Fault prediction method and device, equipment and storage medium

Abstract

The invention discloses a fault prediction method and device, equipment and a storage medium, and relates to the technical field of power batteries, and the fault prediction method comprises the steps: obtaining the multi-modal welding state parameters of target welding equipment in real time, and extracting the time sequence features of the welding state parameters of each modal based on a multi-channel time convolution network; capturing a local time sequence dependency relationship of each welding state parameter and a dependency relationship between the welding state parameters; and then, fusing each time sequence feature based on an attention mechanism feature fusion method to obtain a fused feature capable of representing the importance degree of each time sequence feature, reinforcing the feature strongly related to the fault and weakening the redundant feature, and inputting the fused feature into a preset machine learning model to obtain a fault feature. According to the method, the welding fault prediction result with higher precision can be obtained, so that the early warning efficiency is improved, and the welding quality and the stability and safety of the battery module are improved through fault early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power batteries, and in particular to a fault prediction method, device, equipment and storage medium. BACKGROUND

[0002] With the rapid development of new energy vehicles, the lightweight of the power battery tray has become a key to improving the battery energy density and the performance of the whole vehicle. However, the aluminum alloy welding technology is difficult, and the traditional arc welding process cannot meet the requirements, and the welding quality directly affects the stability and safety of the battery module. The welding equipment of the power battery tray is prone to structural stiffness decline due to thermal deformation after long-term operation, affecting the precision of the welding position.

[0003] At present, the welding process lacks real-time monitoring and fault warning of the running state of the welding equipment, resulting in large stop-line loss and reduced overall productivity. Therefore, how to implement real-time monitoring and fault prediction and warning of the power battery tray welding equipment is a problem to be solved at present. However, the existing welding fault prediction methods are mostly limited to the extraction and analysis of single data, such as vibration signal analysis and fault mode recognition. Single data source cannot fully reflect the welding state of the equipment, and there are problems such as data inconsistency, missing and poor quality. In recent years, although the data fusion technology has gradually developed, there is still a lack of analysis of the correlation between multiple features and the sensitivity of multiple features to the prediction target, resulting in low welding fault prediction accuracy and low warning efficiency, which ultimately reduces the welding quality and the stability and safety of the battery module. SUMMARY

[0004] The embodiments of the present application provide a fault prediction method, device, equipment and storage medium, which can improve the welding fault prediction accuracy and improve the warning efficiency.

[0005] In a first aspect, the embodiments of the present application provide a fault prediction method, which comprises:

[0006] real-time acquisition of the welding state parameters of the target welding equipment in multiple modes;

[0007] extracting the time sequence features of the welding state parameters of each mode based on a preset multi-channel time convolution network;

[0008] fusing each time sequence feature based on an attention mechanism feature fusion method to obtain a fused feature;

[0009] inputting the fused feature into a preset machine learning model to obtain a welding fault prediction result.

[0010] Further, the multi-channel time convolution network has a residual connection structure, and the residual connection structure is used to transfer information across layers in the multi-channel time convolution network.

[0011] Further, the fusion feature is obtained by fusing each time sequence feature based on an attention mechanism feature fusion method, and the fusion feature includes:

[0012] Global information of each time sequence feature is extracted by using global maximum pooling and global average pooling respectively, to obtain first global information and second global information;

[0013] The first global information and the second global information are input into a preset multi-layer perceptron to obtain first output features and second output features respectively;

[0014] The first output features and the second output features are summed element by element to obtain first merged features;

[0015] The first merged features are normalized to obtain second merged features;

[0016] The second merged features are multiplied element by element with the first merged features to obtain the fusion features.

[0017] Further, the fusion feature is obtained by fusing each time sequence feature based on an attention mechanism feature fusion method, and the fusion feature includes:

[0018] Each time sequence feature is weighted and fused based on a preset attention weight matrix to obtain the fusion feature, and the attention weight matrix is automatically learned by a preset neural network to adjust the weight in real time according to the input data.

[0019] Further, before extracting the time sequence features of the welding state parameters of each modality based on the preset multi-channel time convolution network, the fault prediction method further includes:

[0020] The welding state parameters are sequentially denoised, normalized, and cleaned.

[0021] Further, the data cleaning of the welding state parameters includes:

[0022] For any parameter in the welding state parameters, if any parameter exceeds the corresponding preset parameter range, or any parameter exceeds the corresponding preset range, the any parameter is removed; or

[0023] The welding state parameters are cleaned based on a density clustering algorithm.

[0024] Further, after sequentially denoising, normalizing, and cleaning the welding state parameters, the fault prediction method further includes:

[0025] The welding state parameters of each modality are converted into an encoding vector of a unified data dimension and are data-aligned.

[0026] In a second aspect, an embodiment of the present application provides a fault prediction device, the fault prediction device comprising:

[0027] a data acquisition unit configured to acquire, in real time, welding state parameters of a target welding equipment in multiple modes;

[0028] a multi-channel feature extraction unit configured to extract time sequence features of the welding state parameters of each mode based on a preset multi-channel time convolution network;

[0029] a feature fusion unit configured to fuse each time sequence feature based on an attention mechanism feature fusion method to obtain fused features;

[0030] a fault prediction unit configured to input the fused features into a preset machine learning model to obtain a welding fault prediction result.

[0031] In a third aspect, an embodiment of the present application further provides a fault prediction device comprising a processor and a memory, the memory storing a computer program, when the computer program is executed by the processor, the computer program causes the processor to execute steps of any one of the fault prediction methods provided by the embodiments of the present application.

[0032] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, the computer readable storage medium comprising a computer program, when the computer program is executed on an electronic device, the computer program is configured to cause the electronic device to execute steps of any one of the fault prediction methods provided by the embodiments of the present application.

[0033] The present application has the following beneficial effects:

[0034] The fault prediction method of the present application acquires, in real time, welding state parameters of a target welding equipment in multiple modes, extracts time sequence features of the welding state parameters of each mode based on a multi-channel time convolution network, captures local time sequence dependency relationships of each welding state parameter and dependency relationships therebetween, fuses each time sequence feature based on an attention mechanism feature fusion method to obtain fused features capable of representing importance of each time sequence feature, strengthens features strongly related to faults and weakens redundant features, inputs the fused features into a preset machine learning model to obtain a welding fault prediction result with higher precision, thereby improving early warning efficiency, and improving welding quality, stability and safety of a battery module through fault early warning. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0036] Figure 1 is a flowchart of the fault prediction method provided in the embodiments of the present application;

[0037] Figure 2 is another flowchart of the fault prediction method provided in the embodiments of the present application;

[0038] Figure 3 is a data cleaning flowchart of the fault prediction method provided in the embodiments of the present application;

[0039] Figure 4 is a structure and flowchart of the multi-channel time convolution network feature extraction and attention mechanism feature fusion of the fault prediction method provided in the embodiments of the present application;

[0040] Figure 5 is a structure diagram of the fault prediction device provided in the embodiments of the present application;

[0041] Figure 6 is a structure diagram of the fault prediction device provided in the embodiments of the present application. DETAILED DESCRIPTION

[0042] In order to make the person skilled in the art better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0043] The following disclosure provides many different embodiments, or examples, to implement different characteristics of the provided subject matter. Specific examples of components and arrangements are described below to provide a thorough description of embodiments. These are, of course, merely examples and are not intended to be limiting. For example, in the following description, where a first feature is formed on or over a second feature, it is also contemplated that the first feature can be formed in direct contact with the second feature, and that additional features can be formed between the first and second features such that the first and second features are not in direct contact. In addition, the present disclosure can use reference numbers and / or letters in various instances. This repetition is for the purpose of simplicity and clarity and does not in itself dictate a relationship between the various embodiments and / or configurations discussed.

[0044] In addition, the terms "first", "second", and the like, as used in the description do not imply or mean that the indicated technical features are of relative importance or that the indicated technical features are of a certain number. Therefore, the features defined as "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the fact that a person skilled in the art can realize it, and when the combination of technical solutions appears contradictory or cannot be realized, it should be considered that the combination of technical solutions does not exist, nor within the protection scope required by the present application. In addition, the terms "include" and "have" and any variations thereof 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 can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0045] The term "and / or" herein is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B and C can mean including any one or more elements selected from the set consisting of A, B and C.

[0046] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative embodiments to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0047] In addition, in order to better illustrate the present application, numerous specific details are given in the specific embodiments below. Those skilled in the art should understand that the present application can be implemented without certain specific details. In some examples, methods, means, elements and circuits well known to those skilled in the art are not described in detail, in order to highlight the main idea of the present application.

[0048] As described in the foregoing background section, the inventors have found that:

[0049] For real-time monitoring and fault prediction of power battery tray welding equipment, existing welding fault prediction methods are mostly limited to single data extraction and analysis, such as vibration signal analysis and fault mode identification. Single data source is difficult to fully reflect the welding state of the equipment, and there are problems such as data inconsistency, missing and poor quality. In recent years, although the data fusion technology has gradually developed, there is still a lack of analysis of the correlation between multiple features and the sensitivity of multiple features to the prediction target, resulting in low welding fault prediction accuracy and low early warning efficiency, ultimately reducing the welding quality and the stability and safety of the battery module.

[0050] Based on this, the embodiment of the present application proposes a fault prediction method, device, equipment and storage medium, which obtains the multi-modal welding state parameters of the target welding equipment in real time, and extracts the time sequence features of each welding state parameter based on a multi-channel time convolution network, capturing the local time sequence dependence of each welding state parameter and the dependence between them. Then, the attention mechanism feature fusion method is used to fuse each time sequence feature to obtain a fusion feature that can represent the importance of each time sequence feature, which strengthens the features strongly related to faults and weakens the redundant features. By inputting the fusion feature into a pre-set machine learning model, a more accurate welding fault prediction result can be obtained, thereby improving the early warning efficiency and improving the welding quality and the stability and safety of the battery module through fault early warning.

[0051] Specifically, the present embodiment will be described from the perspective of a fault prediction device, which can be integrated in a fault prediction equipment, i.e. the fault prediction method of the present embodiment can be executed by the fault prediction equipment.

[0052] The following will be described in detail with reference to the accompanying drawings. In the present embodiment, the execution subject is taken as an example of the fault prediction equipment. It should be noted that the sequence of the following embodiments is not limited as the preferred sequence of the embodiments. Although the logical sequence is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown in the figure.

[0053] Please refer to Figure 1 and Figure 2 The specific process of the fault prediction method can be as follows:

[0054] Step S101, real-time acquisition of multi-modal welding state parameters of the target welding equipment.

[0055] The target welding equipment is a welding equipment for which welding fault prediction is performed by the embodiment of the present application, such as a battery pack welding equipment, which can be one or multiple. The multi-modal welding state parameters of the target welding equipment refer to various types of parameters in the welding process of the target welding equipment, such as voltage, current, vibration, noise, speed, torque, temperature, human-computer interaction setting, historical fault record, and the like.

[0056] It can be understood that, by acquiring the multi-modal welding state parameters of the target welding equipment in real time, the embodiment of the present application can more comprehensively reflect the state of the target welding equipment compared with the prior solution which is limited to extraction and analysis of single data.

[0057] In some embodiments, a sensor can be used to collect the multi-modal welding state parameters of the target welding equipment.

[0058] In some embodiments, for the multi-modal welding state parameters of the target welding equipment acquired in step S101, the welding state parameters are preprocessed before feature extraction. In this embodiment, the preprocessing of the welding state parameters can include:

[0059] The welding state parameters are sequentially subjected to denoising processing, normalization processing, and data cleaning.

[0060] The denoising processing of the welding state parameters can remove noise data in the welding state parameters and reduce the interference of the noise data on subsequent fault diagnosis; the normalization processing of the welding state parameters can convert the welding state parameters into standard format data, facilitating subsequent data processing; the purpose of the data cleaning of the welding state parameters is to remove abnormal data in the welding state parameters and further remove noise, so that the multi-modal welding state parameters acquired in step S101 are more accurate and reliable, and the accuracy of subsequent fault diagnosis is improved.

[0061] Specifically, the denoising processing of the welding state parameters can include:

[0062] The density of the data feature points of the welding state parameters is calculated, and the formula is as follows:

[0063]

[0064] wherein θ represents the data feature point density; i represents data, i.e., the welding state parameter; k represents the data dimension in the space, a i According to the actual data denoising requirement, a threshold is set for the denoising process. The data in the space is divided into two data groups by the threshold. A two-dimensional graph of the data feature points is established, and the data division boundary is adjusted until the data in one boundary completely meets the noise requirement.

[0065] Then, the welding state parameters after denoising are normalized, and the formula is as follows:

[0066]

[0067] Wherein, j represents the data after denoising, z j represents the normalized standard format data, n j represents the data feature quantity after denoising, represents the data average value, and s represents the data standard deviation.

[0068] After completing the denoising and normalization of the welding state parameters, data cleaning is performed to remove abnormal data. Referring to Figure 3 In some embodiments, the data cleaning of the welding state parameters can include:

[0069] For any parameter in the welding state parameters, if the any parameter exceeds the corresponding preset parameter range, or the any parameter exceeds the corresponding preset range, the any parameter is rejected.

[0070] Specifically, the present embodiment judges whether each welding state parameter after the foregoing denoising and normalization processing exceeds the corresponding parameter range, or exceeds the corresponding sensor range, and if so, the data is considered to be abnormal data and is rejected.

[0071] Alternatively, in some other embodiments, the data cleaning of the welding state parameters can include:

[0072] The welding state parameters are cleaned based on a density clustering algorithm.

[0073] Specifically, the present embodiment uses a density clustering algorithm to identify abnormal data, and feeds back the multi-source heterogeneous data (i.e. multi-modal welding state parameters) in the original data space according to the calculated feature time sequence correlation degree, sets the clustering cluster point of the data and its adjacent data, to determine the clustering attribution of the data, and calculates the data and its clustering cluster:

[0074]

[0075] Wherein, d represents the Euclidean distance of the data and the cluster point in space, o represents the cluster point of the data clustering, l j represents the position of the data j in space, l o represents the position of the cluster point in space.

[0076] The density of the data after clustering is calculated:

[0077]

[0078] wherein θ ′ represents the density of the clustered data space, m represents the data sequence, and x represents the clustering connection point.

[0079] Then, the density clustering algorithm is iterated continuously until the calculated value of the data density reaches the optimum and no longer changes. According to the final data density calculation result, the abnormal data is identified. The boundary value of the abnormality identification is set as:

[0080] θ ′ = μ ± 3σ

[0081] wherein μ represents the mean value of the density, and σ represents the standard deviation of the density. The data whose calculated value of the data density does not belong to the above interval is screened one by one, and the abnormal data identification process based on the density clustering algorithm is completed.

[0082] The identified multi-source heterogeneous abnormal data is cleaned, the missing data points of the abnormal data are analyzed, and the abnormal data is divided into two abnormal types of local missing and long-term missing according to the different time sequence correlation degrees of the abnormal data. The matrix is solved:

[0083] h = (θ ′ δ) c

[0084] wherein h represents the time sequence length of the data missing, and c represents the current dimension of the data.

[0085] The missing part of the abnormal data is filled by establishing a minimum missing function:

[0086] J = (‖h-Y‖) 2

[0087] wherein J represents the missing content filled in the abnormal data, and Y represents the time sequence length of the normal data with the same time sequence correlation degree as the abnormal data.

[0088] The filled data is subjected to abnormal value detection to determine whether the missing degree meets the requirement, and the abnormal data that fails to be filled is returned to be solved by the matrix and filled again.

[0089] Further, in some embodiments, after sequentially performing the denoising processing, the normalization processing and the data cleaning on the welding state parameters, the fault prediction method can further include:

[0090] Converting the welding state parameters of each mode into an encoding vector of a unified data dimension and performing data alignment.

[0091] The embodiment of the present application converts and aligns the aforementioned pretreated multi-modal data, so that the text data and the time series data are both converted into coding vectors of uniform data dimensions, obtaining aligned data with consistent time granularity, which is beneficial to subsequent feature extraction and prediction analysis.

[0092] Specifically, the embodiment of the present application respectively adopts the Bert model for word embedding coding of each feature word contained in each sentence segment of the sentence segmentation and word segmentation information in the text information, obtaining a 1x B-dimensional word coding vector of each feature word, B being the coding dimension size of the word embedding coding of the Bert model. The time series data is segmented and cut according to the set segment length, obtaining each time series data segment of the segmented and cut time series data. The coding dimension size B of the word embedding coding of the sentence segmentation and word segmentation information is taken as the coding dimension size of the autoencoder, and the pre-trained autoencoder is used for coding processing of each time series data segment of the time series data, respectively, to obtain a 1x B-dimensional data segment coding vector of each time series data segment.

[0093] Different systems are respectively synchronized with a standard clock, and time labels are marked in the data, and the time labels are sent to the data platform together with the data. Resampling method is used for data alignment. Resampling is resampling according to the same frequency (time interval) for data with different sampling frequencies, to obtain aligned data with consistent time granularity.

[0094] In step S102, the time series features of the welding state parameters of each modality are extracted based on a preset multi-channel time convolution network.

[0095] The temporal convolutional network (TCN) is a one-dimensional convolutional network structure, which mainly consists of dilated causal convolution and residual connection. The structure is relatively simple, but its accuracy exceeds that of the long short-term memory network (LSTM) and the gated recurrent unit (GRU) in multiple tasks. Due to the introduction of shared convolution kernels in TCN, the model can perform parallel computation to a certain extent, improving the speed of model calculation. When processing time series data in the time dimension, the output y t The model should only depend on the current time t and the previous input, i.e. when actually processing time series prediction problems, the model should ensure causality. Therefore, causal convolution is adopted in the TCN convolution structure to constrain the convolution operation in the time dimension, ensuring that the output y t only depends on the input x0, x1,…, x tThe method solves the problem of information leakage from the future to the past without relying on any future information. The dilated convolution in the dilated causal convolution, also known as the hollow convolution, allows the convolution kernel to sample information at intervals by injecting a specified size of hollow into the standard convolution feature map.

[0096] It can be understood that the time weight of a single sequence lacks understanding of the internal relationship between different time sequence variables when facing the periodic law of cross-domain multi-event and the time sequence law of multivariate time sequence data. Based on this, the embodiment of the present application extracts time sequence features of different modal welding state parameters through a multi-channel TCN network, and uses a stacked structure of dilated convolution to expand the receptive field and capture local time sequence dependency. According to the foregoing, the multi-modal welding state parameters are time sequence data. The embodiment of the present application learns the dependency between multiple variables from different time steps, thereby improving learning efficiency and training ability.

[0097] In a real scene, the multi-modal welding state parameters collected in step S101 usually contain a large amount of redundant information. If the redundant information is treated equally with important information, it will interfere with the performance of the model in extracting information.

[0098] Therefore, in some embodiments, the multi-channel time convolution network has a residual connection structure, and the residual connection structure is used to pass information across layers in the multi-channel time convolution network.

[0099] As shown in Figure 4 , the embodiment of the present application introduces a residual connection structure in the TCN in the deep residual network, so that the network can pass information across layers, improve the information flow of the shallow network features, and solve the gradient disappearance problem caused by the number of layers in the neural network. The embodiment of the present application improves the source language expression method, dynamically selects the source language related information in decoding, thereby greatly improving the shortcomings of the classical Encoder-Decoder framework, paying attention to the key information in the data features and ignoring the redundant information in the data features, thereby improving the model performance.

[0100] In step S103, the attention mechanism feature fusion method is used to fuse each time sequence feature to obtain a fused feature.

[0101] Continuing to refer to Figure 4 , the embodiment of the present application fuses the multiple time sequence features extracted in step S102 based on the attention mechanism feature fusion method to obtain a fused feature, which can represent the importance of each time sequence feature, thereby further strengthening the feature signals (such as abnormal current pulses, temperature mutations, etc.) strongly related to faults and weakening redundant features.

[0102] According to the foregoing, the TCN of the embodiment of the application is one-dimensional convolution, so the mutual information in each sensor dimension is extracted by using the channel attention mechanism in step S103, and the channel attention mechanism needs to be improved. Based on this, in some embodiments, step S103 can include:

[0103] Global information extraction is performed on each time series feature by using global maximum pooling and global average pooling respectively to obtain first global information and second global information; the first global information and the second global information are input into a preset multi-layer perceptron to obtain first output features and second output features respectively; element-wise summation is performed on the first output features and the second output features to obtain first merged features; normalization processing is performed on the first merged features to obtain second merged features; the second merged features and the first merged features are multiplied element by element to obtain fusion features.

[0104] Specifically, the embodiment of the application first performs global information extraction on each time series feature by using global maximum pooling and global average pooling to obtain first global information m and second global information a wherein:

[0105]

[0106] wherein, represents the output of the l-1th layer; l represents the lth layer of the TCN, T represents the time step; i represents the ith channel of the output feature.

[0107] Then, the first global information m l and the second global information a l are respectively input into a multi-layer perceptron, and the multi-layer perceptron includes a linear hidden layer. In order to reduce the parameter overhead in the network, the number of neurons in the hidden layer is k / r, wherein r is the drop rate and k is the number of channels.

[0108] The two groups of output features obtained by the multi-layer perceptron network are merged by element-wise summation, the merged output is normalized by a hard sigmoid activation function, and the final output feature map is obtained by element-wise multiplication with the original input to obtain the final output of the channel attention module, i.e., the fusion features.

[0109] It can be understood that the channel attention mechanism applied in the embodiment of the present application extracts different information in the degraded feature data through max pooling and average pooling. The max pooling extracts the maximum value in each channel, that is, the maximum feedback of information in each channel. When gradient back propagation is performed, the gradient will be fed back to the corresponding maximum place. The average pooling calculates the average value of the data in each channel, extracts the average feedback of all information, and makes up for the problem that the max pooling only focuses on the most important information and completely ignores other less important information. Therefore, the channel attention mechanism of the embodiment of the present application highlights important information and suppresses unimportant information through max pooling and average pooling. Finally, the output features are normalized by the hard sigmoid activation function, so as to represent the importance of information in the TCN output features.

[0110] Alternatively, in other embodiments, step S103 can further include:

[0111] The attention weight matrix is automatically learned by a preset neural network to adjust the weight in real time according to the input data.

[0112] The attention weight matrix is automatically learned by a preset neural network to adjust the weight in real time according to the input data.

[0113] Therefore, the embodiment of the present application can strengthen the feature signals strongly related to the fault and weaken the redundant features by weighting and fusing each time sequence feature through the learnable attention weight matrix.

[0114] Step S104 inputs the fused features into a preset machine learning model to obtain a welding fault prediction result.

[0115] The machine learning model is pre-trained and can output a welding fault prediction result such as fault classification when the input features (such as the fused features) are input.

[0116] It can be understood that the foregoing steps of the embodiment of the present application fuse the multi-modal data to form a multi-modal data feature vector of the target welding equipment, which helps to more accurately perform fault diagnosis of the welding equipment and other applications. The embodiment of the present application takes the fused feature vector as the input of the machine learning model, greatly improves the amount of data used for machine learning modeling, thereby improving the modeling accuracy and further improving the welding fault prediction result.

[0117] In summary, the embodiment of the present invention obtains the multimodal welding state parameters of the target welding equipment in real time, and extracts the time series features of the welding state parameters of each modality based on a multi-channel time convolutional network, thereby capturing the local temporal dependency of each welding state parameter and the dependency between them; then, each time series feature is fused based on the attention mechanism feature fusion method to obtain a fused feature that can characterize the importance of each time series feature, strengthen the features that are strongly correlated with the fault and weaken the redundant features. By inputting the fused feature into a preset machine learning model, a more accurate welding fault prediction result can be obtained, thereby improving the warning efficiency, and improving the welding quality and the stability and safety of the battery module through fault warning.

[0118] This embodiment also provides a fault prediction device, which can be integrated into the fault prediction equipment. Figure 5 As shown, the fault prediction device may include:

[0119] The data acquisition unit 501 is used to acquire multi-modal welding state parameters of the target welding equipment in real time;

[0120] A multi-channel feature extraction unit 502 is used to extract the time series features of the welding state parameters of each mode based on a preset multi-channel time convolution network;

[0121] A feature fusion unit 503 is used to fuse each time series feature based on an attention mechanism feature fusion method to obtain a fused feature;

[0122] The fault prediction unit 504 is used to input the fusion features into a preset machine learning model to obtain a welding fault prediction result.

[0123] like Figure 6 As shown, Figure 6 A schematic diagram of the structure of a fault prediction device provided in an embodiment of the present invention. The fault prediction device 1100 includes a processor 1101 having one or more processing cores, a memory 1102 having one or more computer-readable storage media, and a computer program stored in the memory 1102 and executable on the processor. The processor 1101 is electrically connected to the memory 1102. Those skilled in the art will appreciate that the fault prediction device structure shown in the figure does not limit the fault prediction device, and may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0124] The processor 1101 is the control center of the fault prediction device 1100, connects various parts of the fault prediction device 1100 through various interfaces and lines, performs various functions of the fault prediction device 1100 and processes data by running or loading software programs and / or units stored in the memory 1102 and calling data stored in the memory 1102, thereby monitoring the fault prediction device 1100 as a whole. The processor 1101 can be a processor CPU, a graphics processor GPU, a network processor (NP), etc., and can implement or execute various methods, steps and logic block diagrams disclosed in the embodiments of the present application.

[0125] In the embodiments of the present application, the processor 1101 in the fault prediction device 1100 loads instructions corresponding to processes of one or more application programs into the memory 1102, and runs the application programs stored in the memory 1102 by the processor 1101 to implement various functions according to the following steps. For details, refer to the previous embodiments, which will not be repeated here.

[0126] Optionally, as shown in Figure 6 The fault prediction device 1100 further includes a touch display screen 1103, a radio frequency circuit 1104, an audio circuit 1105, an input unit 1106 and a power supply 1107. The processor 1101 is electrically connected with the touch display screen 1103, the radio frequency circuit 1104, the audio circuit 1105, the input unit 1106 and the power supply 1107 respectively. Those skilled in the art can understand that Figure 6 The structure of the fault prediction device shown in the figure does not constitute a limitation on the fault prediction device, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0127] The touch display screen 1103 can be used to display a graphical user interface and receive operation instructions generated by user acting on the graphical user interface. The touch display screen 1103 can include a display panel and a touch panel. The display panel can be used to display information input by the user or information provided to the user and various graphical user interfaces of the fault prediction device, which can be composed of graphics, text, icons, videos and any combination thereof. Optionally, the display panel can be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. The touch panel can be used to collect touch operations (such as user operations on or near the touch panel using a finger, a stylus or any suitable object or accessory) of the user thereon or therearound, and generate corresponding operation instructions, and the operation instructions execute corresponding programs. Optionally, the touch panel can include two parts of a touch detection device and a touch controller. The touch detection device detects the touch position of the user and detects signals generated by the touch operation, and transmits the signals to the touch controller; the touch controller receives the touch information from the touch detection device, and converts it into touch coordinates, and then sends it to the processor 1101, and can receive commands from the processor 1101 and execute them. The touch panel can cover the display panel, and when the touch panel detects a touch operation thereon or therearound, it transmits to the processor 1101 to determine the type of touch event, and then the processor 1101 provides corresponding visual output on the display panel according to the type of touch event. In the embodiments of the present application, the touch panel and the display panel can be integrated into the touch display screen 1103 to realize input and output functions. However, in some embodiments, the touch panel and the touch panel can realize input and output functions as two independent components. That is, the touch display screen 1103 can also realize input functions as part of the input unit 1106.

[0128] The radio frequency circuit 1104 can be used to transceive radio frequency signals to establish wireless communication with a network device or other fault prediction device, and transceive signals between the network device or other fault prediction device.

[0129] The audio circuit 1105 can be used to provide an audio interface between the user and the fault prediction device through a speaker and a microphone. The audio circuit 1105 can convert the received audio data into an electrical signal and transmit the electrical signal to the speaker for conversion into a sound signal output. On the other hand, the microphone collects a sound signal and converts the sound signal into an electrical signal, which is received by the audio circuit 1105 and converted into audio data. The audio data is output to the processor 1101 for processing, and then transmitted to another fault prediction device through the radio frequency circuit 1104, or output to the memory 1102 for further processing. The audio circuit 1105 can also include a jack for a headset to provide communication between the headset and the fault prediction device.

[0130] The input unit 1106 can be used to receive inputted numbers, character information or user feature information (such as fingerprints, iris, face information, etc.), and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0131] The power supply 1107 is used to supply power to various components of the fault prediction device 1100. Optionally, the power supply 1107 can be logically connected to the processor 1101 through a power management system, so that the power management system can be used to manage charging, discharging, power consumption management and other functions. The power supply 1107 can also include one or more DC or AC power sources, recharging systems, power failure detection circuits, power converters or inverters, power status indicators, and any other components.

[0132] In the above embodiments, the description of each embodiment focuses on different aspects, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0133] Those of ordinary skill in the art can understand that all or part of the steps of the various methods of the above embodiments can be completed by instructions, or by instructions controlling related hardware, which can be stored in a computer readable storage medium and loaded and executed by a processor.

[0134] To this end, an embodiment of the present application provides a computer readable storage medium, which stores a plurality of computer programs capable of being loaded by a processor to execute any fault prediction method provided by an embodiment of the present application. The computer program can execute the steps of the fault prediction method described above, which can be referred to the previous embodiments and will not be described here again.

[0135] The computer readable storage medium can include a read only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0136] Since the computer program stored in the computer readable storage medium can execute any one of the failure prediction methods provided by the embodiments of the present application, the beneficial effects that can be achieved by any one of the failure prediction methods provided by the embodiments of the present application can be achieved, which are described in detail in the foregoing embodiments and will not be repeated here.

[0137] In the foregoing embodiments of the failure prediction device, the computer readable storage medium, the failure prediction apparatus and the computer program product, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments. It can be clearly understood by those skilled in the art that, for the convenience and brevity of description, the specific working process of the foregoing described failure prediction device, computer readable storage medium, computer program product, failure prediction apparatus and corresponding units thereof and the beneficial effects that can be brought can be referred to the description of the failure prediction method in the foregoing embodiments, and will not be repeated here.

[0138] The foregoing provides a detailed description of the failure prediction method, the failure prediction device, the failure prediction apparatus, the computer readable storage medium and the computer program product provided by the embodiments of the present application. The principle and implementation manner of the present application are described by applying specific examples in this paper. The foregoing description of the embodiments is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range will be changed. In view of the foregoing, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A fault prediction method, characterized in that: The method comprises: Real-time acquisition of multi-modal welding state parameters of target welding equipment; Extracting the time series features of the welding state parameters of each mode based on a preset multi-channel time convolutional network; Based on the attention mechanism feature fusion method, each of the time series features is fused to obtain a fused feature; The fusion features are input into a preset machine learning model to obtain welding fault prediction results.

2. The fault prediction method according to claim 1, wherein: The multi-channel temporal convolutional network has a residual connection structure, which is used to transfer information across layers in the multi-channel temporal convolutional network.

3. The fault prediction method according to claim 1, wherein: The feature fusion method based on the attention mechanism fuses each of the time series features to obtain the fused features including: Performing global information extraction on each of the time series features using global maximum pooling and global average pooling to obtain first global information and second global information; Inputting the first global information and the second global information into a preset multilayer perceptron to obtain a first output feature and a second output feature respectively; performing element-by-element summation on the first output feature and the second output feature to obtain a first merged feature; Normalizing the first combined features to obtain second combined features; The second combined feature is multiplied element-by-element by the first combined feature to obtain the fused feature.

4. The fault prediction method according to claim 1, wherein: The feature fusion method based on the attention mechanism fuses each of the time series features to obtain the fused features including: Each of the time series features is weightedly fused based on a preset attention weight matrix to obtain the fused feature. The attention weight matrix is ​​automatically learned through a preset neural network to adjust the weight in real time according to the input data.

5. The fault prediction method according to claim 1, wherein: Before extracting the time series features of the welding state parameters of each mode based on the preset multi-channel time convolution network, the method further includes: The welding state parameters are subjected to denoising, normalization and data cleaning in sequence.

6. The fault prediction method according to claim 5, wherein: The data cleaning of the welding state parameters includes: For any parameter of the welding state parameters, if the parameter exceeds the corresponding preset parameter range, or the parameter exceeds the corresponding preset range, then the parameter is eliminated; or Data cleaning of the welding state parameters is performed based on a density clustering algorithm.

7. The fault prediction method according to claim 5, wherein: After sequentially performing denoising, normalization, and data cleaning on the welding state parameters, the method further includes: The welding state parameters of each mode are converted into encoding vectors of uniform data dimensions and data alignment is performed.

8. A fault prediction device, characterized in that: The fault prediction device comprises: A data acquisition unit, used for acquiring multi-modal welding state parameters of target welding equipment in real time; A multi-channel feature extraction unit, configured to extract the time series features of the welding state parameters of each mode based on a preset multi-channel time convolutional network; A feature fusion unit, configured to fuse each of the time series features based on an attention mechanism feature fusion method to obtain a fused feature; The fault prediction unit is used to input the fusion features into a preset machine learning model to obtain a welding fault prediction result.

9. A fault prediction device, characterized in that: The system comprises a processor and a memory, wherein the memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the fault prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the fault prediction method according to any one of claims 1 to 7.

Images