A method and system for condition monitoring of a harmonic reducer
By installing a temperature sensor at the meshing point of the steel wheel and flexible wheel in the harmonic reducer, combined with a vibration sensor on the inner ring of the output shaft bearing, and using a Gaussian mixture model for data analysis, the problem of low integration in the condition monitoring of the harmonic reducer is solved, achieving accurate condition monitoring and a compact functional layout.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV
- Filing Date
- 2023-06-27
- Publication Date
- 2026-06-23
Smart Images
Figure CN116804592B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of speed reducer condition monitoring, and specifically to a method and system for monitoring the condition of a harmonic speed reducer. Background Technology
[0002] Harmonic reducers are a new type of reducer characterized by high load capacity, large transmission ratio, and smooth transmission. They are widely used in industries such as electronics, aerospace, and robotics. With continuous increases in productivity, the operational requirements for harmonic reducers are becoming increasingly stringent, making condition monitoring of harmonic reducers particularly important.
[0003] Current condition monitoring methods for reducers often rely on vibration signals measured from the output shaft bearing, which originate from the bearing's outer ring. This source is some distance from the fault location and cannot comprehensively reflect bearing fault information. Furthermore, testing and diagnosing rotating internal components often requires specialized auxiliary structures and careful selection of sensor placement and wiring methods, resulting in a dispersed layout of functional components, a less compact design, and low integration. Therefore, it is urgent for those skilled in the art to address these technical issues. Summary of the Invention
[0004] In order to overcome the defects existing in the prior art, the purpose of this invention is to provide a method and system for monitoring the state of a harmonic reducer.
[0005] To achieve the above-mentioned objectives of this invention, this invention provides a method for monitoring the state of a harmonic reducer, comprising the following steps:
[0006] Real-time acquisition of temperature signals at the meshing point of the steel wheel and flexible wheel of the harmonic reducer; and / or, real-time acquisition of vibration signals of the output shaft of the harmonic reducer;
[0007] Data analysis of the temperature signal yields an evaluation index for the temperature state at the meshing point.
[0008] Data analysis of the vibration signal yields the output shaft bearing condition evaluation index D. p ;
[0009] Based on the temperature status evaluation index at the meshing point and / or the output shaft bearing condition evaluation index D p The state of the harmonic reducer is fused and judged.
[0010] This method does not require special auxiliary structures or a large number of sensors. By collecting and analyzing the heat fluctuations at the meshing point of the flexure and steel wheel of the harmonic reducer and the bearing changes of the output shaft, it realizes the condition monitoring of the harmonic reducer and has great practical value.
[0011] In a preferred embodiment of the condition monitoring method for this harmonic reducer, the step of data analysis of the temperature signal is as follows:
[0012] The temperature data for the first n time moments during the detection are denoted as T = [T1, T2, ..., T]. n ], T n The temperature data at time n;
[0013] Set a sliding window of length k. Using this window, select k temperature data points in T with a step size of α (where α is a positive integer), denoted as T' each time. Calculate the average temperature for each T', thus obtaining the temperature mean history data T1 = [T1', T2', ..., T β '], where β is the number of sliding window operations, which is a positive integer, and T β ' is the average temperature of the k temperature data obtained in the βth sliding window;
[0014] Calculate the temperature error obtained from each sliding window operation to obtain the temperature error array ΔT'=[ΔT1',ΔT2',…,ΔT β '], where ΔT β '=T β '-T β-1 ';
[0015] Calculate the cumulative error: The cumulative error is used as an evaluation index for the temperature condition at the meshing point.
[0016] As time goes by, the gears in the transmission components wear down gradually, and the errors increase, leading to a gradual increase in temperature changes that are not easily detected by direct monitoring. This preferred solution accumulates the errors to reflect this gradual trend, thus achieving the purpose of internal condition monitoring.
[0017] In a preferred embodiment of the condition monitoring method for the harmonic reducer, the step of data analysis of the vibration signal is as follows:
[0018] Take the vibration signals from the m stable operating conditions before the test as training samples X1, X2, ..., X m The training samples are preprocessed, and z different features are extracted.
[0019] The parameters of the Gaussian mixture model are determined by training the training samples into the Gaussian mixture model.
[0020] Calculate the Gaussian probability density index P for each training sample. GMM ;
[0021] For P GMM Taking the negative logarithm yields the state index; adding the minimum value gives P. d ;
[0022] P calculated from all training samples d The indicators form a reference cluster;
[0023] Take q vibration signals under stable operation as test samples, preprocess the test samples, calculate the corresponding z feature values, and substitute them into the Gaussian mixture model to calculate the Gaussian probability density index P for each test sample. GMM ', for P GMM Taking the negative logarithm yields the state index; adding the minimum value gives P. d ';
[0024] Calculate P for each test sample d The mean distance from the center of the reference cluster is used to obtain the bearing condition evaluation index D. p .
[0025] This preferred approach overcomes the randomness problem of single-value monitoring and evaluation in traditional methods, and improves the accuracy of monitoring.
[0026] In a preferred embodiment of the state monitoring method for the harmonic reducer, z different features are extracted from the training samples; the training samples are then reduced to 2 to p dimensions, where p is a positive integer greater than 2. The data after each dimensionality reduction is fed into the Gaussian mixture model for training, and the model distribution curve T under the corresponding Gaussian dimension is obtained. The distance D between the data distribution of each dimension and the data distribution of the first dimension is calculated at each dimensionality reduction. All distances D corresponding to each dimensionality reduction are summed to obtain the loss value of the mixture distribution under the current dimension.
[0027] The dimension corresponding to the loss value of the smallest mixture distribution among all dimensions is selected as the optimal dimension of the training sample.
[0028] The parameters of the Gaussian mixture model are determined based on the training samples with the optimal dimension.
[0029] During testing, the test samples are reduced to the optimal dimension for testing.
[0030] This preferred scheme determines the optimal Gaussian dimension by selecting the loss value of the smallest mixture distribution. This overcomes the problem that manually determining the Gaussian dimension can lead to poor results if it is too small, or it can increase the amount of computational data if it is too large, thus causing overfitting.
[0031] In a preferred embodiment of the condition monitoring method for this harmonic reducer, the fusion discrimination step is as follows:
[0032] When the data at time f is input, the result at that time is calculated. and
[0033] Set the cumulative error threshold S Δ ,like The internal temperature of the harmonic reducer is abnormal.
[0034] When the bearing condition evaluation index is greater than the bearing condition evaluation index of the previous calculation time for e consecutive calculation times, e is a positive integer, and the vibration at the output end of the harmonic reducer is abnormal.
[0035] This invention also proposes a condition monitoring system for a harmonic reducer, including a temperature sensor installed at the meshing point of the steel wheel and flexible wheel of the harmonic reducer, and a vibration sensor installed on the inner ring of the bearing on the output shaft of the harmonic reducer. The signal output terminals of both the temperature sensor and the vibration sensor are connected to a processing module, sending corresponding temperature and vibration signals to the processing module. The processing module is communicatively connected to a storage module, which stores at least one executable instruction. This executable instruction causes the processing module to execute the condition monitoring method for the harmonic reducer described above, thereby monitoring the condition of the harmonic reducer. In addition to possessing all the advantages of the aforementioned condition monitoring method for harmonic reducers, this system collects vibration signals from the inner ring of the bearing, which is close to the fault source, providing a more comprehensive reflection of bearing fault information. Furthermore, it eliminates the need for special auxiliary structures and a large number of sensor placements, resulting in simple wiring, concentrated functional component arrangement, compact design, and high integration.
[0036] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0037] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:
[0038] Figure 1 This is a schematic diagram showing the assembly positions of the temperature sensor and the vibration sensor. Detailed Implementation
[0039] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0040] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.
[0041] This invention provides a condition monitoring method for a harmonic reducer, comprising the following steps:
[0042] The temperature signal at the meshing point of the steel wheel and flexible wheel of the harmonic reducer is acquired in real time. In this embodiment, a temperature sensor is installed near the meshing point of the steel wheel and flexible wheel to obtain the most accurate and reliable temperature data at the meshing point of the steel wheel and flexible wheel of the harmonic reducer.
[0043] The vibration signal of the harmonic reducer output shaft is acquired in real time. In this embodiment, a vibration sensor is installed on the inner ring of the bearing near the output shaft of the harmonic reducer to obtain the operating status of the upper bearing of the output shaft of the harmonic reducer, that is, to obtain the vibration signal of the output shaft of the harmonic reducer.
[0044] Data analysis of the temperature signal yields an evaluation index for the temperature state at the meshing point.
[0045] In this embodiment, the steps for data analysis of the temperature signal are as follows:
[0046] The temperature data for the first n time moments during the detection are denoted as T = [T1, T2, ..., T]. n ], T n The temperature data at time n;
[0047] Set a sliding window of length k, and use this sliding window to select k temperature data points in T with a step size of α, where α is a positive integer. Each time, denoted as T', the window selects k temperature data points.
[0048] Let's take k=5 and α=1 as an example:
[0049] Set a sliding window with a length of k = 5. Use this sliding window to select 5 temperature data points in T in sequence with a step size of α = 1, denoted as T'. That is, the first sliding window selects the 1st to 5th temperature data points in T, the second sliding window selects the 2nd to 6th temperature data points in T, and the βth sliding window selects the βth to β+5th temperature data points in T.
[0050] Calculate the average temperature for each T', thus obtaining the temperature mean history data T1=[T1',T2',…,T β '], where β is the number of sliding window operations, which is a positive integer, and T β ' is the average temperature of the k temperature data obtained from the βth sliding window.
[0051] Calculate the temperature error obtained from each sliding window operation to obtain the temperature error array ΔT'=[ΔT1',ΔT2',…,ΔT β '], where ΔT β '=T β '-T β-1 ';
[0052] Calculate the cumulative error: The cumulative error is used as an evaluation index for the temperature condition at the meshing point.
[0053] Data analysis of the vibration signal yields an output shaft bearing condition evaluation index.
[0054] In this embodiment, the steps for data analysis of the vibration signal are as follows:
[0055] Take the vibration signals from the m stable operating conditions before the test as training samples X1, X2, ..., X m The training samples are preprocessed, and z different features are extracted.
[0056] The training samples are reduced to 2 to p dimensions, where p is a positive integer greater than 2. The data after each dimensionality reduction is fed into the Gaussian mixture model for training, and the model distribution curve T under the corresponding Gaussian dimension is obtained. The distance D between the data distribution of each dimension and the data distribution of the first dimension is calculated for each dimensionality reduction. All distances D corresponding to each dimensionality reduction are summed to obtain the loss value of the mixture distribution under the current dimension.
[0057] Taking dimensionality reduction to 4 dimensions as an example: Let the Gaussian dimension be set to 4, and the data distributions in each dimension be T1, T2, T3, and T4, respectively. Then, train the dimensionality-reduced data into a Gaussian mixture model to obtain the model distribution curve T under the 4-dimensional Gaussian. Calculate the distances between the data distributions in each dimension and T1 using Mahalanobis distance (MD): D1 = MD(T, T1), D2 = MD(T, T2), D3 = MD(T, T3), and D4 = MD(T, T4). Calculate the loss value of the mixture distribution in the current dimension: Dloss4 = sum(D1, D2, D3, D4).
[0058] Taking dimensionality reduction to 3 dimensions as an example: Let the Gaussian dimension be set to 3, and the data distributions in each dimension be T1, T2, and T3, respectively. Then, train the dimensionality-reduced data into a Gaussian mixture model to obtain the model distribution curve T under the 3-dimensional Gaussian. Calculate the distances D1 = MD(T, T1), D2 = MD(T, T2), and D3 = MD(T, T3) between the data distributions in each dimension and T1 using Mahalanobis distance MD. Calculate the loss value D_loss3 = sum(D1, D2, D3) for the mixture distribution in the current dimension.
[0059] Similarly, the loss values for mixed distributions reduced to other dimensions can be obtained.
[0060] The dimension corresponding to the loss value of the smallest mixture distribution among all dimensions is selected as the optimal dimension of the training sample.
[0061] The parameters G = [μ1, μ2, ..., μ] of the Gaussian mixture model are determined based on training samples of optimal dimensionality. i ,...,μ z ;σ1,σ2,…,σ i ,...,σ z w1, w2, ..., w i ,...,w z ];μ i , σ i w i represents the mean, variance, and corresponding weight of the i-th single Gaussian element in the Gaussian mixture model function, respectively; where the sum of all weights is 1.
[0062] Calculate the Gaussian probability density index P for each training sample. GMM , N(μ i ,σ i ) represents μ i , σ i It is a single Gaussian function consisting of the mean and variance.
[0063] For P GMM Taking the negative logarithm yields the state index; if P GMM If the calculation equals 0, then the logarithm will become infinity. Therefore, we add a minimum value to each probability value to obtain P. d ,
[0064] P calculated from all training samples d The indicators form a reference cluster.
[0065] The vibration signal of the harmonic reducer running for a period of time during the test is used as the test sample. The number of test samples is q, where q can be 1 or other positive integers greater than 1. The test samples are preprocessed, and the z feature values corresponding to each test sample are calculated. Then, the dimensionality is reduced to the optimal dimension and substituted into the Gaussian mixture model to calculate the Gaussian probability density index P corresponding to each test sample. GMM ', for P GMM Taking the negative logarithm yields the state index; adding the minimum value gives P. d The time period here can be a single sampling point, a sampling time period, or several sampling time periods.
[0066] Calculate P for each test sample d'Distance to the center of the reference cluster. Specifically, first, the center of the reference cluster is determined using an existing algorithm, and then P corresponding to the test sample is calculated.' d The Euclidean distance between each test sample and the center of the reference cluster is used to obtain the P value. d 'Distance to the center of the reference cluster. Then calculate P for all test samples.' d The mean value of ' is used to obtain the bearing condition evaluation index D. p .
[0067] The state of the harmonic reducer is determined by integrating the temperature state evaluation index at the meshing point and the output shaft bearing state evaluation index.
[0068] Specifically, when the data at time f is input, the value at that time is calculated. and
[0069] ① Set the cumulative error threshold S Δ ,judge
[0070] ② There exist e consecutive calculation times in which the current bearing condition evaluation index is greater than the bearing condition evaluation index of the previous calculation time, where e is a positive integer.
[0071] For example, there exists a situation where the bearing condition evaluation index is greater than the bearing condition evaluation index at the previous calculation time for five consecutive calculation times, i.e. In the formula, n-4, n-3, n-2, n-1, and n represent 5 calculation times. It refers to the five bearing condition evaluation indicators with continuous calculation time.
[0072] The condition of the harmonic reducer is evaluated according to the table below:
[0073]
[0074] This application also proposes a condition monitoring system for a harmonic reducer, such as Figure 1 As shown, the device includes a temperature sensor 1 installed at the meshing point of the steel wheel and flexible wheel of the harmonic reducer, and a vibration sensor 2 installed on the inner ring of the bearing on the output shaft of the harmonic reducer. The output end of the temperature sensor transmits data to the processing module via a wire; the signal output end of the vibration sensor transmits the vibration signal to the processing module via a conductive ring and a contact brush. The processing module and a storage module are communicatively connected. The storage module stores at least one executable instruction, which causes the processing module to execute the state monitoring method of the harmonic reducer as described above, thereby realizing the state monitoring of the harmonic reducer.
[0075] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0076] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims
1. A method for monitoring the condition of a harmonic reducer, characterized in that, Includes the following steps: Real-time acquisition of temperature signals at the meshing point of the steel wheel and flexible wheel of the harmonic reducer; and / or, real-time acquisition of vibration signals of the output shaft of the harmonic reducer; Data analysis of the temperature signal yields an evaluation index for the temperature state at the meshing point. The specific steps are as follows: The temperature data from the first n moments during the detection are denoted as... T n The temperature data at time n; Set a sliding window of length k. Using this window, select k temperature data points in time T with a step size of α (where α is a positive integer), denoted as T'. Calculate the average temperature for each T', thus obtaining the temperature mean history data. Where β is the number of sliding window operations, which is a positive integer. The average temperature of the k temperature data obtained by the β-th sliding window; Calculate the temperature error obtained from each window sliding step to obtain a temperature error array. ,in, ; Calculate the cumulative error: The cumulative error is used as an evaluation index of the temperature state at the meshing point. Data analysis of the vibration signal yields the output shaft bearing condition evaluation index. The specific steps are as follows: Take the vibration signals from the m stable operating conditions before the test as training samples. The training samples are preprocessed, and z different features are extracted. The training samples are reduced to 2 to p dimensions respectively, where p is a positive integer greater than 2. The data after each dimensionality reduction is fed into the Gaussian mixture model for training, and the model distribution curve T under the Gaussian of the corresponding dimension is obtained. The distance D between the data distribution of each dimension and the data distribution of the first dimension is calculated at each dimensionality reduction. All distances D corresponding to each dimensionality reduction are summed to obtain the loss value of the mixture distribution under the current dimension. The dimension corresponding to the loss value of the smallest mixture distribution among all dimensions is selected as the optimal dimension of the training sample. The parameters of the Gaussian mixture model are determined based on the training samples with the optimal dimension. During testing, the test samples are reduced to the optimal dimension for testing. The parameters of the Gaussian mixture model are determined by training the training samples into the Gaussian mixture model. Calculate the Gaussian probability density index for each training sample. P GMM ; right P GMM Taking the negative logarithm yields the state index; adding the minimum value gives... P d ; Calculated from all training samples P d The indicators form a reference cluster; Take q vibration signals under stable operation as test samples, preprocess the test samples, calculate the corresponding z feature values, and substitute them into the Gaussian mixture model to calculate the Gaussian probability density index corresponding to each test sample. P GMM ’ ,right P GMM ’ Taking the negative logarithm yields the state index; adding the minimum value gives... P d ’ ; Calculate the corresponding test sample P d ’ The mean distance from the center of the reference cluster is used to obtain the bearing condition evaluation index. D p ; Based on the temperature status evaluation index at the meshing point and / or the output shaft bearing condition evaluation index D p The state of the harmonic reducer is fused and judged.
2. The state monitoring method for a harmonic reducer according to claim 1, characterized in that, The steps of fusion discrimination are as follows: When the data at time f is input, the result at that time is calculated. and ; Set cumulative error threshold ,like If so, the internal temperature of the harmonic reducer will change abnormally. When the bearing condition evaluation index is greater than the bearing condition evaluation index of the previous calculation time for e consecutive calculation times, e is a positive integer, and the vibration at the output end of the harmonic reducer is abnormal.
3. A condition monitoring system for a harmonic reducer, characterized in that, The system includes a temperature sensor located at the meshing point of the steel wheel and flexible wheel of the harmonic reducer, and a vibration sensor located on the inner ring of the bearing on the output shaft of the harmonic reducer. The signal output terminals of both the temperature sensor and the vibration sensor are connected to a processing module to send corresponding temperature and vibration signals to the processing module. The processing module is communicatively connected to a storage module, which stores at least one executable instruction. The executable instruction causes the processing module to execute the state monitoring method for the harmonic reducer as described in claim 1 or 2 to monitor the state of the harmonic reducer.