Method and system for analyzing performance of a metallized film flexible capacitor

Abstract

The application provides a performance analysis method and system of a metalized thin film flexible capacitor, and relates to the technical field of artificial intelligence.In the application, firstly, a target deposition thickness sequence formed in a plating process of a target capacitor is acquired; secondly, first deposition thickness data and a comparison deposition thickness sequence are determined from the target deposition thickness sequence; then, prediction processing is performed based on the first deposition thickness data and plating deposition parameters of the metalized thin film, so that a predicted deposition thickness sequence is obtained; and finally, the predicted deposition thickness sequence and the comparison deposition thickness sequence are compared and analyzed, so that metalized thin film performance data of the target capacitor is obtained.Based on the above, the problem that the reliability of capacitor performance analysis in the prior art is relatively low can be improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a performance analysis method and system for metallized thin-film flexible DC capacitors. Background Technology

[0002] Manufacturing a metallized thin-film flexible DC capacitor, especially a DC-supported capacitor, hinges on improving its energy zoning design to enhance stability and safety. This involves specific design considerations such as the microstructure design of the base film and metal layer, the edge structure design of the metallized film, and microstructure deposition control methods. Precise microstructure deposition control is one of the core technologies for improving the performance of metallized thin-film capacitors. For example, the uniformity of the metal layer on the film can avoid localized electric field concentration and reduce breakdown. Thus, the performance of the capacitor can be determined by analyzing the uniformity of the metal layer; better uniformity indicates better performance. However, in existing technologies, the uniformity of the metal layer is typically tested after deposition. While this allows for efficient performance testing, relying solely on post-deposition testing results may lead to relatively low reliability. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a performance analysis method and system for metallized thin-film flexible capacitors, so as to improve the problem of relatively low reliability of capacitor performance analysis in the prior art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A performance analysis method for a metallized thin-film flexible DC capacitor includes:

[0006] The target deposition thickness sequence formed during the coating process of the target capacitor is obtained, wherein the target deposition thickness sequence includes multiple deposition thickness data corresponding to multiple deposition stages, and each deposition thickness data is used to characterize the thickness of at least one point of the metallization film deposited in the current stage.

[0007] A first deposition thickness data and a comparative deposition thickness sequence are determined from the target deposition thickness sequence, wherein each deposition thickness data in the comparative deposition thickness sequence is located after the first deposition thickness data;

[0008] Based on the first deposition thickness data and the coating deposition parameters of the metallized thin film, a prediction processing is performed to obtain a predicted deposition thickness sequence, wherein each deposition thickness data in the predicted deposition thickness sequence belongs to the prediction result of the deposition stage after the first deposition thickness data.

[0009] By comparing and analyzing the predicted deposition thickness sequence and the comparative deposition thickness sequence, the metallization film performance data of the target capacitor are obtained.

[0010] In a preferred embodiment of the present invention, in the above-described performance analysis method for a metallized thin-film flexible DC capacitor, the step of performing prediction processing based on the first deposition thickness data and the deposition parameters of the metallized thin film to obtain a predicted deposition thickness sequence includes:

[0011] Semantic mining is performed on the first deposition thickness data to output A deposition thickness semantic vectors, wherein the deposition thickness semantic vectors are used to reflect the potential semantic information of the first deposition thickness data;

[0012] The temporal adjustment parameters corresponding to each of the deposition thickness semantic vectors are determined, and based on the temporal adjustment parameters corresponding to each of the deposition thickness semantic vectors, the corresponding deposition thickness semantic vectors in the A deposition thickness semantic vectors are semantically adjusted to output A deposition thickness adjustment vectors. Each of the deposition thickness adjustment vectors is used to predict a deposition thickness data in the required predicted deposition thickness sequence, and the semantic adjustment is used to control the changes between each deposition thickness data in the predicted deposition thickness sequence and the first deposition thickness data.

[0013] Based on the coating deposition parameters, the A deposition thickness adjustment vectors are predicted and the corresponding predicted deposition thickness sequence is output, wherein the predicted deposition thickness sequence includes A deposition thickness data.

[0014] In a preferred embodiment of the present invention, in the above-described performance analysis method for metallized thin-film flexible capacitors, the step of performing semantic mining on the first deposition thickness data and outputting A deposition thickness semantic vectors includes:

[0015] The first deposition thickness data is embedded in a vector space to form a corresponding deposition thickness embedding vector, and the deposition thickness embedding vector is subjected to self-attention processing to form a corresponding deposition thickness semantic vector.

[0016] The deposition thickness semantic vector is expanded to form A deposition thickness semantic vectors, wherein every two deposition thickness semantic vectors in the A deposition thickness semantic vectors are the same.

[0017] In a preferred embodiment of the present invention, in the above-described performance analysis method for metallized thin-film flexible DC capacitors, the steps of determining the timing adjustment parameter corresponding to each of the deposition thickness semantic vectors, and semantically adjusting the corresponding deposition thickness semantic vectors among the A deposition thickness semantic vectors based on the timing adjustment parameter corresponding to each of the deposition thickness semantic vectors, and outputting A deposition thickness adjustment vectors, include:

[0018] A pre-configured reference adjustment parameter is determined, wherein the reference adjustment parameter is used to reflect the magnitude of the difference between adjacent deposition thickness data in the predicted deposition thickness sequence;

[0019] A training adjustment parameters carried in the target thickness prediction network are obtained, and A time-series adjustment parameters are determined based on the reference adjustment parameters and the A training adjustment parameters. The A time-series adjustment parameters and the A deposition thickness semantic vectors have a one-to-one correspondence, and the A training adjustment parameters are formed during the training process of the target thickness prediction network, which belongs to a neural network model.

[0020] Based on each of the A time-series adjustment parameters, the corresponding deposition thickness semantic vector is mapped and adjusted to output A deposition thickness adjustment vectors.

[0021] In a preferred embodiment of the present invention, in the above-described performance analysis method for metallized thin-film flexible capacitors, the step of predicting the A deposition thickness adjustment vectors based on the deposition parameters and outputting the corresponding predicted deposition thickness sequence includes:

[0022] A potential semantic vectors for deposition are determined. The potential semantic vectors for deposition are formed by perturbing the potential semantic vectors in the semantic space where the deposition thickness is located during training. The perturbed vectors follow a normal distribution, and one of the potential semantic vectors for deposition is used to predict a deposition thickness data in the required predicted deposition thickness sequence.

[0023] The deposition parameters are semantically mined to form corresponding deposition parameter semantic vectors. The deposition parameter semantic vectors are used to reflect the semantic information of the deposition parameters. The deposition parameters include at least one of the following: deposition distance, gas flow rate, deposition temperature, deposition bias voltage, deposition atmosphere, sputtering power, and target current.

[0024] The target thickness prediction network predicts the deposition parameter semantic vector and the A potential deposition semantic vectors, and fuses the A deposition thickness adjustment vectors during the prediction process to output the corresponding predicted deposition thickness sequence. The potential semantic vectors in the semantic space of the deposition thickness formed during training are formed during the training process of the target thickness prediction network, which is a neural network model.

[0025] In a preferred embodiment of the present invention, in the above-described performance analysis method for metallized thin-film flexible C capacitors, the step of predicting the deposition parameter semantic vector and the A potential deposition semantic vectors using a target thickness prediction network, and fusing the A deposition thickness adjustment vectors during the prediction process to output the corresponding predicted deposition thickness sequence, includes:

[0026] For each of the A sedimentation potential semantic vectors, the corresponding semantic vector to be predicted is determined based on the sedimentation potential semantic vector;

[0027] The semantic fusion model in the target thickness prediction network is used to fuse the deposition parameter semantic vector and one deposition thickness adjustment vector corresponding to the potential deposition semantic vector among the A deposition thickness adjustment vectors into the semantic vector to be predicted, and output the multi-dimensional fusion vector corresponding to the potential deposition semantic vector.

[0028] The multi-dimensional fusion vector is predicted by the prediction output model in the target thickness prediction network, and a deposition thickness data corresponding to a deposition potential semantic vector is output. The deposition thickness data corresponding to the A deposition potential semantic vectors are combined to form a corresponding predicted deposition thickness sequence.

[0029] In a preferred embodiment of the present invention, in the performance analysis method for the above-mentioned metallized thin-film flexible DC capacitor, the step of determining the corresponding semantic vector to be predicted based on each of the A deposition potential semantic vectors includes:

[0030] The first of the A potential semantic vectors is determined as the corresponding semantic vector to be predicted.

[0031] For each of the A sedimentation latent semantic vectors other than the first sedimentation latent semantic vector, the corresponding concatenated semantic vector is formed by concatenating the concatenated semantic vector with the multi-dimensional fusion vector corresponding to the previous sedimentation latent semantic vector. Then, the concatenated semantic vector is downsampled to form the corresponding semantic vector to be predicted.

[0032] In a preferred embodiment of the present invention, in the performance analysis method for the above-mentioned metallized thin-film flexible C capacitor, the step of fusing the deposition parameter semantic vector and one deposition thickness adjustment vector corresponding to the potential deposition semantic vector among the A deposition thickness adjustment vectors into the semantic vector to be predicted through the semantic fusion model in the target thickness prediction network, and outputting the multi-dimensional fusion vector corresponding to the potential deposition semantic vector, includes:

[0033] The semantic fusion model in the target thickness prediction network includes a first mining branch and a second mining branch, which respectively perform salient feature mining on the semantic vector to be predicted, and output the corresponding first salient semantic vector and second salient semantic vector.

[0034] The deposition parameter semantic vector is fused into the first saliency semantic vector to obtain the corresponding first fused semantic vector;

[0035] One deposition thickness adjustment vector corresponding to the deposition potential semantic vector among the A deposition thickness adjustment vectors is fused into the second saliency semantic vector to obtain the corresponding second fused semantic vector;

[0036] The first fused semantic vector and the second fused semantic vector are added together to output the multi-dimensional fused vector corresponding to the deposited potential semantic vector.

[0037] In a preferred embodiment of the present invention, in the above-described performance analysis method for metallized thin-film flexible DC capacitors, the step of fusing the deposition parameter semantic vector into the first saliency semantic vector to obtain the corresponding first fused semantic vector includes:

[0038] Multiple semantic extractions are performed on the first salient semantic vector in a cascaded manner to form multiple corresponding first extracted semantic vectors. For each semantic extraction, the first extracted semantic vector corresponding to the previous semantic extraction and the deposition parameter semantic vector are concatenated, and the concatenated semantic vector is subjected to self-attention processing. The semantic vector obtained by self-attention processing is then downsampled to form the first extracted semantic vector corresponding to the current semantic extraction. The first extracted semantic vector corresponding to the previous semantic extraction of the first semantic extraction is the first salient semantic vector.

[0039] Multiple semantic diffusions are cascaded on the last first extracted semantic vector to form multiple corresponding first diffuse semantic vectors. For each semantic diffusion, the first diffuse semantic vector corresponding to the previous semantic diffusion is concatenated with a first extracted semantic vector at the corresponding level. The concatenated semantic vector is then subjected to self-attention processing, and the semantic vector obtained from the self-attention processing is then upsampled to form the first diffuse semantic vector corresponding to the current semantic diffusion. The first diffuse semantic vector corresponding to the semantic diffusion preceding the first semantic diffusion is the last first extracted semantic vector.

[0040] The last first diffusion semantic vector is determined as the first fusion semantic vector.

[0041] Based on the above, the present invention also provides a performance analysis system for metallized thin-film flexible DC capacitors, comprising:

[0042] Memory, used to store computer programs;

[0043] A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-described performance analysis method for metallized thin-film flexible DC capacitors.

[0044] This invention provides a method and system for performance analysis of a metallized thin-film flexible CRT capacitor. First, a target deposition thickness sequence is obtained during the deposition process of the target capacitor. Second, a first deposition thickness data and a comparative deposition thickness sequence are determined from the target deposition thickness sequence. Then, based on the first deposition thickness data and the deposition parameters of the metallized thin film, a prediction processing is performed to obtain a predicted deposition thickness sequence. Finally, the predicted deposition thickness sequence and the comparative deposition thickness sequence are compared and analyzed to obtain the metallized thin film performance data of the target capacitor. Based on the above, by predicting the corresponding deposition thickness sequence, it is possible to compare and analyze it as a theoretical value with a comparative deposition thickness sequence as an actual value. This allows for comparison of data at each deposition stage, ensuring a comprehensive analysis of the dynamic process across multiple deposition stages. Therefore, compared to the conventional method of comparing data after deposition is completed, this method is more comprehensive and has higher reliability, thus improving the relatively low reliability of capacitor performance analysis in existing technologies. In addition, since the predicted sedimentary thickness sequence references the first sedimentary thickness data, the initial sedimentary thickness data of the predicted sedimentary thickness sequence and the comparison sedimentary thickness sequence are the same, that is, the initial conditions are the same, which makes the comparison accuracy of subsequent sedimentary thickness data higher and focuses more on the dynamic changes in the subsequent sedimentation process. Attached Figure Description

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0046] Figure 1 This is a structural block diagram of the performance analysis system for a metallized thin-film flexible DC capacitor provided in an embodiment of the present invention.

[0047] Figure 2 This is a schematic flowchart illustrating the performance analysis method for a metallized thin-film flexible DC capacitor provided in an embodiment of the present invention.

[0048] Figure 3 This is a schematic diagram illustrating the fusion process for forming a multi-dimensional fusion vector, as provided in an embodiment of the present invention.

[0049] Figure 4 This is a schematic diagram illustrating the multi-level fusion of semantic vectors provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0052] like Figure 1 As shown, this embodiment of the invention provides a performance analysis system for metallized thin-film flexible CRT capacitors. The performance analysis system may include a memory, a processor, and a performance analysis device for metallized thin-film flexible CRT capacitors.

[0053] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The performance analysis device for the metallized thin-film flexible C capacitor includes at least one software functional module stored in the memory in the form of software or firmware. The processor is used to execute executable computer programs stored in the memory, such as the software functional modules and computer programs included in the performance analysis device for the metallized thin-film flexible C capacitor, to implement the performance analysis method for the metallized thin-film flexible C capacitor provided in this embodiment of the invention.

[0054] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0055] Furthermore, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0056] Furthermore, the performance analysis device for the metallized thin-film flexible CFC may include multiple modules, such as a first module, a second module, a third module, and a fourth module (all of which are software functional modules). The first module is used to acquire a target deposition thickness sequence formed during the coating process of the target capacitor. This target deposition thickness sequence includes multiple deposition thickness data corresponding to multiple deposition stages, with each deposition thickness data characterizing the thickness of at least one point of the metallized thin film deposited in the current stage. The second module is used to determine a first deposition thickness data and a comparative deposition thickness sequence from the target deposition thickness sequence, where each deposition thickness data in the comparative deposition thickness sequence is located after the first deposition thickness data. The third module is used to perform predictive processing based on the first deposition thickness data and the deposition parameters of the metallized thin film to obtain a predicted deposition thickness sequence, where each deposition thickness data in the predicted deposition thickness sequence is a prediction result for the deposition stage following the first deposition thickness data. The fourth module is used to compare and analyze the predicted deposition thickness sequence and the comparative deposition thickness sequence to obtain the metallized thin film performance data of the target capacitor.

[0057] Understandable. Figure 1 The structure shown is for illustrative purposes only. The performance analysis system for the metallized thin-film flexible DC capacitor may also include a ratio Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices, such as devices for acquiring deposition thickness.

[0058] Combination Figure 2 This invention also provides a performance analysis method for a metallized thin-film flexible CRT capacitor applicable to the aforementioned performance analysis system. The method steps defined in the process flow of the performance analysis method for the metallized thin-film flexible CRT capacitor can be implemented by the performance analysis system (hereinafter referred to as the performance analysis system). The following will describe... Figure 2 The specific process shown will be explained in detail.

[0059] Step S110: Obtain the target deposition thickness sequence formed by the target capacitor during the coating process.

[0060] In this embodiment of the invention, the performance analysis system can acquire the target deposition thickness sequence formed during the coating process of the target capacitor. The target deposition thickness sequence includes multiple deposition thickness data points corresponding to multiple deposition stages (which can be arranged in chronological order). Each deposition thickness data point characterizes the thickness of at least one point of the metallized thin film deposited in the current stage (the specific number of points can be selected according to actual needs; for example, more points can be used to ensure accuracy, while fewer points can be used to reduce computational load). Furthermore, the deposition thickness data can be obtained by measuring the reflectivity of the corresponding points using a spectroreflectometer. Since different reflectivities reflect different thicknesses, the measured reflectivity can characterize the corresponding thickness. Thus, by deploying multiple spectroreflectometers (e.g., forming a corresponding detector array), the thickness of multiple points can be monitored. Alternatively, in other embodiments, the film deposition thickness can also be monitored by monitoring changes in the oscillation frequency using a quartz crystal microbalance (QCM), thereby obtaining the corresponding target deposition thickness sequence.

[0061] Step S120: Determine the first deposition thickness data and the comparative deposition thickness sequence from the target deposition thickness sequence.

[0062] In this embodiment of the invention, after obtaining the target deposition thickness sequence, the performance analysis system can determine a first deposition thickness data and a comparative deposition thickness sequence from the target deposition thickness sequence. Each deposition thickness data in the comparative deposition thickness sequence is located after the first deposition thickness data. For example, the first deposition thickness data in the target deposition thickness sequence can be determined as the first deposition thickness data, and then the other deposition thickness data in the target deposition thickness sequence, excluding the first deposition thickness data, can be sorted sequentially to form a corresponding comparative deposition thickness sequence.

[0063] Step S130: Based on the first deposition thickness data and the coating deposition parameters of the metallized thin film, a prediction processing is performed to obtain a predicted deposition thickness sequence.

[0064] In this embodiment of the invention, after determining the first deposition thickness data, the performance analysis system can perform predictive processing based on the first deposition thickness data and the deposition parameters of the metallized thin film to obtain a predicted deposition thickness sequence. Each deposition thickness data point in the predicted deposition thickness sequence is a prediction result for a deposition stage following the first deposition thickness data, i.e., it corresponds one-to-one with the deposition thickness data in the comparative deposition thickness sequence.

[0065] Step S140: Compare and analyze the predicted deposition thickness sequence and the comparative deposition thickness sequence to obtain the metallization film performance data of the target capacitor.

[0066] In this embodiment of the invention, after obtaining the predicted deposition thickness sequence, the performance analysis system can compare and analyze the predicted deposition thickness sequence and the comparative deposition thickness sequence to obtain the metallization film performance data of the target capacitor. For example, the difference between the corresponding deposition thickness data in the predicted deposition thickness sequence and the comparative deposition thickness sequence is calculated, and then the metallization film performance data is determined based on the obtained difference. For instance, a smaller difference indicates a smaller difference between the predicted result and the actual data, suggesting that no abnormalities occurred during the actual deposition process, resulting in a stable deposition process. Therefore, the corresponding deposition effect is as expected, i.e., it has better performance. Conversely, a larger difference indicates that abnormalities may have occurred during the actual deposition process, causing problems in the deposition process. Therefore, the corresponding deposition effect may not meet expectations, i.e., the performance may be poor. Furthermore, the difference can refer to the average of the absolute differences between the deposition thickness values ​​at each point in the two sequences. In addition, the obtained metallization film performance data can be used as a preliminary screening condition. For example, after screening out capacitors with poor performance based on the metallization film performance data, the performance of the capacitor can be further tested in practice, such as by performing a breakdown test, in order to avoid the problem of waste caused by misjudgment.

[0067] Based on the above, by predicting the corresponding deposition thickness sequence, it can be used as a theoretical value to compare with the actual deposition thickness sequence for analysis. This allows for comparison of data at each deposition stage, ensuring a comprehensive analysis of the dynamic process across multiple deposition stages. Therefore, compared to the conventional method of comparing data after deposition is complete, this approach is more comprehensive and reliable, thus addressing the relatively low reliability of capacitor performance analysis in existing technologies. Furthermore, since the predicted deposition thickness sequence references the first deposition thickness data, the initial deposition thickness data of both the predicted and compared sequences are identical, meaning the initial conditions are the same. This results in higher accuracy in comparing subsequent deposition thickness data, focusing more on the dynamic changes during the subsequent deposition process.

[0068] In this embodiment of the invention, the specific implementation of the prediction processing based on the first deposition thickness data and the coating deposition parameters of the metallized thin film in step S120 is not limited and can be selected according to actual needs.

[0069] For example, in an alternative implementation, the first deposition thickness data and the coating deposition parameters can be semantically mined, and then, based on the corresponding semantic mining results, the corresponding predicted deposition thickness sequence can be predicted.

[0070] For example, in another alternative implementation, in order to improve the reliability of the prediction process, the above step S120 may further include the following steps S121, S122 and S123, the details of which are described below.

[0071] Step S121: Perform semantic mining on the first deposition thickness data and output A deposition thickness semantic vectors.

[0072] In this embodiment of the invention, semantic mining can be performed on the first deposition thickness data to output A deposition thickness semantic vectors. These deposition thickness semantic vectors reflect the potential semantic information inherent in the first deposition thickness data. In other words, corresponding potential semantic information can be mined from the first deposition thickness data and represented in vector form to obtain the A deposition thickness semantic vectors. Here, A equals the number of deposition thickness data points in the comparative deposition thickness sequence, or in other words, the number of deposition thickness data points in the predicted deposition thickness sequence to be output.

[0073] Step S122: Determine the timing adjustment parameter corresponding to each of the deposition thickness semantic vectors, and based on the timing adjustment parameter corresponding to each of the deposition thickness semantic vectors, perform semantic adjustment on the corresponding deposition thickness semantic vectors in the A deposition thickness semantic vectors, and output A deposition thickness adjustment vectors.

[0074] In this embodiment of the invention, after obtaining the A sedimentation thickness semantic vectors, a temporal adjustment parameter corresponding to each sedimentation thickness semantic vector can be determined. Based on the temporal adjustment parameter corresponding to each sedimentation thickness semantic vector, the corresponding sedimentation thickness semantic vectors among the A sedimentation thickness semantic vectors are semantically adjusted, outputting A sedimentation thickness adjustment vectors. Each sedimentation thickness adjustment vector is used to predict a sedimentation thickness data point in the required predicted sedimentation thickness sequence. The semantic adjustment is used to control the changes between each sedimentation thickness data point in the predicted sedimentation thickness sequence and the first sedimentation thickness data point. That is, since each sedimentation thickness data point in the predicted sedimentation thickness sequence cannot be the same as the first sedimentation thickness data point, and the sedimentation thickness gradually increases over time (either further along the sequence or further along the deposition stage), the magnitude of the change between the sedimentation thickness data point and the first sedimentation thickness data point becomes increasingly larger. Therefore, it is necessary to determine the corresponding temporal adjustment parameter so that the sedimentation thickness adjustment vector obtained after semantically adjusting the sedimentation thickness semantic vector based on the corresponding temporal adjustment parameter better matches the actual sedimentation thickness, i.e., improving the accuracy of the semantic representation.

[0075] Step S123: Based on the coating deposition parameters, perform prediction processing on the A deposition thickness adjustment vectors and output the corresponding predicted deposition thickness sequence.

[0076] In this embodiment of the invention, after obtaining the A deposition thickness adjustment vectors, the A deposition thickness adjustment vectors can be predicted based on the coating deposition parameters to output a corresponding predicted deposition thickness sequence. The predicted deposition thickness sequence includes A deposition thickness data points. That is, since different deposition progress directly leads to different deposition thicknesses, it is necessary to combine the actual coating deposition parameters for prediction processing to further improve the reliability of the output predicted deposition thickness sequence.

[0077] Alternatively, the specific implementation of semantic mining of the first deposition thickness data in step S121 is not limited. For example, in an alternative implementation, considering that there is generally a correlation between deposition thicknesses at different locations, in order to make the mined deposition thickness semantic vector have a high semantic representation capability, step S121 may further include the following implementable content:

[0078] The first step involves embedding the first deposition thickness data into a vector space to form a corresponding deposition thickness embedding vector. Then, self-attention processing is applied to the deposition thickness embedding vector to form a corresponding deposition thickness semantic vector. Vector space embedding can be implemented using a corresponding word embedding model. For example, the first deposition thickness data may include the deposition thickness at B points (and in some implementations, the location coordinates of these points). Thus, word embedding processing can be performed on each deposition thickness (and location coordinates) to obtain a corresponding word embedding vector. Finally, the B word embedding vectors are combined to form the corresponding deposition thickness embedding vector. The input vector (for example, the dimension of each word embedding vector can be 1*C, and the dimension of the deposition thickness embedding vector can be B*C). In addition, since there are also correlations between the points, the correlations between the points can be mined by performing self-attention processing on the deposition thickness embedding vector carrying the semantic information of the deposition thickness of each point. Furthermore, the result of the self-attention processing can be superimposed on the deposition thickness embedding vector to obtain a deposition thickness semantic vector that can represent both the semantic information of the deposition thickness and the correlation between the points. In this way, the semantic information represented by the deposition thickness semantic vector can be made richer.

[0079] The second step is to expand the deposition thickness semantic vector to form A deposition thickness semantic vectors, wherein every two deposition thickness semantic vectors in the A deposition thickness semantic vectors are the same. In other words, based on the deposition thickness semantic vectors, A-1 more deposition thickness semantic vectors can be copied to obtain A identical deposition thickness semantic vectors.

[0080] Alternatively, the specific implementation of outputting A deposition thickness adjustment vectors in step S122 is not limited. For example, in an alternative implementation, in order to ensure that the output A deposition thickness adjustment vectors can reliably characterize the deposition thickness at each stage or time sequence, step S122 may further include the following implementable content:

[0081] The first step is to determine the pre-configured reference adjustment parameters, which are used to reflect the difference between adjacent deposition thickness data in the predicted deposition thickness sequence. The specific value of the reference adjustment parameters can be configured according to the actual situation. For example, the longer the duration between two adjacent stages, the greater the corresponding difference, that is, the smaller the value of the reference adjustment parameters (belonging to 0-1).

[0082] The second step involves obtaining A training adjustment parameters carried in the target thickness prediction network, and determining A temporal adjustment parameters based on the reference adjustment parameters and the A training adjustment parameters. There is a one-to-one correspondence between the A temporal adjustment parameters and the A deposition thickness semantic vectors, and the A training adjustment parameters are formed during the training process of the target thickness prediction network, which is a neural network model. For example, the training adjustment parameters can be a parameter distribution matrix with the same size as the deposition thickness semantic vector. Thus, each parameter in the training adjustment parameters can be multiplied by the reference adjustment parameters to obtain the corresponding temporal adjustment parameter.

[0083] The third step involves mapping and adjusting the corresponding deposition thickness semantic vector based on each of the A timing adjustment parameters to output A deposition thickness adjustment vectors. For example, since the size of the timing adjustment parameter is the same as the size of the deposition thickness semantic vector, the timing adjustment parameter and the deposition thickness semantic vector can be multiplied bitwise to obtain the corresponding deposition thickness adjustment vector. For instance, the parameter in the first row and first column of the timing adjustment parameter can be multiplied with the parameter in the first row and first column of the deposition thickness semantic vector to obtain the parameter in the first row and first column of the deposition thickness adjustment vector.

[0084] Alternatively, the specific implementation of the prediction processing of the A deposition thickness adjustment vectors in step S123 is not limited. For example, in an alternative implementation, in order to avoid overfitting during the prediction process and to improve the reliability of the prediction process, step S123 may further include the following steps S123a, S123b and S123c, the details of which are as follows.

[0085] Step S123a: Determine A potential semantic vectors for deposition.

[0086] In this embodiment of the invention, A latent semantic vectors for deposition can be determined. These latent semantic vectors are formed by perturbing the latent semantic vectors within the semantic space containing the deposition thickness generated during training. The perturbed vector follows a normal distribution (with a mean of 0), and each latent semantic vector is used to predict a deposition thickness data point in the desired predicted deposition thickness sequence. For example, a random vector following a normal distribution can be generated as the perturbed vector, and then this perturbed vector can be added to the latent semantic vectors within the semantic space containing the deposition thickness (with the same size). It should be noted that in the initial stage of training, A latent semantic vectors can be randomly generated and used as network parameters of the corresponding neural network model, continuously updated during training to form the final latent semantic vector. Since the final latent semantic vector is formed through training, semantic information within the semantic space containing the deposition thickness can be learned. Thus, by using the A latent semantic vectors for deposition, both perturbation can be introduced to improve overfitting, and semantic information within the semantic space containing the deposition thickness can be introduced, avoiding semantic distortion caused by the introduction of perturbation.

[0087] Step S123b: Semantic mining is performed on the coating deposition parameters to form corresponding deposition parameter semantic vectors.

[0088] In this embodiment of the invention, the coating deposition parameters can be semantically mined to form a corresponding deposition parameter semantic vector. The deposition parameter semantic vector reflects the semantic information of the film deposition parameters. These parameters include at least the deposition distance (target-substrate distance, which affects the angle, velocity, and density of particles reaching the substrate during sputtering; for example, if the target and substrate are too close, the substrate receives more sputtered particles, leading to an excessively thick film), gas flow rate (sputtering gas flow rate, the flow rate of the gas used during sputtering (usually argon), which affects the interaction between the gas and the target, as well as the ionization rate during sputtering; for example, when using argon as the sputtering gas, a lower flow rate may result in a lower sputtering rate and an inhomogeneous film, while an excessively high flow rate may generate more ion contamination), deposition temperature (substrate temperature, which significantly affects the film's crystal structure, compositional uniformity, and density), deposition bias (bias voltage, applied to the substrate during sputtering to regulate the energy of the deposited particles), and deposition atmosphere (deposition atmosphere). Atmosphere, the gas environment used in the sputtering process (e.g., vacuum, argon, oxygen, etc., different atmospheres affect the composition and crystallization state of the film), sputtering power (affecting the density and energy of sputtered particles), and target current (the higher the current, the higher the energy of the sputtered particles and the faster the deposition rate) are at least one of the following. Other deposition parameters may also be included in other embodiments. For example, word embedding processing can be performed on each deposition parameter to obtain corresponding word embedding vectors. Then, the word embedding vectors can be combined to form a corresponding combined vector. Finally, self-attention processing can be performed on the combined vectors to enable semantic information mining of each deposition parameter. The combined vector and the result of the self-attention processing are then added to obtain the corresponding deposition parameter vector. For example, the size of each word embedding vector can be 1*X. If there are Y deposition parameters, the size of the deposition parameter semantic vector can be Y*X. For example, for the deposition atmosphere "argon," word embedding processing can yield the corresponding word embedding vector:

[0089] [0.53, 0.85, 0.42, 0.43, 0.10, 0.66, 0.98, 0.83, 0.66, 0.77, 0.86, 0.37, 0.81, 0.15, 0.00, 0.74, 0.45, 0.72, 0.59, ..., 0.76].

[0090] For example, for the sputtering power "100 W", word embedding processing can yield the corresponding word embedding vector:

[0091] [0.23, 0.54, 0.21, 0.11, 0.72, 0.63, 0.39, 0.46, 0.31, 0.26, 0.13, 0.98, 0.04, 0.87, 0.4, 0.18, 0.5, 0.77, 0.98, ..., 0.21].

[0092] Based on this, by combining them, we can obtain the corresponding combined vector:

[0093] {[0.53, 0.85, 0.42, 0.43, 0.10, 0.66, 0.98, 0.83, 0.66, 0.77, 0.86, 0.37, 0.81, 0.15, 0.00, 0.74, 0.45, 0.72, 0.59, ..., 0.76];

[0094] [0.23, 0.54, 0.21, 0.11, 0.72, 0.63, 0.39, 0.46, 0.31, 0.26, 0.13, 0.98, 0.04, 0.87, 0.4, 0.18, 0.5, 0.77, 0.98, ..., 0.21]}.

[0095] Step S123c: The target thickness prediction network is used to predict the deposition parameter semantic vector and the A potential deposition semantic vectors, and the A deposition thickness adjustment vectors are fused during the prediction process to output the corresponding predicted deposition thickness sequence.

[0096] In this embodiment of the invention, after obtaining the deposition parameter semantic vector and the A potential deposition semantic vectors, a target thickness prediction network can be used to predict the deposition parameter semantic vector and the A potential deposition semantic vectors. During the prediction process, the A deposition thickness adjustment vectors are fused to output the corresponding predicted deposition thickness sequence. The potential semantic vectors within the semantic space of the deposition thickness, formed during training, are generated during the training of the target thickness prediction network, which is a neural network model. That is, since the A potential deposition semantic vectors contain perturbation information, the perturbation information can be removed by guiding or fusing the deposition parameter semantic vector and the A deposition thickness adjustment vectors, thereby obtaining semantic vectors with better semantic representation capabilities. This results in a relatively high reliability of the predicted deposition thickness sequence based on these semantic vectors.

[0097] Alternatively, the specific implementation of the output of the corresponding predicted deposition thickness sequence for step S123c is not limited. For example, in an alternative implementation, considering that there is a correlation between adjacent deposition thickness data in the predicted deposition thickness sequence, in order to ensure the effective realization of this correlation and make the reliability of the obtained predicted deposition thickness sequence high, step S123c may further include steps c1, c2 and c3. The details of each step are as follows.

[0098] Step c1: For each of the A sedimentation potential semantic vectors, determine the corresponding semantic vector to be predicted based on the sedimentation potential semantic vector.

[0099] In this embodiment of the invention, for each of the A latent semantic vectors, a corresponding semantic vector to be predicted is determined based on that latent semantic vector. For example, a first semantic vector to be predicted can be determined based on a first latent semantic vector, and a second semantic vector to be predicted can be determined based on a second latent semantic vector. Thus, for A latent semantic vectors, A semantic vectors to be predicted can be determined.

[0100] Step c2: Using the semantic fusion model in the target thickness prediction network, the semantic vector of the deposition parameters and one deposition thickness adjustment vector corresponding to the potential semantic vector of deposition among the A deposition thickness adjustment vectors are fused into the semantic vector to be predicted, and the multi-dimensional fusion vector corresponding to the potential semantic vector of deposition is output.

[0101] In this embodiment of the invention, after obtaining the corresponding semantic vector to be predicted, the semantic fusion model in the target thickness prediction network can be used to fuse the deposition parameter semantic vector and one deposition thickness adjustment vector corresponding to the potential semantic vector among the A deposition thickness adjustment vectors into the semantic vector to be predicted, outputting a multi-dimensional fusion vector corresponding to the potential semantic vector. That is, the semantic fusion model can be an encoding network model that, during the encoding and mining process of the potential semantic vector (i.e., the semantic vector to be predicted), achieves the fusion of the deposition parameter semantic vector and the deposition thickness adjustment vector, thereby enabling the fusion of multi-dimensional semantic information while removing perturbation information. Based on this, for A semantic vectors to be predicted, A multi-dimensional fusion vectors can be obtained.

[0102] Step c3: The multi-dimensional fusion vector is predicted using the prediction output model in the target thickness prediction network to output a deposition thickness data corresponding to a deposition potential semantic vector. The A deposition thickness data corresponding to the A deposition potential semantic vectors are combined to form a corresponding predicted deposition thickness sequence.

[0103] In this embodiment of the invention, after obtaining the multi-dimensional fusion vector, the multi-dimensional fusion vector can be predicted using the prediction output model in the target thickness prediction network to output a deposition thickness data corresponding to the deposition latent semantic vector. Furthermore, the A deposition thickness data corresponding to the A deposition latent semantic vectors are combined to form a corresponding predicted deposition thickness sequence. That is, the prediction output model can be a decoding network model, for example, it can include a fully connected network layer. In this way, the multi-dimensional fusion vector can be fully connected to obtain the corresponding deposition thickness data. Alternatively, it can include a linear mapping function to linearly map the vector obtained from the fully connected processing to obtain the corresponding deposition thickness data. For example, the deposition thickness data can be the deposition thickness at 5 points. Thus, through fully connected processing, a vector of size 1*5 can be obtained, where the 5 parameters correspond to the deposition thickness at 5 points. In other embodiments, self-attention processing can also be performed after the fully connected processing.

[0104] Alternatively, the specific implementation of step c1 above, which determines the semantic vector to be predicted, is not limited. For example, in an alternative implementation, in order to fully consider the influence of the preceding deposition thickness on the subsequent deposition thickness and to ensure that the semantic representation accuracy of the determined semantic vector to be predicted is high, step c1 above may further include the following implementable content:

[0105] The first step is to determine the first sedimentation potential semantic vector among the A sedimentation potential semantic vectors as the corresponding semantic vector to be predicted. In other words, for the first sedimentation potential semantic vector, it can be directly used as the semantic vector to be predicted.

[0106] The second step involves concatenating each of the A latent semantic vectors (excluding the first one) with the corresponding multi-dimensional fusion vector of the previous latent semantic vector to form a concatenated semantic vector. Then, the concatenated semantic vector is downsampled to form a corresponding semantic vector to be predicted. The size of the semantic vector to be predicted can be the same as the size of the latent semantic vectors. Downsampling can be implemented using methods such as convolution and pooling. For example, the second latent semantic vector can be concatenated with the multi-dimensional fusion vector of the first latent semantic vector to form the concatenated semantic vector. In other words, at the second time step, not only the latent semantic vector at the current time step but also the output semantic vector at the first time step are considered.

[0107] Alternatively, the specific implementation of step c2 above, which outputs the multi-dimensional fusion vector corresponding to the deposited latent semantic vector, is not limited. For example, in an alternative implementation, in order to reliably remove perturbation information during the fusion process and ensure that the output multi-dimensional fusion vector has high semantic representation accuracy, step c2 above may further include the following implementable content (in conjunction with...). Figure 3 ):

[0108] The first step involves using the first and second mining branches of the semantic fusion model in the target thickness prediction network to perform saliency feature mining on the semantic vector to be predicted, outputting corresponding first and second saliency semantic vectors. In other words, the first mining branch can be used to mine saliency features from the semantic vector to be predicted to obtain the corresponding first saliency semantic vector, and the second mining branch can be used to mine saliency features from the semantic vector to be predicted to obtain the corresponding second saliency semantic vector. The first and second mining branches use the same saliency feature mining method, but the parameters carried by the mining branches can be different, allowing the first and second mining branches to mine and capture different saliency features in the semantic vector to be predicted, thereby obtaining first and second saliency semantic vectors that can represent different saliency characteristics. The semantic vector is used for semantic analysis. Furthermore, the specific implementation process of salient feature mining can be as follows: Based on the first parameter distribution, second parameter distribution, and third parameter distribution carried by the first mining branch, the deposited parameter semantic vector is mapped (e.g., multiplied) to obtain the corresponding first, second, and third predicted mapping vectors. Then, the dot product result between the transposes of the first and second predicted mapping vectors can be determined, allowing the dot product result to characterize the related semantics between the parameters within the predicted semantic vector. Then, based on the dot product result, a weighted summation operation can be performed on the third predicted mapping vector to obtain the corresponding first salient semantic vector. Since this involves mining related semantics, irrelevant perturbation information can be ignored, thus removing perturbation information to a certain extent. The mining process for the second salient semantic vector is the same.

[0109] The second step is to fuse the deposition parameter semantic vector into the first saliency semantic vector to obtain the corresponding first fused semantic vector. In other words, the deposition parameter vector can be further fused into the first saliency semantic vector to remove the perturbation information and obtain a reliable first fused semantic vector.

[0110] The third step involves fusing one of the sedimentation thickness adjustment vectors corresponding to the sedimentation potential semantic vectors from the A sedimentation thickness adjustment vectors into the second saliency semantic vector to obtain the corresponding second fused semantic vector. In other words, by further fusing the sedimentation thickness adjustment vector into the second saliency semantic vector, the perturbation information can be further removed, thereby obtaining a reliable second fused semantic vector. Based on this, it can be seen that since the other semantic vectors that the first saliency semantic vector and the second saliency semantic vector need to be fused are different (the sedimentation parameter semantic vector and the sedimentation thickness adjustment vector, respectively), the semantic information that the first saliency semantic vector and the second saliency semantic vector need to focus on or represent are different. Based on this, the first excavation branch and the second excavation branch need to excavate separately.

[0111] The fourth step is to add the first fused semantic vector and the second fused semantic vector together to output the multi-dimensional fused vector corresponding to the deposited potential semantic vector.

[0112] Alternatively, the specific method of semantic vector fusion for the second and third steps described above is not limited. For example, in an alternative implementation, to achieve reliable semantic vector fusion through multi-level fusion, the example of fusing the deposition parameter semantic vector into the first saliency semantic vector is given (in conjunction with...). Figure 4 ):

[0113] The first step involves performing multiple concatenated semantic extractions on the first salient semantic vector to form multiple corresponding first extracted semantic vectors. For each semantic extraction, the first extracted semantic vector corresponding to the previous extraction is concatenated with the deposition parameter semantic vector. The concatenated semantic vector undergoes self-attention processing, and the semantic vector obtained from the self-attention processing is then downsampled (through convolution and / or pooling) to form the first extracted semantic vector corresponding to the current semantic extraction. The first extracted semantic vector corresponding to the previous semantic extraction of the first semantic extraction is the first salient semantic vector. For example, in the first semantic extraction, the first salient semantic vector and the deposition parameter semantic vector can be concatenated, followed by self-attention processing and downsampling to obtain the first extracted semantic vector corresponding to the first semantic extraction. Similarly, in the second semantic extraction, the first extracted semantic vector corresponding to the first semantic extraction and the deposition parameter semantic vector can be concatenated, followed by self-attention processing and downsampling to obtain the first extracted semantic vector corresponding to the second semantic extraction.

[0114] The second step involves cascading multiple semantic diffusions on the last first extracted semantic vector to form multiple corresponding first diffused semantic vectors. For each semantic diffusion, the first diffused semantic vector corresponding to the previous semantic diffusion is concatenated with a first extracted semantic vector at the corresponding level. The concatenated semantic vector undergoes self-attention processing, and the semantic vector obtained from the self-attention processing is then upsampled (through transposed convolution and / or interpolation) to form the first diffused semantic vector corresponding to the current semantic diffusion. The first diffused semantic vector corresponding to the semantic diffusion preceding the first semantic diffusion is the last first extracted semantic vector. For example, in the first semantic extraction process, the last first extracted semantic vector can be concatenated with the last first extracted semantic vector, followed by self-attention processing and upsampling to obtain the first diffused semantic vector corresponding to the first semantic diffusion. Similarly, in the second semantic extraction process, the first diffused semantic vector corresponding to the first semantic diffusion can be concatenated with the penultimate extracted semantic vector, followed by self-attention processing and downsampling to obtain the first diffused semantic vector corresponding to the second semantic diffusion.

[0115] The third step is to determine the last first diffusion semantic vector as the first fusion semantic vector.

[0116] In summary, the present invention provides a performance analysis method and system for a metallized thin-film flexible CRT capacitor. First, a target deposition thickness sequence is obtained during the deposition process of the target capacitor. Second, a first deposition thickness data and a comparative deposition thickness sequence are determined from the target deposition thickness sequence. Then, based on the first deposition thickness data and the deposition parameters of the metallized thin film, a prediction processing is performed to obtain a predicted deposition thickness sequence. Finally, the predicted deposition thickness sequence and the comparative deposition thickness sequence are compared and analyzed to obtain the metallized thin film performance data of the target capacitor. Based on the above, by predicting the corresponding deposition thickness sequence, it is possible to compare and analyze it as a theoretical value with a comparative deposition thickness sequence as an actual value. This allows for comparison of data at each deposition stage, ensuring a comprehensive analysis of the dynamic process across multiple deposition stages. Therefore, compared to the conventional method of comparing data after deposition is completed, this method is more comprehensive and has higher reliability, thus improving the relatively low reliability of capacitor performance analysis in existing technologies. In addition, since the predicted sedimentary thickness sequence references the first sedimentary thickness data, the initial sedimentary thickness data of the predicted sedimentary thickness sequence and the comparison sedimentary thickness sequence are the same, that is, the initial conditions are the same, which makes the comparison accuracy of subsequent sedimentary thickness data higher and focuses more on the dynamic changes in the subsequent sedimentation process.

[0117] In the several embodiments provided in this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0118] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0119] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0120] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A performance analysis method for a metallized thin-film flexible DC capacitor, characterized in that, include: The target deposition thickness sequence formed during the coating process of the target capacitor is obtained, wherein the target deposition thickness sequence includes multiple deposition thickness data corresponding to multiple deposition stages, and each deposition thickness data is used to characterize the thickness of at least one point of the metallization film deposited in the current stage. A first deposition thickness data and a comparative deposition thickness sequence are determined from the target deposition thickness sequence, wherein each deposition thickness data in the comparative deposition thickness sequence is located after the first deposition thickness data; Based on the first deposition thickness data and the coating deposition parameters of the metallized thin film, a prediction processing is performed to obtain a predicted deposition thickness sequence, including: performing semantic mining on the first deposition thickness data to output A deposition thickness semantic vectors, wherein the deposition thickness semantic vectors are used to reflect the potential semantic information of the first deposition thickness data; determining the time-series adjustment parameters corresponding to each deposition thickness semantic vector; and, based on the time-series adjustment parameters corresponding to each deposition thickness semantic vector, performing semantic adjustment on the corresponding deposition thickness semantic vectors in the A deposition thickness semantic vectors to output A deposition thickness adjustment vectors, wherein each deposition thickness adjustment vector is used to predict a deposition thickness data in the required predicted deposition thickness sequence, and the semantic adjustment is used to control the changes between each deposition thickness data in the predicted deposition thickness sequence and the first deposition thickness data; and performing prediction processing on the A deposition thickness adjustment vectors based on the coating deposition parameters to output a corresponding predicted deposition thickness sequence, wherein the predicted deposition thickness sequence includes A deposition thickness data, where A is equal to the number of deposition thickness data in the comparison deposition thickness sequence, and each deposition thickness data in the predicted deposition thickness sequence belongs to the prediction result of the deposition stage after the first deposition thickness data; By comparing and analyzing the predicted deposition thickness sequence and the comparative deposition thickness sequence, the metallization film performance data of the target capacitor are obtained.

2. The performance analysis method for metallized thin-film flexible DC capacitors according to claim 1, characterized in that, The step of performing semantic mining on the first deposition thickness data and outputting A deposition thickness semantic vectors includes: The first deposition thickness data is embedded in a vector space to form a corresponding deposition thickness embedding vector, and the deposition thickness embedding vector is subjected to self-attention processing to form a corresponding deposition thickness semantic vector. The deposition thickness semantic vector is expanded to form A deposition thickness semantic vectors, wherein every two deposition thickness semantic vectors in the A deposition thickness semantic vectors are the same.

3. The performance analysis method for metallized thin-film flexible DC capacitors according to claim 1, characterized in that, The steps of determining the timing adjustment parameter corresponding to each of the deposition thickness semantic vectors, and semantically adjusting the corresponding deposition thickness semantic vectors among the A deposition thickness semantic vectors based on the timing adjustment parameter corresponding to each of the deposition thickness semantic vectors, and outputting A deposition thickness adjustment vectors, include: A pre-configured reference adjustment parameter is determined, wherein the reference adjustment parameter is used to reflect the magnitude of the difference between adjacent deposition thickness data in the predicted deposition thickness sequence; A training adjustment parameters carried in the target thickness prediction network are obtained, and A time-series adjustment parameters are determined based on the reference adjustment parameters and the A training adjustment parameters. The A time-series adjustment parameters and the A deposition thickness semantic vectors have a one-to-one correspondence, and the A training adjustment parameters are formed during the training process of the target thickness prediction network, which belongs to a neural network model. Based on each of the A time-series adjustment parameters, the corresponding deposition thickness semantic vector is mapped and adjusted to output A deposition thickness adjustment vectors.

4. The performance analysis method for a metallized thin-film flexible DC capacitor according to claim 1, characterized in that, The step of predicting the A deposition thickness adjustment vectors based on the deposition parameters and outputting the corresponding predicted deposition thickness sequence includes: A potential semantic vectors for deposition are determined. The potential semantic vectors for deposition are formed by perturbing the potential semantic vectors in the semantic space where the deposition thickness is located during training. The perturbed vectors follow a normal distribution, and one of the potential semantic vectors for deposition is used to predict a deposition thickness data in the required predicted deposition thickness sequence. The deposition parameters are semantically mined to form corresponding deposition parameter semantic vectors. The deposition parameter semantic vectors are used to reflect the semantic information of the deposition parameters. The deposition parameters include at least one of the following: deposition distance, gas flow rate, deposition temperature, deposition bias voltage, deposition atmosphere, sputtering power, and target current. The target thickness prediction network predicts the deposition parameter semantic vector and the A potential deposition semantic vectors, and fuses the A deposition thickness adjustment vectors during the prediction process to output the corresponding predicted deposition thickness sequence. The potential semantic vectors in the semantic space of the deposition thickness formed during training are formed during the training process of the target thickness prediction network, which is a neural network model.

5. The performance analysis method for a metallized thin-film flexible DC capacitor according to claim 4, characterized in that, The step of using a target thickness prediction network to predict the deposition parameter semantic vector and the A potential deposition semantic vectors, and fusing the A deposition thickness adjustment vectors during the prediction process to output the corresponding predicted deposition thickness sequence includes: For each of the A sedimentation potential semantic vectors, the corresponding semantic vector to be predicted is determined based on the sedimentation potential semantic vector; The semantic fusion model in the target thickness prediction network is used to fuse the deposition parameter semantic vector and one deposition thickness adjustment vector corresponding to the potential deposition semantic vector among the A deposition thickness adjustment vectors into the semantic vector to be predicted, and output the multi-dimensional fusion vector corresponding to the potential deposition semantic vector. The multi-dimensional fusion vector is predicted by the prediction output model in the target thickness prediction network, and a deposition thickness data corresponding to a deposition potential semantic vector is output. The deposition thickness data corresponding to the A deposition potential semantic vectors are combined to form a corresponding predicted deposition thickness sequence.

6. The performance analysis method for a metallized thin-film flexible DC capacitor according to claim 5, characterized in that, The step of determining the corresponding semantic vector to be predicted based on each of the A sedimentation potential semantic vectors includes: The first of the A potential semantic vectors is determined as the corresponding semantic vector to be predicted. For each of the A sedimentation latent semantic vectors other than the first sedimentation latent semantic vector, the corresponding concatenated semantic vector is formed by concatenating the concatenated semantic vector with the multi-dimensional fusion vector corresponding to the previous sedimentation latent semantic vector. Then, the concatenated semantic vector is downsampled to form the corresponding semantic vector to be predicted.

7. The performance analysis method for a metallized thin-film flexible DC capacitor according to claim 5, characterized in that, The step of fusing the deposition parameter semantic vector and one deposition thickness adjustment vector corresponding to the potential deposition semantic vector among the A deposition thickness adjustment vectors into the semantic vector to be predicted through the semantic fusion model in the target thickness prediction network, and outputting the multi-dimensional fusion vector corresponding to the potential deposition semantic vector, includes: The semantic fusion model in the target thickness prediction network includes a first mining branch and a second mining branch, which respectively perform salient feature mining on the semantic vector to be predicted, and output the corresponding first salient semantic vector and second salient semantic vector. The deposition parameter semantic vector is fused into the first saliency semantic vector to obtain the corresponding first fused semantic vector; One deposition thickness adjustment vector corresponding to the deposition potential semantic vector among the A deposition thickness adjustment vectors is fused into the second saliency semantic vector to obtain the corresponding second fused semantic vector; The first fused semantic vector and the second fused semantic vector are added together to output the multi-dimensional fused vector corresponding to the deposited potential semantic vector.

8. The performance analysis method for a metallized thin-film flexible DC capacitor according to claim 7, characterized in that, The step of fusing the deposition parameter semantic vector into the first saliency semantic vector to obtain the corresponding first fused semantic vector includes: Multiple semantic extractions are performed on the first salient semantic vector in a cascaded manner to form multiple corresponding first extracted semantic vectors. For each semantic extraction, the first extracted semantic vector corresponding to the previous semantic extraction and the deposition parameter semantic vector are concatenated, and the concatenated semantic vector is subjected to self-attention processing. The semantic vector obtained by self-attention processing is then downsampled to form the first extracted semantic vector corresponding to the current semantic extraction. The first extracted semantic vector corresponding to the previous semantic extraction of the first semantic extraction is the first salient semantic vector. Multiple semantic diffusions are cascaded on the last first extracted semantic vector to form multiple corresponding first diffuse semantic vectors. For each semantic diffusion, the first diffuse semantic vector corresponding to the previous semantic diffusion is concatenated with a first extracted semantic vector at the corresponding level. The concatenated semantic vector is then subjected to self-attention processing, and the semantic vector obtained from the self-attention processing is then upsampled to form the first diffuse semantic vector corresponding to the current semantic diffusion. The first diffuse semantic vector corresponding to the semantic diffusion preceding the first semantic diffusion is the last first extracted semantic vector. The last first diffusion semantic vector is determined as the first fusion semantic vector.

9. A performance analysis system for a metallized thin-film flexible DC capacitor, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory is used to execute a computer program stored in the memory to implement the performance analysis method for the metallized thin-film flexible DC capacitor according to any one of claims 1-8.

Images