Vacuum nozzle failure prediction method, device, equipment and storage medium

Abstract

The application discloses a vacuum nozzle failure prediction method, device, equipment and storage medium, and the method comprises the following steps: acquiring preset parameter data of a target vacuum nozzle; inputting first fusion features of the reconstructed preset parameter data into a health classifier, and outputting a state of the target vacuum nozzle based on the health classifier; when the state output result of the target vacuum nozzle is a failure class, inputting second fusion features of the preset parameter data into a failure classifier, and outputting a failure cause of the target vacuum nozzle based on the failure classifier. Unlike the existing prediction method that ignores the time difference between the health state duration and the failure state duration of the vacuum nozzle, the preset parameter data of the target vacuum nozzle is reconstructed in the application, and then the state of the target vacuum nozzle is obtained based on the reconstructed features and the health classifier to determine whether the target vacuum nozzle is failed. Therefore, the application alleviates the negative influence of unbalanced health data distribution and improves the accuracy of the prediction result.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, device, and storage medium for predicting vacuum nozzle failure. Background Technology

[0002] In surface mount technology (SMT) processes, pick-and-place machines are widely used, and each machine is equipped with dozens of vacuum nozzles. These nozzles are not only expensive but also require frequent maintenance. Unexpected nozzle failure can lead to decreased machine efficiency, increased defective products, and even machine shutdown, impacting order fulfillment. Furthermore, nozzle replacement is time-consuming. Therefore, predicting nozzle failure is crucial to ensure timely maintenance and replacement even when orders are not pending.

[0003] Existing prediction methods typically only predict vacuum nozzle failure. However, in order to promptly implement different maintenance plans based on different failure causes, factories also hope to predict the causes of vacuum nozzle failure. Furthermore, existing prediction methods neglect the time difference between the healthy state and the failed state of the vacuum nozzle, resulting in a severely unbalanced distribution of analytical data and consequently low accuracy in prediction results.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a method, apparatus, device, and storage medium for predicting vacuum nozzle failure, aiming to solve the technical problem of low accuracy of prediction results caused by unbalanced data distribution in existing vacuum nozzle failure prediction methods.

[0006] To achieve the above objectives, the present invention provides a method for predicting vacuum nozzle failure, the method comprising: Obtain the preset parameter data of the target vacuum nozzle; The first fusion feature after reconstructing the preset parameter data is input into the health classifier, and the status of the target vacuum nozzle is output based on the health classifier; When the status output result of the target vacuum nozzle is a failure, the preset parameter data is input to the failure classifier, and the failure cause of the target vacuum nozzle is output based on the failure classifier.

[0007] Optionally, before acquiring the preset parameter data of the target vacuum nozzle, the method further includes: Obtain the time-frequency characteristics corresponding to the preset parameter data of the sample vacuum nozzle; Based on the time-frequency characteristics, the second fusion feature of the sample vacuum nozzle is obtained; The second fusion feature of the sample vacuum nozzle is input into the failure classifier, and the failure cause of the sample vacuum nozzle is output based on the failure classifier; The second fusion feature of the sample vacuum nozzle is reconstructed to obtain the first fusion feature of the sample vacuum nozzle; The first fusion feature of the sample vacuum nozzle is input into the health classifier, and the status of the sample vacuum nozzle is output based on the health classifier.

[0008] Optionally, the step of obtaining the time-frequency characteristics corresponding to the preset parameter data of the sample vacuum nozzle includes: The preset parameter data of the sample vacuum nozzle is divided to obtain time-series segment data; The time-series data segment is transformed into a time-frequency image; The time-frequency image is used to extract features through the ResNet18 backbone network to obtain time-frequency features corresponding to the preset parameter data of the sample vacuum nozzle.

[0009] Optionally, before acquiring the time-frequency characteristics corresponding to the preset parameter data of the sample vacuum nozzle, the method further includes: Obtain failure data and remaining usage time of the sample vacuum nozzle; The status identifier of the sample vacuum nozzle is constructed based on the remaining usage time; When the status of the sample vacuum nozzle is identified as a failure state, a failure identifier for the sample vacuum nozzle is constructed based on the failure data.

[0010] Optionally, the step of reconstructing the second fused features of the sample vacuum nozzle to obtain the first fused features of the sample vacuum nozzle includes: Obtain the mean and variance of the second fusion feature of the sample vacuum nozzle corresponding to the state identifier; The mean and variance are smoothed using a Gaussian kernel function; Based on the smoothed mean and variance, the first fusion feature of the sample vacuum nozzle is obtained.

[0011] Optionally, after inputting the second fused feature of the sample vacuum nozzle into the failure classifier and outputting the failure cause of the sample vacuum nozzle based on the failure classifier, the method further includes: Based on the failure causes and failure identifiers of the sample vacuum nozzles, a reweighted loss function is constructed. The failure classifier is updated based on the reweighted loss function.

[0012] Optionally, after inputting the first fused feature of the sample vacuum nozzle into the health classifier and outputting the state of the sample vacuum nozzle based on the health classifier, the method further includes: Based on the state of the sample vacuum nozzle and its state identifier, a loss function is constructed. The health classifier is updated based on the loss function.

[0013] Furthermore, to achieve the above objectives, the present invention also proposes a vacuum nozzle failure prediction device, the vacuum nozzle failure prediction device comprising: The data acquisition module is used to acquire preset parameter data of the target vacuum nozzle; The status output module is used to input the first fusion feature after the preset parameter data is reconstructed into the health classifier, and output the status of the target vacuum nozzle based on the health classifier; The failure cause output module is used to input the preset parameter data into the failure classifier when the status output result of the target vacuum nozzle is a failure, and output the failure cause of the target vacuum nozzle based on the failure classifier.

[0014] Furthermore, to achieve the above objectives, the present invention also proposes a vacuum nozzle failure prediction device, the device comprising: a memory, a processor, and a vacuum nozzle failure prediction program stored in the memory and executable on the processor, the vacuum nozzle failure prediction program being configured to implement the steps of the vacuum nozzle failure prediction method described above.

[0015] In addition, to achieve the above objectives, the present invention also proposes a storage medium storing a vacuum nozzle failure prediction program, wherein when the vacuum nozzle failure prediction program is executed by a processor, it implements the steps of the vacuum nozzle failure prediction method described above.

[0016] This invention discloses a method, apparatus, device, and storage medium for predicting vacuum nozzle failure. The method includes: acquiring preset parameter data of a target vacuum nozzle; inputting a first fusion feature reconstructed from the preset parameter data into a health classifier, and outputting the state of the target vacuum nozzle based on the health classifier; when the output state of the target vacuum nozzle is a failure, inputting a second fusion feature of the preset parameter data into a failure classifier, and outputting the cause of failure of the target vacuum nozzle based on the failure classifier. Unlike existing prediction methods that ignore the duration difference between the healthy state and the failed state of the vacuum nozzle, this invention first reconstructs features from the preset parameter data of the target vacuum nozzle, and then obtains the state of the target vacuum nozzle based on the reconstructed features and the health classifier to determine whether the target vacuum nozzle has failed. Therefore, this invention alleviates the negative impact of unbalanced distribution of health data and improves the accuracy of the prediction results. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of the vacuum nozzle failure prediction device for the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the first embodiment of the vacuum nozzle failure prediction method of the present invention; Figure 3 This is a flowchart illustrating the second embodiment of the vacuum nozzle failure prediction method of the present invention; Figure 4 This is a flowchart illustrating the third embodiment of the vacuum nozzle failure prediction method of the present invention. Figure 5 This is a schematic diagram of the process of the vacuum nozzle failure prediction method of the present invention; Figure 6 This is a structural block diagram of the first embodiment of the vacuum nozzle failure prediction device of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a vacuum nozzle failure prediction device for the hardware operating environment involved in the embodiments of the present invention.

[0021] like Figure 1As shown, the vacuum nozzle failure prediction device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0022] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the vacuum nozzle failure prediction device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0023] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and a vacuum nozzle failure prediction program.

[0024] exist Figure 1 In the vacuum nozzle failure prediction device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the vacuum nozzle failure prediction device of the present invention can be set in the vacuum nozzle failure prediction device. The vacuum nozzle failure prediction device calls the vacuum nozzle failure prediction program stored in the memory 1005 through the processor 1001 and executes the vacuum nozzle failure prediction method provided in the embodiment of the present invention.

[0025] This invention provides a method for predicting vacuum nozzle failure, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the vacuum nozzle failure prediction method of the present invention.

[0026] In this embodiment, the vacuum nozzle failure prediction method includes the following steps: Step S10: Obtain the preset parameter data of the target vacuum nozzle; It should be noted that the executing entity of the method in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer or personal computer, or other electronic devices capable of performing the same or similar functions. Here, the vacuum nozzle failure prediction method provided in this embodiment and the following embodiments will be specifically described using the aforementioned vacuum nozzle failure prediction device (hereinafter referred to as the failure prediction device).

[0027] It should be noted that the aforementioned target vacuum nozzle refers to the vacuum nozzle for which failure status and failure cause prediction are required. The aforementioned preset parameter data can be operational parameters of the vacuum nozzle, such as flow rate, pressure value, and angle. Therefore, this embodiment does not limit the content or number of data points involved in the preset parameter data. The preset parameter data is typically compiled and statistically analyzed by factory personnel based on data from used vacuum nozzles.

[0028] Understandably, the aforementioned preset parameter data can be collected in chronological order and arranged in chronological order for a duration of T, where T is the collection duration. All preset parameter data will be collected at each collection time point. For example, if the preset parameter data includes N head parameters such as the flow rate, pressure value, and angle of the vacuum nozzle, then at the first collection time point, all head parameters such as the flow rate, pressure value, and angle of the vacuum nozzle will be collected, and at the second collection time point, all head parameters such as the flow rate, pressure value, and angle of the vacuum nozzle will still be collected.

[0029] Step S20: Input the first fusion feature after the preset parameter data reconstruction into the health classifier, and output the status of the target vacuum nozzle based on the health classifier; Step S30: When the status output result of the target vacuum nozzle is a failure, the preset parameter data is input to the failure classifier, and the failure cause of the target vacuum nozzle is output based on the failure classifier.

[0030] It should be noted that the classifier described above can be a function or model that maps data in a database to a given category, enabling both classification and prediction of input data. In this embodiment, the health classifier can classify the target vacuum nozzle into two states: healthy and failed. A healthy state indicates that the target vacuum nozzle is still usable, while a failed state indicates that the target vacuum nozzle is about to fail and requires maintenance or replacement. Therefore, the health classifier can be used to predict whether a target vacuum nozzle has failed.

[0031] When the target vacuum nozzle's condition is classified as failure, the failure classifier can categorize the failure into three causes: normal, clogged, and damaged. The normal cause indicates the nozzle failed due to reaching its normal service life; the clogged cause indicates failure due to blockage; and the damaged cause indicates failure due to damage. Therefore, the failure classifier can be used to predict the specific cause of failure of the target vacuum nozzle.

[0032] In practical implementation, if it is necessary to determine whether the target vacuum nozzle has failed, multiple head parameters such as flow rate, pressure value and angle of the target vacuum nozzle can be obtained first. Then, after feature reconstruction of the head parameters of this type of data, the reconstructed features are input into the health classifier. Based on the output result of the health classifier, it is determined whether the target vacuum nozzle has failed. When the status output result of the target vacuum nozzle is failure, the second fused feature of the preset parameter data is input into the failure classifier. Based on the failure cause output by the failure classifier, it is determined whether the target vacuum nozzle has failed normally, failed due to blockage or failed due to damage.

[0033] This embodiment acquires preset parameter data of the target vacuum nozzle; inputs the first fusion feature reconstructed from the preset parameter data into a health classifier, and outputs the state of the target vacuum nozzle based on the health classifier; when the output state of the target vacuum nozzle is a failure, the second fusion feature of the preset parameter data is input into a failure classifier, and the failure reason of the target vacuum nozzle is output based on the failure classifier. Unlike existing prediction methods that ignore the duration difference between the healthy state and the failure state of the vacuum nozzle, this embodiment first reconstructs the features of the preset parameter data of the target vacuum nozzle, and then obtains the state of the target vacuum nozzle based on the reconstructed features and the health classifier to determine whether the target vacuum nozzle has failed. Therefore, this embodiment alleviates the negative impact of unbalanced distribution of health data and improves the accuracy of prediction results.

[0034] Reference Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the vacuum nozzle failure prediction method of the present invention.

[0035] Understandably, the aforementioned health classifier and failure classifier need to be trained on a large number of sample vacuum nozzles before predicting the target vacuum nozzle, generating classification models for the health classifier and failure classifier. Therefore, based on the above... Figure 2 The illustrated embodiment presents a second embodiment of the vacuum nozzle failure prediction method of the present invention.

[0036] It is easy to understand that, in order to facilitate the subsequent classification of target vacuum nozzles, the features of preset parameters of the target vacuum nozzles can be extracted. Then, the target vacuum nozzles can be classified by analyzing the statistical characteristics of the features in the data distribution. Commonly used methods include time-domain feature extraction, frequency-domain feature extraction, and time-frequency feature extraction. However, time-domain feature extraction and frequency-domain feature extraction can only calculate information from one domain, which easily discards important features with high resolution. Therefore, in this embodiment, the preset parameter data can be converted into a time-frequency image by using a time-frequency analysis method, and then the time-frequency features corresponding to the preset parameter data can be extracted by using neural networks or algorithms such as ResNet18 backbone network.

[0037] Therefore, in this embodiment, before step S10, the following steps are also included: Step S01: Obtain the time-frequency characteristics corresponding to the preset parameter data of the sample vacuum nozzle; It should be noted that the process of obtaining the time-frequency features corresponding to the preset parameter data of the sample vacuum nozzle can be as follows: the preset parameter data of the sample vacuum nozzle is divided into time-series segment data; the time-series segment data is transformed into a time-frequency image; and the time-frequency image is used to extract features from the time-frequency image through the ResNet18 backbone network to obtain the time-frequency features corresponding to the preset parameter data of the sample vacuum nozzle.

[0038] This embodiment can use Short Time Fourier Transform (STFT) to convert preset parameter data into a time-frequency image. Of course, other commonly used time-frequency analysis methods, such as Morlet wavelet (MW) and filter-based Hilbert transform (FHT), can also be used. This embodiment does not limit the specific time-frequency analysis method used.

[0039] In practical applications, this embodiment first divides the sample vacuum nozzle into time segments using the sliding pane method, based on preset parameters. The pane size is W, and the step size is S. The formula is shown below:

[0040] Indicates the i-th header parameter From the jth Data bits until All data bits, and denoted as .

[0041] Then the timing segment The two-dimensional time-frequency image is obtained by transforming the image using the Short Time Fourier Transform (STFT). The formula is shown below, where f is the sampling frequency.

[0042] ; Finally, after assembling the obtained time-frequency images into a time-frequency image set according to the corresponding header parameters, the time-frequency image set for the i-th header parameter is... Using the ResNet18 backbone network Feature extraction is performed to obtain the corresponding time-frequency feature set. The formula is as follows: ; Understandably, to facilitate feature reconstruction and verify the accuracy of the classifier's output, the sample vacuum nozzles can be marked first. Therefore, further, before step S01, the following steps are also included: Step S001: Obtain the failure data and remaining usage time of the sample vacuum nozzle; Step S002: Construct a status identifier for the sample vacuum nozzle based on the remaining usage time; Step S003: When the status of the sample vacuum nozzle is marked as failed, construct a failure identifier for the sample vacuum nozzle based on the failure data.

[0043] It should be noted that the aforementioned failure data can refer to the types of failures in the sample vacuum data, including: normal failure, nozzle blockage failure, and nozzle damage failure. The remaining usage time can be the remaining usable time of the sample vacuum data. The aforementioned status indicator can be a manual identifier added to the sample vacuum nozzle by data acquisition personnel during data collection. This identifier can be compared with the output of the health classifier to assess the correctness of the health classifier's output and / or calibrate its output. Status indicators include: healthy status and failure status. Considering the convenience of factory maintenance, this status indicator can be based on the relationship between the remaining usage time of the sample vacuum nozzle and the time approaching failure. This time approaching failure can be manually set by relevant personnel based on experience or estimated from a large number of samples. In practical applications, it is assumed that... This represents the remaining usage time of the sample vacuum nozzle. The time approaching failure of the sample vacuum nozzle is represented by the time approaching failure of the sample vacuum nozzle. < If the vacuum nozzle of the sample is in a faulty state, it is marked as faulty; otherwise, it is in a healthy state.

[0044] Furthermore, when the status of the sample vacuum nozzle is marked as a failure, a failure identifier for the sample vacuum nozzle is constructed based on the aforementioned failure data (i.e., failure types). The failure identifiers include: normal, blocked, and damaged.

[0045] Step S02: Based on the time-frequency features, obtain the second fusion feature of the sample vacuum nozzle; It should be noted that, after obtaining the above time-frequency feature set... Then, the time-frequency feature set needs to be configured according to the type of parameters. The process involves merging all the data corresponding to a single header parameter. The preset parameter formula is shown below: ; Here, n corresponds to the number of header parameters, thus the final result is a feature set that summarizes the features of all header parameters. Feature set This is the second fusion feature mentioned above.

[0046] Step S03: Input the second fusion feature of the sample vacuum nozzle into the failure classifier, and output the failure cause of the sample vacuum nozzle based on the failure classifier; It should be noted that, based on the second fusion feature, a multi-layer fully connected network is used to construct the failure classifier. Output categories are used This means, that is: ; In the output results, This indicates that the failure reason of the output vacuum nozzle is classified as normal. This indicates that the failure of the output vacuum nozzle is due to blockage. This indicates that the failure of the output vacuum nozzle is due to damage.

[0047] Step S04: Reconstruct the second fusion feature of the sample vacuum nozzle to obtain the first fusion feature of the sample vacuum nozzle; After obtaining the second fusion feature, in order to eliminate the data analysis bias caused by the imbalance in the distribution of the health duration and failure duration of the vacuum nozzle, the second fusion feature can be reconstructed before inputting the data into the health classifier for training. Then, the health classifier is constructed based on the reconstructed feature, which is the first fusion feature.

[0048] Furthermore, the steps for feature reconstruction can be: Step S041: Obtain the mean and variance of the second fusion feature of the sample vacuum nozzle corresponding to the status identifier; Step S042: Smooth the mean and variance using a Gaussian kernel function; Step S043: Based on the smoothed mean and variance, obtain the first fusion feature of the sample vacuum nozzle.

[0049] It should be noted that before determining the mean and variance of the features, the first fused feature can be classified based on the status indicators of the sample vacuum nozzles, and then the second fused feature can be classified. Features of health-like labels Calculate their mean values ​​respectively. and variance Among them, when the sample vacuum nozzle When the sample is in a healthy state, the vacuum nozzle of the sample can be identified as such. When the mean and variance are in a healthy state, it can be used to represent the state of the sample. The formulas for determining the mean and variance are as follows: ;

[0050] Furthermore, to facilitate subsequent data processing, this embodiment can use a Gaussian kernel function. For statistical characteristics, mean and variance Smoothing is performed using the following formula: ; ; Then for features Reconstruction is performed to obtain features. .

[0051] ; This yields a new set of features. , This is the first fusion feature mentioned above.

[0052] It should be noted that the first fusion feature of the target vacuum nozzle in the embodiment is also obtained from the second fusion feature of the target vacuum nozzle through the above steps.

[0053] Step S05: Input the first fusion feature of the sample vacuum nozzle into the health classifier, and output the status of the sample vacuum nozzle based on the health classifier.

[0054] It should be noted that, after obtaining the second fusion feature, this embodiment constructs the aforementioned health classifier based on the second fusion feature through a multi-layer fully connected network. The output category is then used. This means, that is:

[0055] In the output results, This indicates that the output vacuum nozzle is in a healthy state. This indicates that the output vacuum nozzle is in a failed state.

[0056] Furthermore, the role of the fully connected network described above is to cascade multiple transformations to achieve the mapping from input to output. Each layer of the neural network is a linear transformation. The transformation result of the previous layer is processed by an activation function and passed to the next layer, forming a multi-layer fully connected neural network. The purpose of the activation function is to perform non-linear operations on the results. In the processing of the fully connected neural network, if the decision rule for data classification determines that the score of a certain input data x in the i-th class is higher than the score in the j-th class, then it is assigned to the i-th class. For example, in the health classifier, if the input sample vacuum nozzle or target vacuum nozzle scores higher in the health class than in the failure class, then the sample vacuum nozzle or target vacuum nozzle is assigned to the health class. The failure classifier processes data in the same way.

[0057] Furthermore, it's understandable that although the input fusion features of the health classifier and the failure classifier are different, and when predicting the target vacuum nozzle using the classifiers, the data results of the health classifier are considered first, and the output results of the failure classifier are considered when the target vacuum nozzle is in a failed state to determine the specific cause of the failure, the amount of input data for the health classifier and the failure classifier is the same. This is because if the failure classifier has very few basic data samples available for analysis, it will easily lead to low accuracy in the data prediction results. In addition, steps S00-S04 are also required when predicting the state and failure cause of the target vacuum nozzle.

[0058] This embodiment acquires the failure data and remaining usage time of a sample vacuum nozzle; constructs a status identifier for the sample vacuum nozzle based on the remaining usage time; when the status identifier of the sample vacuum nozzle is in a failed state, a failure identifier for the sample vacuum nozzle is constructed based on the failure data; the preset parameter data of the sample vacuum nozzle is divided to obtain time-series segment data; the time-series segment data is transformed into a time-frequency image; features are extracted from the time-frequency image using algorithms such as the ResNet18 backbone network to obtain time-frequency features corresponding to the preset parameter data of the sample vacuum nozzle; a second fusion feature of the sample vacuum nozzle is obtained based on the time-frequency features; the second fusion feature of the sample vacuum nozzle is input into a failure classifier, and the failure cause of the sample vacuum nozzle is output based on the failure classifier; the second fusion feature of the sample vacuum nozzle is reconstructed to obtain a first fusion feature of the sample vacuum nozzle; the first fusion feature of the sample vacuum nozzle is input into a health classifier, and the status of the sample vacuum nozzle is output based on the health classifier.

[0059] Therefore, in this embodiment, before predicting the target vacuum nozzle using the health classifier and failure classifier, the health classifier and failure classifier are constructed and trained using the failure data and remaining usage time of the sample vacuum nozzle. A status label is constructed based on the remaining usage time, and feature reconstruction is performed in the feature space to learn the intrinsic relationship between the labels. This mitigates the negative impact of the imbalance in the distribution of health data and improves the reliability and accuracy of the output results of the health classifier and failure classifier.

[0060] Reference Figure 4 , Figure 4 This is a flowchart illustrating the third embodiment of the vacuum nozzle failure prediction method of the present invention, based on the above. Figure 2 Alternatively, as shown in embodiment 3, a third embodiment of the vacuum nozzle failure prediction method of the present invention is proposed. Figure 4 Based on Figure 1 The embodiments shown are examples of the proposed embodiments.

[0061] Understandably, after constructing the health classifier and the failure classifier, it is not only necessary to continuously train them based on a large number of samples, but also to calculate the necessary evaluation metrics based on the output prediction results to evaluate the performance of the classifier, and to correct or update the classifier network, thereby improving the accuracy of the output of the health classifier and failure classifier models.

[0062] Furthermore, as one possible implementation, this embodiment includes the following step after step S03: Step S031: Based on the failure cause and failure identifier of the sample vacuum nozzle, construct a reweighted loss function; It's important to note that a loss function is a function used to measure the degree of inconsistency between the predicted values ​​and the true values ​​of a given classifier. Its output is typically a non-negative real value. The output of the loss function can be used as a feedback signal to adjust the classifier parameters, reducing the loss value corresponding to the current instance, thereby improving the classifier's predictive performance, generalization ability, and robustness against interference.

[0063] Understandably, since vacuum nozzle blockage is far more common than damage among the causes of vacuum nozzle failure, the distribution of failure cause data is severely imbalanced. To mitigate the negative impact of this imbalance, a reweighted loss function can be constructed for the failure classifier to measure the accuracy of the results.

[0064] Step S032: Update the failure classifier model based on the reweighted loss function.

[0065] The reweighted loss function can also be the cross-entropy loss function: ; The failure indicator for the sample vacuum nozzle is now normal. This indicates the probability that the network output is of the normal class. The training set contains the number of occurrences of the normal class. The failure indicator for the sample vacuum nozzle is blockage. This indicates the probability that the network output is of the congested type. To train the frequency of occurrence of the blocking class; The failure indicator for the sample vacuum nozzle is "damaged". This indicates the probability that the network output is corrupted. This is used to measure the frequency of occurrences of the damaged class in the training set. When the failure flag of the sample vacuum nozzle is set to normal, It is 1 if it is 1, otherwise it is 0. and Similarly.

[0066] Similarly, as one possible implementation, this embodiment further includes the following after step S05: Step S051: Construct a loss function based on the state of the sample vacuum nozzle and the state identifier of the sample vacuum nozzle; Step S052: Update the health classifier based on the loss function.

[0067] The loss function for the health classifier can be the cross-entropy loss function, as shown in the following formula: ; in, The status of the sample vacuum nozzle is indicated as healthy. This represents the probability that the network output is classified as healthy. This indicates that the sample vacuum nozzle is in a failed state. This represents the probability that the network output is in the failure class. When the sample is in a healthy state, It is 1 if it is 1, otherwise it is 0. Similarly.

[0068] Furthermore, the overall loss function is: ; That is, each time the classifier network is updated, both the healthy classifier and the failed classifier models are updated together.

[0069] It should be noted that, similarly, after predicting the target vacuum nozzle, the predicted result of the target vacuum nozzle can be compared with the actual state and / or cause of failure, and the health classifier and failure classifier can be updated through loss function and / or reweighted loss function.

[0070] After obtaining the loss function, this embodiment can use the output value of the loss function as a feedback signal to adjust the classifier parameters, thereby improving the classifier's prediction performance on training samples.

[0071] This embodiment constructs a reweighted loss function based on the failure causes and failure identifiers of sample vacuum nozzles; the failure classifier is then updated based on this reweighted loss function. Similarly, a loss function is constructed based on the state and state identifier of the sample vacuum nozzles; the healthy classifier is updated based on this loss function. Therefore, this embodiment not only mitigates the negative impact of imbalanced failure cause data distribution through reweighting, but also improves the predictive performance, generalization ability, and robustness of the classifier models by updating the healthy and failure classifiers through loss functions and / or reweighted loss functions.

[0072] In addition, for ease of understanding reference Figure 5 This explanation does not limit the scope of this solution. Figure 5 This is a schematic diagram of the process of the vacuum nozzle failure prediction method of the present invention. Figure 5 This invention demonstrates the process of predicting the status of a target vacuum nozzle. First, it generates time-frequency images corresponding to each parameter (parameter 1 to parameter N) based on preset parameter data of the target vacuum nozzle. Then, it inputs these time-frequency images into a ResNet18 backbone network to extract the time-frequency features corresponding to each parameter. These time-frequency features are then fused into a second fused feature. Next, the second fused feature is reconstructed to obtain a first fused feature. Finally, the first fused feature is input into a health classifier, which outputs the status of the target vacuum nozzle. When the target vacuum nozzle's status output is a failure, the preset parameter data of the target vacuum nozzle is input into a failure classifier, which outputs the cause of the failure. Furthermore, this invention simultaneously updates the health classifier using a loss function and the failure classifier using a reweighted loss function.

[0073] Furthermore, this embodiment of the invention also proposes a storage medium storing a vacuum nozzle failure prediction program, which, when executed by a processor, implements the steps of the vacuum nozzle failure prediction method described above.

[0074] refer to Figure 6 , Figure 6 This is a structural block diagram of the first embodiment of the vacuum nozzle failure prediction device of the present invention.

[0075] like Figure 6 As shown, the vacuum nozzle failure prediction device proposed in this embodiment of the invention includes: The data acquisition module 601 is used to acquire preset parameter data of the target vacuum nozzle; The status output module 602 is used to input the first fusion feature after the preset parameter data is reconstructed into the health classifier, and output the status of the target vacuum nozzle based on the health classifier; The failure cause output module 603 is used to input the preset parameter data to the failure classifier when the status output result of the target vacuum nozzle is a failure, and output the failure cause of the target vacuum nozzle based on the failure classifier.

[0076] This embodiment acquires preset parameter data of the target vacuum nozzle; inputs the first fusion feature reconstructed from the preset parameter data into a health classifier, and outputs the state of the target vacuum nozzle based on the health classifier; when the output state of the target vacuum nozzle is a failure, the second fusion feature of the preset parameter data is input into a failure classifier, and the failure reason of the target vacuum nozzle is output based on the failure classifier. Unlike existing prediction methods that ignore the duration difference between the healthy state and the failure state of the vacuum nozzle, this embodiment first reconstructs the features of the preset parameter data of the target vacuum nozzle, and then obtains the state of the target vacuum nozzle based on the reconstructed features and the health classifier to determine whether the target vacuum nozzle has failed. Therefore, this embodiment alleviates the negative impact of unbalanced distribution of health data and improves the accuracy of prediction results.

[0077] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0078] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0079] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0080] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for predicting vacuum nozzle failure, characterized in that, The vacuum nozzle failure prediction method includes: Obtain the preset parameter data of the target vacuum nozzle; The first fusion feature after reconstructing the preset parameter data is input into the health classifier, and the status of the target vacuum nozzle is output based on the health classifier; When the status output result of the target vacuum nozzle is a failure, the preset parameter data is input to the failure classifier, and the failure cause of the target vacuum nozzle is output based on the failure classifier. Before acquiring the preset parameter data of the target vacuum nozzle, the process also includes: Obtain the time-frequency characteristics corresponding to the preset parameter data of the sample vacuum nozzle; Based on the time-frequency characteristics, the second fusion feature of the sample vacuum nozzle is obtained; The second fusion feature of the sample vacuum nozzle is input into the failure classifier, and the failure cause of the sample vacuum nozzle is output based on the failure classifier; The second fusion feature of the sample vacuum nozzle is reconstructed to obtain the first fusion feature of the sample vacuum nozzle; The first fusion feature of the sample vacuum nozzle is input into the health classifier, and the status of the sample vacuum nozzle is output based on the health classifier; Before acquiring the time-frequency characteristics corresponding to the preset parameter data of the sample vacuum nozzle, the method further includes: Obtain failure data and remaining usage time of the sample vacuum nozzle; The status identifier of the sample vacuum nozzle is constructed based on the remaining usage time; When the status of the sample vacuum nozzle is identified as a failed state, a failure identifier for the sample vacuum nozzle is constructed based on the failure data. The step of reconstructing the second fusion feature of the sample vacuum nozzle to obtain the first fusion feature of the sample vacuum nozzle includes: Obtain the mean and variance of the second fusion feature of the sample vacuum nozzle corresponding to the state identifier; The mean and variance are smoothed using a Gaussian kernel function; Based on the smoothed mean and variance, the first fusion feature of the sample vacuum nozzle is obtained.

2. The vacuum nozzle failure prediction method as described in claim 1, characterized in that, The step of obtaining the time-frequency characteristics corresponding to the preset parameter data of the sample vacuum nozzle includes: The preset parameter data of the sample vacuum nozzle is divided to obtain time-series segment data; The time-series data segment is transformed into a time-frequency image; The time-frequency image is used to extract features through the ResNet18 backbone network to obtain time-frequency features corresponding to the preset parameter data of the sample vacuum nozzle.

3. The vacuum nozzle failure prediction method as described in claim 1, characterized in that, After inputting the second fusion feature of the sample vacuum nozzle into the failure classifier and outputting the failure cause of the sample vacuum nozzle based on the failure classifier, the method further includes: Based on the failure causes and failure identifiers of the sample vacuum nozzles, a reweighted loss function is constructed. The failure classifier is updated based on the reweighted loss function.

4. The vacuum nozzle failure prediction method as described in claim 1, characterized in that, After inputting the first fusion feature of the sample vacuum nozzle into the health classifier and outputting the state of the sample vacuum nozzle based on the health classifier, the method further includes: Based on the state of the sample vacuum nozzle and its state identifier, a loss function is constructed. The health classifier is updated based on the loss function.

5. A vacuum nozzle failure prediction device, characterized in that, The vacuum nozzle failure prediction device includes: The data acquisition module is used to acquire preset parameter data of the target vacuum nozzle; The status output module is used to input the first fusion feature after the preset parameter data is reconstructed into the health classifier, and output the status of the target vacuum nozzle based on the health classifier; The failure cause output module is used to input the preset parameter data into the failure classifier when the status output result of the target vacuum nozzle is a failure, and output the failure cause of the target vacuum nozzle based on the failure classifier. The status output module is further configured to acquire time-frequency features corresponding to preset parameter data of the sample vacuum nozzle; obtain a second fusion feature of the sample vacuum nozzle based on the time-frequency features; input the second fusion feature of the sample vacuum nozzle into a failure classifier, and output the failure cause of the sample vacuum nozzle based on the failure classifier; reconstruct the second fusion feature of the sample vacuum nozzle to obtain a first fusion feature of the sample vacuum nozzle; input the first fusion feature of the sample vacuum nozzle into a health classifier, and output the status of the sample vacuum nozzle based on the health classifier; The status output module is also used to acquire the failure data and remaining usage time of the sample vacuum nozzle; construct the status identifier of the sample vacuum nozzle based on the remaining usage time; when the status identifier of the sample vacuum nozzle is in a failed state, construct the failure identifier of the sample vacuum nozzle based on the failure data; The state output module is further configured to obtain the mean and variance of the second fusion feature of the sample vacuum nozzle corresponding to the state identifier; smooth the mean and variance using a Gaussian kernel function; and obtain the first fusion feature of the sample vacuum nozzle based on the smoothed mean and variance.

6. A vacuum nozzle failure prediction device, characterized in that, The device includes: a memory, a processor, and a vacuum nozzle failure prediction program stored in the memory and executable on the processor, the vacuum nozzle failure prediction program being configured to implement the steps of the vacuum nozzle failure prediction method as described in any one of claims 1 to 4.

7. A storage medium, characterized in that, The storage medium stores a vacuum nozzle failure prediction program, which, when executed by a processor, implements the steps of the vacuum nozzle failure prediction method as described in any one of claims 1 to 4.

Images