Method and System for Monitoring the Wear Status of Ultrasonic Welding Heads

Abstract

This invention discloses a method for monitoring the wear status of an ultrasonic welding head, comprising: acquiring vibration time-series data of the welding process; extracting time-domain and frequency-domain features of the vibration time-series data from both the time and frequency domains, and forming a multi-dimensional feature vector from the time-domain and frequency-domain features; performing unsupervised clustering on the multi-dimensional feature vector to calculate the similarity between the current wear status of the welding head and the previous wear status; and controlling the wear status of the welding head based on the similarity. This method can reflect the wear status of the ultrasonic welding head in real time, preventing batch welding defects in battery cells due to untimely monitoring of welding head wear; and maximizing the service life of the ultrasonic welding head by judging its wear status.

Description

Technical Field

[0001] This invention relates to the field of welding inspection technology, specifically to a method and system for monitoring the wear status of ultrasonic welding heads. Background Technology

[0002] In the manufacturing of lithium-ion battery cells, ultrasonic welding is a crucial step in the battery production process. The main components of an ultrasonic welding system include a generator, transducer, amplitude transformer, welding head triplet, mold, and mechanism. The ultrasonic welding head is a general term for the end of the ultrasonic wave, and its function is to couple the ultrasonic waves generated by the transducer to the object being welded. Ultrasonic welding utilizes high-frequency vibration waves transmitted through the welding head to the surfaces of the two objects to be welded. Under pressure, the two surfaces rub against each other, forming a fusion between molecular layers. The high-frequency vibration and pressure friction cause wear on the welding head, and this wear accumulates with each welding cycle. As the ultrasonic welding head is in direct contact with the object being welded, the surface smoothness of the welding head directly affects the quality, safety, and final battery capacity consistency of the battery.

[0003] Currently, the lifespan of welding heads is managed on-site by controlling the number of welding operations. The shortcomings are: (1) The lifespan monitoring method is based on the number of welding operations. For safety reasons, the welding head is often not in a state of failure due to wear and tear. Controlling the lifespan by the number of welding operations increases manufacturing costs; (2) Welding heads that have reached the required number of welding operations need to be repaired and reused. However, the condition of the repaired welding head is inconsistent with the initial condition of the brand-new welding head. Using the same standard for control is unreasonable and poses a certain safety risk to the quality of the battery cells; (3) Due to the differences in incoming materials, the on-site process will adjust the welding machine parameters according to the actual situation, resulting in unstable wear and tear of the welding head during each batch of ultrasonic welding. Controlling the lifespan solely by the number of welding operations will increase the quality risk of welding defects.

[0004] In related technologies, the Chinese invention patent application CN111168195A proposes an IPv6-based cluster monitoring and control method for welding machines. This method performs cluster analysis on collected welding production parameters to detect anomalies in welding quality and control the quality of post-weld products. The literature "Cluster Analysis of Welding Structure Quality, Modern Manufacturing Engineering, 2008, No. 2, Zong Pei et al." proposes using cluster analysis to classify the welding quality of various sections of a pressure vessel for problems such as multi-index evaluation, optimization, and classification. However, these technologies all focus on detecting and classifying welding quality, which is not suitable for monitoring the wear state of ultrasonic welding heads. Furthermore, they all employ hierarchical clustering methods, which cannot be corrected once a split or merge is performed, limiting the clustering quality and making them susceptible to interference from outliers.

[0005] Therefore, the lithium battery industry currently lacks an effective way to control the wear and tear of welding heads, and urgently needs a scientific, reasonable and effective way to monitor the wear and tear to replace the method of controlling the lifespan of ultrasonic welding heads solely based on the number of welding operations. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to scientifically, rationally and effectively monitor the wear state of ultrasonic welding heads.

[0007] The present invention solves the above-mentioned technical problems through the following technical means:

[0008] In a first aspect, the present invention proposes a method for monitoring the wear status of an ultrasonic welding head, the method comprising:

[0009] Acquire vibration timing data during the welding process of the welding head;

[0010] The time-domain features and frequency-domain features of the vibration time-series data are extracted from the time-domain direction and the frequency-domain direction, respectively, and the time-domain features and the frequency-domain features are combined to form a multi-dimensional feature vector;

[0011] Unsupervised clustering is performed on the multidimensional feature vectors to calculate the similarity between the current loss state of the welding head and the previous loss state.

[0012] Based on the similarity, the wear status of the welding head is controlled.

[0013] Furthermore, acquiring vibration timing data during the welding process of the welding head includes:

[0014] Vibration timing data of the welding process of the welding head are collected using a triaxial vibration sensor installed on the welding head;

[0015] The maximum frequency acquired by the triaxial sensor is greater than or equal to the frequency of the ultrasonic generator, and the main direction of the triaxial sensor is consistent with the direction of ultrasonic vibration.

[0016] Further, the step of extracting the temporal and frequency features of the vibration time series data from the time domain and frequency domain directions respectively, and combining the temporal and frequency features into a multi-dimensional feature vector, includes:

[0017] The temporal features of the vibration time series data are extracted from the temporal domain direction. The temporal features include the maximum value, standard deviation, mean, and post-sequence features of the downsampled vibration time series data.

[0018] The vibration time series data is subjected to discrete Fourier transform processing to extract the frequency domain features of the vibration time series data. The frequency domain features include the mean vibration frequency, the centroid of the frequency, and the frequency amplitude.

[0019] The time-domain features and the frequency-domain features are combined to form a multi-dimensional feature vector.

[0020] Furthermore, the unsupervised clustering processing of the multidimensional feature vector to calculate the similarity between the current loss state of the welding head and the previous loss state includes:

[0021] The multidimensional feature vector is subjected to unsupervised clustering using a pre-trained self-organizing feature map neural network, and the distance between the newly input multidimensional feature vector and the center of the cluster is calculated.

[0022] The distance value is used as the similarity between the current wear state of the welding head and the previous wear state.

[0023] Furthermore, the control of welding head wear status based on the similarity includes:

[0024] When the similarity is greater than a set threshold, the welding head is determined to be in an abnormal state and an early warning is issued;

[0025] When the similarity is less than or equal to a set threshold, the welding head is determined to be in normal condition.

[0026] Furthermore, the method also includes:

[0027] The historical data of the welding head welding process was divided into several clusters using a clustering method, and the relative distance of each point in the cluster to the cluster center point was calculated.

[0028] The maximum relative distance is selected as the set threshold;

[0029] Where, relative distance = distance from a point within the cluster to the cluster center / median distance of all points within the cluster to the cluster center.

[0030] Secondly, the present invention also proposes an ultrasonic welding head wear status monitoring system, the system comprising:

[0031] The data acquisition module is used to acquire vibration timing data during the welding process of the welding head;

[0032] The feature extraction module is used to extract the time-domain features and frequency-domain features of the vibration time series data from the time-domain direction and the frequency-domain direction, respectively, and to combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector;

[0033] The clustering module is used to perform unsupervised clustering on the multidimensional feature vector and calculate the similarity between the current loss state of the welding head and the previous loss state.

[0034] The control module is used to control the wear status of the welding head based on the similarity.

[0035] Furthermore, the data acquisition module employs a triaxial vibration sensor, which is mounted on the welding head;

[0036] The maximum frequency acquired by the triaxial sensor is greater than or equal to the frequency of the ultrasonic generator, and the principal direction of the triaxial sensor is consistent with the direction of ultrasonic vibration.

[0037] Furthermore, the feature extraction module includes:

[0038] The temporal feature extraction unit is used to extract the temporal features of the vibration time series data from the temporal direction. The temporal features include the maximum value, standard deviation, mean, and the post-sequence features of the downsampled vibration time series data.

[0039] The frequency domain feature extraction unit is used to perform discrete Fourier transform processing on the vibration time series data to extract the frequency domain features of the vibration time series data. The time domain features include the mean vibration frequency, the centroid of the frequency, and the frequency amplitude.

[0040] The feature vector construction unit is used to combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector.

[0041] Furthermore, the clustering processing module includes:

[0042] An unsupervised clustering unit is used to perform unsupervised clustering processing on the multidimensional feature vector using a pre-trained self-organizing feature map neural network, and to calculate the distance between the newly input multidimensional feature vector and the center of the cluster.

[0043] A similarity determination unit is used to use the distance value as the similarity between the current loss state of the welding head and the previous loss state.

[0044] The advantages of this invention are:

[0045] (1) This invention extracts time-domain and frequency-domain features from the vibration time-series data of the welding head during the welding process, and combines the time-domain and frequency-domain features into a multi-dimensional feature vector. Unsupervised clustering is performed on the multi-dimensional feature vector to calculate the similarity between the current loss state of the welding head and the previous loss state. The loss state of the welding head is controlled based on the similarity. It can reflect the loss state of the ultrasonic welding head in real time, and prevent the battery cells from having batch welding defects due to untimely monitoring of welding head loss. The lifespan of the ultrasonic welding head can be maximized by judging the loss state of the ultrasonic welding head. Moreover, by using an unsupervised clustering algorithm, the adjacent relationship can be imposed on the centroid of the cluster. Clusters that are neighbors are more related than clusters that are not neighbors. This relationship is beneficial to the interpretation and visualization of the clustering results.

[0046] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the ultrasonic welding head wear status monitoring method proposed in this invention.

[0048] Figure 2 This is a schematic diagram of the overall process of the ultrasonic welding head wear status monitoring method proposed in this invention;

[0049] Figure 3 This is a schematic diagram of the ultrasonic welding head wear status monitoring system proposed in this invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] like Figures 1 to 2 As shown, the first embodiment of the present invention proposes a method for monitoring the wear status of an ultrasonic welding head, the method comprising the following steps:

[0052] S10. Obtain vibration timing data during the welding process of the welding head;

[0053] S20. Extract the time-domain features and frequency-domain features of the vibration time series data from the time-domain direction and the frequency-domain direction respectively, and combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector;

[0054] S30. Perform unsupervised clustering on the multidimensional feature vector and calculate the similarity between the current loss state of the welding head and the previous loss state.

[0055] S40. Based on the similarity, control the wear status of the welding head.

[0056] This embodiment extracts time-domain and frequency-domain features from the vibration time-series data of the welding head during the welding process, and combines these features into a multi-dimensional feature vector. Unsupervised clustering is then performed on this multi-dimensional feature vector to calculate the similarity between the current and previous wear states of the welding head. This similarity-based control of the welding head wear state allows for real-time monitoring of the ultrasonic welding head's wear state, preventing batch welding defects in battery cells due to untimely monitoring of welding head wear. Furthermore, by determining the wear state of the ultrasonic welding head, its lifespan can be maximized.

[0057] In one embodiment, step S10, acquiring vibration timing data during the welding process of the welding head, specifically includes the following steps:

[0058] Vibration timing data of the welding process of the welding head are collected using a triaxial vibration sensor installed on the welding head;

[0059] The maximum frequency acquired by the triaxial sensor is greater than or equal to the frequency of the ultrasonic generator, and the main direction of the triaxial sensor is consistent with the direction of ultrasonic vibration.

[0060] It should be noted that the timing data of the welding head welding process is the acceleration value that changes over time. In this embodiment, a 3-axis vibration sensor is used to collect timing data in the x, y, and z directions, resulting in three sets of timing data, which can yield data features in three dimensions.

[0061] In one embodiment, step S20: extracting the time-domain features and frequency-domain features of the vibration time-series data from the time-domain direction and the frequency-domain direction respectively, and forming a multi-dimensional feature vector from the time-domain features and the frequency-domain features, specifically includes the following steps:

[0062] S21. Extract the temporal features of the vibration time series data from the time domain direction. The temporal features include the maximum value, standard deviation, mean value, and the post-sequence features of the downsampled vibration time series data.

[0063] It should be noted that the features extracted from the vibration time series data include the maximum value (max), standard deviation (std), and mean (mean), and the position information (index) is obtained, i.e., which point it is located at in the time series. Since the welding time is fixed, the length of the data collected within the welding time is fixed. Therefore, the vibration time series data is downsampled, and the sampling interval is Δn.

[0064] S22. Perform Discrete Fourier Transform on the vibration time series data to extract the frequency domain features of the vibration time series data. The time domain features include the mean vibration frequency, the centroid of the frequency, and the frequency amplitude.

[0065] S23. Combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector.

[0066] It should be noted that the multidimensional feature vector is specifically as follows:

[0067] (frequency mean) x ,.........Center frequency x ,.......Vibration amplitude x ,.......max,std,mean)

[0068] In this embodiment, in addition to peak values, standard deviation and mean values ​​are extracted from the time-domain data features, and mean values, centroid frequency, amplitude, etc., are extracted from the frequency-domain data, making the extraction of time-domain and frequency-domain data features more comprehensive.

[0069] In one embodiment, step S30: performing unsupervised clustering on the multidimensional feature vector to calculate the similarity between the current loss state of the welding head and the previous loss state, specifically includes the following steps:

[0070] S31. Use a pre-trained self-organizing feature map neural network to perform unsupervised clustering on the multidimensional feature vector, and calculate the distance between the newly input multidimensional feature vector and the center of the cluster.

[0071] It should be noted that the multidimensional feature vector described in this embodiment includes a three-dimensional feature vector. Taking the x-axis direction as an example, the feature vector corresponding to the x-axis direction is X = [x1, x2, x3, x4, x5, x6, x7], where x1 is the maximum value of the time series data (m / s²), x2 is the standard deviation of the time series data (m / s²), x3 is the average value of the time series data (m / s²), x4 is the average frequency (kHz), x5 is the centroid frequency (kHz), x6 is the frequency amplitude (m), and x7 is the data after downsampling and the data after the first use of downsampling by the welding head. The feature vectors corresponding to the y-axis and z-axis directions are the same as those corresponding to the x-axis direction.

[0072] S32. The distance value is used as the similarity between the current loss state of the welding head and the previous loss state.

[0073] It should be noted that the similarity between the current wear state and the previous wear state of the welding head is evaluated by calculating the cosine of the angle between the multidimensional feature vectors at the current time and the previous time.

[0074] It should be noted that this embodiment uses a self-organizing feature map (SOM) neural network to perform unsupervised clustering on the multidimensional feature vectors. The input layer of the neural network is set to have 7 inputs and 7 neurons; the output layer has 100 = 10 × 10 (nodes). The process of adjusting the weights and defining the winning domain is as follows:

[0075] (1) Initialization

[0076] Each feature in the input layer is assigned a small random number and then normalized to obtain... Establish the initial winning neighborhood And the learning rate η, where m is the number of neurons in the output layer.

[0077] (2) Accept input

[0078] A random input pattern is selected from the training set and normalized. n represents the number of neurons in the input layer.

[0079] (3) Find the winning node

[0080] calculate and Dot product, find the winning node j with the largest dot product. * If the input pattern is not normalized, the Euclidean distance should be calculated to find the winning node with the smallest distance.

[0081] (4) Define the winning field

[0082] Let the center determine the weight adjustment region at time t, generally the initial neighborhood. Larger, during training It converges over time.

[0083] (5) Adjusting weights

[0084] For the winning field Adjust the weights of all nodes within.

[0085]

[0086] In the formula: w ij (t) represents the weight of neuron i at time j, and α(t,N) represents the training time and the weight of the i-th neuron and the winning neuron j in the neighborhood. * The topological distance N between them is a function.

[0087] (6) End judgment

[0088] When the learning rate α(t) ≤ α minIf the termination condition is not met, the training ends; otherwise, proceed to step (2) to continue training.

[0089] This embodiment uses an unsupervised clustering algorithm to impose adjacency relationships on the centroid of clusters. Clusters that are neighbors are more correlated than clusters that are not neighbors. This relationship is beneficial for the interpretation and visualization of clustering results.

[0090] In one embodiment, in order to effectively avoid the occurrence of "dead neurons" and improve the training accuracy and speed of the SOM neural network, the particle swarm optimization algorithm can be used to optimize the SOM neural network during the first training of the neural network.

[0091] First, for the Particle Swarm Optimization (PSO) algorithm, the SOM neural network is treated as individual particles, and the superposition of the Euclidean distances between the input pattern and the weight vector is considered as the fitness function. Through continuous iteration, the fitness function is minimized. The weight vector of each SOM neural network is updated using the velocity-displacement formula of the PSO algorithm, thus updating the weights of each network. After a certain number of iterations, the weights of the training network are adjusted using the SOM algorithm's weights. The training process of the SOM using the Particle Swarm Optimization algorithm is as follows:

[0092] (1) Randomly set the network weight vector w j , and input sample [x1,x2,x3,x4,x5,x6,x7];

[0093] (2) Using the weight vector w j Initialize the particles for the particle swarm optimization algorithm;

[0094] (3) The superposition of the Euclidean distances between all input vectors and weight vectors is used as the fitness function f(x) of the particle swarm optimization algorithm. i ), x i Given the input sample, optimize the weights through M iterations:

[0095]

[0096] (4) The optimized values ​​of each particle are reset as the connection weights of the neural network, and the SOM neural network algorithm is used for training, iterating n times.

[0097] (5) Repeat steps (2) to (4) until the target number of iterations is reached.

[0098] In one embodiment, step S40: managing the welding head wear status based on the similarity includes the following steps:

[0099] When the similarity is greater than a set threshold, the welding head is determined to be in an abnormal state and an early warning is issued;

[0100] When the similarity is less than or equal to a set threshold, the welding head is determined to be in normal condition.

[0101] In one embodiment, the method further includes:

[0102] The historical data of the welding head welding process was divided into several clusters using a clustering method, and the relative distance of each point in the cluster to the cluster center point was calculated.

[0103] The maximum relative distance is selected as the set threshold;

[0104] Where, relative distance = distance from a point within the cluster to the cluster center / median distance of all points within the cluster to the cluster center.

[0105] It should be noted that points with relatively large distances are selected based on the median distance from each point to the cluster center; because outliers have a very small impact on the median, which can be almost ignored, but have a large impact on the mean, the median is selected.

[0106] In addition, such as Figure 3 As shown, the second embodiment of the present invention proposes an ultrasonic welding head wear status monitoring system, the system comprising:

[0107] The data acquisition module 10 is used to acquire vibration timing data during the welding process of the welding head;

[0108] Feature extraction module 20 is used to extract time-domain features and frequency-domain features of the vibration time series data from the time domain direction and the frequency domain direction respectively, and to combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector;

[0109] Clustering module 30 is used to perform unsupervised clustering on the multidimensional feature vector and calculate the similarity between the current loss state of the welding head and the previous loss state.

[0110] The control module 40 is used to control the wear status of the welding head based on the similarity.

[0111] Time-domain and frequency-domain features are extracted from the vibration time-series data of the welding head during the welding process. These features are then combined into a multi-dimensional feature vector. Unsupervised clustering is performed on this multi-dimensional feature vector to calculate the similarity between the current and previous wear states of the welding head. This similarity-based control of the welding head wear state allows for real-time monitoring of the ultrasonic welding head's wear state, preventing batch welding defects in battery cells due to untimely monitoring of welding head wear. By determining the wear state of the ultrasonic welding head, its lifespan can be maximized.

[0112] In one embodiment, the data acquisition module employs a triaxial vibration sensor, which is mounted on the welding head;

[0113] The maximum frequency acquired by the triaxial sensor is greater than or equal to the frequency of the ultrasonic generator, and the principal direction of the triaxial sensor is consistent with the direction of ultrasonic vibration.

[0114] In one embodiment, the feature extraction module 20 includes:

[0115] The temporal feature extraction unit is used to extract the temporal features of the vibration time series data from the temporal direction. The temporal features include the maximum value, standard deviation, mean, and the post-sequence features of the downsampled vibration time series data.

[0116] The frequency domain feature extraction unit is used to perform discrete Fourier transform processing on the vibration time series data to extract the frequency domain features of the vibration time series data. The time domain features include the mean vibration frequency, the centroid of the frequency, and the frequency amplitude.

[0117] The feature vector construction unit is used to combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector.

[0118] In one embodiment, the clustering processing module 30 includes:

[0119] An unsupervised clustering unit is used to perform unsupervised clustering processing on the multidimensional feature vector using a pre-trained self-organizing feature map neural network, and to calculate the distance between the newly input multidimensional feature vector and the center of the cluster.

[0120] A similarity determination unit is used to use the distance value as the similarity between the current loss state of the welding head and the previous loss state.

[0121] It should be noted that other embodiments or implementation methods of the ultrasonic welding head wear status monitoring system of the present invention can refer to the above-described method embodiments, and will not be repeated here.

[0122] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0123] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0124] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for monitoring the wear status of an ultrasonic welding head, characterized in that, The method includes: Acquire vibration timing data during the welding process of the welding head; The time-domain features and frequency-domain features of the vibration time-series data are extracted from the time-domain direction and the frequency-domain direction, respectively, and the time-domain features and the frequency-domain features are combined to form a multi-dimensional feature vector; Unsupervised clustering is performed on the multidimensional feature vectors to calculate the similarity between the current wear state and the previous wear state of the welding head. This includes using a pre-trained self-organizing feature map neural network to perform unsupervised clustering on the multidimensional feature vectors, calculating the distance between the newly input multidimensional feature vector and the cluster center, and using the distance value as the similarity between the current wear state and the previous wear state of the welding head. Based on the similarity, the wear status of the welding head is controlled.

2. The ultrasonic welding head wear status monitoring method as described in claim 1, characterized in that, The acquisition of vibration timing data during the welding process of the welding head includes: Vibration timing data of the welding process of the welding head are collected using a triaxial vibration sensor installed on the welding head; The maximum frequency acquired by the triaxial vibration sensor is greater than or equal to the frequency of the ultrasonic generator, and the principal direction of the triaxial vibration sensor is consistent with the direction of ultrasonic vibration.

3. The method for monitoring the wear status of an ultrasonic welding head as described in claim 1, characterized in that, The step of extracting time-domain and frequency-domain features of the vibration time-series data from the time-domain and frequency-domain directions respectively, and combining the time-domain and frequency-domain features into a multi-dimensional feature vector, includes: The temporal features of the vibration time series data are extracted from the temporal domain direction. The temporal features include the maximum value, standard deviation, mean, and post-sequence features of the downsampled vibration time series data. The vibration time series data is subjected to discrete Fourier transform processing to extract the frequency domain features of the vibration time series data. The time domain features include the mean vibration frequency, the centroid of the frequency, and the frequency amplitude. The time-domain features and the frequency-domain features are combined to form a multi-dimensional feature vector.

4. The method for monitoring the wear status of an ultrasonic welding head as described in claim 1, characterized in that, The control of welding head wear status based on the similarity includes: When the similarity is greater than a set threshold, the welding head is determined to be in an abnormal state and an early warning is issued; When the similarity is less than or equal to a set threshold, the welding head is determined to be in normal condition.

5. The ultrasonic welding head wear status monitoring method as described in claim 4, characterized in that, The method further includes: The historical data of the welding head welding process was divided into several clusters using a clustering method, and the relative distance of each point in the cluster to the cluster center point was calculated. The maximum relative distance is selected as the set threshold; Wherein, relative distance = distance from a point within the cluster to the cluster center / median distance of all points within the cluster to the cluster center.

6. An ultrasonic welding head wear status monitoring system, characterized in that, The system includes: The data acquisition module is used to acquire vibration timing data during the welding process of the welding head; The feature extraction module is used to extract the time-domain features and frequency-domain features of the vibration time series data from the time-domain direction and the frequency-domain direction, respectively, and to combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector; The clustering module is used to perform unsupervised clustering on the multidimensional feature vector and calculate the similarity between the current loss state of the welding head and the previous loss state. The control module is used to control the wear status of the welding head based on the similarity. The clustering processing module includes: An unsupervised clustering unit is used to perform unsupervised clustering processing on the multidimensional feature vector using a pre-trained self-organizing feature map neural network, and to calculate the distance between the newly input multidimensional feature vector and the center of the cluster. A similarity determination unit is used to use the distance value as the similarity between the current loss state of the welding head and the previous loss state.

7. The ultrasonic welding head wear status monitoring system as described in claim 6, characterized in that, The data acquisition module uses a triaxial vibration sensor, which is mounted on the welding head. The maximum frequency acquired by the triaxial vibration sensor is greater than or equal to the frequency of the ultrasonic generator, and the principal direction of the triaxial vibration sensor is consistent with the direction of ultrasonic vibration.

8. The ultrasonic welding head wear status monitoring system as described in claim 6, characterized in that, The feature extraction module includes: The temporal feature extraction unit is used to extract the temporal features of the vibration time series data from the temporal direction. The temporal features include the maximum value, standard deviation, mean, and the post-sequence features of the downsampled vibration time series data. The frequency domain feature extraction unit is used to perform discrete Fourier transform processing on the vibration time series data to extract the frequency domain features of the vibration time series data. The time domain features include the mean vibration frequency, the centroid of the frequency, and the frequency amplitude. The feature vector construction unit is used to combine the time-domain features and the frequency-domain features into a multi-dimensional feature vector.

Images