A production quality intelligent detection system and method applied to capacitors
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANTONG NANMING ELECTRONICS
- Filing Date
- 2026-04-08
- Publication Date
- 2026-05-29
Smart Images
Figure CN121978592B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of capacitor production quality inspection technology, specifically an intelligent inspection system and method for capacitor production quality. Background Technology
[0002] In the mass production of capacitors, aging treatment is one of the key processes. By applying the rated voltage or slightly higher at a certain temperature, the oxide film inside the capacitor is repaired and improved, while products with inherent defects are prematurely degraded for removal. After aging treatment, the capacitors enter the final testing stage, where electrical parameters such as capacitance, loss tangent, and leakage current are measured to determine their qualification.
[0003] Current aging testing methods primarily employ batch processing, aging capacitors of the same model under identical temperature, voltage, and time conditions, and then using the static electrical parameters at the end of aging as the basis for acceptance. However, this method ignores the dynamic impact of individual capacitor state differences formed in previous processes on the aging process. Some capacitors, despite having potential defects such as microcracks, loose winding, or insufficient impregnation, may temporarily appear to have acceptable electrical parameters under uniform aging conditions, while actual internal damage has already evolved during the aging process. Because current testing focuses only on the final result and fails to capture dynamic characteristics reflecting the evolution of individual states, such as leakage current changes and capacitance recovery trends, these products are difficult to identify effectively and become a major source of early failure during user operation. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent inspection system and method for capacitor production quality, in order to solve the problems mentioned in the background art.
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a smart detection method for the production quality of capacitors, comprising:
[0006] Step S100: Collect the previous process data of each capacitor to be aged, and create a unique identifier for each capacitor; calculate the state stability index based on the previous process data, and dynamically classify the capacitor into several initial state categories based on the state stability index.
[0007] Step S200: Combining the statistical characteristics of each initial state category and the individual state stability index of each capacitor within the category, configure a corresponding personalized initial aging parameter scheme for each capacitor. The personalized initial aging parameter scheme includes differentiated aging temperature range, applied voltage range and reference aging time.
[0008] Step S300: Based on the personalized initial aging parameter scheme, perform aging treatment on a single capacitor at an independent workstation, and continuously collect leakage current and capacitance data during the aging process to obtain real-time aging response characteristics; match the real-time aging response characteristics with the preset standard response characteristics of this category to calculate the aging matching degree; determine whether the capacitor needs to be dynamically migrated to the category based on the aging matching degree; when it deviates from the current category trajectory, automatically migrate the capacitor to the target category and switch to the corresponding aging parameter scheme.
[0009] Step S400: Record the category migration trajectory, parameter adjustment process, and execution time of each stage for each capacitor to form an individualized aging profile; after aging is completed, test the final electrical parameters, and combine the individualized aging profile, aging matching degree, and final electrical parameters to make a comprehensive judgment and output the quality level of each capacitor.
[0010] Furthermore, step S100 includes:
[0011] Step S101: On the capacitor production line, the process parameters of each capacitor to be aged in the previous process stage are obtained through the production management system, including winding tension value, impregnation saturation value and oxide film thickness value; a unique identifier is created for each capacitor, a unique serial number is assigned to it, and the collected raw data is associated with and stored with the serial number; the collected raw data is preprocessed, outliers are removed and noise is smoothed using the moving average method to obtain the preprocessed parameter values;
[0012] Suppose there are n capacitors to be aged, each capacitor corresponding to three process parameter characteristics, denoted by x. ij This represents the j-th original characteristic value of the i-th capacitor, where j=1,2,3 correspond to winding tension, impregnation saturation, and oxide film thickness, respectively. First, the mean μ of the j-th characteristic value across all capacitors is calculated. j and standard deviation σ j Then, Z-score standardization is performed on each feature: x' ij =(x ij -μ j ) / σ j ;x' ij This represents the standardized value of the i-th capacitor based on the j-th feature.
[0013] Step S102: Standardize the value x ij By mapping to the [0,1] interval using range transformation, the probability value p is obtained. ij =[x' ij -min(x' ij )] / [max(x' ij )-min(x' ij )]; where p ijLet x' be the probability percentage of the i-th capacitor on the j-th feature; min(x') ij ) and max(x' ij ) represent the minimum and maximum values of all capacitors after standardization on the j-th feature, respectively;
[0014] Calculate the information entropy of the j-th feature:
[0015] ;
[0016] Among them, E j The information entropy of the j-th feature;
[0017] Calculate the weight of the j-th feature based on information entropy:
[0018] ;
[0019] Among them, w j This represents the weight of the j-th feature;
[0020] The state stability index is obtained by summing the standardized eigenvalues of each capacitor with their corresponding weights.
[0021] ;
[0022] Among them, Z i This represents the stability index of the i-th capacitor;
[0023] Step S103: Assign state stability indices {Z1, Z2, ..., Z} corresponding to all capacitors n As one-dimensional clustering sample points, the K-means clustering algorithm is used to perform initial state category division. Based on the production conditions and product specifications, a pre-set number of clusters K is used, and K state stability indices are randomly selected as initial cluster centers. The Euclidean distance from each sample point to its corresponding cluster center is iteratively calculated, and the sample is assigned to the nearest category. After each iteration, the cluster centers for each category are recalculated until the maximum number of iterations is reached, and clustering is completed after convergence. Each capacitor is assigned an initial category label C. i (0) ∈{1,2,…,K}, and record the cluster centers and sample standard deviations of each category to complete the initial state category division.
[0024] Furthermore, step S200 includes:
[0025] Step S201: Obtain the state stability index, the cluster center of the initial category, and the category number k for each capacitor. Calculate the difference between the individual state stability index and the cluster center of the category to obtain the individual deviation: D. i =Z i -Zc,k ; where D i Z represents the deviation of the i-th capacitor from its category; c,k Let k represent the cluster center of the k-th initial state category; k = {1, 2, ..., K}, where K is the total number of initial state categories.
[0026] Step S202: Based on the initial category number k and category cluster center Z c,k and individual deviation D i Configure personalized initial aging parameter schemes for each capacitor; preset the reference aging temperature T for the k-th category. base,k Reference applied voltage U base,k and the baseline aging time t base,k According to individual deviation D i By correcting the reference parameters, the individual parameters of the capacitor are obtained: U i =U base,k -λ U ·D i ;t i =t base,k +λ t ·D i Among them, U i Apply a voltage t to the i-th capacitor individually. i λ represents the personalized aging time for the i-th capacitor. U and λ t These are the preset voltage correction factor and duration correction factor; the aging temperature uses a category-unified value T. i =T base,k The final initial aging parameter scheme includes the aging temperature range, the applied voltage range, and the personalized aging time.
[0027] Furthermore, step S300 includes:
[0028] Step S301: Based on the personalized initial aging parameter scheme, each capacitor is placed in an independent aging stage and aging is performed according to its set aging temperature range, applied voltage range and personalized aging time. During the aging process, the leakage current value and capacitance value of the capacitor are continuously collected at a fixed sampling frequency and recorded as I(t) and C(t) respectively to form real-time leakage current curve and capacitance curve.
[0029] Step S302: Perform data preprocessing on the real-time acquired leakage current curve and capacitance curve, including outlier removal and moving average filtering. Then, extract the real-time aging response characteristics of the capacitor's internal state evolution from the preprocessed curves, including: leakage current stability value I. s Leakage current peak value I peak Leakage current fall time constant τ, capacitance recovery rate Ch and the capacitance plateau value C p ;
[0030] Among them, the peak leakage current I peak Within a preset time window at the initial stage of voltage application, the maximum value of the leakage current curve is taken as the peak leakage current; the stable leakage current value I... s In the later stages of aging, the average value of all sampling points within the last 10% of the leakage current curve is taken as the stable leakage current value; the leakage current fall time constant τ is obtained through exponential fitting, specifically: I(t) = I s +(I peak -I s )e -t / τ t represents the time variable, specifically the time elapsed from the start of voltage application during the aging process to the current sampling moment; the capacitance plateau value C p In the later stages of aging, the average value of all sampling points within the same time period as the leakage current stability value is taken; the capacitance recovery rate C h It is obtained by extracting the recovery segment of the capacitance curve from the initial value to the plateau value and performing linear fitting on the recovery segment data. The plateau value is the average value of the capacitance curve within a preset time interval after it enters the stable stage.
[0031] Therefore, for the i-th capacitor, a real-time aging response feature vector F is constructed. i =[I s,i ,τ i C h,i C p,i ];F i I s,i τ i C h,i C p,i Let represent the real-time aging response feature vector, leakage current stability value, leakage current fall time constant, capacitance recovery rate, and capacitance plateau value of the i-th capacitor, respectively.
[0032] Step S303: For the k-th initial state category, based on the aging response data of historical qualified samples of this category, calculate the mean of each feature to obtain the category standard response feature vector; calculate the standard deviation of each feature to obtain the corresponding feature standard deviation; after normalizing each feature, calculate the weighted Euclidean distance d between the real-time aging response feature of the i-th capacitor and the standard response feature of its category. i The calculation formula is:
[0033] ;
[0034] Among them, w m F represents the preset weight of the m-th aging response feature;i,m F represents the m-th real-time aging response characteristic value of the i-th capacitor; c,m σ represents the m-th standard response characteristic value of the current category; c,m This represents the standard deviation of the m-th feature in the current category; then, the aging matching degree M is calculated. i =exp(-d i );
[0035] Matching degree M i Compared with the preset migration threshold M th Compare; if M i ≥M th If the capacitor aging process is deemed to be in line with the current category expectations, the current category and aging parameter scheme should be maintained; if M i <M th If the capacitor deviates from the current category trajectory, dynamic category migration needs to be performed.
[0036] Step S304: For the capacitor to be transferred, calculate its real-time response feature vector F. i The matching degree between the capacitor and the standard response feature vectors of all other categories is used to select the category with the highest matching degree as the target category. The capacitor is automatically migrated to the target category and switched to the aging parameter scheme corresponding to the target category. The aging parameters after the switch need to be recalculated based on the individual deviation of the capacitor in the target category. At the same time, the time of this migration, the original category, the target category, and the parameter adjustment data are recorded. During the aging process, step S303 is repeated continuously at fixed time intervals until the personalized aging time of the capacitor is reached. Finally, the complete category migration trajectory of each capacitor in the entire aging process is recorded, including the number of migrations, the migration time, the category before and after the migration, and the parameter adjustment history of each stage.
[0037] Furthermore, step S400 includes:
[0038] Step S401: After the capacitor has completed the personalized aging process, test its final electrical parameters, including at least capacitance, loss tangent and leakage current, and compare these parameters with the specifications of the corresponding capacitor model. If any parameter exceeds the qualified range, the capacitor is directly marked as a defective product.
[0039] Step S402: For capacitors whose final electrical parameters are all qualified, obtain the complete aging process data recorded in step S300, including the number of migrations N. move Minimum matching degree M min and the matching degree is lower than the migration threshold M th Cumulative duration T move ; Calculate the overall stability coefficient of each capacitor: P i =Nmove ×(1-M min )×T move Among them, P i Let represent the overall stability coefficient of the i-th capacitor;
[0040] Step S403: Calculate the overall stability coefficient P for all qualified capacitors. i Perform statistical analysis and calculate its average value P. avg and standard deviation σ P According to P of each capacitor i The quality level is determined by the relationship between the value and the overall distribution: if P i ≤P avg -σ P If P is positive, the capacitor is considered a superior product, indicating that its aging process is stable; if P is negative, the capacitor is considered a superior product. avg -σ P <P i ≤P avg +σ P If P is within the acceptable range, the capacitor is considered a qualified product, indicating that its aging process is within the normal fluctuation range; if P i >P avg +σ P If the capacitor is found to be suspicious, it indicates that there are obvious abnormal fluctuations in its aging process, and further confirmation is required by retesting or short-term life test.
[0041] Step S404: Store the quality grade determination result, individualized aging record, and final electrical parameters of each capacitor in the production quality database. Simultaneously, extract the data of capacitors judged as non-conforming or suspicious products separately, and associate and store their preceding process characteristic data with the migration count, minimum matching degree, and abnormal cumulative duration during the aging process to form a defect feature library. If multiple non-conforming or suspicious products appear consecutively during subsequent production, and their preceding process characteristic distribution matches the abnormal patterns in the defect feature library, an automatic warning will be triggered, reminding process personnel to check whether there is any drift in the preceding process equipment status or process parameters, so that timely adjustments can be made.
[0042] A smart quality inspection system for capacitor production includes an initial classification modeling module, a personalized parameter configuration module, an adaptive aging control module, and a quality assessment module.
[0043] The initial classification modeling module collects the pre-process data of each capacitor to be aged, and creates a unique identifier for each capacitor; it calculates the state stability index based on the pre-process data, and dynamically classifies the capacitor into several initial state categories based on the state stability index.
[0044] The personalized parameter configuration module combines the statistical characteristics of each initial state category and the individual state stability index of each capacitor within the category to configure a corresponding personalized initial aging parameter scheme for each capacitor. The personalized initial aging parameter scheme includes differentiated aging temperature range, applied voltage range and reference aging time.
[0045] The adaptive aging control module performs aging treatment on a single capacitor at an independent workstation based on a personalized initial aging parameter scheme. During the aging process, leakage current and capacitance data are continuously collected to obtain real-time aging response characteristics. The real-time aging response characteristics are matched with the preset standard response characteristics of this category to calculate the aging matching degree. Based on the aging matching degree, it is determined whether the capacitor needs to be dynamically migrated to the category. When it deviates from the current category trajectory, the capacitor is automatically migrated to the target category and switched to the corresponding aging parameter scheme.
[0046] The quality assessment module records the category migration trajectory, parameter adjustment process, and execution time of each stage for each capacitor, forming an individualized aging profile. After aging is completed, the final electrical parameters are tested, and the quality level of each capacitor is output by combining the individualized aging profile, aging matching degree, and final electrical parameters.
[0047] Compared with the prior art, the beneficial effects achieved by the present invention are:
[0048] This invention quantifies process parameters such as winding tension, impregnation saturation, and oxide film thickness into state stability indices through multi-feature fusion and adaptive clustering of previous process data, achieving accurate classification of the initial state of capacitors. This solves the problem of traditional aging processes using uniform parameters that cannot adapt to individual differences. Based on category features and individual deviation, personalized aging parameters are dynamically generated, avoiding under-aging, over-aging, or internal damage caused by uniform aging conditions, thus improving the targeting and rationality of the aging process.
[0049] Furthermore, this invention achieves online assessment and adaptive adjustment of aging status through real-time response feature matching and dynamic category migration during the aging process, overcoming the limitations of traditional detection methods that rely solely on final electrical parameters and cannot identify potential process hazards. Finally, it combines a comprehensive stability coefficient to achieve objective quality grading, while simultaneously associating and storing pre-process data and process anomaly features of defective and suspicious products to form a defect feature library, establishing an anomaly early warning mechanism, enabling reverse tracing of quality issues and process early warning, effectively improving the consistency and reliability of capacitors leaving the factory, and reducing defect rates and rework costs. Attached Figure Description
[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0051] Figure 1 This is a flowchart of a method for intelligent quality inspection applied to capacitor production. Detailed Implementation
[0052] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0053] Please see Figure 1 This invention provides a technical solution: an intelligent detection method for capacitor production quality, comprising:
[0054] Step S100: Collect the previous process data of each capacitor to be aged, and create a unique identifier for each capacitor; calculate the state stability index based on the previous process data, and dynamically classify the capacitor into several initial state categories based on the state stability index.
[0055] Step S200: Combining the statistical characteristics of each initial state category and the individual state stability index of each capacitor within the category, configure a corresponding personalized initial aging parameter scheme for each capacitor. The personalized initial aging parameter scheme includes differentiated aging temperature range, applied voltage range and reference aging time.
[0056] Step S300: Based on the personalized initial aging parameter scheme, perform aging treatment on a single capacitor at an independent workstation, and continuously collect leakage current and capacitance data during the aging process to obtain real-time aging response characteristics; match the real-time aging response characteristics with the preset standard response characteristics of this category to calculate the aging matching degree; determine whether the capacitor needs to be dynamically migrated to the category based on the aging matching degree; when it deviates from the current category trajectory, automatically migrate the capacitor to the target category and switch to the corresponding aging parameter scheme.
[0057] Step S400: Record the category migration trajectory, parameter adjustment process, and execution time of each stage for each capacitor to form an individualized aging profile; after aging is completed, test the final electrical parameters, and combine the individualized aging profile, aging matching degree, and final electrical parameters to make a comprehensive judgment and output the quality level of each capacitor.
[0058] Step S100 includes:
[0059] Step S101: On the capacitor production line, the process parameters of each capacitor to be aged in the previous process stage are obtained through the production management system, including at least the winding tension value, impregnation saturation value and oxide film thickness value; a unique identifier is created for each capacitor, a unique serial number is assigned to it, and the collected raw data is associated with and stored with the serial number; the collected raw data is preprocessed, outliers are removed and noise is smoothed using the moving average method to obtain the preprocessed parameter values;
[0060] Suppose there are n capacitors to be aged, each capacitor corresponding to three process parameter characteristics, denoted by x. ij This represents the j-th original characteristic value of the i-th capacitor, where j=1,2,3 correspond to winding tension, impregnation saturation, and oxide film thickness, respectively. First, the mean μ of the j-th characteristic value across all capacitors is calculated. j and standard deviation σ j Then, Z-score standardization is performed on each feature: x' ij =(x ij -μ j ) / σ j ;x' ij This represents the standardized value of the i-th capacitor based on the j-th feature.
[0061] Step S102: Standardize the value x ij By mapping to the [0,1] interval using range transformation, the probability value p is obtained. ij =[x' ij -min(x' ij )] / [max(x' ij )-min(x' ij )]; where p ij Let x' be the probability percentage of the i-th capacitor on the j-th feature; min(x') ij ) and max(x' ij ) represent the minimum and maximum values of all capacitors after standardization on the j-th feature, respectively;
[0062] Calculate the information entropy of the j-th feature:
[0063] ;
[0064] Among them, E j The information entropy of the j-th feature;
[0065] Calculate the weight of the j-th feature based on information entropy:
[0066] ;
[0067] Among them, w j This represents the weight of the j-th feature;
[0068] The state stability index is obtained by summing the standardized eigenvalues of each capacitor with their corresponding weights.
[0069] ;
[0070] Among them, Z i This represents the stability index of the i-th capacitor;
[0071] Step S103: Assign state stability indices {Z1, Z2, ..., Z} corresponding to all capacitors n As one-dimensional clustering sample points, the K-means clustering algorithm is used to perform initial state category division. Based on the production conditions and product specifications, a pre-set number of clusters K is used, and K state stability indices are randomly selected as initial cluster centers. The Euclidean distance from each sample point to its corresponding cluster center is iteratively calculated, and the sample is assigned to the nearest category. After each iteration, the cluster centers for each category are recalculated until the maximum number of iterations is reached, and clustering is completed after convergence. Each capacitor is assigned an initial category label C. i (0) ∈{1,2,…,K}, and record the cluster centers and sample standard deviations of each category to complete the initial state category division.
[0072] Step S200 includes:
[0073] Step S201: Obtain the state stability index, the cluster center of the initial category, and the category number k for each capacitor. Calculate the difference between the individual state stability index and the cluster center of the category to obtain the individual deviation: D. i =Z i -Z c,k ; where D i Z represents the deviation of the i-th capacitor from its category; c,k Let k represent the cluster center of the k-th initial state category; k = {1, 2, ..., K}, where K is the total number of initial state categories.
[0074] Step S202: Based on the initial category number k and category cluster center Z c,k and individual deviation D i Configure personalized initial aging parameter schemes for each capacitor; preset the reference aging temperature T for the k-th category. base,k Reference applied voltage U base,k and the baseline aging time t base,k According to individual deviation D i By correcting the reference parameters, the individual parameters of the capacitor are obtained: U i =U base,k -λ U·D i ;t i =t base,k +λ t ·D i Among them, U i Apply a voltage t to the i-th capacitor individually. i λ represents the personalized aging time for the i-th capacitor. U and λ t These are the preset voltage correction factor and duration correction factor, respectively; where λ U The value of λ is chosen so that the voltage adjustment does not exceed ±10% of the reference voltage. U ∈[0.05,0.2];λ t The value is chosen to keep the aging time adjustment within ±20% of the base time, λ t ∈[0.2,0.5]; Aging temperature adopts the category-unified value T i =T base,k The final initial aging parameter scheme includes the aging temperature range, the applied voltage range, and the personalized aging time.
[0075] Step S300 includes:
[0076] Step S301: Based on the personalized initial aging parameter scheme, each capacitor is placed in an independent aging stage and aging is performed according to its set aging temperature range, applied voltage range and personalized aging time. During the aging process, the leakage current value and capacitance value of the capacitor are continuously collected at a fixed sampling frequency and recorded as I(t) and C(t) respectively to form real-time leakage current curve and capacitance curve.
[0077] Step S302: Perform data preprocessing on the real-time acquired leakage current curve and capacitance curve, including outlier removal and moving average filtering. Then, extract the real-time aging response characteristics of the capacitor's internal state evolution from the preprocessed curves, including: leakage current stability value I. s Leakage current peak value I peak Leakage current fall time constant τ, capacitance recovery rate C h and the capacitance plateau value C p ;
[0078] Among them, the peak leakage current I peak Within a preset time window at the initial stage of voltage application, the maximum value of the leakage current curve is taken as the peak leakage current; the stable leakage current value I... s In the later stages of aging, the average value of all sampling points within the last 10% of the leakage current curve is taken as the stable leakage current value; the leakage current fall time constant τ is obtained through exponential fitting, specifically: I(t) = I s +(I peak -Is )e -t / τ t represents the time variable, specifically the time elapsed from the start of voltage application during the aging process to the current sampling moment; the capacitance plateau value C p In the later stages of aging, the average value of all sampling points within the same time period as the leakage current stability value is taken; the capacitance recovery rate C h It is obtained by extracting the recovery segment of the capacitance curve from the initial value to the plateau value and performing linear fitting on the recovery segment data; the plateau value is the average value of the capacitance curve within a preset time interval after it enters the stable stage.
[0079] Therefore, for the i-th capacitor, a real-time aging response feature vector F is constructed. i =[I s,i ,τ i C h,i C p,i ];F i I s,i τ i C h,i C p,i Let represent the real-time aging response feature vector, leakage current stability value, leakage current fall time constant, capacitance recovery rate, and capacitance plateau value of the i-th capacitor, respectively.
[0080] Step S303: For the k-th initial state category, based on the aging response data of historical qualified samples of this category, calculate the mean of each feature to obtain the category standard response feature vector; calculate the standard deviation of each feature to obtain the corresponding feature standard deviation; after normalizing each feature, calculate the weighted Euclidean distance d between the real-time aging response feature of the i-th capacitor and the standard response feature of its category. i The calculation formula is:
[0081] ;
[0082] Among them, w m F represents the preset weight of the m-th aging response feature; i,m F represents the m-th real-time aging response characteristic value of the i-th capacitor; c,m σ represents the m-th standard response characteristic value of the current category; c,m This represents the standard deviation of the m-th feature in the current category; then, the aging matching degree M is calculated. i =exp(-d i );
[0083] Matching degree M i Compared with the preset migration threshold M th Compare; if M i ≥M thIf the capacitor aging process is deemed to be in line with the current category expectations, the current category and aging parameter scheme should be maintained; if M i <M th If the capacitor deviates from the current category trajectory, dynamic category migration needs to be performed.
[0084] Step S304: For the capacitor to be transferred, calculate its real-time response feature vector F. i The matching degree between the capacitor and the standard response feature vectors of all other categories is used to select the category with the highest matching degree as the target category. The capacitor is automatically migrated to the target category and switched to the aging parameter scheme corresponding to the target category. The aging parameters after the switch need to be recalculated based on the individual deviation of the capacitor in the target category. At the same time, the time of this migration, the original category, the target category, and the parameter adjustment data are recorded. During the aging process, step S303 is repeated continuously at fixed time intervals until the personalized aging time of the capacitor is reached. Finally, the complete category migration trajectory of each capacitor in the entire aging process is recorded, including the number of migrations, the migration time, the category before and after the migration, and the parameter adjustment history of each stage.
[0085] Step S400 includes:
[0086] Step S401: After the capacitor has completed the personalized aging process, test its final electrical parameters, including at least capacitance, loss tangent and leakage current, and compare these parameters with the specifications of the corresponding capacitor model. If any parameter exceeds the qualified range, the capacitor is directly marked as a defective product.
[0087] Step S402: For capacitors whose final electrical parameters are all qualified, obtain the complete aging process data recorded in step S300, including the number of migrations N. move Minimum matching degree M min and the matching degree is lower than the migration threshold M th Cumulative duration T move ; Calculate the overall stability coefficient of each capacitor: P i =N move ×(1-M min )×T move Among them, P i This represents the overall stability coefficient of the i-th capacitor;
[0088] Step S403: Calculate the overall stability coefficient P for all qualified capacitors. i Perform statistical analysis and calculate its average value P. avg and standard deviation σ P According to P of each capacitor i The quality level is determined by the relationship between the value and the overall distribution: if P i ≤Pavg -σ P If P is positive, the capacitor is considered a superior product, indicating that its aging process is stable; if P is negative, the capacitor is considered a superior product. avg -σ P <P i ≤P avg +σ P If P is within the acceptable range, the capacitor is considered a qualified product, indicating that its aging process is within the normal fluctuation range; if P i >P avg +σ P If the capacitor is found to be suspicious, it indicates that there are obvious abnormal fluctuations in its aging process, and further confirmation is required by retesting or short-term life test.
[0089] Step S404: The quality grade judgment result, individualized aging record, and final electrical parameters of each capacitor are associated and stored in the production quality database. At the same time, the data of capacitors judged as non-conforming and suspicious products are extracted separately, and their previous process characteristic data are associated and stored with the migration number, minimum matching degree, and abnormal cumulative duration during the aging process to form a defect feature library. When multiple non-conforming or suspicious products appear consecutively in the subsequent production process, and their previous process characteristic distribution matches the abnormal pattern in the defect feature library, an early warning prompt is automatically triggered to remind process personnel to check whether there is a drift in the status of previous process equipment or process parameters so that timely adjustments can be made.
[0090] This invention is applied to the mass production of aluminum electrolytic capacitors. For a batch of 1000 aluminum electrolytic capacitors with specifications of 1000μF / 16V to be aged, the production line management system first collects the previous process parameters of the 1000 capacitors in this batch, specifically the winding tension (unit: N), impregnation saturation (unit: %), and oxide film thickness (unit: nm). Each capacitor is assigned a unique serial number C0001-C1000. The original data is stored in association with the serial number. The collected original data is preprocessed: outliers are removed using the 3σ criterion, noise is smoothed using the moving average method, and the window size is set to 5 to obtain the preprocessed data.
[0091] Let n=1000, x ij Let represent the j-th characteristic value of the i-th capacitor, where j=1 corresponds to winding tension, j=2 corresponds to impregnation saturation, and j=3 corresponds to oxide film thickness. Calculate the mean value μ of each characteristic. j and standard deviation σ j Z-score standardization is applied to each feature to obtain x' ij ; will x' ij By mapping to the [0,1] interval through range transformation, the probability value p is obtained. ij ; Calculate the information entropy E of each feature j Based on Ej Calculate weight w j Finally, the following weights were obtained: winding tension weight 0.35, impregnation saturation weight 0.4, and oxide film thickness weight 0.25. The state stability index Z of each capacitor was obtained by weighted summation. i ;
[0092] The preset number of clusters K=3 is used, and the K-means clustering algorithm is employed, with {Z1, Z2, ..., Z...} as the clustering groups. 1000} is a one-dimensional sample, and 3 Z values are randomly selected. i As initial cluster centers, Euclidean distances are iteratively calculated and the cluster centers are updated. The maximum number of iterations is set to 50. After the iterations converge, the clustering is completed, and each capacitor is assigned an initial category label C. i (0) ∈{1,2,3}, record the cluster centers of the three classes as 0.72, 0.51, and 0.33, respectively, and the sample standard deviations as 0.08, 0.07, and 0.09, respectively, thus completing the initial state class division.
[0093] Obtain the Z-axis of each capacitor i Cluster center Z of the category c,k And the category number k, calculate the individual deviation D i Preset baseline aging parameters for 3 initial categories: Category 1 (Z) c,1 =0.72) Reference temperature T base,1 =85℃, reference voltage U base,1 =18V, reference duration t base,1 =2h; Category 2 (Z) c,2 =0.51) Reference temperature T base,2 =80℃, reference voltage U base,2 =17V, reference duration t base,2 =2.5h; Category 3 (Z) c,3 =0.33) Reference temperature T base,3 =75℃, reference voltage U base,3 =16.5V, reference duration t base,3 =3h; Set voltage correction factor λ U =0.1V, duration correction factor λ t =0.3h, according to D i The modification yields personalized parameters, for example, serial number C0001 (category 1, Z). i =0.92, D i =0.2), its personalized parameter is U i =18-0.1×0.2=17.98V,t i =2 + 0.3 × 0.2 = 2.04h, T i=85℃, and finally generate a personalized initial aging parameter scheme for each capacitor.
[0094] 1000 capacitors were placed in independent aging stations and subjected to aging treatment according to their individual parameters. The sampling frequency was set to 1 time / 10s, and the leakage current I(t) and capacitance C(t) were continuously collected to form real-time curves. The curves were preprocessed to extract the real-time aging response characteristics: leakage current peak value I... peak The maximum value within the first 30 seconds of the applied voltage is taken, and the stable leakage current Is is taken as the average value of the samples during the last 10% of the aging period. τ is fitted using the exponential fitting formula I(t) = I s +(I peak -I s )e -t / τ C was obtained. p Take and I s The average capacitance over the same time period, C h The intercepted capacitance increases from the initial value to C. p The recovered segment is obtained through linear fitting, and the feature vector F is constructed. i ;
[0095] Based on historical qualified samples of each category, the standard response feature vector and feature standard deviation are statistically obtained. After feature normalization, the weights w of each feature are set. m Both are 0.25. Calculate the weighted Euclidean distance d. i Thus, the aging matching degree M is obtained. i =exp(-d i ), preset migration threshold M th =0.8; if M i ≥0.8, keep the current category and parameters; if M i <0.8, calculate F i The matching degree with the standard features of other categories is used to migrate to the target category with the highest matching degree. The parameters are recalculated and the migration information is recorded. The matching judgment is repeated every 5 minutes during the aging process until the personalized aging time is reached. Finally, the migration trajectory of each capacitor is recorded. For example, C0056 migrates from category 2 to category 3 at the aging time of 1.2h. The parameters are adjusted synchronously.
[0096] After the aging test, the final electrical parameters of each capacitor were measured: capacitance, loss tangent, and leakage current. These parameters were compared with the standard for this capacitor specification: capacitance 900-1100μF, loss tangent ≤0.15, and leakage current ≤10μA. Any capacitor exceeding these standards was marked as defective. A total of 12 defective capacitors were detected. For the remaining 988 qualified capacitors, the migration number N was obtained. move Minimum matching degree M min Low matching degree cumulative duration T moveCalculate the stability coefficient P of the integrated process. i ;Statistical results show P i The average value P avg =0.32, standard deviation σ P =0.11, classified according to rules: P i ≤0.21 (P) avg -σ P ) is a superior grade product (215 pieces in total), 0.21 <P i ≤0.43 (P) avg +σ P ) are qualified products (683 in total), P i Products with a strength >0.43 are considered suspicious (90 in total). The quality grade, individualized aging records, and final electrical parameters of all capacitors are stored in the production quality database. Data on 12 non-conforming products and 90 suspicious products are extracted and correlated with their previous process characteristics and process anomaly data to form a defect feature library. An early warning is triggered when three or more anomalies occur consecutively and the previous process characteristics match the defect library. In this embodiment, during subsequent production, four suspicious products occurred consecutively, and the winding tension of their previous processes all deviated from the normal range. The system automatically triggered an early warning, reminding process personnel to check the winding equipment and adjust the parameters in a timely manner to achieve closed-loop optimization.
[0097] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. A smart inspection method for capacitor manufacturing quality, characterized in that: The method includes: Step S100: Collect the previous process data of each capacitor to be aged, and create a unique identifier for each capacitor; calculate the state stability index based on the previous process data, and dynamically classify the capacitor into several initial state categories based on the state stability index. Step S200: Combining the statistical characteristics of each initial state category and the individual state stability index of each capacitor within the category, configure a corresponding personalized initial aging parameter scheme for each capacitor. The personalized initial aging parameter scheme includes differentiated aging temperature range, applied voltage range and reference aging time. Step S300: Based on the personalized initial aging parameter scheme, perform aging detection on a single capacitor at an independent workstation, and continuously collect leakage current and capacitance data during the aging process to obtain real-time aging response characteristics; match the real-time aging response characteristics with the preset standard response characteristics of this category to calculate the aging matching degree; determine whether the capacitor needs to be dynamically migrated to the category based on the aging matching degree; when it deviates from the current category trajectory, automatically migrate the capacitor to the target category and switch to the corresponding aging parameter scheme. Step S400: Record the category migration trajectory, parameter adjustment process, and execution time of each stage for each capacitor to form an individualized aging profile; after aging is completed, test the final electrical parameters, and combine the individualized aging profile, aging matching degree, and final electrical parameters to make a comprehensive judgment and output the quality level of each capacitor.
2. The intelligent inspection method for capacitor production quality according to claim 1, characterized in that: Step S100 includes: Step S101: On the capacitor production line, the process parameters of each capacitor to be aged in the previous process stage are obtained through the production management system, including winding tension value, impregnation saturation value and oxide film thickness value; a unique identifier is created for each capacitor, a unique serial number is assigned to it, and the collected raw data is associated with and stored with the serial number; the collected raw data is preprocessed, outliers are removed and noise is smoothed using the moving average method to obtain the preprocessed parameter values; Suppose there are n capacitors to be aged, each capacitor corresponding to three process parameter characteristics, denoted by x. ij This represents the j-th original characteristic value of the i-th capacitor, where j=1,2,3 correspond to winding tension, impregnation saturation, and oxide film thickness, respectively. First, the mean μ of the j-th characteristic value across all capacitors is calculated. j and standard deviation σ j Then, Z-score standardization is performed on each feature: x' ij =(x ij -μ j ) / σ j ;x' ij This represents the standardized value of the i-th capacitor based on the j-th feature.
3. The intelligent inspection method for capacitor production quality according to claim 2, characterized in that: Step S100 further includes: Step S102: Standardize the value x ij By mapping to the [0,1] interval using range transformation, the probability value p is obtained. ij =[x' ij -min(x' ij )] / [max(x' ij )-min(x' ij )]; where p ij Let x' be the probability percentage of the i-th capacitor on the j-th feature; min(x') ij ) and max(x' ij ) represent the minimum and maximum values of all capacitors after standardization on the j-th feature, respectively; Calculate the information entropy of the j-th feature: ； Among them, E j The information entropy of the j-th feature; Calculate the weight of the j-th feature based on information entropy: ； Among them, w j This represents the weight of the j-th feature; The state stability index is obtained by summing the standardized eigenvalues of each capacitor with their corresponding weights. ； Among them, Z i This represents the stability index of the i-th capacitor; Step S103: Assign state stability indices {Z1, Z2, ..., Z} corresponding to all capacitors n As one-dimensional clustering sample points, the K-means clustering algorithm is used to perform initial state category division: The number of clusters K is preset according to the production conditions and product specifications, and K state stability indices are randomly selected as initial cluster centers; the Euclidean distance from each sample point to its corresponding cluster center is iteratively calculated, and the sample is assigned to the nearest category; after each iteration, the cluster centers for each category are recalculated until the maximum number of iterations is reached, and clustering is completed after iterative convergence; each capacitor is assigned an initial category label C. i (0) ∈{1,2,…,K}, and record the cluster centers and sample standard deviations of each category to complete the initial state category division.
4. The intelligent inspection method for capacitor production quality according to claim 3, characterized in that: Step S200 includes: Step S201: Obtain the state stability index, the cluster center of the initial category, and the category number k for each capacitor. Calculate the difference between the individual state stability index and the cluster center of the category to obtain the individual deviation: D. i =Z i -Z c,k ; where D i Z represents the individual deviation of the i-th capacitor within its category; c,k Let k represent the cluster center of the k-th initial state category; k = {1, 2, ..., K}, where K is the total number of initial state categories. Step S202: Based on the initial category number k and category cluster center Z c,k and individual deviation D i Configure personalized initial aging parameter schemes for each capacitor; preset the reference aging temperature T for the k-th category. base,k Reference applied voltage U base,k and the baseline aging time t base,k According to individual deviation D i By correcting the reference parameters, the individual parameters of the capacitor are obtained: U i =U base,k -λ U ·D i ;t i =t base,k +λ t ·D i Among them, U i Apply a voltage t to the i-th capacitor individually. i λ represents the personalized aging time for the i-th capacitor. U and λ t These are the preset voltage correction factor and duration correction factor; the aging temperature adopts the category-unified value T. i =T base,k The final initial aging parameter scheme includes the aging temperature range, the applied voltage range, and the personalized aging time.
5. The intelligent inspection method for capacitor production quality according to claim 1, characterized in that: Step S300 includes: Step S301: Based on the personalized initial aging parameter scheme, each capacitor is placed in an independent aging stage, and aging tests are performed according to its set aging temperature range, applied voltage range and personalized aging time. During the aging process, the leakage current value and capacitance value of the capacitor are continuously collected at a fixed sampling frequency and recorded as I(t) and C(t) respectively to form real-time leakage current curve and capacitance curve. Step S302: Perform data preprocessing on the real-time acquired leakage current curve and capacitance curve, including outlier removal and moving average filtering. Then, extract the real-time aging response characteristics of the capacitor's internal state evolution from the preprocessed curves, including: leakage current stability value I. s Leakage current peak value I peak Leakage current fall time constant τ, capacitance recovery rate C h and the capacitance plateau value C p ; Among them, the peak leakage current I peak Within a preset time window at the initial stage of voltage application, the maximum value of the leakage current curve is taken as the peak leakage current; the stable leakage current value I... s In the later stages of aging, the average value of all sampling points within the last 10% of the leakage current curve is taken as the stable leakage current value; the leakage current fall time constant τ is obtained through exponential fitting, specifically: I(t) = I s +(I peak -I s )e -t / τ t represents the time variable, specifically the time elapsed from the start of voltage application during the aging process to the current sampling moment; the capacitance plateau value C p In the later stages of aging, the average value of all sampling points within the same time period as the leakage current stability value is taken; the capacitance recovery rate C h It is obtained by linearly fitting the data between the initial value and the plateau value of the capacitance curve. The plateau value is the average value of the capacitance curve within a preset time interval after it enters a stable stage. Therefore, for the i-th capacitor, a real-time aging response feature vector F is constructed. i =[I s,i ,τ i C h,i C p,i ];F i I s,i τ i C h,i C p,i Let represent the real-time aging response feature vector, leakage current stability value, leakage current fall time constant, capacitance recovery rate, and capacitance plateau value of the i-th capacitor, respectively.
6. The intelligent inspection method for capacitor production quality according to claim 5, characterized in that: Step S300 further includes: Step S303: For the k-th initial state category, based on the aging response data of historical qualified samples of this category, calculate the mean of each feature to obtain the category standard response feature vector; calculate the standard deviation of each feature to obtain the corresponding feature standard deviation; after normalizing each feature, calculate the weighted Euclidean distance d between the real-time aging response feature of the i-th capacitor and the standard response feature of its category. i The calculation formula is: ； Among them, w m F represents the preset weight of the m-th aging response feature; i,m F represents the real-time aging response characteristic value of the m-th term of the i-th capacitor; c,m σ represents the m-th standard response characteristic value of the current category; c,m The standard deviation of the m-th feature in the current category is represented; then the aging matching degree M is calculated. i =exp(-d i ); Matching degree M i Compared with the preset migration threshold M th Compare; if M i ≥M th If the capacitor aging process is deemed to be in line with the current category expectations, the current category and aging parameter scheme should be maintained; if M i <M th If the capacitor deviates from the current category trajectory, dynamic category migration needs to be performed. Step S304: For the capacitor to be transferred, calculate its real-time response feature vector F. i The matching degree between the capacitor and the standard response feature vectors of all other categories is used to select the category with the highest matching degree as the target category. The capacitor is automatically migrated to the target category and switched to the aging parameter scheme corresponding to the target category. The aging parameters after the switch need to be recalculated based on the individual deviation of the capacitor in the target category. At the same time, the time of this migration, the original category, the target category, and the parameter adjustment data are recorded. During the aging process, step S303 is repeated continuously at fixed time intervals until the personalized aging time of the capacitor is reached. Finally, the complete category migration trajectory of each capacitor in the entire aging process is recorded, including the number of migrations, the migration time, the category before and after the migration, and the parameter adjustment history of each stage.
7. The intelligent inspection method for capacitor production quality according to claim 1, characterized in that: Step S400 includes: Step S401: After the capacitor has completed the personalized aging process, test its final electrical parameters, including at least capacitance, loss tangent and leakage current, and compare these parameters with the specifications of the corresponding capacitor model. If any parameter exceeds the qualified range, the capacitor is directly marked as a defective product. Step S402: For capacitors whose final electrical parameters are all qualified, obtain the complete aging process data recorded in step S300, including the number of migrations N. move Minimum matching degree M min and the matching degree is lower than the migration threshold M th Cumulative duration T move ; Calculate the overall stability coefficient of each capacitor: P i =N move ×(1-M min )×T move Among them, P i This represents the overall stability coefficient of the i-th capacitor; Step S403: Calculate the overall stability coefficient P for all qualified capacitors. i Perform statistical analysis and calculate its average value P. avg and standard deviation σ P According to P of each capacitor i The quality level is determined by the relationship between the value and the overall distribution: if P i ≤P avg -σ P If P is positive, the capacitor is determined to be of superior quality, indicating that the aging test process is stable; if P is negative, the capacitor is determined to be of superior quality. avg -σ P <P i ≤P avg +σ P If P is within the acceptable range, the capacitor is considered a qualified product, indicating that the aging test process is within the normal fluctuation range; if P i >P avg +σ P If the capacitor is deemed suspicious, it indicates that there are significant abnormal fluctuations in the aging test process, and further confirmation by retesting or short-term life test is required. Step S404: The quality grade judgment result, individualized aging record, and final electrical parameters of each capacitor are associated and stored in the production quality database. At the same time, the data of capacitors judged as non-conforming and suspicious products are extracted separately, and their previous process characteristic data are associated and stored with the migration number, minimum matching degree, and abnormal cumulative duration during the aging process to form a defect feature library. When multiple non-conforming or suspicious products appear consecutively in the subsequent production process, and their previous process characteristics match the abnormal characteristics in the defect feature library, an early warning prompt is automatically triggered to remind process personnel to check whether there is a drift in the status of the previous process equipment or process parameters so that timely adjustments can be made.
8. An intelligent inspection system for capacitor production quality, characterized in that: The system includes an initial classification modeling module, a personalized parameter configuration module, an adaptive aging control module, and a quality assessment module. The initial classification modeling module collects the pre-process data of each capacitor to be aged, and creates a unique identifier for each capacitor; it calculates the state stability index based on the pre-process data, and dynamically classifies the capacitor into several initial state categories based on the state stability index. The personalized parameter configuration module combines the statistical characteristics of each initial state category and the individual state stability index of each capacitor within the category to configure a corresponding personalized initial aging parameter scheme for each capacitor. The personalized initial aging parameter scheme includes differentiated aging temperature range, applied voltage range and reference aging time. The adaptive aging control module performs aging treatment on a single capacitor at an independent workstation based on a personalized initial aging parameter scheme, and continuously collects leakage current and capacitance data during the aging process to obtain real-time aging response characteristics. The real-time aging response characteristics are matched with the preset standard response characteristics of this category to calculate the aging matching degree; the aging matching degree is used to determine whether the capacitor needs to be dynamically migrated to the category. When it deviates from the trajectory of this category, the capacitor is automatically migrated to the target category and switched to the corresponding aging parameter scheme. The quality assessment module records the category migration trajectory, parameter adjustment process, and execution time of each stage for each capacitor, forming an individualized aging profile; After aging is completed, the final electrical parameters are tested. The quality grade of each capacitor is output by combining the individualized aging profile, aging matching degree and final electrical parameters.