Intelligent fault diagnosis and self-learning method and system for reducer

Through the combination of Fourier transform, feature extraction and CNN technology, and the particle swarm algorithm, the intelligent fault diagnosis and self-learning of the reducer are realized, the low accuracy problem of traditional fault diagnosis methods is solved, and the robustness of the system is improved and fault detection efficiency.

CN117609869BActive Publication Date: 2024-11-26DOW INTELLIGENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410085658.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-22
Publication Date
2024-11-26
Estimated Expiration
2044-01-22

AI Technical Summary

Technical Problem

Traditional reducer fault diagnosis methods cannot achieve real-time monitoring and intelligent diagnosis, resulting in faults being often discovered after they occur, with low accuracy.

Method used

Fourier transform, frequency domain downsampling, feature extraction and encoding technology combined with convolutional neural network (CNN) for intelligent fault diagnosis of temperature and vibration data, and calculate the optimal control compensation parameters through particle swarm algorithm to realize the reducer Intelligent fault diagnosis and self-learning.

Benefits of technology

It improves the accuracy of reducer fault diagnosis and the robustness of the system, realizes comprehensive monitoring and adaptive optimization of multi-faceted data, and reduces maintenance costs and downtime.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117609869B_ABST
    Figure CN117609869B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of fault diagnosis, and discloses a method and system for intelligent fault diagnosis and self-learning of a reducer, which is used to realize intelligent fault diagnosis of the reducer and improve the accuracy of fault diagnosis of the reducer. The method includes: collecting external influence parameters and frequency domain downsampling of the target reducer to obtain target temperature data and target vibration data; collecting operating state parameters to obtain input speed data and output speed data; processing abnormal values ​​to obtain target torque data and calculate transmission efficiency data; performing feature extraction, feature encoding and vector conversion to obtain temperature transmission relationship vectors and vibration transmission relationship vectors; inputting the temperature transmission relationship vectors and vibration transmission relationship vectors into a reducer fault diagnosis model to perform intelligent fault diagnosis of the reducer and obtain the fault diagnosis result of the reducer; and calculating control compensation parameters to obtain the optimal control compensation parameter combination of the reducer.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to an intelligent fault diagnosis and self-learning method and system for a reducer. Background Art

[0002] With the continuous development of industrial automation, reducers play a vital role in the production process, used to control the speed and output torque of mechanical equipment. However, due to long-term operation and the influence of various working environments, reducers are susceptible to various faults during use, which leads to equipment damage, production interruption and increased maintenance costs.

[0003] Traditional fault diagnosis methods often rely on manual experience or regular maintenance and cannot achieve real-time monitoring and intelligent diagnosis, resulting in faults often being discovered only after they occur, meaning that existing solutions have low accuracy. Summary of the invention

[0004] The present invention provides a speed reducer intelligent fault diagnosis and self-learning method and system, which are used to realize the intelligent fault diagnosis of the speed reducer and improve the accuracy of the speed reducer fault diagnosis.

[0005] The first aspect of the present invention provides an intelligent fault diagnosis and self-learning method for a reducer, which includes: collecting external influencing parameters of a target reducer to be detected to obtain initial temperature data and initial vibration data, and performing Fourier transform and frequency domain downsampling on the initial temperature data and the initial vibration data to obtain target temperature data and target vibration data; collecting operating state parameters of the target reducer to obtain initial speed data and initial torque data, and performing speed classification on the initial speed data to obtain input speed data and output speed data; performing abnormal value processing on the initial torque data to obtain target torque data, and calculating the transmission speed of the target reducer based on the target torque data, the input speed data and the output speed data. efficiency data; extracting features from the transmission efficiency data, the target temperature data and the target vibration data respectively to obtain a plurality of transmission efficiency features, a plurality of target temperature features and a plurality of target vibration features; performing feature coding and vector conversion on the plurality of transmission efficiency features and the plurality of target temperature features to obtain a temperature transmission relationship vector, and performing feature coding and vector conversion on the plurality of transmission efficiency features and the plurality of target vibration features to obtain a vibration transmission relationship vector; inputting the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model to perform intelligent reducer fault diagnosis to obtain a reducer fault diagnosis result; calculating control compensation parameters of the target reducer according to the reducer fault diagnosis result by using a preset particle swarm algorithm to obtain an optimal reducer control compensation parameter combination.

[0006] In combination with the first aspect, in a first implementation method of the first aspect of the present invention, the target reducer to be detected is subjected to external influencing parameter collection to obtain initial temperature data and initial vibration data, and the initial temperature data and the initial vibration data are respectively subjected to Fourier transform and frequency domain downsampling to obtain target temperature data and target vibration data, including: respectively conducting environmental temperature influencing parameter collection on the target reducer to be detected through multiple temperature sensors to obtain temperature sensor signal data of each temperature sensor, and respectively conducting vibration influencing parameter collection on the target reducer through multiple vibration sensors to obtain vibration sensor signal data of each vibration sensor; performing sensor signal fusion and average weighted processing on the temperature sensor signal data of each temperature sensor to obtain initial temperature data, and performing sensor signal fusion and average weighted processing on the vibration sensor signal data of each vibration sensor to obtain initial vibration data; respectively conducting Fourier transform on the initial temperature data and the initial vibration data to obtain temperature frequency domain data and vibration frequency domain data; respectively conducting frequency domain downsampling on the temperature frequency domain data and the vibration frequency domain data to obtain target temperature data and target vibration data.

[0007] In combination with the first aspect, in a second implementation method of the first aspect of the present invention, the initial torque data is processed for outliers to obtain target torque data, and the transmission efficiency data of the target reducer is calculated based on the target torque data, the input speed data and the output speed data, including: determining the target threshold of the initial torque data through a preset DBSCAN algorithm, and performing outlier detection on the initial torque data according to the target threshold to obtain at least one outlier; correcting at least one outlier in the initial torque data to obtain target torque data; calculating the product of the target torque data and the input speed data to obtain input power data, and calculating the product of the target torque data and the output speed data to obtain output power data; performing ratio calculation on the output power data and the input power data to obtain the transmission efficiency data of the target reducer.

[0008] In combination with the first aspect, in a third implementation method of the first aspect of the present invention, the feature extraction is performed on the transmission efficiency data, the target temperature data and the target vibration data respectively to obtain multiple transmission efficiency features, multiple target temperature features and multiple target vibration features, including: obtaining first timestamp data of the transmission efficiency data, and obtaining second timestamp data of the target temperature data, and obtaining third timestamp data of the target vibration data; performing curve fitting on the transmission efficiency data according to the first timestamp data to obtain a transmission efficiency curve, and performing curve fitting on the target temperature data according to the second timestamp data to obtain a target temperature curve, and performing curve fitting on the target vibration data according to the first timestamp data to obtain a transmission efficiency curve. The target vibration data is curve fitted using the three timestamp data to obtain a target vibration curve; curve feature points are extracted from the transmission efficiency curve to obtain multiple candidate efficiency feature points, and curve feature points are extracted from the target temperature curve to obtain multiple candidate temperature feature points, and curve feature points are extracted from the target vibration curve to obtain multiple candidate vibration feature points; feature point screening and feature mapping are performed on the multiple candidate efficiency feature points to generate multiple transmission efficiency features, and feature point screening and feature mapping are performed on the multiple candidate temperature feature points to generate multiple target temperature features, and feature point screening and feature mapping are performed on the multiple candidate vibration feature points to generate multiple target vibration features.

[0009] In combination with the first aspect, in a fourth implementation method of the first aspect of the present invention, the feature encoding and vector conversion of the multiple transmission efficiency characteristics and the multiple target temperature characteristics to obtain a temperature transmission relationship vector, and the feature encoding and vector conversion of the multiple transmission efficiency characteristics and the multiple target vibration characteristics to obtain a vibration transmission relationship vector, include: feature encoding the multiple transmission efficiency characteristics and the multiple target temperature characteristics according to a preset first feature encoding rule to obtain multiple first efficiency feature encoding values ​​and multiple temperature feature encoding values; feature normalizing and vector conversion of the multiple first efficiency feature encoding values ​​and the multiple temperature feature encoding values ​​to obtain a temperature transmission relationship vector; feature encoding the multiple transmission efficiency characteristics and the multiple target vibration characteristics according to a second feature encoding rule to obtain multiple second efficiency feature encoding values ​​and multiple temperature feature encoding values; feature normalizing and vector conversion of the multiple second efficiency feature encoding values ​​and the multiple temperature feature encoding values ​​to obtain a vibration transmission relationship vector.

[0010] In combination with the first aspect, in a fifth implementation of the first aspect of the present invention, the temperature transmission relationship vector and the vibration transmission relationship vector are input into a preset reducer fault diagnosis model to perform intelligent reducer fault diagnosis to obtain a reducer fault diagnosis result, including: inputting the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model, the reducer fault diagnosis model including a first CNN network, a second CNN network and a weight output layer; performing reducer fault prediction on the temperature transmission relationship vector through the first CNN network to obtain a first reducer fault prediction result, the first CNN network including three convolutional pooling layers and one fully connected layer; performing reducer fault prediction on the vibration transmission relationship vector through the second CNN network The system predicts the fault of the reducer by using the coefficient vector to obtain the fault prediction result of the second reducer, wherein the second CNN network includes three convolutional pooling layers and one fully connected layer; the temperature influence weight analysis is performed on the target reducer to obtain the temperature influence weight, and the vibration influence weight analysis is performed on the target reducer to obtain the vibration influence weight; the temperature influence weight and the first reducer fault prediction result are weighted by the weight output layer to obtain the first weighted fault prediction result, and the vibration influence weight and the second reducer fault prediction result are weighted to obtain the second weighted fault prediction result; the first weighted fault prediction result and the second weighted fault prediction result are fused to obtain the reducer fault diagnosis result.

[0011] In combination with the first aspect, in a sixth implementation of the first aspect of the present invention, the preset particle swarm algorithm is used to calculate the control compensation parameters of the target reducer according to the reducer fault diagnosis result to obtain the optimal reducer control compensation parameter combination, including: calibrating the compensation parameter control range of the target reducer according to the reducer fault diagnosis result to obtain the compensation parameter control range; defining the function parameters of the compensation parameter control range to obtain the corresponding target function parameters, and setting the function parameters of the preset initial random generation function through the target function parameters to obtain the target random generation function; performing random operation on the initial value of the compensation parameter of the target reducer through the target random generation function. The method comprises the following steps: generating a particle by a machine to obtain a corresponding random initial value set; performing particle association generation on the random initial value set through the compensation parameter control range to obtain a particle population, and performing population segmentation on the particle population to obtain a plurality of sub-particle populations; matching fitness calculation functions on each of the sub-particle populations to obtain a fitness calculation function corresponding to each of the sub-particle populations; performing fitness calculation on each of the sub-particle populations through the fitness calculation function corresponding to each of the sub-particle populations to obtain a particle fitness set corresponding to the particle population; iteratively calculating the particle fitness set until a preset condition is met, thereby generating an optimal reducer control compensation parameter combination of the target reducer.

[0012] The second aspect of the present invention provides an intelligent fault diagnosis and self-learning system for a reducer, and the intelligent fault diagnosis and self-learning system for a reducer includes: an acquisition module, which is used to collect external influencing parameters of a target reducer to be detected, obtain initial temperature data and initial vibration data, and perform Fourier transform and frequency domain downsampling on the initial temperature data and the initial vibration data, respectively, to obtain target temperature data and target vibration data; a classification module, which is used to collect operating status parameters of the target reducer, obtain initial speed data and initial torque data, and perform speed classification on the initial speed data to obtain input speed data and output speed data; a calculation module, which is used to perform abnormal value processing on the initial torque data to obtain target torque data, and calculate the transmission efficiency data of the target reducer based on the target torque data, the input speed data and the output speed data. an extraction module, used to perform feature extraction on the transmission efficiency data, the target temperature data and the target vibration data respectively, to obtain a plurality of transmission efficiency features, a plurality of target temperature features and a plurality of target vibration features; a conversion module, used to perform feature encoding and vector conversion on the plurality of transmission efficiency features and the plurality of target temperature features, to obtain a temperature transmission relationship vector, and to perform feature encoding and vector conversion on the plurality of transmission efficiency features and the plurality of target vibration features, to obtain a vibration transmission relationship vector; a diagnosis module, used to input the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model to perform intelligent reducer fault diagnosis, to obtain a reducer fault diagnosis result; a compensation module, used to calculate the control compensation parameters of the target reducer according to the reducer fault diagnosis result through a preset particle swarm algorithm, to obtain an optimal reducer control compensation parameter combination.

[0013] The third aspect of the present invention provides an intelligent fault diagnosis and self-learning device for a reducer, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the intelligent fault diagnosis and self-learning device for the reducer executes the above-mentioned intelligent fault diagnosis and self-learning method for the reducer.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, and when the computer-readable storage medium is run on a computer, the computer executes the above-mentioned intelligent fault diagnosis and self-learning method for the reducer.

[0015] In the technical solution provided by the present invention, the target reducer is subjected to external influence parameter collection and frequency domain downsampling to obtain target temperature data and target vibration data; the operating state parameter collection is performed to obtain input speed data and output speed data; abnormal value processing is performed to obtain target torque data and calculate transmission efficiency data; feature extraction, feature encoding and vector conversion are performed to obtain temperature transmission relationship vector and vibration transmission relationship vector; the temperature transmission relationship vector and the vibration transmission relationship vector are input into the reducer fault diagnosis model to perform intelligent fault diagnosis of the reducer to obtain the fault diagnosis result of the reducer; control compensation parameter calculation is performed to obtain the optimal control compensation parameter combination of the reducer. The present invention can realize the monitoring of various data of the reducer, including temperature, vibration, speed, torque, etc., by comprehensively collecting external influence parameters and combining operating state parameters. This helps to form a more comprehensive evaluation of the operating state of the reducer. Intelligent fault diagnosis is performed on temperature and vibration data by using deep learning technology, especially convolutional neural network (CNN). By learning complex patterns in the model data, potential reducer faults can be more accurately identified. The particle swarm algorithm is used to analyze the fault diagnosis results by self-learning and calculate the optimal control compensation parameter combination of the reducer. Such adaptability can continuously optimize the system according to the actual operating conditions, improving the robustness and performance of the entire system. Through the fusion of multiple temperature sensors and vibration sensors, as well as Fourier transform and frequency domain downsampling of the data, the obtained data is more representative and stable, providing a more reliable input for subsequent analysis. Through curve fitting, feature point extraction, encoding and vector conversion, representative features can be efficiently extracted from the original data to form temperature transmission relationship vectors and vibration transmission relationship vectors, which helps to reduce data dimensions and improve processing efficiency, thereby realizing intelligent fault diagnosis of reducers and improving the accuracy of reducer fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of an embodiment of a method for intelligent fault diagnosis and self-learning of a reducer in an embodiment of the present invention;

[0017] Figure 2 A flow chart of calculating transmission efficiency data in an embodiment of the present invention;

[0018] Figure 3 is a flow chart of feature extraction in an embodiment of the present invention;

[0019] Figure 4 It is a flow chart of feature encoding and vector conversion in an embodiment of the present invention;

[0020] Figure 5 A schematic diagram of an embodiment of an intelligent fault diagnosis and self-learning system for a reducer in an embodiment of the present invention;

[0021] Figure 6 It is a schematic diagram of an embodiment of the intelligent fault diagnosis and self-learning device for the reducer in the embodiment of the present invention. DETAILED DESCRIPTION

[0022] The embodiment of the present invention provides a method and system for intelligent fault diagnosis and self-learning of a reducer, which is used to realize intelligent fault diagnosis of a reducer and improve the accuracy of fault diagnosis of a reducer. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the intelligent fault diagnosis and self-learning method of the reducer in the embodiment of the present invention includes:

[0024] S101, collecting external influencing parameters of a target reducer to be detected to obtain initial temperature data and initial vibration data, and performing Fourier transform and frequency domain downsampling on the initial temperature data and initial vibration data respectively to obtain target temperature data and target vibration data;

[0025] It is understandable that the execution subject of the present invention may be a reducer intelligent fault diagnosis and self-learning system, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0026] Specifically, the target reducer to be tested is respectively subjected to environmental temperature impact parameter collection through multiple temperature sensors to obtain temperature sensing signal data of each temperature sensor. Each temperature sensor records temperature sensing signal data at different positions. These sensors can be distributed in different areas to capture temperature changes at different positions of the target reducer. For example, one temperature sensor is installed on the housing of the reducer, while another is located inside it. The data generated by these temperature sensors not only provides diversity, but can also be used for more comprehensive temperature analysis. Then, the target reducer is respectively subjected to vibration impact parameter collection through multiple vibration sensors to obtain vibration sensing signal data of each vibration sensor. These sensors can be installed at different positions to capture vibration information in different directions and parts. For example, one vibration sensor is located near the input shaft of the reducer, while another is located near the output shaft. These vibration sensors record vibration sensing signal data, which contains the vibration mode of the reducer during operation. For temperature sensing signal data, the data of multiple sensors can be fused and averaged to obtain initial temperature data. This helps to eliminate the errors of individual sensors and provide more accurate overall temperature data. Similarly, for vibration sensor signal data, fusion and average weighted processing can be used to generate initial vibration data to reduce noise and improve data consistency. Then, the initial temperature data and initial vibration data are Fourier transformed to obtain temperature frequency domain data and vibration frequency domain data. Fourier transform decomposes the data into different frequency components, which helps to analyze the temperature and vibration changes in different frequency ranges. These frequency domain data can provide more information about the state of the reducer, such as vibration peaks at specific frequencies are associated with specific fault types. In order to reduce the complexity and storage requirements of the data, the temperature frequency domain data and vibration frequency domain data are frequency-domain downsampled. This step helps to reduce the size of the data set while retaining key frequency domain information. Frequency domain downsampling can be performed according to specific needs to ensure that the retained data can still be effectively used for subsequent analysis and fault diagnosis.

[0027] S102, collecting operating state parameters of the target reducer to obtain initial speed data and initial torque data, and classifying the initial speed data to obtain input speed data and output speed data;

[0028] Specifically, in order to collect the initial speed data and initial torque data, corresponding sensors need to be installed on the target reducer. Usually, rotating mechanical equipment such as reducers are equipped with rotation speed sensors and torque sensors. These sensors can measure and record the speed and torque values ​​of the reducer in real time. The speed sensor usually works based on the principle of photoelectric encoder or Hall effect sensor, and is used to monitor the movement speed of the rotating shaft. The torque sensor can use the principle of strain measurement, capacitance or resistance measurement, etc., to measure the torque on the output shaft of the reducer. For example, the server installs a speed sensor and a torque sensor on its input and output shafts respectively. These sensors will collect and record the data of the reducer in real time during operation. The data obtained includes the speed and torque values ​​of the time series. These data are usually continuous and recorded at a very high sampling frequency to obtain detailed operating status information. For example, recording a data point every millisecond can obtain very detailed operating status data. Speed ​​classification is performed to obtain input speed data and output speed data. The purpose of this classification is to distinguish the recorded data into the speed of the input shaft and the speed of the output shaft for further analysis and fault diagnosis. Usually, this classification can be achieved by analyzing the characteristics of the speed data. The speed of the input shaft usually has a high frequency and amplitude of change, while the speed of the output shaft changes relatively smoothly. By applying signal processing techniques such as filtering, Fourier transform, or using model identification methods, these two data with different characteristics can be separated. This will generate input speed data and output speed data. For example, assume that the input speed of the reducer usually varies between 500RPM and 3000RPM, while the output speed varies between 50RPM and 300RPM. By analyzing the speed data, data points in these two ranges can be identified and classified into input and output speed data. This classification process can be implemented using automated algorithms to ensure accuracy and efficiency. Through this step, the server obtains the input speed data and output speed data of the target reducer, which will be used for subsequent analysis and fault diagnosis. The accurate acquisition and classification of input speed data and output speed data are very important for monitoring the operating status of the reducer and detecting potential faults, which helps to improve the performance and reliability of the equipment while reducing downtime and maintenance costs.

[0029] S103, performing abnormal value processing on the initial torque data to obtain target torque data, and calculating the transmission efficiency data of the target reducer according to the target torque data, the input speed data, and the output speed data;

[0030] It should be noted that the server uses DBSCAN (density clustering algorithm) to determine the target threshold for the initial torque data. DBSCAN is a clustering algorithm that can divide data into different clusters based on the density of data points. In this case, the server uses DBSCAN to help determine outliers in the torque data. For example, assume that the torque data of the reducer should vary between 100Nm and 500Nm under normal circumstances. Through the DBSCAN algorithm, the server determines a target threshold so that the torque data is considered normal within this range. Data points outside this range will be considered outliers. The server performs outlier detection on the initial torque data and identifies at least one outlier based on the target threshold. These outliers represent unusual torque data points and are caused by faults or other abnormal conditions. For example, there is a data point that measures a torque of 1000Nm, which is identified as an outlier. Subsequently, for at least one outlier in the torque data, the server performs correction to obtain the target torque data. Correction can be done in different ways, depending on the cause of the outlier. Typically, the server replaces the outlier with the average of the surrounding data points to ensure data consistency and accuracy. The server calculates the product of the target torque data and the input speed data to obtain the input power data. The product of torque and speed represents the input power of the machine, that is, the energy input. For example, if the target torque data is 500Nm and the input speed is 1000RPM, then the input power will be 500Nm*1000RPM=500000Nm / min. Then, the server calculates the product of the target torque data and the output speed data to obtain the output power data. This represents the output power of the machine, that is, the energy provided by the machine when performing work. For example, if the target torque data is 500Nm and the output speed is 100RPM, then the output power will be 500Nm*100RPM=50000Nm / min. The server calculates the ratio of the output power data and the input power data to obtain the transmission efficiency data of the target reducer. The transmission efficiency is the ratio between the input power and the output power, usually expressed as a percentage. For example, if the input power is 500000Nm / min and the output power is 50000Nm / min, the transmission efficiency will be (50000 / 500000)*100%=10%.

[0031] S104, performing feature extraction on the transmission efficiency data, the target temperature data, and the target vibration data respectively to obtain a plurality of transmission efficiency features, a plurality of target temperature features, and a plurality of target vibration features;

[0032] Specifically, the server obtains the timestamp data of the transmission efficiency data, the target temperature data, and the target vibration data. The timestamp data is used to associate information from different data sources to ensure time consistency. Generally, these timestamp data correspond to the time point of data collection. For example, the first timestamp data of the transmission efficiency data corresponds to the start time of the data recording, the second timestamp data of the target temperature data corresponds to the time of the temperature data recording, and the third timestamp data of the target vibration data corresponds to the time of the vibration data. The server uses these timestamp data to perform curve fitting. For the transmission efficiency data, the server uses the first timestamp data to perform curve fitting on the transmission efficiency data to obtain a transmission efficiency curve. Curve fitting can use different mathematical models, such as linear regression, polynomial fitting, or exponential fitting, to adapt to the characteristics of the data. Similarly, the server uses the second timestamp data to perform curve fitting on the target temperature data to obtain a target temperature curve. The temperature curve can reflect the trend of temperature changes over time. At the same time, the server also uses the third timestamp data to perform curve fitting on the target vibration data to obtain a target vibration curve. The vibration curve can help the server understand the changes in vibration. Subsequently, the server extracts feature points from the curve. The server analyzes the shape of the curve to find key features therein. These feature points may include the maximum value, minimum value, slope, curvature, rate of change, etc. of the curve. For example, from the transmission efficiency curve, the server extracts the maximum and minimum values ​​of the efficiency, which reflect the performance peak and lowest point during the operation of the reducer. From the temperature curve, the server extracts the rate of change of temperature to understand the speed of temperature change. From the vibration curve, the server extracts the maximum amplitude and frequency of vibration to understand the intensity and frequency characteristics of vibration. For the extracted multiple candidate feature points, the server performs feature point screening and feature mapping to generate multiple transmission efficiency features, multiple target temperature features, and multiple target vibration features. This step can be achieved by setting feature thresholds, screening conditions, or using machine learning algorithms. For example, for transmission efficiency data, the server screens out the maximum efficiency value and the minimum efficiency value, and then maps them into two transmission efficiency features. For target temperature data, the server screens out feature points with a temperature change rate greater than a certain threshold and maps them into target temperature features. For target vibration data, the server screens out feature points with a vibration frequency higher than a certain threshold and maps them into target vibration features.

[0033] S105, performing feature coding and vector conversion on multiple transmission efficiency features and multiple target temperature features to obtain a temperature transmission relationship vector, and performing feature coding and vector conversion on multiple transmission efficiency features and multiple target vibration features to obtain a vibration transmission relationship vector;

[0034] Specifically, the server performs feature encoding on multiple transmission efficiency features and multiple target temperature features according to a preset first feature encoding rule. Feature encoding is to map the original feature data to a new representation form, usually in a digital way. Different feature encoding rules can be applied to different types of feature data to better capture their important information. After encoding, the server obtains multiple first efficiency feature encoding values ​​and multiple temperature feature encoding values. For these encoded feature values, the server performs feature normalization and vector conversion. Feature normalization is to adjust the values ​​of different features to a similar scale range to ensure that they have similar weights. Normalization can use methods such as minimum-maximum scaling or Z-score standardization. For example, if the server has multiple transmission efficiency feature encoding values ​​of different scales, they are scaled to between 0 and 1 by performing minimum-maximum scaling to ensure that they have similar importance. Similarly, for the temperature feature encoding value, the server uses standardization to adjust it to a distribution with a mean of 0 and a standard deviation of 1. According to the second feature encoding rule, the server performs feature encoding on multiple transmission efficiency features and multiple target vibration features to obtain multiple second efficiency feature encoding values ​​and multiple vibration feature encoding values. These encoding values ​​may have different encoding rules and scales from the first feature encoding values. For example, for the transmission efficiency feature, the server uses binary encoding to represent the efficiency of different states as 0 and 1. For the target vibration feature, the server uses amplitude encoding to map different vibration amplitudes to positive values. For these second feature encoding values, the server also performs feature normalization and vector conversion to obtain the vibration transmission relationship vector.

[0035] S106, inputting the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model to perform intelligent reducer fault diagnosis to obtain a reducer fault diagnosis result;

[0036] Specifically, the preset reducer fault diagnosis model includes a first CNN network, a second CNN network, and a weight output layer. These networks are used to process temperature and vibration transmission relationship vectors for fault prediction. The first CNN network performs reducer fault prediction on the temperature transmission relationship vector to obtain a first reducer fault prediction result. The first CNN network generally includes a convolution layer, a pooling layer, and a fully connected layer for extracting features and performing classification. The network can analyze the temperature transmission relationship vector to predict the state and performance of the reducer. Then, the vibration transmission relationship vector is used to predict the reducer fault, and the second reducer fault prediction result is obtained. The second CNN network also includes a convolution layer, a pooling layer, and a fully connected layer for analyzing the vibration transmission relationship vector to perform fault prediction. For example, if the temperature transmission relationship vector indicates that the temperature change of the reducer is abnormal, the first CNN network can identify this abnormality and perform fault prediction. Similarly, if the vibration transmission relationship vector indicates that the vibration of the reducer is abnormal, the second CNN network can perform corresponding fault prediction. The server performs temperature influence weight analysis and vibration influence weight analysis on the target reducer to obtain temperature influence weight and vibration influence weight. These weights reflect the degree of influence of temperature and vibration on the performance of the reducer. For example, by analyzing historical data and the training results of the model, the server determines that the weight of the influence of temperature on the performance of the reducer is 0.7, while the weight of the influence of vibration is 0.3. This shows that temperature has a greater impact on the reducer. Through the weight output layer, the server weights the prediction results of the temperature impact weight and the first reducer fault prediction result to obtain the first weighted fault prediction result. Similarly, the server weights the prediction results of the vibration impact weight and the second reducer fault prediction result to obtain the second weighted fault prediction result. Through result fusion, the server combines the first weighted fault prediction result and the second weighted fault prediction result to obtain the final reducer fault diagnosis result. This result can indicate the status of the reducer, such as "high temperature fault" and "vibration fault", which helps to detect problems early and take appropriate maintenance measures to improve the reliability and performance of the reducer.

[0037] S107. Calculate the control compensation parameters of the target reducer according to the reducer fault diagnosis result by using a preset particle swarm algorithm to obtain an optimal reducer control compensation parameter combination.

[0038] Specifically, the server calibrates the compensation parameter control range of the target reducer according to the reducer fault diagnosis result. The parameter range that needs to be adjusted is determined according to the diagnosis result. For example, if the fault diagnosis result shows that the reducer has a high temperature problem, the server needs to adjust the parameters of the cooling system, such as the fan speed or the coolant flow rate. The server defines the function parameters of the compensation parameter control range to obtain the corresponding target function parameters. These parameter definitions will be used to measure the performance and effect of different parameter combinations. For example, if the server adjusts the fan speed, the server defines the parameter range as 1000RPM to 2000RPM, and then maps this range to the target function parameters, such as the temperature reduction speed or power consumption. Subsequently, through the target function parameters, the server sets the initial random generation function to generate a set of random initial values. These random initial values ​​represent different compensation parameter combinations. For example, the server uses a random generation function to generate random initial values ​​of the fan speed between 1000RPM and 2000RPM. Through the compensation parameter control range, the server performs particle association generation on the random initial value set to obtain a particle population. These particles represent different parameter combinations, which will be used for further optimization and calculation. For example, if the server has 10 particles, each particle represents a different fan speed value. Then, the server divides the particle population into multiple sub-particle populations. Each sub-particle population will contain several particles for parallel fitness calculation. Each sub-particle population will be matched with a fitness calculation function. The fitness calculation function is a function that measures the performance of a parameter combination and is calculated based on the objective function parameters. These fitness calculation functions will be matched with each sub-particle population to evaluate the performance of each particle. For example, for different fan speed values, the fitness calculation function will evaluate the performance of each speed value, such as the temperature reduction speed and power consumption. Through the fitness calculation function, the server obtains the fitness value of each sub-particle population. These fitness values ​​reflect the performance and effect of different parameter combinations. The server iteratively calculates the particle fitness value until the preset conditions are met, and generates the optimal reducer control compensation parameter combination of the target reducer. The algorithm will continuously adjust the parameter combination until the optimal solution is found, that is, the parameter combination that best suits the current situation. For example, if the goal of a server is to achieve a temperature drop in the shortest possible time, the algorithm will iteratively try different fan speed values ​​to find the most effective speed value that will achieve a rapid temperature drop.

[0039] In the embodiment of the present invention, by comprehensively collecting external influencing parameters and combining operating state parameters, the method can realize the monitoring of various aspects of reducer data, including temperature, vibration, speed, torque, etc. This helps to form a more comprehensive evaluation of the operating state of the reducer. Using deep learning technology, especially convolutional neural network (CNN), intelligent fault diagnosis is performed on temperature and vibration data. By learning complex patterns in the model data, potential reducer faults can be more accurately identified. Using particle swarm algorithm, the fault diagnosis results are analyzed by self-learning, and the optimal reducer control compensation parameter combination is calculated. Such adaptability can continuously optimize the system according to the actual operating conditions, improving the robustness and performance of the entire system. Through the fusion of multiple temperature sensors and vibration sensors, as well as Fourier transform and frequency domain downsampling processing of data, the obtained data is more representative and stable, providing more reliable input for subsequent analysis. By means of curve fitting, feature point extraction, encoding and vector conversion, representative features can be efficiently extracted from the original data, forming temperature transmission relationship vectors and vibration transmission relationship vectors, which helps to reduce the data dimension and improve processing efficiency, thereby realizing intelligent fault diagnosis of the reducer and improving the accuracy of reducer fault diagnosis.

[0040] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0041] (1) Collecting the environmental temperature influence parameters of the target reducer to be inspected through multiple temperature sensors to obtain temperature sensor signal data of each temperature sensor, and collecting the vibration influence parameters of the target reducer through multiple vibration sensors to obtain vibration sensor signal data of each vibration sensor;

[0042] (2) performing sensor signal fusion and average weighted processing on the temperature sensor signal data of each temperature sensor to obtain initial temperature data, and performing sensor signal fusion and average weighted processing on the vibration sensor signal data of each vibration sensor to obtain initial vibration data;

[0043] (3) Performing Fourier transform on the initial temperature data and initial vibration data respectively to obtain temperature frequency domain data and vibration frequency domain data;

[0044] (4) Downsample the temperature frequency domain data and the vibration frequency domain data in the frequency domain to obtain target temperature data and target vibration data.

[0045] Specifically, the target reducer to be tested is respectively collected for the environmental temperature impact parameters through multiple temperature sensors. Each temperature sensor is responsible for monitoring the temperature changes at different locations or under different environmental conditions. These sensors are distributed in different parts of the reducer to fully cover the temperature data. For example, for an industrial reducer, the server will install multiple temperature sensors, including one located near the motor, one located in the gearbox, and one located in the bearing. These sensors will continuously monitor the temperature and generate temperature sensing signal data. Similarly, the target reducer is respectively collected for the vibration impact parameters through multiple vibration sensors. These sensors can monitor the vibration conditions of various parts of the reducer to more comprehensively understand the state of the reducer. For example, one vibration sensor is located at the upper part of the gearbox, another is located at the lower part, and another is located at the bearing. These sensors will monitor the vibration and generate vibration sensing signal data. In order to obtain more accurate initial temperature data and initial vibration data, the server performs sensor signal fusion and average weighted processing on the data of each sensor. This is to reduce the noise in the sensor data and ensure the reliability of the final data. Sensor signal fusion can merge the data of multiple sensors into an overall data set, thereby improving the consistency of the data. Average weighted processing helps remove outliers or unstable data points to obtain smoother data. For example, if there are three temperature sensors, each reporting a different temperature value, the server averages and weights them to obtain a more stable and reliable initial temperature data. Subsequently, the server performs Fourier transform on the initial temperature data and initial vibration data. Fourier transform is a mathematical tool used to convert time domain data into frequency domain data. This step helps analyze the spectral characteristics of temperature and vibration signals to better understand their nature. For temperature data, Fourier transform can help the server understand the periodic changes in temperature, such as temperature changes during day and night or temperature fluctuations caused by periodic cooling system operation. For vibration data, Fourier transform can reveal the frequency components of vibration, which is very important for detecting faults or abnormal vibrations. To simplify data processing and improve computational efficiency, the server performs frequency domain downsampling on temperature frequency domain data and vibration frequency domain data. The server only retains the key information in the spectrum, reducing the redundancy and complexity of the data while retaining information useful for fault diagnosis and performance optimization. For example, if the frequency domain data of a server contains hundreds of frequency points, frequency domain downsampling can reduce it to tens of the most important frequency points, thereby reducing the data volume and computational complexity while retaining key information.

[0046] In a specific embodiment, if Figure 2 As shown, the process of executing step S103 may specifically include the following steps:

[0047] S201, determining a target threshold of initial torque data by using a preset DBSCAN algorithm, and performing outlier detection on the initial torque data according to the target threshold to obtain at least one outlier;

[0048] S202, correcting at least one outlier in the initial torque data to obtain target torque data;

[0049] S203, calculating the product of the target torque data and the input speed data to obtain input power data, and calculating the product of the target torque data and the output speed data to obtain output power data;

[0050] S204, calculating the ratio of the output power data to the input power data to obtain the transmission efficiency data of the target reducer.

[0051] Specifically, the server uses a preset DBSCAN algorithm (density-based spatial clustering algorithm) to determine the target threshold of the initial torque data. The DBSCAN algorithm can help the server find core points and outliers in the data set. A core point is a point with enough data points around it, while an outlier is an isolated point. By analyzing the data set, DBSCAN can automatically determine an appropriate target threshold, which will be used for subsequent outlier detection. For example, suppose there is a set of initial torque data, which includes normal working data and an outlier (due to some fault or abnormal situation). After analyzing these data, the DBSCAN algorithm determines a target threshold to distinguish between normal data and outliers. The server uses the target threshold to perform outlier detection on the initial torque data. The purpose of outlier detection is to identify and mark data points that are significantly different from normal data. These data points represent potential faults or abnormal situations. Typically, the distance or similarity of the data point is compared with the target threshold to determine whether it is an outlier. At least one outlier in the initial torque data is corrected to obtain the target torque data. Outliers are generated due to sensor failure, measurement error, or other reasons. Therefore, these outliers need to be corrected to ensure the consistency and accuracy of the data. For example, if there is an outlier in the server's initial torque data due to a sensor failure, the server estimates the correct torque value by comparing it with the values ​​of the surrounding data points. Then, the server calculates the product of the target torque data and the input speed data to obtain the input power data. The input power represents the power required to drive the reducer. For example, assuming that the target torque data is 100Nm and the input speed data is 1000RPM, then the input power data is 100Nm*1000RPM=100000W (watts). The server calculates the product of the target torque data and the output speed data to obtain the output power data. The output power represents the power output from the reducer to perform a specific task. For example, if the target torque data is 100Nm and the output speed data is 500RPM, then the output power data is 100Nm*500RPM=50000W. The server calculates the transmission efficiency data of the target reducer by dividing the output power data by the input power data. The transmission efficiency represents the efficiency of energy conversion and transmission inside the reducer. For example, in this embodiment, the input power is 100000 W, the output power is 50000 W, and the transmission efficiency is 50%, which means that the reducer converts 50% of the input power into useful output power when working.

[0052] In a specific embodiment, if Figure 3 As shown, the process of executing step S104 may specifically include the following steps:

[0053] S301, acquiring first timestamp data of transmission efficiency data, acquiring second timestamp data of target temperature data, and acquiring third timestamp data of target vibration data;

[0054] S302, performing curve fitting on the transmission efficiency data according to the first timestamp data to obtain a transmission efficiency curve, performing curve fitting on the target temperature data according to the second timestamp data to obtain a target temperature curve, and performing curve fitting on the target vibration data according to the third timestamp data to obtain a target vibration curve;

[0055] S303, extracting curve feature points from the transmission efficiency curve to obtain a plurality of candidate efficiency feature points, extracting curve feature points from the target temperature curve to obtain a plurality of candidate temperature feature points, and extracting curve feature points from the target vibration curve to obtain a plurality of candidate vibration feature points;

[0056] S304, performing feature point screening and feature mapping on multiple candidate efficiency feature points to generate multiple transmission efficiency features, performing feature point screening and feature mapping on multiple candidate temperature feature points to generate multiple target temperature features, and performing feature point screening and feature mapping on multiple candidate vibration feature points to generate multiple target vibration features.

[0057] Specifically, the server obtains the first timestamp data of the transmission efficiency data, the second timestamp data of the target temperature data, and the third timestamp data of the target vibration data. The timestamp is key information for identifying the data acquisition time. The server uses the first timestamp data to curve fit the transmission efficiency data to obtain a transmission efficiency curve. Curve fitting is a mathematical technique used to find the best fitting curve between data points to better understand the trend and change of the data. Then, the server uses the second timestamp data to curve fit the target temperature data to obtain a target temperature curve. Similarly, curve fitting can help the server understand the changing trend of temperature data. The server uses the third timestamp data to curve fit the target vibration data to obtain a target vibration curve. The curve of the vibration data can reveal the intensity and frequency of the vibration. Subsequently, the server extracts curve feature points of the transmission efficiency curve, the target temperature curve, and the target vibration curve. The server looks for key feature points on the curve to describe the characteristics of the curve. For example, for the transmission efficiency curve, the server extracts feature points such as the maximum efficiency point, the minimum efficiency point, and the efficiency change rate. For the target temperature curve, the server extracts feature points such as the temperature peak and the fluctuation frequency. For the target vibration curve, the server extracts feature points such as the vibration amplitude and the vibration frequency. The server screens and maps these extracted feature points to generate multiple transmission efficiency features, multiple target temperature features, and multiple target vibration features. These features will be used in subsequent fault diagnosis and self-learning processes.

[0058] In a specific embodiment, if Figure 4 As shown, the process of executing step S105 may specifically include the following steps:

[0059] S401, feature encoding a plurality of transmission efficiency features and a plurality of target temperature features according to a preset first feature encoding rule to obtain a plurality of first efficiency feature encoding values ​​and a plurality of temperature feature encoding values;

[0060] S402, performing feature normalization and vector conversion on a plurality of first efficiency feature code values ​​and a plurality of temperature feature code values ​​to obtain a temperature transmission relationship vector;

[0061] S403, feature encoding the plurality of transmission efficiency features and the plurality of target vibration features according to a second feature encoding rule to obtain a plurality of second efficiency feature encoding values ​​and a plurality of temperature feature encoding values;

[0062] S404 , performing feature normalization and vector conversion on the plurality of second efficiency feature code values ​​and the plurality of temperature feature code values ​​to obtain a vibration transmission relationship vector.

[0063] Specifically, the server performs feature encoding on multiple transmission efficiency features and multiple target temperature features according to a preset first feature encoding rule to obtain multiple first efficiency feature encoding values ​​and multiple temperature feature encoding values. Encoding is the process of converting original features into digital representations so that the computer can understand and process these features. Each feature has a different encoding method, and the appropriate encoding method is selected according to the nature of the data and the needs of the problem. The server performs feature normalization and vector conversion on multiple first efficiency feature encoding values ​​and multiple temperature feature encoding values ​​to obtain a temperature transmission relationship vector. Feature normalization is to scale the values ​​of different features to the same scale range to ensure that they contribute equally to the model. Vector conversion combines these feature encoding values ​​into a feature vector for subsequent analysis. For example, the server normalizes the encoded features and scales them between 0 and 1. The server performs feature encoding on multiple transmission efficiency features and multiple target vibration features according to the second feature encoding rule to obtain multiple second efficiency feature encoding values ​​and multiple temperature feature encoding values. According to the second feature encoding rule, the server selects different encoding methods to better represent the relationship between features. The server performs feature normalization and vector conversion on the multiple second efficiency feature coding values ​​and the multiple vibration feature coding values ​​to obtain a vibration transmission relationship vector.

[0064] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0065] (1) Inputting the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model, the reducer fault diagnosis model includes a first CNN network, a second CNN network and a weight output layer;

[0066] (2) The first CNN network is used to predict the speed reducer fault based on the temperature transmission relationship vector to obtain the first speed reducer fault prediction result. The first CNN network includes three convolutional pooling layers and one fully connected layer.

[0067] (3) The second CNN network is used to predict the reducer fault based on the vibration transmission relationship vector to obtain the second reducer fault prediction result. The second CNN network includes three convolutional pooling layers and one fully connected layer.

[0068] (4) Performing a temperature influence weight analysis on the target reducer to obtain the temperature influence weight, and performing a vibration influence weight analysis on the target reducer to obtain the vibration influence weight;

[0069] (5) The temperature influence weight and the first reducer fault prediction result are weighted by the weight output layer to obtain a first weighted fault prediction result, and the vibration influence weight and the second reducer fault prediction result are weighted to obtain a second weighted fault prediction result;

[0070] (6) The first weighted fault prediction result and the second weighted fault prediction result are fused to obtain the reducer fault diagnosis result.

[0071] Specifically, the reducer fault diagnosis model includes a first CNN network, a second CNN network, and a weight output layer. The purpose of these networks and layers is to perform fault diagnosis prediction and weight analysis on the input transmission relationship vector. The first CNN network is used to predict the reducer fault on the temperature transmission relationship vector. The first CNN network usually includes a convolution layer and a pooling layer, as well as a fully connected layer. These layers perform feature extraction and fault prediction on the input transmission relationship vector. The second CNN network is used to predict the reducer fault on the vibration transmission relationship vector. The second CNN network is similar to the first CNN network, but focuses on the analysis of vibration data. The two networks can work in parallel to improve the fault diagnosis performance. The temperature impact weight analysis is performed on the target reducer to obtain the temperature impact weight. This is to understand the degree of influence of temperature on fault diagnosis. Similarly, the vibration impact weight analysis is performed to obtain the vibration impact weight. These two steps can use various methods, such as feature importance analysis or weight analysis algorithm. Subsequently, the temperature impact weight and the first reducer fault prediction result are weighted by the weight output layer to obtain the first weighted fault prediction result. Similarly, the vibration impact weight and the second reducer fault prediction result are weighted to obtain the second weighted fault prediction result. These weights can be determined based on their importance to better consider the contribution of temperature and vibration to the fault. The first weighted fault prediction result and the second weighted fault prediction result are fused to obtain the reducer fault diagnosis result. The result fusion can adopt a variety of methods, such as voting, weighted average or rule-based decision-making, to determine the final fault diagnosis result.

[0072] In a specific embodiment, the process of executing step S107 may specifically include the following steps:

[0073] (1) Calibrate the compensation parameter control range of the target reducer according to the reducer fault diagnosis result to obtain the compensation parameter control range;

[0074] (2) Define the function parameters of the compensation parameter control range to obtain the corresponding target function parameters, and set the function parameters of the preset initial random generation function through the target function parameters to obtain the target random generation function;

[0075] (3) The initial values ​​of the compensation parameters of the target reducer are randomly generated through the target random generation function to obtain the corresponding random initial value set;

[0076] (4) Generate particle associations for the random initial value set through the compensation parameter control range to obtain a particle population, and then divide the particle population into multiple sub-particle populations;

[0077] (5) Match the fitness calculation function of each sub-particle population respectively to obtain the fitness calculation function corresponding to each sub-particle population;

[0078] (6) Through the fitness calculation function corresponding to each sub-particle population, the fitness of each sub-particle population is calculated respectively to obtain the particle fitness set corresponding to the particle population;

[0079] (7) The particle fitness set is iteratively calculated until the preset conditions are met, and the optimal reducer control compensation parameter combination of the target reducer is generated.

[0080] Specifically, the compensation parameter control range of the target reducer is calibrated according to the reducer fault diagnosis result. According to the fault diagnosis result, the range of the control parameter that needs to be adjusted is determined. For example, if the fault diagnosis result shows that the flow rate of lubricating oil needs to be increased to reduce friction, the range of the lubricating oil flow rate can be calibrated as a part of the increase. The calibration of this range can be determined according to actual conditions and experience. Function parameters are defined for the compensation parameter control range. The parameters that need to be adjusted are defined as the input parameters of the function to obtain the corresponding objective function parameters. These parameters will be used in the subsequent optimization process. Then, the preset initial random generation function is set with the objective function parameters to obtain the target random generation function. This function will generate a random combination of compensation parameters according to the defined parameter range. The target random generation function is used to randomly generate the initial value of the compensation parameter of the target reducer to obtain the corresponding random initial value set. These random initial values ​​will be used as the starting point of the particle swarm algorithm. The random initial value set is generated by particle association through the compensation parameter control range to obtain a particle population. Particle association generation can generate particles according to the parameter range to ensure that they are within a reasonable range. The particle population is divided into populations to obtain multiple sub-particle populations. Each sub-particle population will be responsible for finding the optimal parameter combination to meet the preset conditions. The fitness calculation function is matched for each sub-particle population to obtain the fitness calculation function corresponding to each sub-particle population. The fitness calculation function is used to evaluate the performance of the parameter combination to determine which parameter combinations are closer to the optimal solution. Through the fitness calculation function corresponding to each sub-particle population, the fitness calculation is performed on each sub-particle population to obtain the particle fitness set corresponding to the particle population. These fitness values ​​will be used to guide the search of particles in the parameter space. The particle fitness set is iteratively calculated until the preset conditions are met. During the iteration process, the particles will continuously adjust their parameter combinations to find the best compensation parameter combination. Once the preset conditions are met, the optimal reducer control compensation parameter combination of the target reducer is generated.

[0081] The above describes the intelligent fault diagnosis and self-learning method of the reducer in the embodiment of the present invention. The following describes the intelligent fault diagnosis and self-learning system of the reducer in the embodiment of the present invention. Figure 5In one embodiment of the invention, an intelligent fault diagnosis and self-learning system for a reducer includes:

[0082] An acquisition module is used to acquire external influencing parameters of a target reducer to be detected, obtain initial temperature data and initial vibration data, and respectively perform Fourier transformation and frequency domain downsampling on the initial temperature data and the initial vibration data to obtain target temperature data and target vibration data;

[0083] A classification module, used to collect operating state parameters of the target reducer to obtain initial speed data and initial torque data, and to classify the initial speed data to obtain input speed data and output speed data;

[0084] a calculation module, configured to perform abnormal value processing on the initial torque data to obtain target torque data, and calculate the transmission efficiency data of the target reducer according to the target torque data, the input speed data, and the output speed data;

[0085] An extraction module, used to perform feature extraction on the transmission efficiency data, the target temperature data, and the target vibration data, respectively, to obtain a plurality of transmission efficiency features, a plurality of target temperature features, and a plurality of target vibration features;

[0086] a conversion module, configured to perform feature coding and vector conversion on the plurality of transmission efficiency features and the plurality of target temperature features to obtain a temperature transmission relationship vector, and perform feature coding and vector conversion on the plurality of transmission efficiency features and the plurality of target vibration features to obtain a vibration transmission relationship vector;

[0087] A diagnosis module, used for inputting the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model to perform intelligent reducer fault diagnosis and obtain a reducer fault diagnosis result;

[0088] The compensation module is used to calculate the control compensation parameters of the target reducer according to the reducer fault diagnosis result through a preset particle swarm algorithm to obtain an optimal reducer control compensation parameter combination.

[0089] Through the cooperation of the above components, through the comprehensive collection of external influencing parameters, combined with the operating status parameters, this method can realize the monitoring of various aspects of reducer data, including temperature, vibration, speed, torque, etc. This helps to form a more comprehensive evaluation of the operating status of the reducer. Using deep learning technology, especially convolutional neural network (CNN), intelligent fault diagnosis of temperature and vibration data is performed. Through the model learning data of complex patterns, potential reducer faults can be more accurately identified. Using particle swarm algorithm, the fault diagnosis results are analyzed by self-learning, and the optimal combination of reducer control compensation parameters is calculated. Such adaptability can continuously optimize the system according to the actual operating conditions, improving the robustness and performance of the entire system. Through the fusion of multiple temperature sensors and vibration sensors, as well as Fourier transform and frequency domain downsampling processing of the data, the obtained data is more representative and stable, providing a more reliable input for subsequent analysis. Through curve fitting, feature point extraction, encoding and vector conversion, representative features can be efficiently extracted from the original data to form temperature transmission relationship vectors and vibration transmission relationship vectors, which helps to reduce data dimensions and improve processing efficiency, thereby realizing intelligent fault diagnosis of reducers and improving the accuracy of reducer fault diagnosis.

[0090] above Figure 5 The intelligent fault diagnosis and self-learning system of the reducer in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The intelligent fault diagnosis and self-learning device of the reducer in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0091] Figure 6 It is a structural schematic diagram of a reducer intelligent fault diagnosis and self-learning device provided by an embodiment of the present invention. The reducer intelligent fault diagnosis and self-learning device 600 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 610 (for example, one or more processors) and a memory 620, and one or more storage media 630 (for example, one or more mass storage devices) storing application programs 633 or data 632. Among them, the memory 620 and the storage medium 630 can be short-term storage or permanent storage. The program stored in the storage medium 630 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the reducer intelligent fault diagnosis and self-learning device 600. Furthermore, the processor 610 can be configured to communicate with the storage medium 630, and execute a series of instruction operations in the storage medium 630 on the reducer intelligent fault diagnosis and self-learning device 600.

[0092] The reducer intelligent fault diagnosis and self-learning device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input and output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 6 The structure of the intelligent fault diagnosis and self-learning device for reducers shown does not constitute a limitation of the intelligent fault diagnosis and self-learning device for reducers, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0093] The present invention also provides an intelligent fault diagnosis and self-learning device for a reducer, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the intelligent fault diagnosis and self-learning method for the reducer in the above-mentioned embodiments.

[0094] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the reducer intelligent fault diagnosis and self-learning method.

[0095] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0096] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0097] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent fault diagnosis and self-learning of a reducer, characterized in that: The intelligent fault diagnosis and self-learning method for the reducer includes: Collect external influencing parameters of the target reducer to be detected to obtain initial temperature data and initial vibration data, and perform Fourier transform and frequency domain downsampling on the initial temperature data and the initial vibration data to obtain target temperature data and target vibration data; Collecting operating state parameters of the target reducer to obtain initial speed data and initial torque data, and classifying the initial speed data to obtain input speed data and output speed data; Performing abnormal value processing on the initial torque data to obtain target torque data, and calculating the transmission efficiency data of the target reducer according to the target torque data, the input speed data, and the output speed data; Performing feature extraction on the transmission efficiency data, the target temperature data, and the target vibration data respectively to obtain a plurality of transmission efficiency features, a plurality of target temperature features, and a plurality of target vibration features; The plurality of transmission efficiency characteristics and the plurality of target temperature characteristics are feature-encoded and vector-converted to obtain a temperature transmission relationship vector, and the plurality of transmission efficiency characteristics and the plurality of target vibration characteristics are feature-encoded and vector-converted to obtain a vibration transmission relationship vector; specifically comprising: feature-encoding the plurality of transmission efficiency characteristics and the plurality of target temperature characteristics according to a preset first feature encoding rule to obtain a plurality of first efficiency characteristic encoding values ​​and a plurality of temperature characteristic encoding values; feature-normalizing and vector-converting the plurality of first efficiency characteristic encoding values ​​and the plurality of temperature characteristic encoding values ​​to obtain a temperature transmission relationship vector; feature-encoding the plurality of transmission efficiency characteristics and the plurality of target vibration characteristics according to a second feature encoding rule to obtain a plurality of second efficiency characteristic encoding values ​​and a plurality of vibration characteristic encoding values; feature-normalizing and vector-converting the plurality of second efficiency characteristic encoding values ​​and the plurality of vibration characteristic encoding values ​​to obtain a vibration transmission relationship vector; the vector conversion is to combine different vectors into one feature vector; The temperature transmission relationship vector and the vibration transmission relationship vector are input into a preset reducer fault diagnosis model to perform intelligent reducer fault diagnosis, and obtain a reducer fault diagnosis result; specifically including: inputting the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model, the reducer fault diagnosis model includes a first CNN network, a second CNN network and a weight output layer; performing reducer fault prediction on the temperature transmission relationship vector through the first CNN network to obtain a first reducer fault prediction result, the first CNN network includes three convolutional pooling layers and one fully connected layer; performing reducer fault prediction on the vibration transmission relationship vector through the second CNN network , obtaining a second reducer fault prediction result, the second CNN network includes three convolutional pooling layers and one fully connected layer; performing a temperature influence weight analysis on the target reducer to obtain a temperature influence weight, and performing a vibration influence weight analysis on the target reducer to obtain a vibration influence weight; weighting the temperature influence weight and the first reducer fault prediction result through the weight output layer to obtain a first weighted fault prediction result, and weighting the vibration influence weight and the second reducer fault prediction result to obtain a second weighted fault prediction result; fusing the first weighted fault prediction result and the second weighted fault prediction result to obtain a reducer fault diagnosis result; The preset particle swarm algorithm is used to calculate the control compensation parameters of the target reducer according to the reducer fault diagnosis result to obtain the optimal reducer control compensation parameter combination; specifically, the following steps are used: according to the reducer fault diagnosis result, the compensation parameter control range of the target reducer is calibrated to obtain the compensation parameter control range; the function parameters of the compensation parameter control range are defined to obtain the corresponding target function parameters, and the function parameters of the preset initial random generation function are set by the target function parameters to obtain the target random generation function; the initial value of the compensation parameter of the target reducer is randomly generated by the target random generation function to obtain the corresponding random initial value. an initial value set; performing particle association generation on the random initial value set through the compensation parameter control range to obtain a particle population, and performing population segmentation on the particle population to obtain a plurality of sub-particle populations; performing fitness calculation function matching on each of the sub-particle populations to obtain a fitness calculation function corresponding to each of the sub-particle populations; performing fitness calculation on each of the sub-particle populations through the fitness calculation function corresponding to each of the sub-particle populations to obtain a particle fitness set corresponding to the particle population; performing iterative calculation on the particle fitness set until a preset condition is met, thereby generating an optimal reducer control compensation parameter combination of the target reducer.

2. The intelligent fault diagnosis and self-learning method for reducer according to claim 1 is characterized in that: The external influencing parameters of the target reducer to be detected are collected to obtain initial temperature data and initial vibration data, and the initial temperature data and the initial vibration data are respectively subjected to Fourier transformation and frequency domain downsampling to obtain target temperature data and target vibration data, including: Collecting the environmental temperature influence parameters of the target reducer to be detected through a plurality of temperature sensors to obtain the temperature sensor signal data of each temperature sensor, and collecting the vibration influence parameters of the target reducer through a plurality of vibration sensors to obtain the vibration sensor signal data of each vibration sensor; Performing sensor signal fusion and average weighted processing on the temperature sensor signal data of each temperature sensor to obtain initial temperature data, and performing sensor signal fusion and average weighted processing on the vibration sensor signal data of each vibration sensor to obtain initial vibration data; Performing Fourier transform on the initial temperature data and the initial vibration data respectively to obtain temperature frequency domain data and vibration frequency domain data; The temperature frequency domain data and the vibration frequency domain data are respectively downsampled in the frequency domain to obtain target temperature data and target vibration data.

3. The intelligent fault diagnosis and self-learning method for reducer according to claim 1 is characterized in that: The performing abnormal value processing on the initial torque data to obtain target torque data, and calculating the transmission efficiency data of the target reducer according to the target torque data, the input speed data, and the output speed data, includes: Determine a target threshold of the initial torque data by a preset DBSCAN algorithm, and perform outlier detection on the initial torque data according to the target threshold to obtain at least one outlier; Correcting at least one outlier in the initial torque data to obtain target torque data; Calculating the product of the target torque data and the input speed data to obtain input power data, and calculating the product of the target torque data and the output speed data to obtain output power data; The ratio of the output power data to the input power data is calculated to obtain the transmission efficiency data of the target reducer.

4. The intelligent fault diagnosis and self-learning method for reducer according to claim 1 is characterized in that: The feature extraction is performed on the transmission efficiency data, the target temperature data and the target vibration data respectively to obtain a plurality of transmission efficiency features, a plurality of target temperature features and a plurality of target vibration features, including: Acquire first timestamp data of the transmission efficiency data, acquire second timestamp data of the target temperature data, and acquire third timestamp data of the target vibration data; Performing curve fitting on the transmission efficiency data according to the first timestamp data to obtain a transmission efficiency curve, performing curve fitting on the target temperature data according to the second timestamp data to obtain a target temperature curve, and performing curve fitting on the target vibration data according to the third timestamp data to obtain a target vibration curve; Extracting curve feature points from the transmission efficiency curve to obtain a plurality of candidate efficiency feature points, extracting curve feature points from the target temperature curve to obtain a plurality of candidate temperature feature points, and extracting curve feature points from the target vibration curve to obtain a plurality of candidate vibration feature points; Feature point screening and feature mapping are performed on the multiple candidate efficiency feature points to generate multiple transmission efficiency features, and feature point screening and feature mapping are performed on the multiple candidate temperature feature points to generate multiple target temperature features, and feature point screening and feature mapping are performed on the multiple candidate vibration feature points to generate multiple target vibration features.

5. An intelligent fault diagnosis and self-learning system for a reducer, characterized in that: The intelligent fault diagnosis and self-learning system for the reducer includes: An acquisition module is used to acquire external influencing parameters of a target reducer to be detected, obtain initial temperature data and initial vibration data, and respectively perform Fourier transformation and frequency domain downsampling on the initial temperature data and the initial vibration data to obtain target temperature data and target vibration data; A classification module, used to collect operating state parameters of the target reducer to obtain initial speed data and initial torque data, and to classify the initial speed data to obtain input speed data and output speed data; a calculation module, configured to perform abnormal value processing on the initial torque data to obtain target torque data, and calculate the transmission efficiency data of the target reducer according to the target torque data, the input speed data, and the output speed data; An extraction module, used to perform feature extraction on the transmission efficiency data, the target temperature data, and the target vibration data, respectively, to obtain a plurality of transmission efficiency features, a plurality of target temperature features, and a plurality of target vibration features; A conversion module, used for performing feature coding and vector conversion on the multiple transmission efficiency characteristics and the multiple target temperature characteristics to obtain a temperature transmission relationship vector, and performing feature coding and vector conversion on the multiple transmission efficiency characteristics and the multiple target vibration characteristics to obtain a vibration transmission relationship vector; specifically comprising: performing feature coding on the multiple transmission efficiency characteristics and the multiple target temperature characteristics according to a preset first feature coding rule to obtain a plurality of first efficiency feature coding values ​​and a plurality of temperature feature coding values; performing feature normalization and vector conversion on the plurality of first efficiency feature coding values ​​and the plurality of temperature feature coding values ​​to obtain a temperature transmission relationship vector; performing feature coding on the multiple transmission efficiency characteristics and the multiple target vibration characteristics according to a second feature coding rule to obtain a plurality of second efficiency feature coding values ​​and a plurality of vibration feature coding values; performing feature normalization and vector conversion on the plurality of second efficiency feature coding values ​​and the plurality of vibration feature coding values ​​to obtain a vibration transmission relationship vector; the vector conversion is to combine different vectors into one feature vector; The diagnosis module is used to input the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model to perform intelligent reducer fault diagnosis and obtain a reducer fault diagnosis result; specifically comprising: inputting the temperature transmission relationship vector and the vibration transmission relationship vector into a preset reducer fault diagnosis model, the reducer fault diagnosis model comprising a first CNN network, a second CNN network and a weight output layer; performing reducer fault prediction on the temperature transmission relationship vector through the first CNN network to obtain a first reducer fault prediction result, the first CNN network comprising three convolutional pooling layers and one fully connected layer; performing reducer fault prediction on the vibration transmission relationship vector through the second CNN network Fault prediction is performed to obtain a second reducer fault prediction result, wherein the second CNN network includes three convolutional pooling layers and one fully connected layer; a temperature influence weight analysis is performed on the target reducer to obtain a temperature influence weight, and a vibration influence weight analysis is performed on the target reducer to obtain a vibration influence weight; the temperature influence weight and the first reducer fault prediction result are weighted by the weight output layer to obtain a first weighted fault prediction result, and the vibration influence weight and the second reducer fault prediction result are weighted to obtain a second weighted fault prediction result; the first weighted fault prediction result and the second weighted fault prediction result are fused to obtain a reducer fault diagnosis result; The compensation module is used to calculate the control compensation parameters of the target reducer according to the reducer fault diagnosis result through a preset particle swarm algorithm to obtain the optimal reducer control compensation parameter combination; specifically includes: calibrating the compensation parameter control range of the target reducer according to the reducer fault diagnosis result to obtain the compensation parameter control range; defining the function parameters of the compensation parameter control range to obtain the corresponding target function parameters, and setting the function parameters of the preset initial random generation function through the target function parameters to obtain the target random generation function; randomly generating the initial value of the compensation parameter of the target reducer through the target random generation function to obtain the corresponding A set of random initial values; performing particle association generation on the random initial value set through the compensation parameter control range to obtain a particle population, and performing population segmentation on the particle population to obtain multiple sub-particle populations; performing fitness calculation function matching on each of the sub-particle populations to obtain a fitness calculation function corresponding to each of the sub-particle populations; performing fitness calculation on each of the sub-particle populations through the fitness calculation function corresponding to each of the sub-particle populations to obtain a particle fitness set corresponding to the particle population; performing iterative calculation on the particle fitness set until a preset condition is met, thereby generating an optimal reducer control compensation parameter combination for the target reducer.

6. An intelligent fault diagnosis and self-learning device for a reducer, characterized in that: The intelligent fault diagnosis and self-learning device for reducer comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the reducer intelligent fault diagnosis and self-learning device executes the reducer intelligent fault diagnosis and self-learning method as described in any one of claims 1-4.

7. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the intelligent fault diagnosis and self-learning method for the reducer as described in any one of claims 1-4 is implemented.