An antenna pedestal transmission system running-in quality abnormality detection method

Abstract

At present, there is no effective method for evaluating the running-in quality of the antenna pedestal orientation transmission system in the industry, which relies on the professional knowledge and engineering experience of operators and inspectors, and it is difficult to realize the accurate control of the running-in quality, which is not conducive to the digitization of the assembly process. The application provides an antenna pedestal transmission system running-in quality anomaly detection method, which obtains the data of the running-in process through a vibration acceleration sensor, extracts time domain, frequency domain, wavelet domain and other characteristics, trains the anomaly detection unit under each working condition, and uses weight to weight the prediction results of the anomaly detection unit to judge whether the running-in quality is abnormal.

Description

Technical Field

[0001] This invention belongs to the field of equipment testing, and in particular relates to a method for detecting abnormal running-in quality of an antenna mount transmission system. Background Technology

[0002] The antenna mount azimuth transmission system consists of a motor, a reducer, a gear pair, and other structural components. Its assembly quality directly affects the smoothness, service life, and transmission noise of the entire transmission system.

[0003] Existing evaluation methods suffer from limitations such as relying on single evaluation indicators and expert experience, lacking objective judgment standards, and being unsuitable for mass production. Research indicates that the industry currently lacks effective methods for evaluating the running-in quality of antenna mount azimuth transmission systems. Methods that depend on the professional knowledge and engineering experience of operators and inspectors make it difficult to achieve precise control over running-in quality and hinder the digitalization of the assembly process.

[0004] Therefore, this paper proposes a running-in quality assessment method for antenna mount transmission systems. This method overcomes the shortcomings of existing methods that rely on expert experience and lack objective evaluation standards. It is of great significance for achieving precise control of running-in quality and digitizing the running-in process. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes a method for detecting anomalies in the running-in quality of an antenna mount transmission system. This method uses a vibration acceleration sensor to acquire data from the running-in process and extracts features in the time domain, frequency domain, and wavelet domain. Anomaly detection units are trained for each operating condition, and the prediction results of these units are weighted and fused to determine whether there are any anomalies in the running-in quality. Specifically, the method includes the following steps:

[0006] 1) Obtain the vibration acceleration signal X of the conventional antenna mount system under normal conditions under M different operating conditions. k , k = 1, 2, ..., M.

[0007] 2) Calculate and extract the vibration acceleration signal X k Its time-domain characteristics include mean, variance, skewness, root mean square, kurtosis, peak index, impulse index, waveform index, etc.

[0008] 3) Perform a Fast Fourier Transform on the acquired vibration acceleration signal to obtain f. k .

[0009] 4) Calculate and extract the vibration acceleration signal X k The frequency domain characteristics.

[0010] 5) Regarding the vibration acceleration signal X k Perform N-level wavelet packet decomposition to obtain 2 NA decomposition coefficient vector {d1,d2,...,d2} N}, calculate the energy e of the wavelet packet decomposition coefficient vector di. i : Where L represents the length of the wavelet decomposition coefficient vector of the N-level signal, which is 1 / 2 of the total signal length. N i = 1, 2, ..., 2 N .

[0011] 6) Calculate the energy distribution characteristics E of the wavelet packet decomposition coefficients at the i-th level. i :

[0012] 7) Constructing a feature vector: Combine the extracted time-domain features, frequency-domain features, and wavelet packet coefficient energy distribution features into a feature vector.

[0013] 8) Based on the health data features of multiple working conditions, support vector description models (SVDD) are trained using the feature data of a single working condition. Given k working conditions, the SVDD model is trained using the health data feature set of the i-th working condition. The resulting model is denoted as SVDD. i For the health feature set of k working conditions, k SVDD models were trained.

[0014] 9) Use k SVDD models to predict the health data feature set of the working condition, and obtain the prediction results O1, O2,...,Ok.

[0015] 10) Weight initialization and result fusion: Initialize weights w = {w1, w2, ..., wk}, and fuse the prediction results of k SVDD models: O_pred = w1O1 + w2O2 + ... + wkOk.

[0016] 11) Weight update and result fusion: Compare the predicted result O_pred with the actual result, calculate the prediction error, and perform backpropagation to update the weight w.

[0017] 12) After the termination condition is met, the final weight w* is obtained.

[0018] 13) Prediction of unknown vibration acceleration signals: For unknown vibration acceleration signals, truncation and feature extraction are performed respectively. The prediction results of SVDD1, SVDD2, ..., SVDDk are weighted by weight w* to obtain the final prediction result.

[0019] The beneficial effects of this invention are as follows:

[0020] 1. The model is trained using only normal data from the antenna mount running-in process, avoiding the problems of missing or imbalanced samples in traditional models.

[0021] 2. It can detect anomalies in the antenna mount running-in process under multiple operating conditions, and is not limited to a single operating condition. Attached Figure Description

[0022] Figure 1 This is a flowchart of the feature extraction process.

[0023] Figure 2 This is a schematic diagram of wavelet packet decomposition.

[0024] Figure 3 This is a flowchart of the model training process.

[0025] Figure 4 This is a flowchart for anomaly detection.

[0026] Figure 5 This is a basic diagram of an antenna mount.

[0027] Figure 6 This is a schematic diagram of the SVDD1 detection results.

[0028] Figure 7 This is a schematic diagram of the SVDD2 detection results.

[0029] Figure 8 This is a schematic diagram of the SVDD3 detection results.

[0030] Figure 9 This is a schematic diagram of the final test results. Detailed Implementation

[0031] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] The basic components of the antenna mount are shown in the attached figure. Figure 5 As shown, it includes a motor reducer, a bus ring, an azimuth synchronization mechanism, a pitch assembly, and a rotary hinge. An acceleration sensor is placed at the motor reducer to acquire vibration acceleration signals during the antenna mount's run-in process.

[0033] Feature extraction, as shown in the appendix Figure 1 and 2 As shown, the features include time-domain features, frequency-domain features, and wavelet-domain features. When extracting wavelet-domain features, follow the attached... Figure 2 Wavelet packet decomposition was performed, with the selected mother wavelet being 'db4', and the decomposition level being 3.

[0034] The SVDD model training has three operating conditions, denoted as Condition 1, Condition 2, and Condition 3. The support vector description model is trained using normal data from Condition 1 to obtain SVDD1; the support vector description model is trained using normal data from Condition 2 to obtain SVDD2; and the support vector description model is trained using normal data from Condition 3 to obtain SVDD3.

[0035] Weight initialization and update: Initialize weight w = {1,1,1}. Use SVDD1, SVDD2 and SVDD3 to detect the normal data of working conditions 1 to 3 respectively. Record the detection results as O1, O2 and O3. Apply weights to the detection results to obtain the fused detection result. Then, backpropagate based on the fused detection result to update the weight w until the termination condition is met.

[0036] The test set, consisting of normal and abnormal data from operating conditions 1-3, was used to validate the trained model. The results are attached. Figure 6 -Appendix Figure 9 In the test set, the detection accuracy of a single SVDD model was less than 92%, while the detection accuracy of the present invention reached 98.9%, demonstrating the effectiveness of the present invention.

[0037] This invention is not limited to the specific embodiments described above, and various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made to the above embodiments based on the technical essence of this invention should be included within the scope of protection of this invention.

Claims

1. A method for detecting abnormal running-in quality of an antenna mount transmission system, characterized in that: Specifically, the following steps are included: 1) Obtain the vibration acceleration signal X of the conventional antenna mount system under normal conditions under M different operating conditions. k k=1,2,…,M; 2) Calculate and extract the vibration acceleration signal X k The temporal characteristics; 3) Perform a Fast Fourier Transform on the acquired vibration acceleration signal to obtain f. k ; 4) Calculate and extract the vibration acceleration signal X k Frequency domain characteristics; 5) Regarding the vibration acceleration signal X k Perform N-level wavelet packet decomposition to obtain Decomposition coefficient vector Calculate the energy of the wavelet packet decomposition coefficient vector di. Where L represents the length of the wavelet decomposition coefficient vector of the N-level signal, which is 1 / 2 of the total signal length. N , 6) Calculate the energy distribution characteristics of the wavelet packet decomposition coefficients at the i-th level. ; 7) Constructing a feature vector: Combine the extracted time-domain features, frequency-domain features, and wavelet packet decomposition coefficient energy distribution features into a feature vector; 8) Based on the health data features under multiple working conditions, use the health data features under a single working condition to train the Support Vector Description Model (SVDD) respectively, and obtain k SVDD models; 9) Using k SVDD models, the health data feature set of the working condition is predicted, and the prediction results are O1, O2, ..., Ok; 10) Initialize weights w={w1,w2,...,wk} and fuse the prediction results of k SVDD models: O_pred=w1O1+w2O2+…+wkOk; 11) Weight update and result fusion: Compare the predicted result O_pred with the actual result, calculate the prediction error, and perform backpropagation to update the weight w; 12) After the termination condition is met, the final weight is obtained. 13) Prediction of unknown vibration acceleration signals: For unknown vibration acceleration signals, truncation and feature extraction are performed, and weights are used... The prediction results of SVDD1, SVDD2, ..., SVDDk are weighted to obtain the final prediction result.

Images