A small thin film capacitor production device and method

Abstract

The application provides a small thin film capacitor production process and method, relates to the capacitor detection technical field, and comprises a data acquisition module, which is used for acquiring quality evaluation parameters of the thin film capacitor under a current key process; a quality detection module, which is used for analyzing the quality evaluation parameters to determine whether the quality of the thin film capacitor is qualified; a directed graph construction module, which is used for constructing a directed graph of each quality evaluation parameter in the current and next key process; a feature reconstruction module, which is used for reconstructing a feature matrix of each parent node in the directed graph; and a warning module, which is used for warning and determining the thin film capacitor processed by using the next key process parameter. The application realizes real-time quality detection of the thin film capacitor in the production process by collecting the quality evaluation coefficient of the thin film capacitor in real time and combining the evaluation standard, so that the unqualified thin film capacitor can be removed in the subsequent key process, thereby saving the production and manufacturing resources.

Description

Technical Field

[0001] This invention relates to the field of capacitor testing technology, specifically to a manufacturing process and method for small film capacitors. Background Technology

[0002] Film capacitors consist of two conducting plates sandwiched between an insulating dielectric layer. When a voltage is applied between the two plates, the film capacitor stores charge, playing a crucial role in circuits for tuning, bypassing, coupling, and filtering. With the rapid development of electronic information technology and the increasingly fast pace of digital electronic product upgrades, the production and sales of consumer electronics, primarily flat-screen TVs, laptops, and digital cameras, continue to grow, driving the growth of the film capacitor industry.

[0003] To ensure a high yield rate for capacitors, manufacturers conduct quality inspections on them before they leave the factory. However, traditional capacitors typically rely on manual sampling and offline testing for post-production inspection. Even if substandard film capacitors are detected, it is difficult to remedy the situation and remove them from production in a timely manner, thus failing to save production resources.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a manufacturing process and method for small film capacitors to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A manufacturing process for a small film capacitor includes: The data acquisition module is used to collect quality assessment parameters of the film capacitor under the current key process, which includes winding, welding, energizing and aging performed in sequence. The quality inspection module analyzes the quality assessment parameters under the current key process based on the set evaluation criteria to determine whether the quality of the film capacitor after the current key process is qualified, and performs directed graph construction after the judgment result is qualified. The directed graph construction module, for the current critical process (excluding the last critical process), takes its corresponding quality assessment parameters as parent nodes and the quality assessment parameters in the next critical process as child nodes. Based on the assessment criteria, it determines the critical approximation degree of the quality assessment parameters as node feature values, and combines the process parameters of the next critical process to determine the correlation between parent and child nodes. Based on the correlation, it determines the directed edges between nodes, thereby constructing a directed graph. The feature reconstruction module, based on the directed edges in the directed graph, determines the neighboring nodes of the parent node relative to each child node, aggregates the feature values ​​of the neighboring nodes to obtain supplementary information of the parent node relative to each child node, and constructs the parent node feature matrix, which includes the parent node feature values, the correlation between the parent node and each child node, and the supplementary information of the parent node relative to each child node. The early warning module includes a prediction model that integrates and analyzes the feature matrices of each parent node and the process parameters of the next critical process to output the predicted values ​​of the child node features, and an alarm unit that analyzes the predicted values ​​of the child node features based on evaluation criteria to make an alarm judgment when the film capacitor is processed using the process parameters of the next critical process.

[0007] Furthermore, the types of quality assessment parameters are determined based on the key process type, as follows: For the critical winding process, the quality assessment parameters collected include alignment error, thickness uniformity, and the proportion of surface defects, including bubbles, wrinkles, and damage. For the critical welding process, the quality assessment parameters collected include the percentage of welding defects and the contact resistance failure rate. Welding defects include incomplete welds, detached welds, and cracks. For the critical process of enabling, the quality assessment parameters collected include initial leakage current, initial capacitance, initial loss angle, and initial equivalent series resistance. For the critical aging process, the quality assessment parameters collected include leakage current degradation rate, capacitance degradation rate, loss angle degradation rate, equivalent series resistance degradation rate, and service life prediction.

[0008] Furthermore, the evaluation criteria are determined based on the type of quality evaluation parameters, as follows: When the current critical process is winding, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: Alignment error shall not exceed the alignment error threshold; Rule 2: The thickness range shall not exceed the thickness range threshold, and the thickness coefficient of variation shall not exceed the thickness coefficient of variation threshold. Rule 3: The proportion of surface defects shall not exceed the threshold for the proportion of surface defects; When the current critical process is welding, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The proportion of welding defects shall not exceed the threshold for the proportion of welding defects; Rule 2: The contact resistance failure rate shall not exceed the failure rate threshold; When the current critical process is to enable, the evaluation criteria include the following rules. If the following rules are met at the same time, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The initial leakage current shall not exceed the leakage current threshold. Rule 2: The initial capacitance shall not be less than the capacitance threshold. Rule 3: The initial loss angle shall not exceed the loss angle threshold. Rule 4: The initial equivalent series resistance shall not exceed the equivalent resistance threshold. When the current critical process is aging, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The rate of leakage current degradation shall not exceed the leakage current degradation threshold. Rule 2: The rate of capacity degradation shall not exceed the capacity degradation threshold; Rule 3: The rate of loss angle deterioration shall not exceed the loss angle deterioration threshold; Rule 4: The rate of degradation of the equivalent series resistance shall not exceed the threshold value for degradation of the equivalent resistance. Rule 5: The predicted service life value shall not be less than the service life threshold.

[0009] Furthermore, the logic for obtaining the critical approximation degree is as follows: for a quality assessment parameter in the evaluation standard that applies the rule of not being greater than the corresponding threshold, its critical approximation degree is characterized by the ratio of its value to the corresponding threshold. For a quality assessment parameter in the evaluation standard that applies the rule of not being less than the corresponding threshold, its relative error with the corresponding threshold is calculated, and the product of the relative error and -1 is taken as the critical approximation degree of this quality assessment parameter.

[0010] Furthermore, the correlation is characterized by the absolute value of the Pearson correlation coefficient. For any parent node and any child node, they are defined as the target parent node and the target child node, respectively. The logic for determining the correlation between the two is as follows: the critical process to which the current critical process belongs is defined as the first benchmark process, and the critical process to which the next critical process belongs is defined as the second benchmark process. Film capacitors whose process parameters in the second benchmark process are similar to those in the next critical process in historical production are used as candidate samples. Then, candidate samples, except for the target parent node, whose quality assessment parameter values ​​collected under the first benchmark process are similar to those collected under the current critical process are selected as reference samples. Pearson analysis is performed on the characteristic values ​​of the target parent node and the target child node of the reference samples to determine the absolute value of the Pearson correlation coefficient between the target parent node and the target child node under the process parameters of the next critical process.

[0011] Furthermore, the logic for establishing directed edges is as follows: if the correlation between a parent node and a child node is greater than the correlation threshold, then a directed edge is established from the parent node to the child node; otherwise, no directed edge is established between them.

[0012] Furthermore, for any parent node, its feature matrix has three columns and the same number of rows as the number of child nodes. The elements in the first column are the feature values ​​of the parent node, the elements in the second column are the correlations between the parent node and each child node, and the elements in the third column are the supplementary information of the parent node relative to each child node. Specifically, for any parent node and any child node, they are defined as the target parent node and the target child node, respectively. The logic for determining the supplementary information of the target parent node relative to the target child node is as follows: We count the other parent nodes that have directed edges with the target child node, excluding the target parent node, and use them as the neighbor nodes of the target parent node relative to the target child node. With the cumulative value equal to 1 as a constraint, we scale the relevance between each neighbor node and the target child node proportionally to obtain the relevance weight of each neighbor node and the target child node. We calculate the product of the feature value of each neighbor node and its corresponding relevance weight and sum them up to provide supplementary information about the target parent node relative to the target child node.

[0013] Furthermore, the prediction model includes three prediction sub-models selected and applied based on the current key process type: a first prediction sub-model applied when the current key process is winding, a second prediction sub-model applied when the current key process is welding, and a third prediction sub-model applied when the current key process is empowerment. Each prediction sub-model includes an input layer, which receives the feature matrices of each parent node and the process parameters of the next key process; a hidden layer, which extracts features from the feature matrices of each parent node and the process parameters of the next key process; and an output layer, which outputs the predicted feature values ​​of each child node. The alarm unit works as follows: it analyzes the predicted values ​​of sub-nodes based on the evaluation criteria to deduce the predicted values ​​of quality evaluation parameters for the next critical process, and combines the evaluation criteria to determine whether the quality of the film capacitor after the next critical process is qualified. If the result is unqualified, an alarm is triggered.

[0014] Furthermore, the key processes are defined as including only winding, welding, and energizing, and the quality assessment parameters and standards corresponding to aging are merged into the quality assessment parameters and standards corresponding to energizing. The prediction model includes two prediction sub-models selected based on the current key process type: the first prediction sub-model applied when the current key process is winding, and the second prediction sub-model applied when the current key process is welding.

[0015] An automated quality inspection method for film capacitors, used in the aforementioned small film capacitor manufacturing process, includes the following steps: Step S1: Collect quality assessment parameters of the film capacitor under the current key process, the key process including winding, welding, energizing and aging in sequence; Step S2: Based on the set evaluation criteria, analyze the quality evaluation parameters under the current key process to determine whether the quality of the film capacitor after the current key process is qualified, and perform directed graph construction after the judgment result is qualified. Step S3: For the current critical process except the last critical process, take each quality assessment parameter corresponding to it as the parent node and each quality assessment parameter in the next critical process as the child node. Determine the critical approximation degree of the quality assessment parameter based on the assessment standard as the node feature value. Combine the process parameters of the next critical process to determine the correlation between the parent node and the child node. Determine the directed edges between the nodes based on the correlation to construct a directed graph. Step S4: Based on the directed edges in the directed graph, determine the neighboring nodes of the parent node relative to each child node, aggregate the feature values ​​of the neighboring nodes to obtain the supplementary information of the parent node relative to each child node, and construct the parent node feature matrix, which includes the parent node feature values, the correlation between the parent node and each child node, and the supplementary information of the parent node relative to each child node. Step S5: Based on the prediction model, the feature matrix of each parent node and the process parameters of the next key process are fused and analyzed to output the predicted value of the child node features. The predicted value of the child node features is then analyzed based on the evaluation criteria to make an alarm judgment on the use of the process parameters of the next key process to process the film capacitor.

[0016] Compared with the prior art, the beneficial effects of the present invention are: The present invention relates to a small film capacitor manufacturing process and method. By real-time acquisition of the quality assessment coefficient of the film capacitor during the current critical process and combining it with the assessment standard, real-time quality detection is achieved during the film capacitor production process. This allows for the elimination of substandard film capacitors in subsequent critical processes, thereby saving production resources. Furthermore, by integrating and analyzing the directed graph and the process parameters of the next critical process, the quality assessment parameters of the film capacitor in the next critical process are predicted. This prediction helps determine whether the quality of the film capacitor processed using the next critical process parameters is qualified. If the quality is not qualified, an alarm is triggered in time to prompt the staff to adjust the process parameters of the next critical process, thus avoiding the problem of wasting production resources by manufacturing substandard film capacitors. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the detection system modules in this invention; Figure 2 This is a schematic diagram of the detection method in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example 1: Please see Figure 1 This invention provides a manufacturing process for small film capacitors, comprising: The data acquisition module is used to collect quality assessment parameters of the film capacitor under the current key process, which includes winding, welding, energizing and aging performed in sequence. The types of quality assessment parameters are determined based on the types of key processes, as follows: For the critical winding process, the quality assessment parameters collected include alignment error, thickness uniformity, and the proportion of surface defects, including bubbles, wrinkles, and damage. As one implementation method, the maximum value of the shortest distance from each point on the edge of the insulating material in the film capacitor to the electrode material is used as the alignment error. The logic for obtaining this error is as follows: a side image of the film capacitor after winding is captured by a high-definition camera; the edges of the insulating material and the electrode material of the film capacitor are identified based on the Canny edge detection algorithm; the shortest pixel distance from each point on the edge of the insulating material to the edge of the electrode material is calculated using image processing software; and the maximum value of the shortest pixel distance is converted into the actual distance based on the resolution and scale of the side image, which is used as the alignment error. The larger the value, the worse the alignment accuracy of the insulating material and the electrode material. The Canny edge detection algorithm and related image processing software can all be existing technologies, which will not be elaborated here. As one implementation method, thickness uniformity includes thickness range and thickness variation coefficient. The acquisition logic is as follows: multiple laser thickness gauges are installed on the side or above the winding machine to collect the thickness values ​​of each point after the film capacitor is wound. The difference between the maximum and minimum thickness values ​​is taken as the thickness range. The standard deviation and mean of the thickness values ​​at each point are calculated, and the ratio of the two is taken as the thickness variation coefficient. The thickness range is used to measure the extreme thickness difference of the film capacitor, and the thickness variation coefficient is used to measure the overall thickness non-uniformity of the film capacitor. If both are too large, it indicates that there is a risk of inconsistent electrical characteristics after the film capacitor is wound. As one implementation method, the logic for obtaining the surface defect ratio is as follows: based on the high-definition camera to capture the surface image of the film capacitor after winding, the defect area on the surface of the film capacitor is identified based on the Canny edge detection algorithm, which includes bubble area, wrinkle area and broken area. This is the prior art and will not be elaborated. The ratio of the total number of pixels in all defect areas to the number of pixels in the surface image is calculated as the surface defect ratio. The larger the value, the worse the surface quality of the film capacitor after winding. For the critical welding process, the quality assessment parameters collected include the percentage of welding defects and the contact resistance failure rate. Welding defects include incomplete welds, detached welds, and cracks. As one implementation method, the logic for obtaining the welding defect ratio is as follows: Based on the surface image of the thin film capacitor after welding, a high-definition camera is used to capture the image. Based on the image analysis algorithm (such as edge detection, morphological operation or feature extraction), the number of solder joints with poor solder joints, desoldering and cracks on the surface of the thin film capacitor is identified. The ratio of the number of solder joints with poor solder joints, desoldering and cracks on the surface of the thin film capacitor is calculated as the welding defect ratio. The larger the value, the worse the welding quality of the thin film capacitor. The identification of solder joints with poor solder joints, desoldering and cracks on the surface of the thin film capacitor based on the image analysis algorithm is the prior art and will not be elaborated here. As one implementation method, the logic for obtaining the contact resistance failure rate is as follows: Several solder joints are randomly selected as sample solder joints within the welded film capacitor. The contact resistance of each sample solder joint is measured using a micro-ohmmeter or multimeter. Sample solder joints whose contact resistance is outside the preset acceptable range are considered defective. The ratio of the number of defective sample solder joints to the number of sample solder joints is calculated as the contact resistance failure rate. An excessively high value will lead to increased power consumption, shortened lifespan, and increased failure risk for the film capacitor. The specific value of the sample solder joints is determined based on the circuit complexity and number of solder joints of the film capacitor, generally 5-10 is preferred. The acceptable resistance range is determined based on the application type of the film capacitor. For example, in general electronic equipment, the acceptable resistance range can be 0-5 milliohms, while in special fields such as aerospace, medical, and automotive, the requirements for the acceptable resistance range are further tightened, and the value can be 0-1 milliohms. For the critical process of enabling, the quality assessment parameters collected include initial leakage current, initial capacitance, initial loss angle, and initial equivalent series resistance. As one implementation method, the logic for obtaining the initial leakage current is as follows: a set test voltage is applied across the film capacitor to energize it to the test voltage. Then, a leakage current meter is used to measure the leakage current after the film capacitor has been energized, which is used as the initial leakage current. The test voltage is generally set to the rated voltage of the film capacitor or 1.5 times the rated voltage. As one implementation method, the logic for obtaining the initial capacitance is as follows: connect an LCR meter or capacitance meter to the energized film capacitor to obtain the capacitance value of the energized film capacitor as the initial capacitance value. As one implementation method, the logic for obtaining the initial loss angle is as follows: connect an LCR meter or impedance analyzer to the energized film capacitor to obtain the loss angle of the energized film capacitor as the initial loss angle. Here, it is preferable to use an LCR meter so as to obtain the initial capacitance and initial loss angle of the energized film capacitor at the same time. As one implementation method, the logic for obtaining the initial equivalent series resistance is as follows: connect an LCR meter or impedance analyzer to the energized film capacitor to obtain the equivalent series resistance of the energized film capacitor as the initial equivalent series resistance. Here, it is preferable to use an LCR meter so as to simultaneously obtain the initial capacitance, initial loss angle and initial equivalent series resistance of the energized film capacitor. For the critical aging process, the quality assessment parameters collected include leakage current degradation rate, capacitance degradation rate, loss angle degradation rate, equivalent series resistance degradation rate, and service life prediction. As one implementation method, the logic for obtaining the leakage current degradation rate is as follows: calculate the difference between the final value of the leakage current and the initial leakage current, and use the ratio of this difference to the aging time as the leakage current degradation rate. The larger the value, the faster the degradation rate of the film capacitor, which means that the performance stability of the film capacitor is worse. The final value of the leakage current is the leakage current of the film capacitor after aging is completed. The method for collecting the final value of the leakage current is the same as that for the initial leakage current, and will not be described in detail here. As one implementation method, the logic for obtaining the capacitance degradation rate is as follows: calculate the difference between the initial capacitance and the final capacitance value, and calculate the ratio of this difference to the rated capacitance of the film capacitor as the capacitance value decrease rate. The ratio of this capacitance value decrease rate to the aging time is taken as the capacitance degradation rate. The larger the value, the faster the degradation rate of the film capacitor, which means that the performance stability of the film capacitor is worse. The final capacitance value is the capacitance of the film capacitor after aging is completed. The method for collecting the final capacitance value is the same as that for the initial capacitance, and will not be described in detail here. As one implementation method, the logic for obtaining the loss angle degradation rate is as follows: calculate the difference between the final loss angle value and the initial loss angle, and use the ratio of this difference to the aging time as the loss angle degradation rate. The larger the value, the faster the degradation rate of the thin film capacitor, which means that the performance stability of the thin film capacitor is worse. The final loss angle value is the loss angle of the thin film capacitor after aging is completed. The method for collecting the final loss angle value is the same as that for the initial loss angle, and will not be described in detail here. As one implementation method, the logic for obtaining the equivalent series resistance degradation rate is as follows: calculate the difference between the final value of the equivalent series resistance and the initial equivalent series resistance, and use the ratio of this difference to the aging time as the equivalent series resistance degradation rate. The larger the value, the faster the degradation rate of the film capacitor, which means that the performance stability of the film capacitor is worse. The final value of the equivalent series resistance is the equivalent series resistance of the film capacitor after aging is completed. The method for collecting the final value of the equivalent series resistance is the same as that for the initial equivalent series resistance, and will not be described in detail here. As one implementation method, the logic for obtaining the service life prediction value is as follows: a fault judgment benchmark is predefined, such as the leakage current exceeding twice the rated value or the capacitance decreasing by more than 20% as the fault judgment benchmark for the film capacitor. During the aging test, the leakage current and capacitance of the film capacitor are collected at equal time intervals. The aging time-leakage current curve and the aging time-capacity curve are plotted using a polynomial fitting method. The aging time corresponding to the leakage current first reaching twice the rated value in the aging time-leakage current curve is taken as the first prediction time. The aging time corresponding to the capacitance first decreasing to 80% of the initial capacitance in the aging time-capacity curve is taken as the second prediction time. The minimum value of the first prediction time and the second prediction time is selected as the service life prediction value. The equal time interval can be 50 hours, 75 hours, 100 hours, etc. The rated value of the leakage current of the film capacitor is determined according to its application type, and can generally be set between 0.1 microamps and 5 microamps. The polynomial fitting of the aging time-leakage current curve is a conventional technical means, which will not be elaborated here. Furthermore, for film capacitors in the same production batch, several film capacitors are randomly selected as sample film capacitors in the current key process. Based on the above method, the quality assessment parameters of each sample film capacitor in the current key process are obtained, and the average value is taken as the quality assessment parameter of the film capacitors in the current key process. This enables real-time detection of the quality of film capacitors in the same production batch without the need to collect quality assessment parameters for each film capacitor, thus reducing workload.

[0021] The quality inspection module analyzes the quality assessment parameters under the current key process based on the set evaluation criteria to determine whether the quality of the film capacitor after the current key process is qualified, and performs directed graph construction after the judgment result is qualified. The evaluation criteria are determined based on the types of quality evaluation parameters, as follows: When the current critical process is winding, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The alignment error shall not exceed the alignment error threshold. The specific alignment error threshold shall be set by the staff according to the actual situation. For example, it can be set to 0.1mm. This will effectively evaluate the alignment accuracy between the insulating material and the electrode material in the wound film capacitor and identify film capacitors with substandard alignment accuracy. Rule 2: The thickness range shall not exceed the thickness range threshold, and the thickness coefficient of variation shall not exceed the thickness coefficient of variation threshold. The thickness range threshold and the thickness coefficient of variation threshold shall be set by the staff according to the actual situation. For example, the thickness range threshold can be set to 0.05mm and the thickness coefficient of variation threshold can be set to 5%. This is to effectively evaluate the thickness uniformity of the wound film capacitor and identify film capacitors with non-compliant thickness uniformity. Rule 3: The proportion of surface defects shall not exceed the surface defect proportion threshold. The specific surface defect proportion threshold shall be set by the staff according to the actual situation. For example, it can be set to 2% to effectively evaluate the surface defects of the wound film capacitor and identify film capacitors with substandard surface quality. When the current critical process is welding, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The proportion of welding defects shall not exceed the welding defect proportion threshold. The specific welding defect proportion threshold shall be set by the staff according to the actual situation, such as 3%, so as to effectively evaluate the welding quality of the film capacitor after welding and identify the film capacitors with substandard welding quality. Rule 2: The contact resistance failure rate shall not exceed the failure rate threshold. The specific failure rate threshold shall be set by the staff according to the actual situation. For example, it can be set to 5% to effectively evaluate the welding quality of the film capacitor after welding and identify film capacitors with substandard welding quality. When the current critical process is to enable, the evaluation criteria include the following rules. If the following rules are met at the same time, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The initial leakage current shall not exceed the leakage current threshold. The specific leakage current threshold shall be set by the staff according to the actual situation. It can be set to one-thousandth of the rated working voltage of the film capacitor. For example, if the rated working voltage of the film capacitor is 50 volts, the leakage current threshold can be set to 50 microamps. This will effectively evaluate the electrical parameters of the film capacitor after it has been energized and identify film capacitors whose electrical parameters do not meet the standards. Rule 2: The initial capacitance shall not be less than the capacitance threshold. The capacitance threshold shall be set by the staff according to the actual situation. It can be set between 90% and 95% of the rated capacitance of the film capacitor. This is to effectively evaluate the electrical parameters of the film capacitor after it has been energized and identify film capacitors whose electrical parameters do not meet the standards. Rule 3: The initial loss angle shall not be greater than the loss angle threshold. The specific loss angle threshold shall be set by the staff according to the actual situation. For general applications, it can be set to 0.5°. In special fields such as aerospace, medical, and automotive, it can be further strictly set to 0.2°. This is to effectively evaluate the electrical parameters of the empowered film capacitor and identify film capacitors with substandard electrical parameters. Rule 4: The initial equivalent series resistance shall not exceed the equivalent resistance threshold. The equivalent resistance threshold shall be set by the staff according to the actual situation. The unit may be set to milliohms, and the value may be set to the product of the rated operating voltage of the film capacitor and a value between 1 and 5. For example, when the rated operating voltage of the film capacitor is 50 volts, the equivalent resistance threshold may be set between 1 and 5 milliohms. This is to effectively evaluate the electrical parameters of the film capacitor after it has been energized and identify film capacitors whose electrical parameters do not meet the standards. When the current critical process is aging, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The leakage current degradation rate shall not exceed the leakage current degradation threshold. The specific leakage current degradation threshold shall be set by the staff according to the actual situation. For example, it can be set to 0.5 microamps per 100 hours. This will effectively evaluate the electrical parameter stability of the film capacitor and identify film capacitors whose electrical parameter stability does not meet the standard. Rule 2: The rate of capacitance degradation shall not exceed the capacitance degradation threshold. The specific capacitance degradation threshold shall be set by the staff according to the actual situation. For example, it can be set to 0.5% per 100 hours. This will effectively assess the electrical parameter stability of the film capacitor and identify film capacitors whose electrical parameter stability does not meet the standard. Rule 3: The rate of loss angle degradation shall not exceed the loss angle degradation threshold. The specific loss angle degradation threshold shall be set by the staff according to the actual situation. For example, it can be set to 0.1° per 100 hours. This will effectively evaluate the electrical parameter stability of the film capacitor and identify film capacitors whose electrical parameter stability does not meet the standard. Rule 4: The rate of degradation of the equivalent series resistance shall not exceed the equivalent resistance degradation threshold. The specific equivalent resistance degradation threshold shall be set by the staff according to the actual situation. For example, it can be set to 0.1 milliohms per 100 hours to effectively evaluate the electrical parameter stability of the film capacitor and identify film capacitors whose electrical parameter stability does not meet the standard. Rule 5: The predicted service life value shall not be less than the service life threshold. The specific value shall be set by the staff according to the actual situation. For example, it can be set to 2000 hours for general applications, and 5000 hours for special fields such as aerospace, medical, and automotive. This is to effectively evaluate the service life of film capacitors and identify film capacitors that do not meet the service life standard.

[0022] The directed graph construction module, for the current critical process (excluding the last critical process), takes its corresponding quality assessment parameters as parent nodes and the quality assessment parameters in the next critical process as child nodes. Based on the assessment criteria, it determines the critical approximation degree of the quality assessment parameters as node feature values, and combines the process parameters of the next critical process to determine the correlation between parent and child nodes. Based on the correlation, it determines the directed edges between nodes, thereby constructing a directed graph. The logic for obtaining the critical approximation degree is as follows: For quality assessment parameters in the evaluation criteria that are subject to the rule of not exceeding the corresponding threshold, including alignment error, thickness range, thickness coefficient of variation, surface defect ratio, welding defect ratio, contact resistance failure rate, initial leakage current, initial loss angle, initial equivalent series resistance, leakage current degradation rate, capacitance degradation rate, loss angle degradation rate, and equivalent series resistance degradation rate, the critical approximation degree is characterized by the ratio of the parameter to the corresponding threshold. This converts each quality assessment parameter into a critical approximation degree with a value between 0 and 1. The larger the value, the worse the quality assessment parameter of the film capacitor. For example, for the quality assessment parameter of alignment error, the critical approximation degree is the ratio of the alignment error to the alignment error threshold. The quality assessment parameters are scaled to a critical approximation degree with an absolute value between 0 and 1, thereby unifying the dimensions of each quality assessment parameter and facilitating subsequent model analysis. For quality assessment parameters that apply the rule of "not less than the corresponding threshold" in the evaluation criteria, such as the initial capacitance and predicted service life, the logic for obtaining the critical approximation degree is as follows: calculate the relative error between the parameter and the corresponding threshold, and use the product of this relative error and -1 as the critical approximation degree of this quality assessment parameter. The smaller the value, the better the quality assessment parameter of the film capacitor. This is consistent with the logic of the critical approximation degree of quality assessment parameters that apply the rule of "not greater than the corresponding threshold" mentioned above. For example, for the quality assessment parameter of initial capacitance value, first calculate the relative error between the initial capacitance value and the capacitance threshold, and use the product of this relative error and -1 as the critical approximation degree of the initial capacitance value. Scale the quality assessment parameter to the critical approximation degree with an absolute value between 0 and 1, thereby unifying the dimensions of each quality assessment parameter and facilitating subsequent model analysis. The correlation is characterized by the absolute value of the Pearson correlation coefficient. For any parent node and any child node, they are defined as the target parent node and the target child node, respectively. The logic for determining the correlation between the two is as follows: the critical process to which the current critical process belongs is defined as the first benchmark process, and the critical process to which the next critical process belongs is defined as the second benchmark process. Film capacitors with process parameters similar to those of the second benchmark process and the next critical process in historical production are used as candidate samples. Then, candidate samples, except for the target parent node, whose quality assessment parameter values ​​collected under the first benchmark process are similar to those collected under the current critical process are selected as reference samples. Pearson analysis is performed on the characteristic values ​​of the target parent node and the target child node of the reference samples to determine the absolute value of the Pearson correlation coefficient between the target parent node and the target child node under the process parameters of the next critical process. It should be noted that for process parameters belonging to state parameters (such as switch states), similarity is defined as consistency of state parameters. For process parameters and quality assessment parameters belonging to numerical parameters (such as pressure and temperature), similarity is defined as a relative error of no more than 5%. This is to ensure that the selected reference samples can focus on the target parent node and target child node, remove the adverse interference of process parameters and other quality assessment parameters except for the target parent node, and accurately describe the correlation between the target parent node and target child node under the current state. The calculation of Pearson correlation coefficient is a conventional technique for those skilled in the art and will not be elaborated here. As one implementation method, the process parameters for the key winding process include, but are not limited to, winding speed, winding tension, winding diameter and winding angle; the process parameters for the key welding process include, but are not limited to, welding temperature, welding time and welding pressure; and the process parameters for the key energizing process include, but are not limited to, energizing voltage, energizing time and energizing frequency. The logic for establishing directed edges is as follows: if the correlation between a parent node and a child node is greater than the correlation threshold, it indicates that the two have a strong correlation, and a directed edge from the parent node to the child node is established; otherwise, it indicates that the correlation between the two is poor, and a directed edge is not established between them. The correlation threshold is generally set between 0.2 and 0.4 to ensure that directed edges are established between parent nodes and child nodes with strong correlation. The specific value can be set by the staff according to the actual situation, and no restrictions are imposed here. The feature reconstruction module, based on the directed edges in the directed graph, determines the neighboring nodes of the parent node relative to each child node, aggregates the feature values ​​of the neighboring nodes to obtain supplementary information of the parent node relative to each child node, and constructs the parent node feature matrix, which includes the parent node feature values, the correlation between the parent node and each child node, and the supplementary information of the parent node relative to each child node. For any parent node, its feature matrix has three columns and the number of rows is the same as the number of child nodes. The elements in the first column are the feature values ​​of the parent node, the elements in the second column are the correlations between the parent node and each child node, and the elements in the third column are the supplementary information of the parent node relative to each child node. Specifically, for any parent node and any child node, they are defined as the target parent node and the target child node, respectively. The logic for determining the supplementary information of the target parent node relative to the target child node is as follows: We count the other parent nodes (excluding the target parent node) that have directed edges with the target child node, and these are considered as the neighbor nodes of the target parent node relative to the target child node. Since both the target parent node and its neighbor nodes relative to the target child node have directed edges with the target child node, this indicates a strong correlation between the target parent node and its neighbor nodes relative to the target child node. In other words, the influence of the target parent node and its neighbor nodes relative to the target child node on the target child node is complementary. If we only consider the influence of the target parent node itself on the target child node in the feature matrix of the target parent node, the final constructed feature matrices of each parent node will be independent, failing to fully reflect this relationship. Since they are complementary, it is necessary to introduce supplementary information by combining the feature values ​​of neighboring nodes. Specifically, with the cumulative value equal to 1 as a constraint, the correlation between each neighboring node and the target child node is scaled proportionally to obtain the correlation weight between each neighboring node and the target child node. The product of the feature value of each neighboring node and its corresponding correlation weight is calculated and summed to serve as supplementary information of the target parent node relative to the target child node. The weighted fusion method for obtaining supplementary information can effectively consider the strength of the correlation between each neighboring node and the target child node, ensuring the comprehensiveness and rationality of the supplementary information. This supplements the influence of its neighboring nodes on the target child node in the feature matrix of the target parent node, thereby improving the accuracy of subsequent model predictions. It should be noted that when there is no directed edge between the target parent node and the target child node, it indicates that the correlation between the target parent node and the target child node is poor. This means that the feature value of the target parent node is not an important factor to consider when predicting the feature value of the target child node, and there is no need to introduce supplementary information. In this case, it is determined that the target parent node has no neighboring nodes relative to the target child node, and the aggregation information of the target parent node relative to the target child node is zero. When there is a directed edge between the target parent node and the target child node, but there are no other parent nodes besides the target parent node that have directed edges with the target child node, it means that the feature value of the target parent node is the only important factor to consider when predicting the feature value of the target child node, and there is no need to introduce supplementary information. In this case, the aggregation information of the target parent node relative to the target child node is also zero. The early warning module includes a prediction model that integrates and analyzes the feature matrices of each parent node and the process parameters of the next critical process to output the predicted values ​​of the child node features, and an alarm unit that analyzes the predicted values ​​of the child node features based on evaluation criteria to make an alarm judgment when the film capacitor is processed using the process parameters of the next critical process. The prediction model includes three prediction sub-models selected based on the current key process type: the first prediction sub-model applied when the current key process is winding, the second prediction sub-model applied when the current key process is welding, and the third prediction sub-model applied when the current key process is empowerment. Each prediction sub-model includes an input layer, which receives the feature matrices of each parent node and the process parameters of the next key process; a hidden layer, which extracts features from the feature matrices of each parent node and the process parameters of the next key process; and an output layer, which outputs the predicted feature values ​​of each child node. It should be noted that 6-8 hidden layers can be set, and the number of nodes in each layer can be set according to the actual situation, such as 64, 128 or 256 nodes, and the ReLU function can be used to improve the nonlinearity of the model. The process of training the prediction sub-model is as follows: From the historically manufactured film capacitors, select those that have passed the quality judgment after the first benchmark process as film capacitor samples. Retrieve the quality assessment parameter values ​​collected under the first benchmark process and the quality assessment parameter values ​​collected under the second benchmark process for each film capacitor sample, as well as the process parameters of the second benchmark process used when manufacturing each film capacitor sample. Use the above method to obtain the parent node feature matrix and child node feature values ​​of each film capacitor sample. Combine the process parameters of the second benchmark process to form a sample set. Divide the sample set into a training set, a test set, and a validation set in a ratio of 70:15:15. Set the batch size and training period. The batch size can be set between 32 and 256, and the training period can be set between 30 and 50. The parent node feature matrix of each thin-film capacitor sample in the training set and the process parameters of the second benchmark process are used as inputs, and the child node feature values ​​of the corresponding thin-film capacitor samples are used as output labels to train the prediction sub-model. The mean squared error is used as the loss function. During training, the model parameters are updated through backpropagation to minimize the loss function. Specifically, the model parameters can be updated through optimization algorithms such as Adam and SGD. During training, the model hyperparameters, such as learning rate, batch size, and number of hidden layers, are adjusted through the validation set to optimize the model performance. The initial value of the learning rate can be set between 0.001 and 0.01. After reaching the predetermined training period, the test set is input into the prediction sub-model for performance testing. If the relative error is less than the preset threshold, the training is considered complete. Otherwise, the model hyperparameters are adjusted and retraining is performed. The preset threshold can be set between 3% and 8% to ensure that the model after training has a good simulation fitting ability. The specific value can be set by the staff according to the actual situation and is not restricted here. The alarm unit works as follows: it analyzes the predicted values ​​of sub-nodes based on the evaluation criteria to deduce the predicted values ​​of quality evaluation parameters for the next critical process, and judges whether the quality of the film capacitor after the next critical process is qualified in combination with the evaluation criteria. When the judgment result is unqualified, an alarm is triggered to remind the staff to adjust the process parameters of the next critical process in a timely manner to avoid wasting production resources. Furthermore, since the film capacitor has already been manufactured under the critical process of aging, the critical process of aging is mainly to detect whether the film capacitor is qualified. Adjusting the process parameters of the critical process of aging cannot change the quality of the film capacitor. Therefore, the alarm unit mentioned above can only avoid the waste of resources caused by unnecessary aging tests in the critical process of aging, but cannot avoid the waste of resources caused by manufacturing unqualified film capacitors. Therefore, the critical process is defined as only including winding, welding and aging, and the quality assessment parameters and assessment standards corresponding to aging are merged into the quality assessment parameters and assessment standards corresponding to aging. The prediction model includes two prediction sub-models selected based on the current critical process type. They are the first prediction sub-model applied when the current critical process is winding and the second prediction sub-model applied when the current critical process is welding. Compared with reducing the waste of resources caused by unnecessary aging tests, this setting can comprehensively analyze whether the overall quality of the film capacitor is qualified when the next critical process is aging, fundamentally further reducing the waste of production and manufacturing resources.

[0023] It should be noted that the previous text has already given a conversion scheme for converting quality assessment parameters into their critical approximation degree based on the evaluation criteria. When the predicted value of the sub-node (the predicted value of the critical approximation degree of the quality assessment parameter under the next key process) is known, the predicted value of the quality assessment parameter under the next key process can be obtained by reverse derivation according to the conversion scheme. The specific reverse derivation process will not be repeated here. The description of judging whether the quality is qualified based on the evaluation criteria has been discussed in the previous text and will not be repeated here.

[0024] Example 2: Please see Figure 2 This invention provides an automated quality inspection method for thin-film capacitors, used in the small thin-film capacitor manufacturing process described in Embodiment 1 above, comprising the following steps: Step S1: Collect quality assessment parameters of the film capacitor under the current key process, the key process including winding, welding, energizing and aging in sequence; Step S2: Based on the set evaluation criteria, analyze the quality evaluation parameters under the current key process to determine whether the quality of the film capacitor after the current key process is qualified, and perform directed graph construction after the judgment result is qualified. Step S3: For the current critical process except the last critical process, take each quality assessment parameter corresponding to it as the parent node and each quality assessment parameter in the next critical process as the child node. Determine the critical approximation degree of the quality assessment parameter based on the assessment standard as the node feature value. Combine the process parameters of the next critical process to determine the correlation between the parent node and the child node. Determine the directed edges between the nodes based on the correlation to construct a directed graph. Step S4: Based on the directed edges in the directed graph, determine the neighboring nodes of the parent node relative to each child node, aggregate the feature values ​​of the neighboring nodes to obtain the supplementary information of the parent node relative to each child node, and construct the parent node feature matrix, which includes the parent node feature values, the correlation between the parent node and each child node, and the supplementary information of the parent node relative to each child node. Step S5: Based on the prediction model, the feature matrix of each parent node and the process parameters of the next key process are fused and analyzed to output the predicted value of the child node features. The predicted value of the child node features is then analyzed based on the evaluation criteria to make an alarm judgment on the use of the process parameters of the next key process to process the film capacitor.

[0025] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0026] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0027] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0028] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A small-scale thin-film capacitor production apparatus, characterized in that, include: The data acquisition module is used to collect quality assessment parameters of the film capacitor under the current key process, which includes winding, welding, energizing and aging. The quality inspection module analyzes the quality assessment parameters under the current key process based on the set evaluation criteria to determine whether the quality of the film capacitor after the current key process is qualified, and performs directed graph construction after the judgment result is qualified. The directed graph construction module, for the current critical process (excluding the last critical process), takes its corresponding quality assessment parameters as parent nodes and the quality assessment parameters in the next critical process as child nodes. Based on the assessment criteria, it determines the critical approximation degree of the quality assessment parameters and uses it as the node feature value. Combined with the process parameters of the next critical process, it determines the correlation between parent and child nodes, and determines the directed edges between nodes based on the correlation, thereby constructing a directed graph. The feature reconstruction module, based on the directed edges in the directed graph, determines the neighboring nodes of the parent node relative to each child node, aggregates the feature values ​​of the neighboring nodes to obtain supplementary information of the parent node relative to each child node, and constructs the parent node feature matrix, which includes the parent node feature values, the correlation between the parent node and each child node, and the supplementary information of the parent node relative to each child node. The early warning module includes a prediction model that integrates and analyzes the feature matrices of each parent node and the process parameters of the next critical process to output the predicted values ​​of the child node features, and an alarm unit that analyzes the predicted values ​​of the child node features based on evaluation criteria to make an alarm judgment when the film capacitor is processed using the process parameters of the next critical process. For quality assessment parameters that apply the rule of "not greater than the corresponding threshold" in the evaluation criteria, their critical approximation degree is characterized by the ratio of their value to the corresponding threshold. For quality assessment parameters that apply the rule of "not less than the corresponding threshold" in the evaluation criteria, the relative error between their value and the corresponding threshold is calculated, and the product of the relative error and -1 is taken as the critical approximation degree of this quality assessment parameter. The correlation is characterized by the absolute value of the Pearson correlation coefficient. For any parent node and any child node, they are defined as the target parent node and the target child node, respectively. The logic for determining the correlation between the two is as follows: the critical process to which the current critical process belongs is defined as the first benchmark process, and the critical process to which the next critical process belongs is defined as the second benchmark process. Film capacitors with process parameters similar to those of the second benchmark process and the next critical process in historical production are used as candidate samples. Then, candidate samples, except for the target parent node, whose quality assessment parameter values ​​collected under the first benchmark process are similar to those collected under the current critical process are selected as reference samples. Pearson analysis is performed on the characteristic values ​​of the target parent node and the target child node of the reference samples to determine the absolute value of the Pearson correlation coefficient between the target parent node and the target child node under the process parameters of the next critical process.

2. The small film capacitor production apparatus according to claim 1, characterized in that: The types of quality assessment parameters are determined based on the types of key processes, as follows: For the critical winding process, the quality assessment parameters collected include alignment error, thickness uniformity, and the proportion of surface defects, including bubbles, wrinkles, and damage. For the critical welding process, the quality assessment parameters collected include the percentage of welding defects and the contact resistance failure rate. Welding defects include incomplete welds, detached welds, and cracks. For the critical process of enabling, the quality assessment parameters collected include initial leakage current, initial capacitance, initial loss angle, and initial equivalent series resistance. For the critical aging process, the quality assessment parameters collected include leakage current degradation rate, capacitance degradation rate, loss angle degradation rate, equivalent series resistance degradation rate, and service life prediction.

3. The small film capacitor production apparatus according to claim 2, characterized in that: The evaluation criteria are determined based on the types of quality evaluation parameters, as follows: When the current critical process is winding, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: Alignment error shall not exceed the alignment error threshold; Rule 2: The thickness range shall not exceed the thickness range threshold, and the thickness coefficient of variation shall not exceed the thickness coefficient of variation threshold. Rule 3: The proportion of surface defects shall not exceed the threshold for the proportion of surface defects; When the current critical process is welding, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The proportion of welding defects shall not exceed the threshold for the proportion of welding defects; Rule 2: The contact resistance failure rate shall not exceed the failure rate threshold; When the current critical process is to enable, the evaluation criteria include the following rules. If the following rules are met at the same time, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The initial leakage current shall not exceed the leakage current threshold. Rule 2: The initial capacitance shall not be less than the capacitance threshold. Rule 3: The initial loss angle shall not exceed the loss angle threshold. Rule 4: The initial equivalent series resistance shall not exceed the equivalent resistance threshold. When the current critical process is aging, the evaluation criteria include the following rules. If the following rules are met simultaneously, the film capacitor processed by the current critical process is deemed to be of qualified quality; otherwise, it is deemed to be unqualified. Rule 1: The rate of leakage current degradation shall not exceed the leakage current degradation threshold. Rule 2: The rate of capacity degradation shall not exceed the capacity degradation threshold; Rule 3: The rate of loss angle deterioration shall not exceed the loss angle deterioration threshold; Rule 4: The rate of degradation of the equivalent series resistance shall not exceed the threshold value for degradation of the equivalent resistance. Rule 5: The predicted service life value shall not be less than the service life threshold.

4. The small film capacitor production apparatus according to claim 1, characterized in that: The logic for establishing directed edges is as follows: if the correlation between a parent node and a child node is greater than the correlation threshold, then a directed edge is established from the parent node to the child node; otherwise, no directed edge is established between them.

5. The small film capacitor production apparatus according to claim 1, characterized in that: For any parent node, Its feature matrix has three columns and the number of rows is the same as the number of child nodes. The elements in the first column are the feature values ​​of the parent node, the elements in the second column are the correlations between the parent node and each child node, and the elements in the third column are the supplementary information of the parent node relative to each child node. Specifically, for any parent node and any child node, they are defined as the target parent node and the target child node, respectively. The logic for determining the supplementary information of the target parent node relative to the target child node is as follows: We count the other parent nodes that have directed edges with the target child node, excluding the target parent node, and use them as the neighbor nodes of the target parent node relative to the target child node. With the cumulative value equal to 1 as a constraint, we scale the relevance between each neighbor node and the target child node proportionally to obtain the relevance weight of each neighbor node and the target child node. We calculate the product of the feature value of each neighbor node and its corresponding relevance weight and sum them up to provide supplementary information about the target parent node relative to the target child node.

6. The small film capacitor production apparatus according to claim 1, characterized in that: The prediction model includes three prediction sub-models selected based on the current key process type: the first prediction sub-model applied when the current key process is winding, the second prediction sub-model applied when the current key process is welding, and the third prediction sub-model applied when the current key process is empowerment. Each prediction sub-model includes an input layer, which receives the feature matrices of each parent node and the process parameters of the next key process; a hidden layer, which extracts features from the feature matrices of each parent node and the process parameters of the next key process; and an output layer, which outputs the predicted feature values ​​of each child node. The alarm unit works as follows: based on the evaluation criteria, it analyzes the predicted values ​​of sub-nodes, deduces the predicted values ​​of quality evaluation parameters for the next critical process, and combines the evaluation criteria to determine whether the quality of the film capacitor after the next critical process is qualified. If the result is unqualified, an alarm is triggered.

7. The small film capacitor production apparatus according to claim 6, characterized in that: The key processes are defined as including only winding, welding and energizing. The quality assessment parameters and standards corresponding to aging are merged into the quality assessment parameters and standards corresponding to energizing. The prediction model includes two prediction sub-models selected based on the current key process type: the first prediction sub-model applied when the current key process is winding and the second prediction sub-model applied when the current key process is welding.

8. A method for producing a small film capacitor, using the small film capacitor production apparatus according to any one of claims 1-7, characterized in that, Includes the following steps: Step S1: Collect quality assessment parameters of the film capacitor under the current key process, which includes winding, welding, energizing and aging; Step S2: Based on the set evaluation criteria, analyze the quality evaluation parameters under the current key process to determine whether the quality of the film capacitor after the current key process is qualified, and perform directed graph construction after the judgment result is qualified. Step S3: For the current critical process except the last critical process, take each quality assessment parameter corresponding to it as the parent node and each quality assessment parameter in the next critical process as the child node. Determine the critical approximation degree of the quality assessment parameter based on the assessment standard as the node feature value. Combine the process parameters of the next critical process to determine the correlation between the parent node and the child node. Determine the directed edges between the nodes based on the correlation to construct a directed graph. Step S4: Based on the directed edges in the directed graph, determine the neighboring nodes of the parent node relative to each child node, aggregate the feature values ​​of the neighboring nodes to obtain the supplementary information of the parent node relative to each child node, and construct the parent node feature matrix, which includes the parent node feature values, the correlation between the parent node and each child node, and the supplementary information of the parent node relative to each child node. Step S5: Based on the prediction model, the feature matrix of each parent node and the process parameters of the next key process are fused and analyzed to output the predicted value of the child node features. The predicted value of the child node features is then analyzed based on the evaluation criteria to make an alarm judgment on the use of the process parameters of the next key process to process the film capacitor.

Images