A noise detection method and system for speed reducers

Through the multi-state detection and convolutional neural network model combined with historical data, noise detection is carried out for different working states and usage scenarios of the reducer, which solves the problem of large detection errors in the existing technology, and achieves more efficient and accurate fault diagnosis and Noise assessment.

CN118706446BActive Publication Date: 2024-11-08ZHEJIANG SANKAI MECHANICAL & ELECTRICAL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411205270.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-08
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

The existing reducer noise detection technology cannot accurately detect different working conditions and usage scenarios, resulting in large detection errors and lack of personalized evaluation for different reducers.

Method used

Multi-state detection, Mel frequency transformation, convolutional neural network model and nonlinear computing model are used, combined with historical data and vibration data, and the corresponding strategies are implemented to improve detection accuracy and efficiency through comparison of abnormality degree evaluation values ​​and standard thresholds.

Benefits of technology

It improves the accuracy and reliability of reducer fault diagnosis, enhances the robustness and generalization capabilities of the model, and provides an intelligent, precise and efficient noise detection solution, suitable for different scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118706446B_ABST
    Figure CN118706446B_ABST
Patent Text Reader

Abstract

The invention discloses a noise detection method and system applied to a reducer, and relates to the technical field of noise detection. The system comprises a data acquisition and preprocessing module, an audio feature extraction module, a neural network model construction module, a comprehensive analysis module and a result display module. The technical key points are as follows: different calculation models are built for different reducers, and corresponding data are used to calculate an abnormality evaluation value to ensure the accuracy of the abnormality evaluation value, and the abnormality evaluation value is corrected by historical usage data in a nonlinear correction calculation model to more accurately evaluate the abnormal risk degree of the reducer; by comprehensively utilizing historical data, establishing a nonlinear calculation model, setting a standard threshold, formulating different strategies and performing data preprocessing, the accuracy, reliability and efficiency of reducer fault diagnosis are effectively improved, and at the same time, a more intelligent, precise and efficient technical solution is provided for noise detection of the reducer.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of noise detection, and in particular to a noise detection method and system applied to a reducer. Background Art

[0002] Noise detection refers to the collection, analysis and evaluation of sound through acoustic measurement equipment to determine the intensity, frequency, waveform and other characteristics of the sound, so as to identify and locate the noise source; for reducer noise detection, the existing solutions mainly include: sound level meter measurement: use a sound level meter to measure the noise sound pressure level of the reducer when it is working, and determine whether it meets the relevant noise standards; spectrum analysis: use a spectrum analyzer to perform frequency analysis on the noise and identify the main noise sources, such as gear meshing, bearing rolling, motor vibration, etc.; sound source location: use a sound source locator to locate the noise and determine the location of the noise source in order to carry out targeted noise reduction; noise simulation: use finite element analysis software to simulate the noise of the reducer, simulate the noise characteristics of the reducer when it is working, and provide a reference for noise reduction design.

[0003] The existing application publication number is: CN111610023A, and the document named "A reducer noise evaluation method, device and handheld reducer noise evaluation instrument" points out: It solves the problem of low reducer detection quality and inability to evaluate reducer performance in the prior art. The present invention includes setting the speed parameter of the reducer for collection, inputting standard sample data, setting preprocessing parameters, and setting evaluation standards; acquiring and storing the sound signal, vibration signal and speed signal of the reducer; performing equal-angle resampling on the vibration signal, and performing order analysis on the resampled data to obtain order analysis data; performing FFT transformation on the order analysis data to generate an order diagram of the data to be tested, and obtaining an evaluation grade after comparison; the present invention realizes high-quality detection of reducer noise and obtains the evaluation result of the reducer. Although it can integrate sound signals and vibration signals, it does not complete targeted noise detection for reducers running under different conditions, and the accuracy of the detection results needs to be improved.

[0004] In combination with the above documents and prior art, when traditionally detecting noise on a reducer, the working state is not considered first. The noise generated during normal operation is only collected, and the subsequent analysis and processing process is completed based on the noise data. The model is then trained to ensure the accuracy of the subsequent analysis results. However, before this, some potential faults or abnormalities may only appear in a single or several working states. If the reducer only maintains a single working state for subsequent detection, it is easy to cause large detection errors. Secondly, for reducers moving in different situations, for example, for a newly produced reducer and a reducer that has been in use, there needs to be a distinction or difference in the evaluation data considered by the two. If the same method or process is used for detection, it is also easy to cause detection errors. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the shortcomings of the prior art, the present invention provides a noise detection method and system for reducers, builds different calculation models for different reducers, and uses corresponding data to calculate the abnormality evaluation value to ensure the accuracy of the abnormality evaluation value; through the comprehensive use of historical data, establishment of nonlinear calculation models, setting standard thresholds, formulating different strategies and performing data preprocessing, such technical means, the accuracy, reliability and efficiency of reducer fault diagnosis are effectively improved, and at the same time, a more intelligent, precise and efficient technical solution is provided for noise detection of reducers, which can cope with different scenarios, has strong applicability, and solves the problems raised in the background technology.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A noise detection system applied to a reducer, comprising:

[0010] The data acquisition and preprocessing module collects the noise data of the reducer when it is running in different working states in a soundproof environment, and preprocesses the noise data; the working states of the reducer include no-load, full-load, acceleration and deceleration;

[0011] The audio feature extraction module uses the Mel frequency conversion formula to convert the audio in the noise data into Mel frequency, calculates and generates the Mel spectrum, extracts the MFCC features, and forms a feature matrix;

[0012] The neural network model building module selects the convolutional neural network as the model architecture. After receiving the feature matrix, it extracts the local features of the audio data through multiple convolutional layers, then processes the feature dimensions through the pooling layer, and finally uses the fully connected layer to map the extracted features to the category space, and uses the output layer to output the prediction results;

[0013] The comprehensive analysis module determines whether the reducer has normal noise or abnormal noise based on the prediction results. When it is determined to be normal noise, no response action is taken; when it is determined to be abnormal noise, the determination and analysis mechanism is triggered under the condition of determining the noise type. The process is as follows: S101, determine whether there is a use record of the reducer to be detected;

[0014] S102. If there is a usage record for the reducer to be tested, the historical usage data of the reducer is obtained, and a nonlinear correction calculation model is built in combination with the Mel frequency and vibration data obtained within a predetermined time period T to generate an abnormality evaluation value;

[0015] S103. If there is no usage record for the reducer to be tested, a nonlinear interactive calculation model is constructed based on the Mel frequency and vibration data obtained within a predetermined time period T to generate an abnormality evaluation value;

[0016] S104, comparing the abnormality evaluation value with a standard threshold preset by a statistical analysis method, and executing a corresponding strategy according to the comparison result.

[0017] Furthermore, the noise data includes the sound signal of the reducer during operation, including but not limited to the sound pressure level, audio frequency and timbre; the process of preprocessing the noise data includes but is not limited to:

[0018] Resampling: Convert all audio to a fixed sampling rate;

[0019] Re-channel: A mono audio file is converted to stereo by copying the signal of the first channel to the second channel;

[0020] Truncation: All audio intervals are adjusted to the same length.

[0021] Furthermore, the Mel frequency transformation formula is as follows:

[0022] ;

[0023] In the formula, Mel(f) represents the value of frequency f on the Mel frequency scale, f represents ordinary audio, 2595 represents a constant, and 700 represents a constant, which represents the sensitivity of the human ear to the sound frequency.

[0024] Furthermore, in the convolutional neural network, it includes:

[0025] Input layer: receives the feature matrix as the input of the model;

[0026] Several convolutional layers: use convolution kernels to perform convolution operations on the input feature matrix to extract local features, including but not limited to frequency change trends and pitch changes;

[0027] Pooling layer: reduces the dimension of the feature matrix output by the convolutional layer;

[0028] Fully connected layer: maps the feature vector output by the pooling layer to the category space. Each neuron is connected to all neurons in the previous layer and activated using a nonlinear activation function.

[0029] Output layer: Outputs the predicted value based on the classification results.

[0030] Further, in S102 and S103, the historical usage data of the reducer includes the number of failures and the number of maintenances;

[0031] The vibration data represents the vibration acceleration, and the vibration data is obtained synchronously when the Mel frequency is obtained, and is pre-stored in a temporary storage space built into the system. When the judgment and analysis mechanism is triggered, the vibration data in the temporary storage space is synchronously retrieved. If the judgment and analysis mechanism is not triggered, the temporary storage space is cleared.

[0032] Furthermore, in S102 and S103, the process of generating the abnormality evaluation value is as follows:

[0033] First, the Mel frequency and vibration data are obtained within a predetermined time period T, and the Mel frequency and vibration data are dimensionlessly processed to generate a standardized noise stability index and a standardized vibration stability index, the formulas are as follows:

[0034] (1);

[0035] (2);

[0036] In formula (1), represents the standardized noise stability index, t represents the number of the corresponding moment, and the moment represents a time point or a time period, t=1, 2, ..., T, T is a positive integer, Indicates the Mel frequency corresponding to the current moment, Represents the average value of the Mel frequency at all times within the time period T;

[0037] In formula (2), represents the standardized vibration stability index, t represents the number of the corresponding moment, and the moment represents a time point or a time period, t=1, 2, ..., T, T is a positive integer, Indicates the corresponding vibration data at the current moment, Represents the average value of vibration data at all times within the time period T;

[0038] Secondly, the formula for generating the abnormality assessment value is as follows:

[0039] ;

[0040] In the formula, Adav represents the abnormality assessment value, are weight coefficients, They are also weight coefficients, and the value range of each weight coefficient is 0~1. gc and jc represent the number of failures and maintenance times respectively. NUR means when there is no usage record, and EUR means when there is a usage record.

[0041] Furthermore, the process of executing the corresponding strategy according to the comparison results is as follows:

[0042] When the abnormality assessment value does not exceed the standard threshold, the monitoring and maintenance strategy is executed, the corresponding reducer is marked, and it is included in the long-term monitoring plan for inspection and maintenance according to the established cycle;

[0043] When the abnormality assessment value exceeds the standard threshold, the priority repair strategy is executed, and several reducers whose abnormality assessment values ​​exceed the standard threshold are sorted according to priority. Under the condition that the abnormality assessment value exceeds the standard threshold, the abnormality assessment value is positively correlated with the priority, and staff are arranged to repair and replace each reducer according to priority.

[0044] Furthermore, the process of obtaining the preset standard threshold is as follows:

[0045] Based on historical data: Analyze the abnormality evaluation value data of the reducer in past operation to obtain the range of abnormality evaluation values ​​under normal noise and the range of abnormality evaluation values ​​when abnormal noise occurs;

[0046] Statistical analysis: After preliminary analysis, the abnormality evaluation values ​​of the reducer under normal noise and abnormal noise are collected; the collected data are statistically analyzed to determine the mean and standard deviation of the abnormality evaluation values ​​under normal noise; the standard threshold is set according to the statistical results, and the calculation formula of the standard threshold is as follows:

[0047] ;

[0048] In the formula, By represents the standard threshold, It represents the average value of the abnormality evaluation value under normal noise, S represents the multiple, and the value of S is 2~3. Represents the standard deviation of the abnormality assessment value under normal noise.

[0049] Furthermore, it also includes a result display module, which displays the prediction results and the results of judging whether each reducer has normal noise or abnormal noise in a table in a first page scrolling on the display end, and synchronously displays the results of comparing the abnormality degree assessment value with the standard threshold and the corresponding executed strategy in a second page scrolling on the display end.

[0050] A noise detection method applied to a reducer comprises the following specific steps:

[0051] S1. In a soundproof environment, collect the noise data of the reducer when it is running under different working conditions, and pre-process the noise data; the working conditions of the reducer include no-load, full-load, acceleration and deceleration;

[0052] S2, using the Mel frequency conversion formula to convert the audio in the noise data into Mel frequency, and calculate and generate the Mel frequency spectrum, extract the MFCC features, and form a feature matrix;

[0053] S3. Select convolutional neural network as the model architecture. After receiving the feature matrix, extract the local features of the audio data through multiple convolutional layers, then process the feature dimensions through the pooling layer, and finally use the fully connected layer to map the extracted features to the category space, and use the output layer to output the prediction results.

[0054] S4. Determine whether the reducer has normal noise or abnormal noise based on the prediction results. If it is determined to be normal noise, no response action is taken; if it is determined to be abnormal noise, the judgment and analysis mechanism is triggered under the condition of determining the noise type. The process is as follows:

[0055] Determine whether there is a usage record for the reducer to be tested;

[0056] If there is a usage record for the reducer to be tested, the historical usage data of the reducer is obtained, and the Mel frequency and vibration data obtained within the predetermined time period T are combined to build a nonlinear correction calculation model to generate an abnormality evaluation value; if there is no usage record for the reducer to be tested, the Mel frequency and vibration data obtained within the predetermined time period T are combined to build a nonlinear interaction calculation model to generate an abnormality evaluation value;

[0057] Compare the abnormality assessment value with the standard threshold value preset by statistical analysis method, and execute the corresponding strategy according to the comparison result;

[0058] S5. The prediction results and the results of judging whether each reducer has normal noise or abnormal noise are displayed in a table on a first page scrolling on the display end, and the results of comparing the abnormality evaluation value with the standard threshold and the corresponding strategy executed are synchronously displayed on a second page scrolling on the display end.

[0059] (III) Beneficial effects

[0060] The present invention provides a noise detection method and system for a reducer, which has the following beneficial effects:

[0061] This solution effectively improves the accuracy and reliability of reducer fault diagnosis through multi-state detection, soundproof environment, Mel frequency transformation, data enhancement, and processing technologies, and enhances the robustness and generalization ability of the model, which facilitates the effectiveness of subsequent noise detection work;

[0062] This solution adopts a reasonable network structure, such as convolutional layer, pooling layer, fully connected layer, etc., which can effectively reduce the training time and computing resource consumption of the model, improve the training efficiency of the model, and enhance the generalization ability of the model. It provides a more efficient, accurate and reliable technical solution for the diagnosis of reducer abnormality categories, and has high application value.

[0063] This solution builds different calculation models for different reducers, and uses corresponding data to calculate the abnormality assessment value to ensure the accuracy of the abnormality assessment value. In the nonlinear correction calculation model, the abnormality assessment value is corrected by historical usage data to more accurately assess the abnormality risk level of the reducer;

[0064] This solution effectively improves the accuracy, reliability and efficiency of reducer fault diagnosis by comprehensively utilizing historical data, establishing nonlinear calculation models, setting standard thresholds, formulating different strategies and performing data preprocessing. It also provides a more intelligent, precise and efficient technical solution for reducer noise detection, which can cope with different scenarios and has strong applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a schematic diagram of the modular structure of a noise detection system applied to a reducer in the present invention;

[0066] Figure 2 It is a schematic diagram of the overall process of a noise detection method applied to a reducer in the present invention;

[0067] Figure 3 Refer to the figure for the actual working of the noise detection of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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] Research and development concept:

[0070] When testing the noise of the reducer, first use a microphone to collect the sound of the product when it is running. Specifically, a condenser microphone with a dual condenser sound capsule design can easily capture sounds from all angles, with good recording effects and low costs. The microphone collects sound data to achieve environmental perception. After analyzing the sound data, the mechanical part is mainly based on PLC to make corresponding controls to manipulate the valves at the functional testing station.

[0071] The design of the software part includes data collection and preprocessing, audio feature extraction, data loading, construction of a neural network model, and display of results after comprehensive analysis. The noise data collected from the reducer is processed as described above, and the classification model finally constructed can successfully distinguish normal noise from abnormal noise such as wear, aging, and insufficient lubrication. Through real-time monitoring and data feedback, problems can be discovered and maintained in a timely manner, thereby improving the safety and efficiency of the equipment.

[0072] Example 1

[0073] See also Figure 1 and Figure 3 , this embodiment provides a noise detection system for a reducer, the system includes various functional modules that run in sequence, namely, a data acquisition and preprocessing module, an audio feature extraction module, a neural network model building module, a comprehensive analysis module and a result display module. The application scenario of the system is: placing a number of reducers in a separate soundproof chamber for separate synchronous noise detection, and uniformly displaying the detection results on a display terminal;

[0074] Data acquisition and preprocessing module:

[0075] In a soundproof environment, collect the noise data of the reducer when it is running in different working states, and pre-process the noise data; the working states of the reducer include no-load, full-load, acceleration and deceleration states;

[0076] No-load: It means the reducer rotates at a constant speed after being powered on normally;

[0077] Full load: means that the output end of the reducer is connected to the damper, and the rotation resistance set by the damper is the standard resistance value received by the reducer during actual operation, which is obtained through advance experimental tests or based on historical test experiments;

[0078] Acceleration: It means that after the reducer is powered on normally, it will complete uniform acceleration according to the acceleration program built into the established PLC;

[0079] Deceleration: It means that after the reducer is powered on normally, it will complete uniform deceleration according to the deceleration program built into the established PLC;

[0080] By running the reducer, it passes through these four working states in sequence, thus ensuring the rationality and effectiveness of the subsequent test results. Since some fault hazards or abnormalities may only appear in a single or several working states, if the reducer is only kept in a single working state for subsequent testing, it is easy to cause a large test error;

[0081] Noise data includes the sound signals of the reducer during operation, including but not limited to sound pressure level, audio frequency and timbre;

[0082] In a soundproof environment, the reducer to be tested is placed in a soundproof chamber, so that the collection environment should be controlled as much as possible to reduce the interference of background noise, thereby ensuring the accuracy of the noise data;

[0083] The process of preprocessing the noisy data is as follows:

[0084] Resampling: Convert all audio data to a fixed sampling rate (e.g. 16kHz) to ensure data consistency;

[0085] Re-channel: Mono audio files can be converted to stereo by copying the signal of the first channel to the second channel for subsequent processing;

[0086] Padding or truncation: adjust the intervals of all audios to the same length (for example, 1 second) to avoid subsequent processing problems caused by length differences;

[0087] In addition, it also includes: frequency masking: randomly masking certain frequencies in the spectrum diagram to simulate the real auditory environment and enhance the model's robustness to noise of different frequencies; time masking: randomly selecting time intervals on the time axis for masking to improve the model's adaptability to time changes; time shifting: randomly shifting the audio signal to the left or right to improve data diversity and enhance classification performance.

[0088] Audio feature extraction module:

[0089] The Mel frequency conversion formula is used to convert the audio in the noise data into Mel frequency, which can better simulate the human ear's perception of sound frequency, calculate and generate the Mel frequency spectrum, extract the MFCC (Mel frequency cepstral coefficient) features, form a feature matrix, and provide input for subsequent modeling;

[0090] Among them, the Mel frequency transformation formula is as follows:

[0091] ;

[0092] In the formula, Mel(f) represents the value of frequency f on the Mel frequency scale, f represents ordinary audio, and the unit is Hertz (Hz), 2595 represents a constant used to align the Mel frequency scale with the audio scale, log10 represents the logarithm with base 10, and 700 represents a constant representing the sensitivity of the human ear to the sound frequency. Below 700 Hz, the human ear is more sensitive to frequency changes, while above 700 Hz, it is less sensitive to frequency changes;

[0093] Explanation of the formula: This formula maps ordinary audio to a nonlinear Mel frequency scale, which is consistent with the way the human ear perceives sound frequency. In the lower frequency range, the Mel frequency scale is almost linear with the ordinary audio scale, while in the higher frequency range, the Mel frequency scale becomes smoother; in other words, for low-frequency sounds, the human ear is more sensitive to audio changes, while for high-frequency sounds, the perception of audio changes is slower;

[0094] Summary: Mel frequency is a nonlinear frequency scale based on human ear perception. It can better reflect the way the human ear perceives sound frequency, so it has a wide range of applications in audio processing and feature extraction.

[0095] In addition, when calculating and generating the Mel-spectrogram, the Chroma and Spectral Contrast features can also be extracted to form a feature matrix, which will not be described in detail here. A built-in data loader is also configured between the audio feature extraction module and the neural network model construction module for data loading and batch processing; the specific content is: write a custom DataLoader, use the PyTorch or TensorFlow framework to load the preprocessed noise data, and perform batch processing and data enhancement operations (such as random shearing and scaling) at the same time.

[0096] Specifically, this embodiment adopts:

[0097] Multi-state detection: The solution is designed with four working states: no-load, full-load, acceleration, and deceleration. The reducer is allowed to pass through these states in sequence, so that the fault characteristics under different states can be more comprehensively captured, avoiding the errors caused by single-state detection; Soundproof environment: By placing the reducer in a soundproof chamber, background noise interference is reduced, the accuracy of noise data collection is improved, and errors are reduced;

[0098] Mel frequency conversion: Mel frequency conversion is used to convert audio data into Mel frequency, which is more in line with the human ear's perception of sound frequency. MFCC features are extracted to more accurately identify sound features. Data enhancement: Random shearing, scaling and other data enhancement operations are performed in the data loader to improve data diversity, enhance the generalization ability of the model, reduce the model's dependence on specific noise, and reduce errors.

[0099] In general, this solution effectively improves the accuracy and reliability of reducer fault diagnosis through technical means such as multi-state detection, soundproof environment, Mel frequency transformation, data enhancement, and processing, and enhances the robustness and generalization ability of the model, which facilitates the effectiveness of subsequent noise detection work.

[0100] Neural network model building blocks:

[0101] Convolutional neural network (CNN) is selected as the main model architecture, which has strong feature learning ability and can effectively process high-dimensional audio data. After receiving the feature matrix, it extracts local features of the audio data through multiple convolutional layers, then processes the feature dimensions through the pooling layer, and finally maps the extracted features to the category space through the fully connected layer, and uses the output layer to output the prediction results;

[0102] The following is a detailed explanation and examples of each layer:

[0103] 1. Input Layer:

[0104] Function: Receive the preprocessed audio feature matrix as the input of the model;

[0105] For example, suppose we use Mel Frequency Cepstral Coefficients (MFCC) as audio features, extract 13 MFCC coefficients, and convert them into 128 frames. Then the dimension of the feature matrix received by the input layer is (13, 128).

[0106] 2. Multiple Convolutional Layers:

[0107] Function: Use the convolution kernel to perform convolution operation on the input feature matrix to extract local features;

[0108] It should be noted that local features include, but are not limited to, frequency change trends and pitch changes;

[0109] For example, suppose we use a 3*3 convolution kernel with a stride of 1 to perform a convolution operation on the input feature matrix; each convolution kernel can extract a specific type of feature, such as frequency change trend, pitch change, etc.; the first convolution layer: input dimension (13, 128), use 32 3*3 convolution kernels, output dimension (32, 126); the second convolution layer: input dimension (32, 126), use 64 3*3 convolution kernels, output dimension (64, 124); each convolution layer is usually followed by a nonlinear activation function, such as the ReLU function, to enhance the nonlinear expression ability of the model;

[0110] 3. Pooling Layer:

[0111] Function: Reduce the dimension of the feature matrix output by the convolution layer, retain important feature information, and reduce computational complexity;

[0112] Example: Common pooling operations include Max Pooling and Average Pooling; Max Pooling: Use a 2*2 pooling window with a step size of 2, and select the maximum value in each region as the output; Average Pooling: Use a 2*2 pooling window with a step size of 2, and calculate the average value in each region as the output; For example, use Max Pooling on the output (64, 124) of the second convolutional layer, and the output dimension is (64, 62);

[0113] 4. Fully Connected Layer:

[0114] Function: Map the feature vector output by the pooling layer to the category space. Each neuron is connected to all neurons in the previous layer and activated using a nonlinear activation function.

[0115] For example, the pooling layer output (64, 62) is expanded into a one-dimensional vector (3968), and then a fully connected layer is used to map the vector into a 10-dimensional vector space, representing 10 different noise types. The noise types are listed as follows: normal noise, excessive wear, poor lubrication, imbalance, and bearing failure.

[0116] Excessive wear, poor lubrication, imbalance, and bearing failure are abnormal noises. 1. Excessive wear: Long-term operation or excessive load may cause increased wear of the internal parts of the reducer, thereby increasing noise and vibration. 2. Poor lubrication: Lack of proper lubrication will lead to increased friction, which in turn will generate more noise and vibration. 3. Imbalance: Due to improper installation, damaged or loose parts, the reducer may become unbalanced, resulting in increased vibration. 4. Bearing failure: Bearing damage or failure is one of the common faults of the reducer, which will cause a significant increase in noise and vibration.

[0117] 5. Output Layer:

[0118] Function: Output the predicted value according to the classification result;

[0119] For example, the 10-dimensional vector output by the fully connected layer is normalized using the Softmax function to obtain a 10-dimensional probability distribution. Each dimension represents the predicted probability of a noise type. Finally, the category with the highest probability is selected as the prediction result.

[0120] In practical applications, the network structure of CNN can be adjusted according to specific data and tasks, such as adding or reducing convolutional layers and pooling layers, using different convolution kernel sizes and pooling window sizes, and selecting different activation functions and loss functions.

[0121] Specifically, this embodiment adopts:

[0122] Convolutional neural network: The convolutional neural network can automatically learn the features in the audio data without manually designing feature extraction methods, which improves the efficiency and accuracy of feature extraction; Multi-layer convolution and pooling: Through multiple convolutional layers and pooling layers, different levels of features in the audio data can be extracted, such as frequency change trends and pitch changes, so as to more comprehensively identify fault characteristics;

[0123] Fully connected layer and Softmax: Use the fully connected layer to map the extracted features to the category space and use the Softmax function for normalization to output the predicted probability of each category, which improves the generalization ability of the model and enables it to better identify unknown noise types;

[0124] In general, this solution adopts a reasonable network structure, such as convolutional layers, pooling layers, fully connected layers, etc., which can effectively reduce the training time and computing resource consumption of the model, improve the training efficiency of the model, and enhance the generalization ability of the model. It provides a more efficient, accurate and reliable technical solution for the diagnosis of reducer abnormality categories and has high application value.

[0125] Comprehensive analysis module:

[0126] The reducer is judged as normal noise or abnormal noise according to the prediction result. If it is judged as normal noise, no response action is taken; if it is judged as abnormal noise, the judgment and analysis mechanism is triggered under the condition of determining the noise type;

[0127] S101, determining whether there is a usage record of the reducer to be tested;

[0128] The judgment process is as follows:

[0129] Define usage records: clarify what is the "usage record" of the reducer. It only needs to obtain whether the reducer has been installed and used. Basic information such as installation time and installation location can be recorded.

[0130] Select the record storage method: Database: Use the database to store records for easy data management and query; File system: Use the file system to store records, such as creating a text file named with the reducer serial number to record relevant information; Label: Put a label on the reducer to record relevant information;

[0131] Automatic judgment method: Different judgment methods can be selected according to different recording methods: Database query: query the database through the reducer serial number or other identification information to determine whether there is a record; File system query: query the file system through the reducer serial number or other identification information to determine whether there is a corresponding file; Label recognition: use image recognition technology to identify the label on the reducer and obtain relevant information;

[0132] Implementation method: Programming: Use programming language to write programs to implement automatic judgment logic; Use third-party libraries: Use specialized database management libraries, file system operation libraries or image recognition libraries; Hardware devices: Use hardware devices with data collection and recognition functions, such as barcode scanners, RFID identifiers, etc.;

[0133] Example: Assume that a database is used to store usage records, and the reducer serial number is used as the primary key;

[0134] import sqlite3

[0135] def check_record(serial_number):

[0136] conn = sqlite3.connect('gearbox_records.db')

[0137] cursor = conn.cursor()

[0138] cursor.execute("SELECT* FROM records WHERE serial_number = ?",(serial_number,))

[0139] result = cursor.fetchone()

[0140] conn.close()

[0141] if result:

[0142] return True # record exists

[0143] else:

[0144] return False # No record exists

[0145] serial_number = '123456789'

[0146] if check_record(serial_number):

[0147] print(f"Reducer {serial_number} has usage record")

[0148] else:

[0149] print(f"Reducer {serial_number} does not have any usage record");

[0150] S102. If there is a usage record for the reducer to be tested, the historical usage data of the reducer is obtained, and a nonlinear correction calculation model is built in combination with the Mel frequency and vibration data obtained within a predetermined time period T to generate an abnormality evaluation value;

[0151] S103. If there is no usage record for the reducer to be tested, a nonlinear interactive calculation model is constructed based on the Mel frequency and vibration data obtained within a predetermined time period T to generate an abnormality evaluation value;

[0152] Wherein, in S102 and S103, the historical usage data of the reducer includes the number of failures and the number of maintenances;

[0153] The vibration data represents the vibration acceleration. The vibration acceleration of the reducer in various states can be measured by using a vibration sensor. The vibration data is obtained synchronously when the Mel frequency is obtained and pre-stored in the temporary storage space built into the system. When the judgment and analysis mechanism is triggered, the vibration data in the temporary storage space is retrieved synchronously. If the judgment and analysis mechanism is not triggered, the temporary storage space is cleared to facilitate the storage of the corresponding vibration data of the next reducer to be detected, thereby ensuring the effective use of the temporary storage space.

[0154] In S102 and S103, the process of generating the abnormality evaluation value is as follows:

[0155] First, the Mel frequency and vibration data are obtained within a predetermined time period T, and the Mel frequency and vibration data are preprocessed, that is, dimensionless, to generate a standardized noise stability index and a standardized vibration stability index. The formula is as follows:

[0156] (1);

[0157] (2);

[0158] In formula (1), represents the standardized noise stability index, t represents the number of the corresponding moment, and the moment represents a time point or a time period, t=1, 2, ..., T, T is a positive integer, Indicates the Mel frequency corresponding to the current moment, Represents the average value of the Mel frequency at all times within the time period T;

[0159] In formula (2), represents the standardized vibration stability index, t represents the number of the corresponding moment, and the moment represents a time point or a time period, t=1, 2, ..., T, T is a positive integer, Indicates the corresponding vibration data at the current moment, Represents the average value of vibration data at all times within the time period T;

[0160] It should be noted that the preset time period T covers all working states of the reducer, namely no-load, full-load, acceleration and deceleration states;

[0161] Secondly, the formula for generating the abnormality assessment value is as follows:

[0162] ;

[0163] In the formula, Adav represents the abnormality evaluation value. The larger the abnormality evaluation value, the higher the abnormality of the corresponding reducer. are weight coefficients used to adjust the relative importance of noise, vibration, and their interactions in the evaluation. They are also weight coefficients, and the value range of each weight coefficient is 0~1. gc and jc represent the number of failures and the number of maintenances respectively. 10*gc / jc represents the correction value. Under normal circumstances, the number of failures is always less than the number of maintenances. When making corrections, it is necessary to add a constant 10 to make adjustments to ensure the rationality of the correction value. The existence of usage records / the absence of usage records are both limiting conditions for the corresponding formulas.

[0164] It should be noted that the weight coefficient is determined by the coefficient of variation method, which is a method of assigning weights to each indicator based on the degree of variation between the current value of each evaluation indicator and the target value. If the numerical difference of a certain indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich information for distinguishing, and thus the indicator should be given a larger weight; conversely, if the numerical difference of each evaluated object on a certain indicator is small, then the ability of this indicator to distinguish the evaluated objects is weak, and thus the indicator should be given a smaller weight. This method directly uses the information contained in each indicator to obtain the weight of the indicator through calculation, so it is objective;

[0165] In practical applications, these weights can be set equal as needed (such as , where k is the total number of weight coefficients, here k=3) to simulate the operation of the corresponding calculation model, and still retain the interaction terms; similarly, a data-driven method can be considered to determine the weights, but here we still use the method of directly setting the weights and emphasize the importance of nonlinear interactions; in order to simplify the explanation, in the end, a simplified but still complex non-weighted (or equal-weighted) evaluation formula is obtained, which can be expressed as: abnormality evaluation value = 1 / 3×standardized noise stability index + 1 / 3×standardized vibration stability index + 1 / 3×(standardized noise stability index×standardized vibration stability index);

[0166] This formula combines the standardized noise, standardized vibration and the interaction terms between them in an equal-weighted manner to evaluate the abnormality of the reducer. It should be noted that the weight coefficient (1 / 3) here is a reflection of equal weight, but in actual applications it may need to be adjusted according to specific circumstances. At the same time, the existence of the interaction term increases the complexity of the formula, enabling it to more comprehensively reflect the performance of the reducer under different conditions.

[0167] S104, comparing the abnormality evaluation value with a standard threshold value preset by using a statistical analysis method;

[0168] When the abnormality assessment value does not exceed the standard threshold, it means that the abnormality of the corresponding reducer is low, and the monitoring and maintenance strategy is implemented, the corresponding reducer is marked, and it is included in the long-term monitoring plan, that is, for example, a 3-year monitoring plan is set, and inspections and maintenance are carried out according to the established cycle to prevent the deterioration of potential abnormal problems;

[0169] When the abnormality assessment value exceeds the standard threshold, it means that the abnormality of the corresponding reducer is high. The priority repair strategy is implemented, and the reducers with abnormality assessment values ​​exceeding the standard threshold are sorted according to priority. Under the condition that the abnormality assessment value exceeds the standard threshold, the abnormality assessment value is positively correlated with the priority. The staff is arranged to repair and replace each reducer according to the priority, and handle it according to the fault type to avoid the expansion of the fault;

[0170] The process of presetting the standard threshold is as follows:

[0171] Based on historical data: Analyze the abnormality evaluation value data of the reducer in the past operation (if there is no use record for the reducer, analyze the reducer of the same model with use record) to understand the range of abnormality evaluation value under normal noise and the range of abnormality evaluation value when abnormal noise occurs. The calculation of abnormality evaluation value under normal noise adopts the same formula as that for generating abnormality evaluation value under the condition of no use record;

[0172] Statistical analysis: After preliminary analysis, collect the abnormality evaluation values ​​of the reducer under normal noise and abnormal noise; perform statistical analysis on the collected data to determine the mean and standard deviation of the abnormality evaluation values ​​under normal noise; set the standard threshold value based on the statistical results, and set the standard threshold value to the mean value of the abnormality evaluation value under normal noise plus a certain multiple of the standard deviation (such as 2 or 3 times the standard deviation); and the calculation formula of the standard threshold value is as follows:

[0173] ;

[0174] In the formula, By represents the standard threshold, It represents the average value of the abnormality evaluation value under normal noise, S represents the multiple, and the value of S is 2~3. Represents the standard deviation of the abnormality assessment value under normal noise;

[0175] Example:

[0176] Assume that through statistical analysis, it is determined that the average value of the abnormality assessment value of the reducer under normal noise is 50, and the standard deviation is 10; in order to ensure timely response when a large abnormality occurs, the standard threshold can be set to the average value plus 2 times the standard deviation, that is, 70; thus, when the abnormality assessment value exceeds 70, it means that the reducer is seriously abnormal and requires immediate maintenance; and an assessment value below 70 indicates that the reducer is always abnormal or has sudden abnormalities, and can be included in the long-term monitoring plan for regular inspection and maintenance; of course, this standard threshold is only an example, and the specific value can also be set and adjusted according to actual conditions.

[0177] Specifically, this solution builds different calculation models for different reducers (i.e., whether they have been used or not), and uses corresponding data to calculate the abnormality assessment value to ensure the accuracy of the abnormality assessment value. In the nonlinear correction calculation model, the abnormality assessment value is corrected by historical usage data to more accurately assess the abnormality risk level of the reducer.

[0178] A standard threshold was set to divide the reducer abnormality diagnosis results into two situations: low abnormality and high abnormality. Corresponding strategies were formulated for the two situations to effectively prevent the expansion of faults and improve maintenance efficiency. During this period, temporary storage space was used to store vibration data, and it was retrieved and cleared according to needs, effectively managing data and improving data utilization.

[0179] In general, this solution effectively improves the accuracy, reliability and efficiency of reducer fault diagnosis by comprehensively utilizing historical data, establishing nonlinear calculation models, setting standard thresholds, formulating different strategies and performing data preprocessing. At the same time, it also provides a more intelligent, precise and efficient technical solution for reducer noise detection, which can cope with different scenarios and has strong applicability.

[0180] Results display module:

[0181] The prediction results and the results of judging whether each reducer has normal noise or abnormal noise are displayed in a table on a first scrolling page on the display end, and the results of comparing the abnormality evaluation value with the standard threshold and the corresponding strategy executed are synchronously displayed on a second scrolling page on the display end;

[0182] Used to display the results clearly, making it easier for staff to observe and record.

[0183] Example 2

[0184] See also Figure 2Based on Example 1, this embodiment further provides a noise detection method for a reducer, comprising the following specific steps: S1, in a soundproof environment, collecting noise data of the reducer when it is running in different working states, and preprocessing the noise data; the working states of the reducer include no-load, full-load, acceleration and deceleration;

[0185] S2, using the Mel frequency conversion formula to convert the audio in the noise data into Mel frequency, and calculate and generate the Mel frequency spectrum, extract the MFCC features, and form a feature matrix;

[0186] S3. Select convolutional neural network as the model architecture. After receiving the feature matrix, extract the local features of the audio data through multiple convolutional layers, then process the feature dimensions through the pooling layer, and finally use the fully connected layer to map the extracted features to the category space, and use the output layer to output the prediction results.

[0187] S4. Determine whether the reducer has normal noise or abnormal noise based on the prediction results. If it is determined to be normal noise, no response action is taken; if it is determined to be abnormal noise, the judgment and analysis mechanism is triggered under the condition of determining the noise type. The process is as follows:

[0188] Determine whether there is a usage record for the reducer to be tested;

[0189] If there is a usage record for the reducer to be tested, the historical usage data of the reducer is obtained, and the Mel frequency and vibration data obtained within the predetermined time period T are combined to build a nonlinear correction calculation model to generate an abnormality evaluation value; if there is no usage record for the reducer to be tested, the Mel frequency and vibration data obtained within the predetermined time period T are combined to build a nonlinear interaction calculation model to generate an abnormality evaluation value;

[0190] Compare the abnormality assessment value with the standard threshold value preset by statistical analysis method, and execute the corresponding strategy according to the comparison result;

[0191] S5. The prediction results and the results of judging whether each reducer has normal noise or abnormal noise are displayed in a table on a first page scrolling on the display end, and the results of comparing the abnormality evaluation value with the standard threshold and the corresponding strategy executed are synchronously displayed on a second page scrolling on the display end.

[0192] In the application, the several formulas involved are all calculated by taking their numerical values ​​after removing the dimensions, and the formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent actual situation. Some coefficients or weights in the formulas are set by technical personnel in this field according to actual conditions, so they will not be elaborated here.

[0193] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0194] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0195] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A noise detection system applied to a reducer, characterized in that: The system includes: The data acquisition and preprocessing module collects the noise data of the reducer when it is running in different working states in a soundproof environment, and preprocesses the noise data; the working states of the reducer include no-load, full-load, acceleration and deceleration; The audio feature extraction module uses the Mel frequency conversion formula to convert the audio in the noise data into Mel frequency, calculates and generates the Mel spectrum, extracts the MFCC features, and forms a feature matrix; The neural network model building module selects the convolutional neural network as the model architecture. After receiving the feature matrix, it extracts the local features of the audio data through multiple convolutional layers, then processes the feature dimensions through the pooling layer, and finally uses the fully connected layer to map the extracted features to the category space, and uses the output layer to output the prediction results; The comprehensive analysis module determines whether the reducer has normal noise or abnormal noise based on the prediction results. When it is determined to be normal noise, no response action is taken; when it is determined to be abnormal noise, the determination and analysis mechanism is triggered under the condition of determining the noise type. The process is as follows: S101, determine whether there is a use record of the reducer to be detected; S102. If there is a usage record for the reducer to be tested, the historical usage data of the reducer is obtained, and a nonlinear correction calculation model is built in combination with the Mel frequency and vibration data obtained within a predetermined time period T to generate an abnormality evaluation value; S103. If there is no usage record for the reducer to be tested, a nonlinear interactive calculation model is constructed based on the Mel frequency and vibration data obtained within a predetermined time period T to generate an abnormality evaluation value; S104, comparing the abnormality evaluation value with a standard threshold preset by a statistical analysis method, and executing a corresponding strategy according to the comparison result.

2. A noise detection system for a reducer according to claim 1, characterized in that: The noise data includes the sound signals of the reducer during operation, including but not limited to the sound pressure level, audio frequency and timbre; the process of preprocessing the noise data includes but is not limited to: Resampling: Convert all audio to a fixed sampling rate; Re-channel: A mono audio file is converted to stereo by copying the signal of the first channel to the second channel; Truncation: All audio intervals are adjusted to the same length.

3. The noise detection system for a reducer according to claim 2, characterized in that: Mel frequency conversion formula is as follows: ; In the formula, Mel(f) represents the value of frequency f on the Mel frequency scale, f represents ordinary audio, 2595 represents a constant, and 700 represents a constant, which represents the sensitivity of the human ear to the sound frequency.

4. The noise detection system for a reducer according to claim 3 is characterized in that: In convolutional neural networks, these include: Input layer: receives the feature matrix as the input of the model; Several convolutional layers: use convolution kernels to perform convolution operations on the input feature matrix to extract local features, including but not limited to frequency change trends and pitch changes; Pooling layer: reduces the dimension of the feature matrix output by the convolutional layer; Fully connected layer: maps the feature vector output by the pooling layer to the category space. Each neuron is connected to all neurons in the previous layer and activated using a nonlinear activation function. Output layer: Outputs the predicted value based on the classification results.

5. A noise detection system for a reducer according to claim 4, characterized in that: In S102 and S103, the historical usage data of the reducer includes the number of failures and the number of maintenances; The vibration data represents the vibration acceleration, and the vibration data is obtained synchronously when the Mel frequency is obtained, and is pre-stored in a temporary storage space built into the system. When the judgment and analysis mechanism is triggered, the vibration data in the temporary storage space is synchronously retrieved. If the judgment and analysis mechanism is not triggered, the temporary storage space is cleared.

6. A noise detection system for a reducer according to claim 5, characterized in that: In S102 and S103, the process of generating the abnormality evaluation value is as follows: First, the Mel frequency and vibration data are obtained within a predetermined time period T, and the Mel frequency and vibration data are dimensionlessly processed to generate a standardized noise stability index and a standardized vibration stability index, the formulas are as follows: ; ; In formula (1), represents the standardized noise stability index, t represents the number of the corresponding moment, and the moment represents a time point or a time period, t=1, 2, ..., T, T is a positive integer, Indicates the Mel frequency corresponding to the current moment, Represents the average value of the Mel frequency at all times within the time period T; In formula (2), represents the standardized vibration stability index, t represents the number of the corresponding moment, and the moment represents a time point or a time period, t=1, 2, ..., T, T is a positive integer, Indicates the corresponding vibration data at the current moment, Represents the average value of vibration data at all times within the time period T; Secondly, the formula for generating the abnormality assessment value is as follows: ; In the formula, Adav represents the abnormality assessment value, are weight coefficients, They are also weight coefficients, and the value range of each weight coefficient is 0~1. gc and jc represent the number of failures and maintenance times respectively. NUR means when there is no usage record, and EUR means when there is a usage record.

7. The noise detection system for a reducer according to claim 6, characterized in that: The process of executing the corresponding strategy according to the comparison results is as follows: When the abnormality assessment value does not exceed the standard threshold, the monitoring and maintenance strategy is executed, the corresponding reducer is marked, and it is included in the long-term monitoring plan for inspection and maintenance according to the established cycle; When the abnormality assessment value exceeds the standard threshold, the priority repair strategy is executed, and several reducers whose abnormality assessment values ​​exceed the standard threshold are sorted according to priority. Under the condition that the abnormality assessment value exceeds the standard threshold, the abnormality assessment value is positively correlated with the priority, and staff are arranged to repair and replace each reducer according to priority.

8. The noise detection system for a reducer according to claim 7, characterized in that: The process of obtaining the preset standard threshold is as follows: Based on historical data: Analyze the abnormality evaluation value data of the reducer in past operation to obtain the range of abnormality evaluation values ​​under normal noise and the range of abnormality evaluation values ​​when abnormal noise occurs; Statistical analysis: After preliminary analysis, the abnormality evaluation values ​​of the reducer under normal noise and abnormal noise are collected; The collected data is statistically analyzed to determine the mean and standard deviation of the abnormality evaluation value under normal noise; the standard threshold is set according to the statistical results, and the calculation formula of the standard threshold is as follows: ; In the formula, By represents the standard threshold, It represents the average value of the abnormality evaluation value under normal noise, S represents the multiple, and the value of S is 2~3. Represents the standard deviation of the abnormality assessment value under normal noise.

9. A noise detection system for a reducer according to claim 8, characterized in that: It also includes a result display module, which displays the prediction results and the results of judging whether each reducer has normal noise or abnormal noise in a table on a first page scrolling on the display end, and synchronously displays the results of comparing the abnormality evaluation value with the standard threshold and the corresponding executed strategy on a second page scrolling on the display end.

10. A noise detection method for a reducer, using the system according to any one of claims 1 to 9, characterized in that: The steps include: S1. In a soundproof environment, collect the noise data of the reducer when it is running under different working conditions, and pre-process the noise data; the working conditions of the reducer include no-load, full-load, acceleration and deceleration; S2, using the Mel frequency conversion formula to convert the audio in the noise data into Mel frequency, and calculate and generate the Mel frequency spectrum, extract the MFCC features, and form a feature matrix; S3. Select convolutional neural network as the model architecture. After receiving the feature matrix, extract the local features of the audio data through multiple convolutional layers, then process the feature dimensions through the pooling layer, and finally use the fully connected layer to map the extracted features to the category space, and use the output layer to output the prediction results. S4. Determine whether the reducer has normal noise or abnormal noise based on the prediction results. If it is determined to be normal noise, no response action is taken; if it is determined to be abnormal noise, the judgment and analysis mechanism is triggered under the condition of determining the noise type. The process is as follows: Determine whether there is a usage record for the reducer to be tested; If there is a usage record for the reducer to be tested, the historical usage data of the reducer is obtained, and the Mel frequency and vibration data obtained within the predetermined time period T are combined to build a nonlinear correction calculation model to generate an abnormality evaluation value; if there is no usage record for the reducer to be tested, the Mel frequency and vibration data obtained within the predetermined time period T are combined to build a nonlinear interaction calculation model to generate an abnormality evaluation value; Compare the abnormality assessment value with the standard threshold value preset by statistical analysis method, and execute the corresponding strategy according to the comparison result; S5. The prediction results and the results of judging whether each reducer has normal noise or abnormal noise are displayed in a table on a first page scrolling on the display end, and the results of comparing the abnormality evaluation value with the standard threshold and the corresponding strategy executed are synchronously displayed on a second page scrolling on the display end.