Negative pressure loop fault early warning method, device and equipment and readable storage medium

Abstract

The invention discloses a negative pressure loop fault early warning method, device and equipment and a readable storage medium. The negative pressure loop fault early warning method comprises the steps that real-time to-be-tested negative pressure value data of a to-be-tested negative pressure suction nozzle on battery negative pressure formation equipment in the using process are acquired; and inputting the to-be-tested negative pressure value data into the trained negative pressure fault prediction model to obtain the predicted fault time of the to-be-tested negative pressure suction nozzle. According to the method and the device, the real-time to-be-tested negative pressure value data of the to-be-tested negative pressure suction nozzle is input into the trained negative pressure fault prediction model, so that the predicted fault time of the to-be-tested negative pressure suction nozzle can be obtained; the predicted fault time obtained through the trained negative pressure fault prediction model is more accurate than the fault time determined manually according to the material and use experience of the suction nozzle, so that an operator can be guided to judge the current actual state of the negative pressure loop and replace the negative pressure suction nozzle in time; the technical problems of waste caused by advanced replacement of the suction nozzle and battery damage caused by delayed replacement of the suction nozzle are solved.

Description

Technical Field

[0001] The present application relates to the technical field of negative pressure formation of batteries, and particularly to a method, device, equipment and readable storage medium for warning of negative pressure circuit faults. Background Art

[0002] Currently, in the production process of lithium batteries, negative pressure needs to be pumped and discharged during the formation stage. The active substances inside the battery are used to discharge gas under negative pressure, thereby changing the pore shape of the battery material, improving the electrochemical performance of the battery, increasing the charging and discharging speed, and extending the service life of the battery. One of the basic conditions for using negative pressure is the negative pressure nozzle. If the nozzle fails, it will cause safety problems such as overpressure, overcurrent, or air leakage, which will damage the battery.

[0003] In related technologies, current equipment or production lines regularly replace the negative pressure nozzles according to the material and usage experience of the nozzles, such as replacing them once every three months, etc. However, on the one hand, this approach may result in replacing the nozzle even when it is not damaged, causing waste. On the other hand, some nozzles will be damaged in advance, resulting in a decrease in the negative pressure effect during the production of lithium batteries, leading to quality problems in the batteries produced during this period or the need for additional rework.

[0004] Therefore, it is necessary to design a method for warning of negative pressure circuit faults to overcome the above problems. Summary of the Invention

[0005] The present application provides a method, device, equipment and readable storage medium for warning of negative pressure circuit faults, which can solve the technical problems of waste caused by replacing the nozzle in advance and damage to the battery caused by delaying the replacement of the nozzle in related technologies.

[0006] In a first aspect, an embodiment of the present application provides a method for warning of negative pressure circuit faults, and the method for warning of negative pressure circuit faults includes:

[0007] Obtain real-time data of the negative pressure value to be tested of the negative pressure nozzle to be tested during the use of the battery negative pressure formation equipment;

[0008] Input the data of the negative pressure value to be tested into the trained negative pressure fault prediction model to obtain the predicted fault time of the negative pressure nozzle to be tested.

[0009] In combination with the first aspect, in an implementation manner, the negative pressure fault prediction model is obtained by the following method:

[0010] Based on the negative pressure value data during the use of negative pressure in the source domain and the time when the negative pressure nozzle fails, as well as the negative pressure value data during the use of negative pressure in the target domain, the negative pressure fault prediction model is trained; wherein, the source domain is the battery negative pressure formation equipment that has been put into use, and the target domain is the newly produced battery negative pressure formation equipment.

[0011] In combination with the first aspect, in one embodiment, the negative pressure fault prediction model is trained based on the negative pressure value data during the negative pressure use process of the source domain, the time when the negative pressure suction nozzle fails, and the negative pressure value data during the negative pressure use process of the target domain, including:

[0012] Input the negative pressure value data during the negative pressure use process of the source domain into the initial source domain negative pressure fault prediction model to obtain a calculated fault time and the characteristics of the change in the source domain negative pressure value;

[0013] Obtain the target domain negative pressure fault prediction model based on the source domain negative pressure fault prediction model;

[0014] Input the negative pressure value data during the negative pressure use process of the target domain into the target domain negative pressure fault prediction model to obtain the characteristics of the change in the target domain negative pressure value;

[0015] Calculate the loss value according to the calculated fault time, the time when the negative pressure suction nozzle fails, the characteristics of the change in the source domain negative pressure value, and the characteristics of the change in the target domain negative pressure value, and update the parameters of the source domain negative pressure fault prediction model according to the loss value until the training ends. The latest source domain negative pressure fault prediction model is the trained negative pressure fault prediction model.

[0016] In combination with the first aspect, in one embodiment, the step of inputting the negative pressure value data during the negative pressure use process of the source domain into the initial source domain negative pressure fault prediction model to obtain a calculated fault time and the characteristics of the change in the source domain negative pressure value includes:

[0017] Extract the characteristics of the change in the negative pressure value from the negative pressure value data in the source domain negative pressure fault prediction model to obtain the characteristics of the change in the source domain negative pressure value;

[0018] Project the characteristics of the change in the source domain negative pressure value into a low-dimensional space and maximize the separation of the normal characteristics and the fault characteristics to obtain the characteristics after projection in the source domain space;

[0019] Output the calculated fault time according to the mapping relationship between the characteristics after projection in the source domain space and the fault time.

[0020] In combination with the first aspect, in one embodiment, the step of extracting the characteristics of the change in the negative pressure value from the negative pressure value data in the source domain negative pressure fault prediction model to obtain the characteristics of the change in the source domain negative pressure value includes:

[0021] In the source domain negative pressure fault prediction model, extract the local change characteristics of the negative pressure value from the negative pressure value data through a convolutional neural network, and then obtain the overall change characteristics of the negative pressure value through a self-attention mechanism network to obtain the characteristics of the change in the source domain negative pressure value.

[0022] In combination with the first aspect, in one implementation, calculating a loss value based on the calculated value of the fault time, the time when the negative pressure suction nozzle fails, the characteristics of the source domain negative pressure value change, and the characteristics of the target domain negative pressure value change, and updating the parameters of the source domain negative pressure fault prediction model according to the loss value until the training is completed, and obtaining the latest source domain negative pressure fault prediction model, which is the trained negative pressure fault prediction model, includes:

[0023] Calculating a cross-entropy loss value based on the calculated value of the fault time and the time when the negative pressure suction nozzle fails;

[0024] Calculating a center distance loss value based on the time when the negative pressure suction nozzle fails and the characteristics after the source domain space projection; wherein, the characteristics after the source domain space projection are obtained by projecting the characteristics of the source domain negative pressure value change into a low-dimensional space and maximizing the separation of normal characteristics and fault characteristics;

[0025] Calculating a deep alignment loss value based on the characteristics after the source domain space projection and the characteristics after the target domain space projection; wherein, the characteristics after the target domain space projection are obtained by projecting the characteristics of the target domain negative pressure value change into a low-dimensional space and maximizing the separation of normal characteristics and fault characteristics;

[0026] Updating the parameters of the source domain negative pressure fault prediction model according to the cross-entropy loss value, the center distance loss value, and the calculated deep alignment loss value until the training is completed, and obtaining the latest source domain negative pressure fault prediction model, which is the trained negative pressure fault prediction model.

[0027] In combination with the first aspect, in one implementation, obtaining the target domain negative pressure fault prediction model based on the source domain negative pressure fault prediction model includes:

[0028] Reading the model parameters of the source domain negative pressure fault prediction model and writing the model parameters into the model of the target domain to obtain the target domain negative pressure fault prediction model.

[0029] In the second aspect, an embodiment of the present application provides a negative pressure circuit fault warning device. The negative pressure circuit fault warning device includes: an acquisition module, where the acquisition module is used to acquire real-time data of the negative pressure value to be tested of the negative pressure suction nozzle to be tested during the use process of the battery negative pressure formation equipment; a prediction module, where the prediction module is used to input the data of the negative pressure value to be tested into the trained negative pressure fault prediction model to obtain the predicted fault time of the negative pressure suction nozzle to be tested.

[0030] In the third aspect, an embodiment of the present application provides a battery negative pressure formation equipment. The battery negative pressure formation equipment includes a processor, a memory, and a negative pressure circuit fault warning program stored on the memory and executable by the processor. When the negative pressure circuit fault warning program is executed by the processor, the steps of the above-mentioned negative pressure circuit fault warning method are implemented.

[0031] Fourthly, an embodiment of the present application provides a computer-readable storage medium, on which a negative pressure loop fault warning program is stored. When the negative pressure loop fault warning program is executed by a processor, the steps of the above-mentioned negative pressure loop fault warning method are realized.

[0032] The beneficial effects brought by the technical solution provided by the embodiment of the present application include:

[0033] By inputting the real-time data of the negative pressure value to be tested of the negative pressure suction nozzle during use into the trained negative pressure fault prediction model, the predicted fault time of the negative pressure suction nozzle to be tested can be obtained, and the predicted fault time obtained by the trained negative pressure fault prediction model is more accurate than the fault time determined manually according to the material and usage experience of the suction nozzle, which can guide the operator to judge the current actual state of the negative pressure loop and replace the negative pressure suction nozzle in time, solving the technical problems of waste caused by early replacement of the suction nozzle and damage to the battery caused by delayed replacement of the suction nozzle in the related art. Description of the Drawings

[0034] Figure 1 is a schematic flowchart of a negative pressure loop fault warning method provided by an embodiment of the present application;

[0035] Figure 2 is a schematic structural diagram of a negative pressure fault prediction model provided by an embodiment of the present application;

[0036] Figure 3 is a schematic diagram of the negative pressure value change feature extraction process provided by an embodiment of the present application;

[0037] Figure 4 is a schematic diagram of the feature space projection process provided by an embodiment of the present application;

[0038] Figure 5 is a schematic flowchart of offline training and online diagnosis provided by an embodiment of the present application;

[0039] Figure 6 is a schematic diagram of the trend of the accuracy rate of the test set provided by an embodiment of the present application;

[0040] Figure 7 is a schematic diagram of the trend of the loss value provided by an embodiment of the present application. Detailed Embodiments

[0041] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0042] First, some technical terms in this application are explained to facilitate the understanding of this application by those skilled in the art.

[0043] Source domain: The area where the knowledge of negative pressure change is accumulated, representing the battery negative pressure forming equipment or production line that has been put into use.

[0044] Target domain: Representing the newly produced battery negative pressure forming equipment or production line.

[0045] To make the purpose, technical solution and advantages of this application clearer, the following will further describe the embodiments of this application in detail in conjunction with the accompanying drawings.

[0046] In a first aspect, an embodiment of this application provides a method for warning of negative pressure circuit faults.

[0047] In one embodiment, referring to Figure 1 , Figure 1 is a schematic flowchart of the first embodiment of the negative pressure circuit fault warning method of this application. As Figure 1 shown, the negative pressure circuit fault warning method includes:

[0048] S1: Obtain the real-time data of the negative pressure value to be tested of the negative pressure suction nozzle on the battery negative pressure forming equipment during use.

[0049] S2: Input the data of the negative pressure value to be tested into the trained negative pressure fault prediction model to obtain the predicted fault time of the negative pressure suction nozzle to be tested.

[0050] Among them, the battery negative pressure forming equipment can be newly produced equipment or production line, or equipment or production line that has been put into use. The negative pressure fault prediction model can be installed in the newly produced equipment or the equipment that has been put into use to predict the predicted fault time of the negative pressure suction nozzle. Of course, the battery negative pressure forming equipment in this embodiment is preferably newly produced equipment, that is, preferably place the trained negative pressure fault prediction model in the newly produced battery negative pressure forming equipment for use.

[0051] In this embodiment, by inputting the real-time negative pressure value data of the negative pressure nozzle to be tested during use into the trained negative pressure fault prediction model, the predicted fault time of the negative pressure nozzle to be tested can be directly obtained. Moreover, the predicted fault time obtained through the trained negative pressure fault prediction model is more accurate than the fault time determined manually according to the material and usage experience of the nozzle, which can guide the operator to judge the actual state of the negative pressure circuit and replace the negative pressure nozzle in time, solving the technical problems of waste caused by early replacement of the nozzle and battery damage caused by delayed replacement of the nozzle in the related art.

[0052] Further, in one embodiment, the negative pressure fault prediction model can be obtained by the following method: based on the negative pressure value data during the use of the negative pressure in the source domain, the time when the negative pressure nozzle fails, and the negative pressure value data during the use of the negative pressure in the target domain, the negative pressure fault prediction model is trained; wherein, the source domain is the battery negative pressure forming equipment that has been put into use, and the target domain is the newly produced battery negative pressure forming equipment. This embodiment mainly collects the negative pressure value data during the use of the negative pressure from the delivered equipment or production line (source domain) and the newly delivered equipment (target domain).

[0053] In this embodiment, since the battery negative pressure forming equipment that has been put into use has been used for a period of time and has accumulated certain knowledge of negative pressure changes, a large amount of negative pressure value data and the time data when the negative pressure nozzle fails in the battery negative pressure forming equipment that has been put into use can be used to learn and identify the possibility and time of the negative pressure circuit failing, and then the negative pressure fault prediction model is trained. And because the negative pressure fault prediction model needs to be used in the newly produced battery negative pressure forming equipment, not only the data of the battery negative pressure forming equipment that has been put into use but also the data of the newly produced battery negative pressure forming equipment are combined during the training of the negative pressure fault prediction model, so that the trained negative pressure fault prediction model can be effectively used across domains in the target domain, solving the problem that the negative pressure fault prediction model is not applicable or inaccurate in the new equipment in the related art.

[0054] Of course, in other embodiments, if the negative pressure fault prediction model is to be used in the battery negative pressure forming equipment that has been put into use, the negative pressure value data of the target domain may not be required during the training of the negative pressure fault prediction model.

[0055] For example, when obtaining the negative pressure value data during the use of the negative pressure in the source domain and the time when the negative pressure nozzle fails, the negative pressure value X = {x1, x2,..., x n} during the production process can be first obtained from the equipment (source domain) at the production site, recorded once per second, the total number of negative pressure values is n, and the negative pressure values when the negative pressure circuit fails in the n negative pressure value data are E = {e1, e2,..., em}, where the number of negative pressure values is m. Set the length of each negative pressure value feature segment to l, and k time periods with faults T = {t1, t2,..., t k}.

[0056] Divide the normal negative pressure values X according to the length of each negative pressure value feature segment, and place the relative times of normal and fault occurrences into different time periods T to obtain:

[0057]

[0058] Among them, V is the sub-segment of negative pressure values segmented by length l, I is the total number of sub-segments of negative pressure values, and Y is the time period of fault occurrence corresponding to this sub-segment of negative pressure values. Among them, y i takes a value from {t1, t2,..., t k}.

[0059] By using the above method, the negative pressure value data V during the use of negative pressure in the source domain and the corresponding time Y when the negative pressure suction nozzle fails are obtained; when the negative pressure value data is v1, the corresponding failure time is y1; when the negative pressure value data is v2, the corresponding failure time is y2; when the negative pressure value data is v i , the corresponding failure time is y i .

[0060] Furthermore, in one embodiment, as shown in Figure 2 , the training of the negative pressure fault prediction model based on the negative pressure value data during the use of negative pressure in the source domain, the time when the negative pressure suction nozzle fails, and the negative pressure value data during the use of negative pressure in the target domain may include:

[0061] Step a: Input the negative pressure value data during the use of negative pressure in the source domain into the initial source domain negative pressure fault prediction model to obtain the calculated fault time and the change characteristics of the source domain negative pressure value. That is, in this embodiment, when training the model, an initial source domain negative pressure fault prediction model (i.e., a model of the relationship between negative pressure value change and fault) will be constructed first. By training and adjusting the parameters of this initial source domain negative pressure fault prediction model, the trained negative pressure fault prediction model is finally obtained. In this initial source domain negative pressure fault prediction model, the input of the model is the sub-segment of negative pressure values V, the output is the possible fault time (calculated fault time) y, and the change characteristics of the source domain negative pressure value can also be obtained through this source domain negative pressure fault prediction model. Figure 2 Among them, V source represents the sub-segment of negative pressure values in the source domain.

[0062] Step b: Obtain the target domain negative pressure fault prediction model based on the source domain negative pressure fault prediction model. In this embodiment, the target domain negative pressure fault prediction model may not need to be redesigned, and the target domain negative pressure fault prediction model can be constructed using the source domain negative pressure fault prediction model. Of course, in other embodiments, the target domain negative pressure fault prediction model may also be obtained without using the source domain negative pressure fault prediction model, such as redesigning a target domain negative pressure fault prediction model. In this embodiment, an initial source domain negative pressure fault prediction model is first constructed, and an initial target domain negative pressure fault prediction model can be obtained through this initial source domain negative pressure fault prediction model. After the parameters of the source domain negative pressure fault prediction model are updated later, the target domain negative pressure fault prediction model can also be updated according to the updated source domain negative pressure fault prediction model. Figure 2 In, Model source represents the source domain negative pressure fault prediction model, and Model target represents the target domain negative pressure fault prediction model.

[0063] Step c: Input the negative pressure value data during the use of the target domain into the target domain negative pressure fault prediction model to obtain the target domain negative pressure value change characteristics. After obtaining the target domain negative pressure fault prediction model, the target domain negative pressure value change characteristics can be obtained using the target domain negative pressure fault prediction model. Figure 2 In, V target represents the negative pressure value data (i.e., the negative pressure value sub-segment) of the target domain. Among them, steps b and c can be carried out synchronously with step a, or step by step. Synchronous execution saves more time and improves efficiency.

[0064] Step d: Calculate the loss value based on the calculated value of the fault time, the time when the negative pressure nozzle fails, the source domain negative pressure value change characteristics, and the target domain negative pressure value change characteristics, and update the parameters of the source domain negative pressure fault prediction model according to the loss value until the training ends. The latest source domain negative pressure fault prediction model obtained is the trained negative pressure fault prediction model. Among them, when calculating the loss value, it can be directly calculated using the calculated value of the fault time, the time when the negative pressure nozzle fails, the source domain negative pressure value change characteristics, and the target domain negative pressure value change characteristics, or the source domain negative pressure value change characteristics and the target domain negative pressure value change characteristics can be transformed to calculate the loss value. For example, after the source domain negative pressure value change characteristics and the target domain negative pressure value change characteristics are projected in space to obtain the projected characteristics, the projected characteristics are used to calculate the loss value.

[0065] In this embodiment, during the first time, the negative pressure value data of the source domain is input into the initial source domain negative pressure fault prediction model to obtain the calculated fault time and the change characteristics of the source domain negative pressure value. Then, the negative pressure value data of the target domain is input into the initial target domain negative pressure fault prediction model to obtain the change characteristics of the target domain negative pressure value. Next, the loss value is calculated based on the operation results of the two models, and the parameters of the source domain negative pressure fault prediction model are updated according to the loss value. Moreover, the target domain negative pressure fault prediction model is synchronously updated according to the updated source domain negative pressure fault prediction model. Then, the negative pressure value data of the source domain is input into the updated source domain negative pressure fault prediction model for training again, and the negative pressure value data of the target domain is input into the updated target domain negative pressure fault prediction model for training. The loss value is calculated again, and the source domain negative pressure fault prediction model and the target domain negative pressure fault prediction model are continuously updated; that is, steps a-d are repeated above until the training ends after convergence, that is, the training can end when the calculated loss value no longer decreases.

[0066] In this embodiment, the source domain and the target domain are respectively modeled to capture the subtle changes of the negative pressure loop in different states. By using the negative pressure value data of the source domain and the target domain to train in the two models and calculate the loss value, the model can be updated and optimized according to the loss value, so that the optimized model can be effectively used across domains from the source domain to the target domain, and the model can be put into use in new equipment or production lines at a lower cost.

[0067] Further, in one embodiment, as shown in Figures 3 to 4 inputting the negative pressure value data of the source domain during the negative pressure use process into the initial source domain negative pressure fault prediction model to obtain the calculated fault time and the change characteristics of the source domain negative pressure value includes:

[0068] Step a1: Extract the change characteristics of the negative pressure value from the negative pressure value data in the source domain negative pressure fault prediction model to obtain the change characteristics of the source domain negative pressure value.

[0069] Step a2: Project the change characteristics of the source domain negative pressure value into a low-dimensional space and maximize the separation of the normal characteristics and the fault characteristics to obtain the characteristics after projection in the source domain space.

[0070] Step a3: Output the calculated fault time according to the mapping relationship between the characteristics after projection in the source domain space and the fault time. In this embodiment, the change characteristics of the negative pressure value are first extracted from the negative pressure value data, so that the change characteristics of the negative pressure value can be refined, and the change laws of the negative pressure value during normal and faulty conditions can be found. Then, space projection is performed, which can transform the state where the low-dimensional characteristics are inseparable into a separable state, which is more conducive to the convergence of the function fitting in step a3.

[0071] In this embodiment, both the source domain negative pressure fault prediction model and the target domain negative pressure fault prediction model mainly include three parts, namely, the negative pressure change feature extraction module, the feature space projection module, and the function fitting module. Among them, the negative pressure change feature extraction module can extract the local and / or global change features of the negative pressure value; the feature space projection module can be composed of a multi-layer fully connected neural network, which is responsible for projecting the source domain negative pressure value change features (i.e., Feature global ) into a low-dimensional space and maximizing the separation of normal features and fault features. Refer to Figure 4 as shown, which is the feature space projection process. Among them, FC represents a certain layer of fully connected neural network, g represents the number of fully connected neural networks, and Feature final represents the feature after space projection. The function fitting module is composed of a single layer of fully connected neural network, which is responsible for fitting the internal relationship between Feature final and the fault time y, that is, the calculated value of the fault time is:

[0072]

[0073] where p output represents the fitting parameter, b output represents the fitting bias term parameter, represents matrix multiplication.

[0074] Furthermore, in one embodiment, extracting the negative pressure value change features from the negative pressure value data in the source domain negative pressure fault prediction model to obtain the source domain negative pressure value change features includes: in the source domain negative pressure fault prediction model, extracting the local change features of the negative pressure value from the negative pressure value data through a convolutional neural network, and then obtaining the extraction of all change features of the negative pressure value through a self-attention mechanism network to obtain the source domain negative pressure value change features.

[0075] Refer to Figure 3 as shown for the extraction process of the negative pressure value change features. In this embodiment, the negative pressure change feature extraction is composed of a multi-layer convolutional neural network and a self-attention mechanism network. Among them, the convolutional neural network is responsible for extracting the local change features of the negative pressure value, and the self-attention mechanism network is responsible for extracting the all change features of the negative pressure value. Feature local in the figure represents the local change feature of the negative pressure value obtained through the convolutional neural network, Feature global represents the global change feature of the negative pressure value obtained through the self-attention mechanism network, and Q, K, and M represent the correlation projections between the local feature and the global feature. Assume that the convolutional kernel size of the convolutional neural network is 1×C, the number of channels is H, and the convolutional kernel moves one negative pressure value each time. Then the total number of movements S = l - C + 1, and the local feature Feature local is:[[]] ​

[0076]

[0077] Among them, p c represents the convolutional kernel parameter, which is self-learned during the model training process to capture the best local features; v is the negative pressure value sub-segment. Subsequently, Feature local is projected onto the Q, K, and M vector sets respectively. First, matrix multiplication is performed between the Q and K vectors to obtain the self-attention matrix. The self-attention matrix is multiplied by the M vector to obtain the feature correlation and importance representation, and finally, spatial projection is performed to obtain Feature global :

[0078]

[0079] Among them, represents matrix multiplication.

[0080] In this embodiment, the convolutional neural network can extract the local change features of the negative pressure value from the negative pressure value data. Furthermore, through the self-attention mechanism network, the overall change features of the negative pressure value can be obtained, and the change features of the source domain negative pressure value can be obtained. Thanks to the design of the local and global feature capture structures for negative pressure changes in the model, the corresponding features can be effectively extracted without manual intervention, which is different from the current practice that relies on the operator's experience.

[0081] Furthermore, in one embodiment, calculating the loss value according to the calculated value of the fault time, the time when the negative pressure nozzle fails, the change features of the source domain negative pressure value, and the change features of the target domain negative pressure value, and updating the parameters of the source domain negative pressure fault prediction model according to the loss value until the training ends, and obtaining the latest source domain negative pressure fault prediction model, which is the trained negative pressure fault prediction model, may include:

[0082] Step d1: Calculate the cross-entropy loss value according to the calculated value of the fault time and the time when the negative pressure nozzle fails.

[0083] Step d2: Calculate the center distance loss value according to the time when the negative pressure nozzle fails and the features after projection in the source domain; among them, the features after projection in the source domain are obtained by projecting the change features of the source domain negative pressure value into a low-dimensional space and maximizing the separation between normal features and fault features. The features after projection in the source domain can be obtained in the source domain negative pressure fault prediction model.

[0084] Step d3: Calculate the deep alignment loss value according to the features after projection in the source domain and the features after projection in the target domain; among them, the features after projection in the target domain are obtained by projecting the change features of the target domain negative pressure value into a low-dimensional space and maximizing the separation between normal features and fault features. The features after projection in the target domain can be obtained in the target domain negative pressure fault prediction model.

[0085] Step d4: Update the parameters of the source domain negative pressure fault prediction model according to the cross-entropy loss value, the center distance loss value, and the calculated depth alignment loss value until the training is completed. The latest source domain negative pressure fault prediction model obtained is the trained negative pressure fault prediction model.

[0086] In this embodiment, the loss value includes a total of three sub-loss values, namely the cross-entropy loss value L cross_entropy , the center distance loss value L center_distance , and the depth alignment loss value L depth_alignment , which are respectively:

[0087] Cross-entropy loss value:

[0088]

[0089] where y i is the true label (i.e., the value in the time Y when the negative pressure nozzle corresponding to the negative pressure value segment of the source domain obtained at the beginning fails), is the calculated fault time value obtained through the source domain negative pressure fault prediction model.

[0090] Center distance loss value:

[0092]

[0093] where r1 and r2 are two constraint boundaries, represents the center of the fault time of the y i th negative pressure value segment, represents L2 regularization, k is the number of time periods with faults, and w, z ∈ {1,..., k}.

[0094] Depth alignment loss value:

[0095]

[0096] where ω represents the number of Feature final , U and D are the diagonalized matrices of the covariance matrices of Feature final in the source domain and the target domain respectively, σ is the eigenvalue of the diagonalized matrix of U, and μ is the eigenvalue of the diagonalized matrix of D. The final loss function is:

[0097] Loss = L cross_entropy + α·L center_distance + β·L depth_alignment

[0098] Among them, α and β are loss trade-off parameters. According to this loss function, the total loss value can be calculated.

[0099] In this embodiment, the cross-entropy loss value, the center distance loss value, and the depth alignment loss value are calculated, and these three loss values are used to optimize the model. Among them, the cross-entropy loss value can guide the model parameters to move in the direction of the correct y value; the center distance loss value can separate the final features of each t type in the fault time period as far as possible. Simply put, taking each t as a class, the feature distance between classes is maximized. In this way, on the one hand, it makes the subsequent function fitting easier to converge, and on the other hand, it makes the model more robust; the depth alignment loss value can solve the alignment problem of the negative pressure change characteristics between the source domain and the target domain. Since there is no actual y label data in the target domain, by aligning the features of the two domains, the target domain can use the labels of the source domain for learning.

[0100] Further, in one embodiment, obtaining the target domain negative pressure fault prediction model based on the source domain negative pressure fault prediction model may include: reading the model parameters of the source domain negative pressure fault prediction model and writing the model parameters into the model of the target domain to obtain the target domain negative pressure fault prediction model. That is, in this embodiment, the target domain negative pressure fault prediction model may have the same structure as the source domain negative pressure fault prediction model, and the parameters are shared between the two models. After obtaining the source domain negative pressure fault prediction model, the target domain negative pressure fault prediction model can be directly copied.

[0101] See Figure 5 As shown, in the present application, the stage of training the model can be the offline training stage. In the offline training stage, after the initial model is established, the negative pressure value sub-segments of the source domain and the target domain are used for training and learning. In the model, the negative pressure change characteristics can be obtained from the negative pressure value, feature space projection and feature numerical fitting are performed to obtain the calculated fault time value. After the offline training is completed, a trained negative pressure fault prediction model will be obtained; then, the known negative pressure value data can be used to test the trained negative pressure fault prediction model to test the prediction accuracy of the model; finally, the online diagnosis stage can be entered, and the trained model is placed in the device, and the real-time negative pressure value data is obtained and input into the trained negative pressure fault prediction model to obtain the predicted fault time of the negative pressure suction nozzle to be tested, so as to guide the operator to perform equipment maintenance work.

[0102] A specific embodiment is given below:

[0103] Step 1: Collect the negative pressure value data during the use of negative pressure from the delivered equipment or production line (source domain) and the new equipment to be delivered (target domain).

[0104] We obtained the negative pressure value data during the production process from the sites of two actual production lines, labeled as A1 and A2 respectively. Set A1 as the source domain, which contains 1079 segments of normal production data and 261 segments of abnormal negative pressure data. Set A2 as the target domain, which contains 30 segments of normal production data and 69 segments of abnormal negative pressure data. Set the length of each negative pressure value feature segment as l = 120, and perform negative pressure value sub-segment cutting for k = 10 faulty time periods. Finally, the number of negative pressure value sub-segment training sets for A1 is 1.44 million, the number of validation sets is 0.16 million, and the number of test sets is 0.4 million. For A2, the number of training sets is 12,000, the number of test sets is 4,000, and the number of online test sets is 2,000.

[0105] Step 2: Model the negative pressure values of the source domain and the target domain respectively to capture the subtle changes of the negative pressure loop in different states.

[0106] Create the model and loss function according to the aforementioned principle description, which are the source domain negative pressure fault prediction model Model source and the target domain negative pressure fault prediction model Model target . The trade-off parameters α of the loss function are 0.5 and β is 2. According to the process of the offline training stage, input the training sets and validation sets of A1 and the training set data of A2 into the model for 400 rounds of training, with the input quantity of 320 each time. Perform annealing operation when half of the training is completed. Use the test sets of A1 and A2 to prove the prediction accuracy of the model, as Figure 6 shown. The prediction accuracy of the A1 test set is 97.01%, and the prediction accuracy of the A2 test set is 96.02% (the test trend curve of A1 is above that of A2). As Figure 7 shown, the loss value also continuously decreases with the progress of training, proving that the model can effectively identify the subtle laws of negative pressure changes and gradually converge. Finally, the loss value of the training set is 0.6351 and that of the validation set is 0.726.

[0107] Step 3: Put the trained model into actual production environment for use.

[0108] Use the 20,000 online test set negative pressure value sub-segments reserved from A2 as the simulation verification of the online production environment. According to the description of the online diagnosis stage, input the online test data into the model, and the obtained accuracy is 96.1%, proving its use value in the actual production environment.

[0109] Based on the technical defects in the related technology, this application invented a negative pressure loop fault warning method. By modeling the negative pressure changes in the production process, it learns from the negative pressure change characteristics and identifies the possibility and time of the negative pressure loop failure, which can guide the operator to judge the current actual state of the negative pressure loop and solve problems such as waste caused by early replacement and damage to batteries caused by delayed replacement.

[0110] In a second aspect, an embodiment of the present application further provides a negative pressure circuit fault warning device, and the negative pressure circuit fault warning device includes: an acquisition module, which is configured to acquire real-time data of a negative pressure value to be tested of a negative pressure suction nozzle to be tested during the use of a battery negative pressure forming device; a prediction module, which is configured to input the data of the negative pressure value to be tested into a trained negative pressure fault prediction model to obtain a predicted fault time of the negative pressure suction nozzle to be tested.

[0111] Among them, the function realization of each module in the above negative pressure circuit fault warning device corresponds to each step in the above embodiment of the negative pressure circuit fault warning method, and its function and implementation process will not be elaborated here one by one.

[0112] In a third aspect, an embodiment of the present application provides a battery negative pressure forming device. In the embodiment of the present application, the battery negative pressure forming device may include a processor, a memory, a communication interface, and a communication bus.

[0113] Among them, the communication bus can be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0114] The communication interface includes interfaces such as input / output (I / O) interfaces, physical interfaces, and logical interfaces for realizing the interconnection of components inside the battery negative pressure forming device, and interfaces for realizing the interconnection of the battery negative pressure forming device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, an optical fiber interface, an ATM interface, etc.; the user device can be a display screen (Display), a keyboard (Keyboard), etc.

[0115] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical memory, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0116] The processor can be a general-purpose processor, which can call the negative pressure loop fault warning program stored in the memory and execute the negative pressure loop fault warning method provided in the embodiments of the present application. For example, the general-purpose processor can be a central processing unit (CPU). Among them, the method executed when the negative pressure loop fault warning program is called can refer to the various embodiments of the negative pressure loop fault warning method of the present application, which will not be elaborated here.

[0117] Fourthly, the embodiments of the present application further provide a computer-readable storage medium.

[0118] The negative pressure loop fault warning program is stored on the readable storage medium of the present application. When the negative pressure loop fault warning program is executed by a processor, the steps of the negative pressure loop fault warning method as described above are implemented.

[0119] Among them, the method implemented when the negative pressure loop fault warning program is executed can refer to the various embodiments of the negative pressure loop fault warning method of the present application, which will not be elaborated here.

[0120] It should be noted that the serial numbers of the above embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0121] The terms "including" and "having" and any variations thereof in the specification, claims and drawings of the present application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices. The descriptions of terms such as "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit that "first", "second" and "third" are of different types.

[0122] In the description of the embodiments of the present application, terms such as "exemplary", "for example" or "for instance" are used to indicate examples, illustrations or explanations. Any embodiment or design solution described as "exemplary", "for example" or "for instance" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of terms such as "exemplary", "for example" or "for instance" is intended to present relevant concepts in a specific manner.

[0123] In the description of the embodiments of the present application, unless otherwise specified, " / " means "or". For example, A / B may mean A or B. The "and / or" in the text is only a relationship describing the associated objects, indicating that there can be three relationships. For example, A and / or B may mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of the present application, "a plurality of" means two or more than two.

[0124] In some of the processes described in the embodiments of the present application, there are a plurality of operations or steps that appear in a specific order. However, it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or may be executed in parallel. The serial numbers of the operations are only used to distinguish different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in order or in parallel, and these operations or steps may be combined.

[0125] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium as described above (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions for causing a terminal device to execute the methods described in the various embodiments of the present application.

[0126] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, is equally included in the patent protection scope of the present application.

Claims

1. A method for warning of negative pressure circuit faults, characterized in that, The negative pressure circuit fault warning method includes: Obtaining real-time negative pressure value data to be tested of the negative pressure suction nozzle to be tested during the use of the battery negative pressure forming equipment; Inputting the negative pressure value data to be tested into the trained negative pressure fault prediction model to obtain the predicted fault time of the negative pressure suction nozzle to be tested.

2. The negative pressure circuit fault warning method according to claim 1, wherein, The negative pressure fault prediction model is obtained by the following method: Based on the negative pressure value data during the use of negative pressure in the source domain, the time when the negative pressure suction nozzle fails, and the negative pressure value data during the use of negative pressure in the target domain, the negative pressure fault prediction model is trained; wherein, the source domain is the battery negative pressure forming equipment that has been put into use, and the target domain is the newly produced battery negative pressure forming equipment.

3. The negative pressure circuit fault warning method according to claim 2, characterized in that, The training of the negative pressure fault prediction model based on the negative pressure value data during the use of negative pressure in the source domain, the time when the negative pressure suction nozzle fails, and the negative pressure value data during the use of negative pressure in the target domain includes: Inputting the negative pressure value data during the use of negative pressure in the source domain into the initial source domain negative pressure fault prediction model to obtain the calculated fault time value and the negative pressure value change characteristics of the source domain; Obtaining the target domain negative pressure fault prediction model based on the source domain negative pressure fault prediction model; Inputting the negative pressure value data during the use of negative pressure in the target domain into the target domain negative pressure fault prediction model to obtain the negative pressure value change characteristics of the target domain; Calculating the loss value according to the calculated fault time value, the time when the negative pressure suction nozzle fails, the negative pressure value change characteristics of the source domain, and the negative pressure value change characteristics of the target domain, and updating the parameters of the source domain negative pressure fault prediction model according to the loss value until the training ends, and the latest source domain negative pressure fault prediction model obtained is the trained negative pressure fault prediction model.

4. The negative pressure circuit fault warning method according to claim 3, characterized in that The inputting the negative pressure value data during the use of negative pressure in the source domain into the initial source domain negative pressure fault prediction model to obtain the calculated fault time value and the negative pressure value change characteristics of the source domain includes: Extracting the negative pressure value change characteristics from the negative pressure value data in the source domain negative pressure fault prediction model to obtain the negative pressure value change characteristics of the source domain; Projecting the negative pressure value change characteristics of the source domain into a low-dimensional space and maximizing the separation of normal characteristics and fault characteristics to obtain the characteristics after projection in the source domain space; Outputting the calculated fault time value according to the mapping relationship between the characteristics after projection in the source domain space and the fault time.

5. The negative pressure circuit fault warning method according to claim 4, characterized in that, The extracting the negative pressure value change characteristics from the negative pressure value data in the source domain negative pressure fault prediction model to obtain the negative pressure value change characteristics of the source domain includes: In the source domain negative pressure fault prediction model, the local change characteristics of the negative pressure value are extracted from the negative pressure value data through a convolutional neural network, and then the overall change characteristics of the negative pressure value are obtained through a self-attention mechanism network to obtain the negative pressure value change characteristics of the source domain.

6. The negative pressure circuit fault warning method according to claim 3, wherein The calculating the loss value according to the calculated fault time value, the time when the negative pressure suction nozzle fails, the negative pressure value change characteristics of the source domain, and the negative pressure value change characteristics of the target domain, and updating the parameters of the source domain negative pressure fault prediction model according to the loss value until the training ends, and the latest source domain negative pressure fault prediction model obtained is the trained negative pressure fault prediction model includes: Calculating the cross-entropy loss value according to the calculated fault time value and the time when the negative pressure suction nozzle fails; Calculate the central distance loss value according to the time when the negative pressure suction nozzle fails and the characteristics after the projection in the source domain space; wherein, the characteristics after the projection in the source domain space are obtained by projecting the variation characteristics of the source domain negative pressure value into a low-dimensional space and maximizing the separation between the normal characteristics and the fault characteristics. Calculate the depth alignment loss value according to the characteristics after the projection in the source domain space and the characteristics after the projection in the target domain space; wherein, the characteristics after the projection in the target domain space are obtained by projecting the variation characteristics of the target domain negative pressure value into a low-dimensional space and maximizing the separation between the normal characteristics and the fault characteristics. Update the parameters of the source domain negative pressure fault prediction model according to the cross-entropy loss value, the central distance loss value, and the calculated depth alignment loss value until the training ends, and the latest source domain negative pressure fault prediction model obtained is the trained negative pressure fault prediction model.

7. The negative pressure circuit fault warning method according to claim 3, wherein The obtaining of the target domain negative pressure fault prediction model based on the source domain negative pressure fault prediction model includes: Read the model parameters of the source domain negative pressure fault prediction model and write the model parameters into the model of the target domain to obtain the target domain negative pressure fault prediction model.

8. A negative pressure circuit fault warning device, characterized in that, The negative pressure circuit fault warning device includes: An acquisition module, which is used to acquire real-time data of the negative pressure value to be tested of the negative pressure suction nozzle to be tested during the use of the battery negative pressure forming equipment. A prediction module, which is used to input the data of the negative pressure value to be tested into the trained negative pressure fault prediction model to obtain the predicted fault time of the negative pressure suction nozzle to be tested.

9. A battery negative pressure formation device, characterized in that, The battery negative pressure forming equipment includes a processor, a memory, and a negative pressure circuit fault warning program stored on the memory and executable by the processor. When the negative pressure circuit fault warning program is executed by the processor, the steps of the negative pressure circuit fault warning method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that, A negative pressure circuit fault warning program is stored on the computer-readable storage medium. When the negative pressure circuit fault warning program is executed by the processor, the steps of the negative pressure circuit fault warning method according to any one of claims 1 to 7 are implemented.

Images