A reducer assembly system and method

Abstract

The present invention relates to a reducer assembly system and method, belonging to the field of intelligent assembly technology, the method comprising: collecting multi-dimensional performance data of the reducer; a pre-built intelligent assembly model calculating assembly evaluation indicators and / or the average value of the assembly evaluation indicators based on the multi-dimensional performance data; the intelligent assembly model judging whether the assembly is qualified based on the comparison result of the average value of the assembly evaluation indicators and a pre-set target value score; wherein, a comprehensive performance evaluation function model based on weight distribution is provided in the intelligent assembly model to calculate the assembly evaluation indicators and / or the average value of the assembly evaluation indicators. The intelligent assembly model provided by the present invention reduces the reliance on manual judgment and realizes the automated evaluation of the overall performance of the reducer. It provides a basis for quantitative evaluation for different application scenarios and solves the problem that traditional testing methods cannot provide comprehensive performance evaluation.

Description

Technical Field

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

[0002] Precision reducers are key components for achieving high-precision control and are widely used in fields such as industrial automation, robotics, and aerospace. Their performance directly impacts the operating accuracy and stability of equipment. These fields place varying demands on reducers. For example, industrial automation emphasizes efficiency and noise control, while robotics prioritizes accuracy and vibration suppression.

[0003] While existing assembly testing methods can provide specific parameters such as transmission accuracy, efficiency, vibration, and noise, comprehensive evaluation of the reducer's overall performance still relies on manual effort. This approach has significant limitations, making it difficult to quickly adjust and optimize the performance of reducers that fail to meet specific application requirements. For example, in high-performance scenarios, even a slight return error can affect the overall performance of the device.

[0004] To overcome these issues, existing technologies have proposed intelligent testing and evaluation systems. These systems should be able to automatically analyze test data and quickly pinpoint the specific causes of substandard reducer performance, enabling targeted improvement measures. This not only helps improve production efficiency but also ensures high-quality final products that meet the stringent requirements of diverse application scenarios. This approach can more effectively support the continued development and technological advancement of various industries.

[0005] In different application scenarios, the performance requirements for precision reducers vary, focusing on key indicators such as transmission accuracy, efficiency, vibration, and noise. However, the current lack of a comprehensive method for evaluating these key performance indicators makes it difficult to obtain a comprehensive quantitative evaluation of assembled reducers. While existing testing methods can provide independent data on transmission accuracy, efficiency, vibration, and noise, these data are not integrated into a unified evaluation system, resulting in the need for manual judgment to evaluate the overall performance of reducers.

[0006] For example, CN117870545A discloses an analysis system for the assembly of motor reducers for industrial robot arms, comprising: a drive control system, a measurement system, and a test platform. The drive control system can communicate with the measurement system to collect data during the test process; the drive control system further communicates with the test platform to control the posture of the test platform, thereby adjusting the state of the motor under test and the reducer under test. The drive control system includes: a controller, a connecting line, and a driver; the measurement system includes: a laser centering instrument and a three-dimensional force sensor; and the test platform includes: a base, a parallel platform, a motor mounting plate, a transparent bellows, and a reducer mounting plate. This technical solution is used in an analysis system for the assembly of motor reducers for industrial robot arms, and is used to collect and test data on the motor under test and the reducer under test in different assembly states. However, this technical solution can only measure the posture and current data of the reducer, and cannot collect multi-performance data in an all-round manner.

[0007] The lack of a systematic approach to directly linking various performance indicators to assembly parameters makes it difficult to efficiently and quickly adjust reducers that fail to meet specific requirements. The ideal solution would be to provide an analytical method that assigns weights to these key performance indicators and, through intelligent fusion technology, quantitatively evaluates reducer performance under different requirements. This would help technicians quickly determine the optimal assembly adjustment plan based on specific application requirements, thereby improving work efficiency and product quality.

[0008] Therefore, to overcome the limitations of existing technologies, a smart assembly system for reducers is needed that can intelligently integrate a comprehensive evaluation system of multiple performance indicators to support efficient and rapid assembly and adjustment of reducers. This will not only reduce human intervention but also ensure that the reducer achieves optimal performance in various application environments.

[0009] In addition, on the one hand, there are differences in understanding among those skilled in the art; on the other hand, the applicant studied a large number of documents and patents when making the present invention, but due to space limitations, not all details and contents are listed in detail. However, this does not mean that the present invention does not have the characteristics of these prior arts. On the contrary, the present invention already has all the characteristics of the prior art, and the applicant reserves the right to add relevant prior art to the background technology. Summary of the Invention

[0010] In different application scenarios, the performance requirements for precision reducers vary, focusing on key indicators such as transmission accuracy, efficiency, vibration, and noise. However, the current lack of a comprehensive method for evaluating these key performance indicators makes it difficult to obtain a comprehensive quantitative evaluation of assembled reducers. While existing testing methods can provide independent data on transmission accuracy, efficiency, vibration, and noise, these data are not integrated into a unified evaluation system, resulting in the need for manual judgment to evaluate the overall performance of reducers.

[0011] How to improve the intelligent assembly method to reduce the dependence on manual labor in evaluating the overall performance of the reducer is a technical problem to be solved by the present invention.

[0012] In response to the deficiencies of the prior art, the present invention provides a reducer assembly method from a first aspect, the method comprising: collecting multi-dimensional performance data of the reducer; a pre-built intelligent assembly model calculating assembly evaluation indicators and / or the average value of the assembly evaluation indicators based on the multi-dimensional performance data; the intelligent assembly model judging whether the assembly is qualified based on a comparison result of the average value of the assembly evaluation indicators and a pre-set target value score; wherein, a comprehensive performance evaluation function model based on weight distribution is provided in the intelligent assembly model to calculate the assembly evaluation indicators and / or the average value of the assembly evaluation indicators.

[0013] The intelligent assembly model provided by this invention reduces reliance on manual judgment and enables automated evaluation of the overall performance of the reducer. It provides a basis for quantitative evaluation in different application scenarios and solves the problem that traditional testing methods cannot provide comprehensive performance evaluation.

[0014] According to a preferred embodiment, the steps of constructing an intelligent assembly model include: constructing a comprehensive performance evaluation function model based on weight distribution based on multi-dimensional performance indicators, and determining the weight coefficient in the comprehensive performance evaluation function model, wherein a judgment matrix is ​​constructed based on the relative scale value of the multi-dimensional performance data, and the weight coefficient is calculated; after determining the weight coefficient, the assembly evaluation index of the reducer is calculated based on the comprehensive performance evaluation function model; data enhancement is performed on the curve of the assembly evaluation index and a data set sample is formed; and artificial intelligence model training is performed based on the data set sample to form an intelligent assembly model.

[0015] The present invention can adjust the importance of various performance indicators according to different application scenarios, ensuring that the final assembly evaluation indicators can accurately reflect the actual working conditions and requirements of the reducer.

[0016] According to a preferred embodiment, the method further includes normalizing the multi-dimensional performance data before determining the weight coefficients to limit the raw multi-dimensional performance data to the range of [0, 1], thereby accelerating convergence during AI model training. Normalization prevents certain high-volume data from having an excessive impact on the model during training, thereby ensuring the stability of the training process and accelerating model convergence.

[0017] According to a preferred embodiment, the comprehensive performance evaluation function model is: α = ω1ET + ω2η + ω3S + ω4DB; where α represents the assembly evaluation index, ET represents the transmission error, η represents the transmission efficiency, S represents the vibration signal, DB represents the noise signal, and ω1, ω2, ω3, and ω4 represent weighting coefficients, respectively. The comprehensive performance evaluation function model constructed by the present invention allows each multidimensional performance indicator (transmission error, transmission efficiency, vibration signal, noise signal) to be assigned different weights according to actual needs, forming a comprehensive assembly evaluation index. This approach not only considers the impact of individual performance parameters but also takes into account the interactions between them, providing a more accurate measurement standard for the overall performance of the reducer.

[0018] According to a preferred embodiment, the step of determining the weight coefficient of the comprehensive performance evaluation function model includes: calculating the nth power of each row in the judgment matrix to obtain an n-dimensional vector: ; Normalize the vector to a weight vector, the weight coefficient is: ;in, represents the weight coefficient, Represents the relative scale value in the judgment matrix; It represents the nth root of the product of all elements in the j-th row of the judgment matrix.

[0019] The weight coefficient calculation method of the present invention ensures the scientific and rational weight distribution. This method can dynamically adjust the weight of each performance parameter according to the actual situation, so that the final assembly evaluation index can better reflect the actual condition of the reducer, while also enhancing the flexibility and adaptability of the evaluation system.

[0020] According to a preferred embodiment, the step of performing data enhancement on the curve of the assembly evaluation index includes: performing data enhancement on the curve of the assembly evaluation index and calculating: ; ; ; f represents the signal data obtained after enhancement; p represents slicing of the assembly evaluation index α, q represents offsetting of the assembly evaluation index α, r represents front-to-back mirror flipping of the assembly evaluation index α, and α represents the assembly evaluation index; q a Indicates the starting point offset; p a Indicates the slice length; pT Indicates the number of signal sampling points when the output end of the reducer rotates for one cycle.

[0021] By performing data augmentation on the curves of the assembly evaluation metrics, the quantity and diversity of sample data are increased, helping to improve the learning and generalization capabilities of the AI ​​model. This is crucial for improving the model's performance when faced with new data, further enhancing the reliability and accuracy of the model's predictions.

[0022] According to a second aspect, the present invention provides a reducer assembly system comprising a sensing unit and a server. The sensing unit is used to collect multi-dimensional performance data of the reducer; the server has an intelligent assembly model pre-built within it. The intelligent assembly model calculates assembly evaluation indicators and / or average values ​​of the assembly evaluation indicators based on the multi-dimensional performance data; and determines whether the assembly is qualified based on a comparison of the average values ​​of the assembly evaluation indicators with pre-set target values. The intelligent assembly model includes a comprehensive performance evaluation function model based on weight distribution to calculate the assembly evaluation indicators and / or average values ​​of the assembly evaluation indicators.

[0023] The advantages of the system are that it automates the evaluation of the overall performance of the reducer, reducing the need for manual intervention and improving assembly efficiency and accuracy. Furthermore, by introducing a comprehensive performance evaluation function model based on weight distribution, the different performance emphases of the reducer in different application scenarios are fully considered, thus ensuring the comprehensiveness and accuracy of the assembly qualification analysis.

[0024] According to a preferred embodiment, the intelligent assembly model is configured as follows: constructing a comprehensive performance evaluation function model based on weight distribution based on multi-dimensional performance indicators, determining the weight coefficient in the comprehensive performance evaluation function model, wherein a judgment matrix is ​​constructed based on the relative scale value of the multi-dimensional performance data, and the weight coefficient is calculated; after determining the weight coefficient, the assembly evaluation index of the reducer is calculated based on the comprehensive performance evaluation function model.

[0025] The intelligent assembly model dynamically adjusts the weights of various performance parameters based on their importance to better reflect the actual operating conditions of the reducer. This not only improves the scientific nature and rationality of the assembly evaluation indicators, but also enhances the adaptability and flexibility of the system. By constructing a judgment matrix and calculating weight coefficients, it ensures a more rational distribution of weights for each performance parameter, thereby improving the accuracy and reliability of assembly qualification analysis.

[0026] According to a preferred embodiment, the intelligent assembly model is also configured to: normalize the multidimensional performance data before determining the weight coefficient to limit the range of the original multidimensional performance data to [0,1], so that the convergence in the process of training the artificial intelligence model is faster.

[0027] Normalization ensures that data of varying magnitudes does not disproportionately impact the final results. It also accelerates the convergence of AI model training, improving the stability and accuracy of the intelligent assembly model. This improves the overall performance of the intelligent assembly model by ensuring that it can quickly learn and make reliable predictions even with complex and changing data sets, thereby optimizing the assembly qualification analysis process.

[0028] According to a preferred embodiment, the comprehensive performance evaluation function model is: α=ω1ET+ω2η+ω3S+ω4DB; wherein α represents the assembly evaluation index, ET represents the transmission error, η represents the transmission efficiency, S represents the vibration signal, DB represents the noise signal, and ω1, ω2, ω3 and ω4 represent weight coefficients respectively. The comprehensive performance evaluation function model helps to comprehensively quantify the key performance indicators of the reducer and provides a unified evaluation system. By reasonably setting the various weight coefficients, the assembly evaluation standards can be customized according to different application requirements to ensure that the evaluation results have both wide applicability and accuracy. This not only simplifies the process of assembly qualification analysis, but also improves the reliability and consistency of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a simplified schematic diagram of the module connection relationship of the reducer assembly system provided by the present invention;

[0030] Figure 2 It is a schematic flow chart of the reducer assembly method provided by the present invention;

[0031] Figure 3 It is a logical diagram of the evaluation and correction guidance of the reducer assembly system provided by the present invention;

[0032] Figure 4 It is a schematic diagram of the connection structure of the intelligent assembly test platform for the reducer provided by the present invention.

[0033] List of reference numerals:

[0034] 100: Intelligent assembly test platform; 110: Drive motor; 120: First torque sensor; 130: Input encoder; 140: Reducer; 150: Output encoder; 160: Second torque sensor; 170: Load motor; 180: Server; 190: Noise sensor; 200: Vibration sensor. DETAILED DESCRIPTION

[0035] The following is a detailed description with reference to the accompanying drawings.

[0036] The intelligent testing and evaluation system provided by existing technologies is capable of automatically analyzing test data and quickly identifying the exact cause of the poor performance of the reducer 140, thereby implementing precise improvement strategies. This not only helps improve manufacturing efficiency but also ensures that the final product meets high quality standards and adapts to the strict specifications of various application scenarios.

[0037] However, the requirements for various performance indicators of the reducer 140, such as transmission accuracy, transmission efficiency, vibration level, and noise level, vary depending on the application environment. Currently, however, there is no comprehensive method for evaluating these important performance indicators, making it difficult to obtain a comprehensive and accurate evaluation of the assembled reducer 140. Although existing testing tools can provide individual data on transmission accuracy, transmission efficiency, transmission vibration, and transmission noise, this information is not integrated into a unified evaluation framework. Therefore, the overall performance evaluation of the reducer 140 still relies on manual analysis.

[0038] Due to the lack of an effective method to directly associate various performance indicators with assembly parameters, it is difficult to efficiently and quickly adjust the reducer 140 that does not meet specific requirements.

[0039] To address the shortcomings of the prior art, the present invention provides a reducer assembly system and method. The present invention also provides an intelligent assembly and testing platform 100 for reducers. The present invention also provides a server 180 for processing data from the reducer 140 and evaluating and determining whether the assembly is qualified. Preferably, the server 180 is also used to analyze reducer 140 faults and provide corresponding corrective solutions. Preferably, the present invention provides a storage medium for storing the operating program of the server 180.

[0040] Preferably, the server 180 can also be replaced by physical hardware such as a dedicated integrated chip, a remote server, or a server group that has data processing capabilities or runs an artificial intelligence model.

[0041] like Figure 1 and Figure 4 As shown, the present invention provides a reducer assembly system, including a sensing unit and a server 180.

[0042] The sensor unit is used to collect multi-dimensional performance data of the reducer 140. Figure 1As shown, the sensing unit includes a first torque sensor 120, a second torque sensor 160, a noise sensor 190, and a vibration sensor 200. Preferably, the first torque sensor 120 is connected to the input end of the reducer 140 via the input end encoder 130. The second torque sensor 160 is connected to the output end of the reducer 140 via the output end encoder 150. The first torque sensor 120 and the second torque sensor 160 are used to collect position data and torsional force data of the reducer 140 to analyze the transmission accuracy and transmission efficiency of the reducer 140. The noise sensor 190 is used to collect noise data of the reducer 140 for transmission noise analysis of the reducer 140. The vibration sensor 200 is used to collect vibration data of the reducer 140 for transmission vibration analysis of the reducer 140. Preferably, during collection, the vibration sensor 200 is connected to the assembled reducer 140 through a dedicated hole.

[0043] The present invention can also provide an intelligent assembly test platform 100, such as Figure 4 As shown. On the intelligent assembly test platform 100, the first torque sensor 120 is connected to the drive motor 110. The second torque sensor 160 is connected to the load motor 170. Each sensor in the sensing unit is connected to the server 180 via a wired or wireless connection for data transmission. In other words, the multi-dimensional performance data of the reducer 140 of the present invention includes position data (angle), torsional force data (torque), noise data, and vibration data.

[0044] Preferably, if Figure 2 As shown, various sensor detections and experimental data collection were performed.

[0045] Specifically, before assembly begins, experimental data for known fault types of the reducer 140 is collected. Specifically, the sensor unit on the intelligent assembly test platform 100 detects signals such as angle, torque, vibration, and noise of the assembled reducer 140. In other words, before testing, experiments are conducted on the reducer 140 with known error types, and the sensor unit detects signals such as angle, torque, vibration, and noise.

[0046] Preferably, an intelligent assembly model is pre-built inside the server 180. The construction principle of the intelligent assembly model is: by processing the four signals of angle signal, torque signal, vibration signal and noise signal, the corresponding four analysis indicators are calculated and normalized to obtain the assembly evaluation indicator α=ω1ε+ω2η+ω3S+ω4DB. The assembly evaluation indicators are used as data sets, and artificial intelligence models such as convolutional neural network models are trained through deep learning to realize the identification of fault types of different reducers 140. The artificial intelligence model formed by training is an intelligent assembly model. Figure 3As shown, the four analysis indicators are: transmission accuracy analysis indicator, transmission efficiency analysis indicator, transmission vibration analysis indicator and transmission noise analysis indicator.

[0047] The smart assembly model is configured as follows:

[0048] S1: Calculate an assembly evaluation index and / or an average value of the assembly evaluation index based on the multi-dimensional performance data.

[0049] Preferably, a comprehensive performance evaluation function model based on weight distribution is provided in the intelligent assembly model to calculate the assembly evaluation index and / or the average value of the assembly evaluation index.

[0050] S2: Determine whether the assembly is qualified based on the comparison result of the average value of the assembly evaluation index and the preset target value score.

[0051] Preferably, the steps of constructing the intelligent assembly model include:

[0052] S31: Construct a comprehensive performance evaluation function model based on weight distribution based on multi-dimensional performance indicators, and determine the weight coefficient in the comprehensive performance evaluation function model.

[0053] According to a preferred embodiment, the comprehensive performance evaluation function model is:

[0054] α=ω1ET+ω2η+ω3S+ω4DB.

[0055] Where α represents the assembly evaluation index, ET represents the transmission error, η represents the transmission efficiency, S represents the vibration signal, DB represents the noise signal, and ω1, ω2, ω3, and ω4 represent weighting coefficients. Vibration signal S and noise signal DB are directly collected by vibration sensor 200 and noise sensor 190, respectively.

[0056] Preferably, the transmission accuracy of this type is analyzed based on the angle test data collected by the first torque sensor 120 and the second torque sensor 160 .

[0057] The calculation formula of transmission error ET is:

[0058] ET(t)=θ2(t)-θ1(t) / i.

[0059] θ1 represents the angular displacement of the input shaft, θ2 represents the angular displacement of the output shaft, and i represents the theoretical transmission ratio, that is, the design speed ratio between the input shaft and the output shaft.

[0060] Preferably, the transmission efficiency is analyzed based on the input / output torque test data collected by the first torque sensor 120 and the second torque sensor 160 .

[0061] The calculation formula of transmission efficiency η is: η=(T out / T in )×100%.

[0062] T out T represents the torque applied to the output end by the reducer 140; in Represents the torque applied to the input.

[0063] Preferably, a storage unit is provided in the server 180. The vibration and noise data collected by the vibration sensor 200 and the noise sensor 190 are directly stored in the storage unit of the server 180 for subsequent use by the intelligent assembly model.

[0064] Preferably, before determining the weight coefficient, the multidimensional performance data is normalized to limit the range of the original multidimensional performance data to [0,1], so that the convergence in the training process of the artificial intelligence model is faster, the performance of the intelligent assembly model is improved, and the dimensional expression is converted into a dimensionless expression, so that indicators of different units or magnitudes can be substituted into the assembly evaluation index function to evaluate and analyze the assembly of the reducer 140.

[0065] Preferably, the normalized calculation formula is:

[0066] X=(X i -X min ) / (X max -X min ).

[0067] Among them, X represents the normalized data, X i Represents the original multi-dimensional performance data. X i It includes four types of data: transmission error ET, transmission efficiency η, vibration signal S and noise signal DB. max With X min Respectively represent the maximum and minimum values ​​of the original multi-dimensional performance data.

[0068] Preferably, it is necessary to determine the weight coefficients corresponding to the transmission accuracy, transmission efficiency, vibration data and noise data of the reducer 140, and the comprehensive performance evaluation function model of the reducer 140 can be weighted and balanced according to the reducer data indicators with different requirements.

[0069] Preferably, a judgment matrix is ​​constructed based on the relative scale values ​​of the multi-dimensional performance data, and weight coefficients are calculated.

[0070] S311: Constructing a weight hierarchy model of the reducer 140 .

[0071] The target layer is the assembled reducer 140, the criterion layer is the transmission accuracy, transmission efficiency, vibration data and noise data of the assembled reducer 140, and the solution layer is the assembled qualified reducer 140 and the assembled unqualified reducer 140.

[0072] The AHP method can optimally systematize the complex assembly process, breaking it down into multiple hierarchical structures: the top layer is the target layer, such as the overall performance of a qualified reducer 140; the middle layer is the criteria layer, covering data such as transmission accuracy, transmission efficiency, transmission vibration, and transmission noise; and the bottom layer is the solution layer or indicator layer, specifically the technical parameters of reducer 140. This hierarchical structure helps clearly display and process the relationships between the various analysis indicators, ensuring that no key factors are overlooked during the decision-making process.

[0073] By treating the assembly process of reducer 140 as a system and breaking it down into different levels and criteria, the AHP method can comprehensively consider the interrelationships between various assembly indicators. This method provides a systematic evaluation framework that is particularly suitable for complex system assembly such as reducer 140 because it ensures a balance between various performance analysis indicators. This systematic evaluation not only improves the transparency of the assembly process but also enhances the ability to control the final assembly quality.

[0074] S312: Construct a judgment matrix based on the relative importance of indicators.

[0075] The judgment matrix scale ranges from integers 1-9 and the reciprocals of 1-9. The magnitude of the values ​​indicates the relative importance of two indicators among transmission accuracy, transmission efficiency, vibration level, and noise level. The larger the scale, the higher the importance of the former, and a scale of 1 indicates that they are equally important.

[0076] Construct an n×n judgment matrix P:

[0077] .

[0078] a ij Indicates relative scale value. a ij Satisfy the following conditions: (1) a ij >0, (2)a ij =1 / a ji , (3) a ii ,a jj =0.

[0079] There are 4 criteria layer indicators, so n=4.

[0080] For example, construct a 4×4 judgment matrix P:

[0081] .

[0082] For example, when assembling a harmonic reducer, high transmission accuracy, high transmission efficiency, good vibration level, and normal noise level are required. The intelligent assembly model can clearly determine the scale between these four performance indicators of the harmonic reducer required for assembly. The scale of transmission accuracy to transmission efficiency is 1, the scale of transmission accuracy to vibration level is 5, and the scale of transmission accuracy to noise level is 9. According to Satty's 1-9 scale, all relative scale values ​​can be obtained. ij .

[0083] From this, we can see that the judgment matrix will change accordingly based on the user's performance requirements.

[0084] S313: Calculate the weight coefficient based on the square root method.

[0085] Calculate the nth power of each row in the judgment matrix to obtain an n-dimensional vector: . Normalize the vector to a weight vector, and the weight coefficient is: .in, represents the weight coefficient, Represents the relative scale value in the judgment matrix; It represents the nth root of the product of all elements in the j-th row of the judgment matrix.

[0086] For example, calculating the fourth power of each row in the judgment matrix yields a 4-dimensional vector: . Normalize the vector to a weight vector, and the weight coefficient is: .

[0087] Preferably, whether the consistency of the matrix is ​​passed is evaluated based on the CR value.

[0088] .

[0089] CI stands for consistency ratio. RI is the random consistency index obtained by Satty simulation 1000 times and can be obtained by looking up the table.

[0090] First, calculate the maximum eigenvalue of the judgment matrix: .

[0091] For example, when n=4, .

[0092] (Pω) i It means the judgment matrix is ​​multiplied by the standardized weights and accumulated row by row.

[0093] Then, calculate the consistency ratio CI value: .

[0094] For example, when n=4, .

[0095] When CR<0.1, it means that the consistency of the judgment matrix is ​​considered to be within the allowable range.

[0096] When the CR value is ≥0.1, it means that there is a logical error in the judgment matrix.

[0097] For example, the scale of transmission error ET relative to vibration data S is 3 (indicating that transmission accuracy is slightly more important than transmission vibration), and the scale of transmission error ET relative to noise data DB is 1 / 3 (indicating that transmission noise is slightly more important than transmission accuracy). When judging transmission accuracy relative to transmission noise, according to the above logic, transmission noise should be more important than transmission vibration. If, when constructing the judgment matrix, the scale of vibration data S relative to noise data DB is incorrectly entered as 3 (transmission vibration is slightly more important than transmission noise), then a logical error has occurred.

[0098] When calculating the weight distribution, the present invention adopts the AHP hierarchical analysis method to calculate the weight coefficient corresponding to the analysis index. Its advantages are: no need for a large number of data samples to support, nor need to consider the mutual influence between the data. It only needs to consider the required indicators for the assembly of the reducer 140. Starting from the essence of the requirements for the performance of the reducer, the logic is clearer than that of the general quantitative method. The hierarchical analysis method regards the research object as a system and makes decisions according to the thinking mode of decomposition, comparison, judgment and synthesis. It is particularly suitable for multi-objective and multi-criteria system evaluation such as the performance evaluation of the reducer. The integration of the deep learning method of the intelligent assembly model of the present invention associates the data of various sensor units with the performance indicators of the reducer. While realizing the classification of assembly errors, the trained data set is combined with the weight distribution to realize intelligent assembly evaluation.

[0099] S32: After the weight coefficient is determined, the assembly evaluation index of the speed reducer 140 is calculated based on the comprehensive performance evaluation function model.

[0100] After calculating the weight coefficient ω according to actual needs, the weight coefficient is substituted into the comprehensive performance evaluation function model: α=ω1ε+ω2η+ω3S+ω4DB, and the assembly evaluation index α can be obtained as a curve of the synthetic data of the four original performance data.

[0101] Perform mean processing on α: .

[0102] represents the average value of the assembly evaluation index α, Represents the assembly evaluation index α of each dimension.

[0103] If the objects tested during the test are all reducers 140 of known fault types, there is no need to perform assembly qualification judgment.

[0104] S33: Perform data enhancement on the curve of the assembly evaluation index and form a data set sample.

[0105] After slicing, shifting, and flipping the assembly evaluation index α, the sample f is obtained.

[0106] Preferably, data enhancement is performed on the curve of the assembly evaluation index to calculate:

[0107] ;

[0108] ;

[0109] .

[0110] f represents the signal data obtained after enhancement; p represents slicing of the assembly evaluation index α, q represents offsetting of the assembly evaluation index α, r represents mirror flipping of the assembly evaluation index α, and α represents the assembly evaluation index; q a Indicates the starting point offset; p a Indicates the slice length; p T Indicates the number of signal sampling points when the output end of the reducer rotates one circle, that is, one cycle.

[0111] It means that the slices at the corresponding positions are spliced ​​end to end in sequence to obtain the final signal data sample f.

[0112] S34: Perform artificial intelligence model training based on data set samples to form an intelligent assembly model.

[0113] The sample order is disrupted by random seeds. The sample library consisting of several signal data samples f is divided into training set, validation set, and test set in a ratio of 7:2:1 to obtain the training data set for the artificial intelligence model.

[0114] like Figure 2 and Figure 3 As shown, when a deep learning neural network model is used as an artificial intelligence model, the deep learning neural network model performs deep learning tasks by processing a data set. The deep learning neural network model architecture includes components such as a convolutional layer, a pooling layer, a ReLU activation function, a Dropout layer, a fully connected layer, and a Softmax layer. These components work together on the data set to optimize parameters and extract data features. After completing the above steps and the model training is completed, the classification performance of the model is evaluated through the test set, thereby achieving accurate classification of different types of faults of the reducer 140, and finally obtaining an optimized pre-trained model, namely the intelligent assembly model.

[0115] Specifically, during the training process, the present invention selects the Adam optimizer as the optimization algorithm, which can effectively adjust the learning rate and accelerate the convergence process. At the same time, the cross entropy loss function is used as a standard to measure the difference between the predicted results and the actual labels. In order to enhance the generalization ability of the model, a Dropout layer is set after the convolution layer to randomly discard a part of the neurons to reduce overfitting. Then, the Flatten layer is used to convert the multidimensional feature map into a one-dimensional vector to facilitate subsequent fully connected layer processing. Finally, the Softmax layer is responsible for calculating the probability distribution corresponding to each category and outputting the final classification result. In this way, the entire process not only improves the accuracy of the model in identifying the fault type of the reducer 140, but also ensures that the model has good generalization performance and is suitable for fault detection tasks in actual application scenarios.

[0116] After completing the training, verification and testing of the artificial intelligence model, the assembly of the reducer 140 will be officially carried out.

[0117] After the reducer 140 is assembled, the intelligent assembly model performs an assembly evaluation index function α and its mean based on the multi-dimensional performance data collected by the sensor unit. Calculation.

[0118] In the intelligent assembly model, the model parameters, namely the target value scores, are set in advance.

[0119] The target value score ξ is set according to different performance requirements of the speed reducer 140 .

[0120] ,

[0121] Among them, ET Cv ,η Cv 、S Cv 、DB Cv They respectively represent indicator data sets of the reducer 140 under standard operating conditions.

[0122] The following table lists the target value functions used to calculate the target value scores under different performance requirements.

[0123]

[0124] The intelligent assembly model determines whether the assembly is qualified based on the average value and target value score of the assembly evaluation index.

[0125] Preferably, when When the intelligent assembly model determines that the assembly is qualified, the product will be shipped out, otherwise it will fail.

[0126] Preferably, when the classification result is unqualified, the intelligent assembly model classifies the test data of the reducer 140 and determines the classification of the error type. Specifically, the intelligent assembly model collects data of the assembly evaluation index α and classifies and determines the error type.

[0127] During the training of the artificial intelligence model, since experimental data of the reducer 140 with faults of different error types is used, the intelligent assembly model can learn the error type during the training process, thereby determining the classification of the error type and outputting the classification result of the error type.

[0128] Preferably, if Figure 3 As shown in the figure, after obtaining the classification results, the intelligent assembly model can now determine the error type of reducer 140 assembly. If it is determined to be a gear machining accuracy error, the gear needs to be returned to the factory for reprocessing. If it is determined to be an assembly error, the classification of different assembly error types will continue. If it is determined to be an assembly center distance error, the center distance between important meshing components needs to be adjusted. If it is determined to be a bearing preload error, the bearing preload size needs to be adjusted during the installation of reducer 140.

[0129] After the speed reducer 140 is assembled, the intelligent assembly model proposes a corresponding correction plan for the assembled product of the speed reducer 140 according to the actual situation.

[0130] The intelligent assembly testing platform 100 for the reducer 140 and its intelligent assembly model of the present invention reconstruct a comprehensive performance evaluation function model and intelligently assess the conformity of the reducer 140 assembly, achieving precise classification of error types and providing targeted correction solutions, significantly improving the quality of the assembled product and production efficiency. Furthermore, with data accumulation and iterative model optimization, the predictive capability and robustness of the intelligent assembly model continue to improve, further promoting continuous improvement in production, reducing overall costs, and enhancing the company's market competitiveness.

[0131] It should be noted that the above-mentioned specific embodiments are exemplary, and those skilled in the art can come up with various solutions inspired by the disclosure of the present invention, and these solutions also fall within the scope of the disclosure of the present invention and fall within the scope of protection of the present invention. Those skilled in the art should understand that the present invention specification and its drawings are illustrative and do not constitute a limitation on the claims. The scope of protection of the present invention is defined by the claims and their equivalents. The present invention specification contains multiple inventive concepts, such as "preferably" and "according to a preferred embodiment", which means that the corresponding paragraph discloses an independent concept, and the applicant reserves the right to file a divisional application based on each inventive concept.

Claims

1. A method for assembling a reducer, characterized in that: The method comprises: Collecting multi-dimensional performance data of the reducer (140) including position data, torsional force data, noise data and vibration data; The pre-built intelligent assembly model calculates the assembly evaluation index and / or the average value of the assembly evaluation index based on the multi-dimensional performance data; The intelligent assembly model determines whether the assembly is qualified based on a comparison result of the average value of the assembly evaluation index and a preset target value score; The intelligent assembly model is provided with a comprehensive performance evaluation function model based on weight distribution to calculate the assembly evaluation index and / or the average value of the assembly evaluation index. The intelligent assembly model can dynamically adjust the weights of different performance parameters according to their importance to reflect the actual working conditions of the reducer. The comprehensive performance evaluation function model is: α=ω1ET+ω2η+ω3S+ω4DB; Wherein, α represents the assembly evaluation index, ET represents the transmission error, η represents the transmission efficiency, S represents the vibration signal, DB represents the noise signal, ω1, ω2, ω3 and ω4 represent the weight coefficients respectively; Substituting weight coefficients into the comprehensive performance evaluation function model to obtain a curve of an assembly evaluation index, performing data enhancement on the curve of the assembly evaluation index and forming a data set sample; The curve of the assembly evaluation index is calculated by data enhancement: ； ； ； f represents the signal data obtained after enhancement; p represents slicing of the assembly evaluation index α, q represents offsetting of the assembly evaluation index α, r represents mirror flipping of the assembly evaluation index α, and α represents the assembly evaluation index; q a Indicates the starting point offset; p a Indicates the slice length; p T Indicates the number of signal sampling points when the output end of the reducer rotates for one cycle.

2. The method according to claim 1, characterized in that The steps of constructing the intelligent assembly model include: Construct a comprehensive performance evaluation function model based on weight distribution based on multi-dimensional performance indicators, Determining weight coefficients in the comprehensive performance evaluation function model, wherein a judgment matrix is ​​constructed based on relative scale values ​​of multi-dimensional performance data, and the weight coefficients are calculated; After determining the weight coefficient, calculating an assembly evaluation index of the reducer (140) based on the comprehensive performance evaluation function model; Artificial intelligence model training is performed based on the data set samples to form an intelligent assembly model.

3. The method according to claim 2, characterized in that The method further comprises: Before determining the weight coefficient, the multi-dimensional performance data is normalized to limit the range of the original multi-dimensional performance data to [0, 1], so that the convergence in the process of training the artificial intelligence model is faster.

4. The method according to claim 2 or 3, characterized in that The step of determining the weight coefficient of the comprehensive performance evaluation function model includes: Calculate the nth power of each row in the judgment matrix to obtain an n-dimensional vector: ; Normalize the vector into a weight vector, and the weight coefficient is: ; in, represents the weight coefficient, Represents the relative scale value in the judgment matrix; It represents the nth root of the product of all elements in the j-th row of the judgment matrix.

5. A reducer assembly system, characterized in that: include: A sensing unit for collecting multi-dimensional performance data of the reducer (140) including position data, torsional force data, noise data, and vibration data; The server (180) has a pre-built intelligent assembly model. The intelligent assembly model calculates an assembly evaluation index and / or an average value of the assembly evaluation index based on the multi-dimensional performance data; and determines whether the assembly is qualified based on a comparison result of the average value of the assembly evaluation index and a preset target value score; The intelligent assembly model is provided with a comprehensive performance evaluation function model based on weight distribution to calculate the assembly evaluation index and / or the average value of the assembly evaluation index. The intelligent assembly model can dynamically adjust the weights of different performance parameters according to their importance to reflect the actual working conditions of the reducer. The comprehensive performance evaluation function model is: α=ω1ET+ω2η+ω3S+ω4DB; Wherein, α represents the assembly evaluation index, ET represents the transmission error, η represents the transmission efficiency, S represents the vibration signal, DB represents the noise signal, ω1, ω2, ω3 and ω4 represent the weight coefficients respectively; Substituting weight coefficients into the comprehensive performance evaluation function model to obtain a curve of an assembly evaluation index, performing data enhancement on the curve of the assembly evaluation index and forming a data set sample; The curve of the assembly evaluation index is calculated by data enhancement: ； ； ； f represents the signal data obtained after enhancement; p represents slicing of the assembly evaluation index α, q represents offsetting of the assembly evaluation index α, r represents mirror flipping of the assembly evaluation index α, and α represents the assembly evaluation index; q a Indicates the starting point offset; p a Indicates the slice length; p T Indicates the number of signal sampling points when the output end of the reducer rotates for one cycle.

6. The system according to claim 5, characterized in that The intelligent assembly model is configured as follows: Construct a comprehensive performance evaluation function model based on weight distribution based on multi-dimensional performance indicators, Determining weight coefficients in the comprehensive performance evaluation function model, wherein a judgment matrix is ​​constructed based on relative scale values ​​of multi-dimensional performance data, and the weight coefficients are calculated; After the weight coefficient is determined, the assembly evaluation index of the speed reducer (140) is calculated based on the comprehensive performance evaluation function model.

7. The system according to claim 5 or 6, characterized in that The intelligent assembly model is further configured to: Before determining the weight coefficient, the multi-dimensional performance data is normalized to limit the range of the original multi-dimensional performance data to [0, 1], so that the convergence in the process of training the artificial intelligence model is faster.

Images