction robot bearing to be evaluated and repeat steps S2 and S3. The obtained feature vector is used as a test set and input into the trained self-organizing map network evaluation model to calculate the performance degradation index DI, and the degree of performance degradation of the construction robot bearing is determined according to the degradation starting threshold.

2. The construction robot bearing performance degradation assessment method according to claim 1, characterized in that: The time domain characteristics of the bearing vibration signal S(t) extracted in step S21 can be expressed by the formula: F4(t)=max{S i (t)}-min{S i (t)} (8) In formulas (5) to (12), F1(t) to F8(t) are respectively represented by mean, root mean square, variance, peak-to-peak value, kurtosis, skewness, kurtosis, and standard deviation, {S i (t)} i＝1,2,...,n Represents the i-th sampling sequence of the vibration signal S(t) at time t.

3. The construction robot bearing performance degradation assessment method according to claim 1, characterized in that: The specific implementation method of the lumped empirical mode decomposition in step S22 comprises the following steps: S221, setting two parameters of lumped empirical mode decomposition, the white noise amplitude a is set to 0.2 times the standard deviation of the original data, the lumped average number M is set to 100, and the original signal x(t) is input; S222, add a white noise sequence n of a certain amplitude to the original signal x(t) m (t), we get the signal x m (t), x m (t) = x(t) + n m (t); S223, after adding noise to the signal x m (t) Perform empirical mode decomposition (EMD) to obtain N IMF components a i,m (i=1,2,...,N); S224, repeat step S223 until white noise sequences with different amplitudes are added to x(t) M times; S225. The IMF obtained from each decomposition is The lumped average calculation is performed, and the calculation result is the N IMF components obtained by the lumped empirical mode decomposition.

4. The construction robot bearing performance degradation assessment method according to claim 1, characterized in that: The calculation formula of the effective IMF component selection method based on the correlation coefficient method in step S22 is: In the formula, r i is the correlation coefficient between the ith IMF component and the original vibration signal, x i represents the normalized value of the ith IMF component, y i is the normalized value of the original vibration signal, and the standard deviation λ of the N correlation coefficient values ​​is calculated as the threshold to filter out the IMF components with higher correlation, that is, the correlation coefficient r is selected. i >λ, eliminating r i The IMF component of <λ, the calculation formula of λ is:

5. The construction robot bearing performance degradation assessment method according to claim 1, characterized in that: In step S3, the initial feature set obtained in S2 is firstly Normalization processing is performed to unify the feature scale, and then the principal component analysis method is used to fuse and reduce the dimension of the normalized initial feature set to reduce feature redundancy. The principal components are sorted according to the contribution rate, and n principal components with a cumulative contribution rate greater than 90% and the smallest dimension are selected as the input feature vector of the self-organizing map network evaluation model.

6. The construction robot bearing performance degradation assessment method according to claim 1, characterized in that: The specific implementation method of step S4 self-organizing map network training comprises the following steps: S41, set the number of input neurons n, the number of output layer neurons m, and the maximum number of iterations l max , learning rate α; S42, set the initial weights of the input layer and the output layer to random values ​​close to 0, and convert the input vector X = [x1, x2, ..., x n ] and the weight vector W corresponding to the output layer neurons j =[w j1 ,w j2 ,...,w jn ](1≤j≤m) is normalized to and S43. Calculate input vector and the weight vector of each neuron Euclidean distance S44. Select the neuron with the smallest Euclidean distance as the winning neuron and record it as W j * , update the weights of the winning neuron and its neighboring neurons: W j (t+1)=W j (t) (16) In the formula, the learning rate α is (0, 1], t is the training time, and α decreases as the learning dimension increases, that is, as the learning proceeds, the weights will transition from coarse adjustment to fine adjustment; S45, increase the number of iterations l = l + 1, when l > l max Or α<α min When , the network training ends, otherwise returns to step S43 to continue executing the network training.

7. The construction robot bearing performance degradation assessment method according to claim 1, characterized in that: The specific implementation of step S5 includes the following steps: S51, collecting the bearing vibration signal of the construction robot to be evaluated and repeating steps S2 and S3; S52, taking the feature vector obtained in step S51 as a test set and inputting it into the trained self-organizing map network evaluation model, calculating the Euclidean distance with the best matching unit weight vector as the performance degradation index DI, and the calculation formula is: IN=||Yw BMU || (17) Where Y represents the input feature vector of the SOM network, w BMU Represents the weight vector of the best matching unit corresponding to the input vector, |||| represents the Euclidean distance calculation. When the DI value is large, it means that the current state of the bearing is very different from the healthy state; conversely, the smaller the index value, the smaller the difference between the current state of the bearing and the healthy state. S53. Set a degradation starting threshold, and evaluate the performance degradation status of the bearing through quantitative analysis of degradation indicators. If the DI value is less than the degradation starting threshold, it means that the bearing is in a healthy stage; if the DI value is greater than the degradation starting threshold, it means that the bearing is in a fault degradation stage, and the larger the DI value, the more serious the bearing fault.

8. The construction robot bearing performance degradation assessment method according to claim 7, characterized in that: The degradation starting threshold in step S53 is determined by the commonly used 3σ criterion: assuming that the training data conforms to the Gaussian normal distribution N(μ,σ), the probability value of the real-time new data belonging to the interval [μ-3σ,μ+3σ] is 99.73%, where μ and σ are the mean and standard deviation of the statistical data, respectively. Once the performance index value DI at time t is not within this range, it means that the bearing has seriously deviated from the normal state. Therefore, time t can be used as the starting point of bearing degradation.

Images