Thin film capacitor life intelligent prediction method fusing multi-dimensional data

Abstract

The application discloses a film capacitor life intelligent prediction method fusing multi-dimensional data and relates to the technical field of capacitor reliability. The application innovatively integrates three types of core data by constructing a multi-dimensional data fusion model: one is the process parameters of the whole process of film capacitor manufacturing; two is the real-time monitoring operation condition data; three is a large batch history failure database based on an acceleration factor. A machine learning algorithm is used to dynamically associate the above multi-source heterogeneous data, a nonlinear life attenuation model is established, the contribution degree of process defects to electrical performance degradation is quantified, and early life warning is realized. Compared with traditional methods, the present technology improves the prediction accuracy by more than 40%, can identify progressive failures caused by metal layer micro scratches, self-healing effect accumulation, and hidden factors of gold spraying contact degradation, and provides accurate basis for preventive maintenance of power electronic equipment.

Description

Technical Field

[0001] This invention relates to the field of capacitor reliability technology, and in particular to an intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data. Background Technology

[0002] As a core energy storage component in power electronic systems, the accuracy of thin-film capacitor lifetime prediction directly affects equipment reliability. Current mainstream prediction methods in this field rely on accelerated aging tests and statistical models, such as applying a fixed voltage and current rate and collecting capacity decay data to establish a Weibull distribution or Arrhenius model. However, these methods have significant limitations: First, the difference between the accelerated test load spectrum and actual operating conditions leads to prediction bias; for example, high-frequency ripple in photovoltaic inverters is not effectively simulated. Second, key defects in the manufacturing process (such as film wrinkles caused by uneven winding tension and coating cracks caused by slitting burrs) are not included in the lifetime assessment system. Furthermore, traditional external sensors cannot monitor internal state changes in real time, such as the location of self-healing discharge and the degradation process of the metal layer. These problems result in prediction errors in existing technologies generally exceeding 30%, and they cannot locate weak areas inside the capacitor.

[0003] The typical scheme of "A Method for Predicting the Lifespan of a Thin-Film Capacitor" disclosed in Chinese Patent CN103543346A includes the temperature and humidity of the usage environment, the current waveform, voltage waveform, and frequency of the capacitor, and simulates the working environment conditions of the capacitor using a programmable temperature and humidity control box. A dedicated signal generator produces the same current and voltage waveforms as when the capacitor is operating. Through accelerated testing and calculation of the changes in capacitor parameters before and after the test, theoretical and empirical formulas are used to predict the lifespan of the capacitor. This invention offers short testing time and accurate and reliable test results.

[0004] Based on existing technological bottlenecks, this patent needs to address three core issues: 1) How to achieve cross-temporal and spatial correlation between manufacturing process defects and operating conditions, such as mapping the winding tension gradient to the stress concentration factor of the metal layer; 2) How to construct a physical mechanism-driven life decay model to replace purely data-driven black-box prediction, such as quantifying the impact of burr height on crack propagation rate based on fracture mechanics formulas; 3) How to achieve life visualization and traceability, breaking through the limitations of traditional numerical displays, such as the combination of the mechanical dial and process coding. These needs are particularly urgent in high-temperature and high-humidity application scenarios. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a method for intelligent prediction of thin-film capacitor lifetime that integrates multi-dimensional data to solve the problem. Existing life prediction methods typically rely on a single accelerated aging test or a simplified mathematical model, which makes it difficult to accurately reflect the failure mechanisms under actual complex operating conditions, and also fails to fully incorporate key process parameters of the manufacturing process.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for intelligent prediction of the lifetime of thin-film capacitors by integrating multi-dimensional data, which includes the following steps: Step 1: Embed a distributed fiber optic sensor array during the capacitor core winding process to collect the winding tension gradient, film misalignment, and interlayer stress distribution in real time; Step 2: Perform terahertz wave scanning on the slit metallized film to construct a three-dimensional topological map of the edge burrs and correlate it with the coating thickness distribution; Step 3: Deposit a nanoscale piezoelectric sensing layer on the gold-sprayed end face to simultaneously monitor the impact energy distribution and adhesion strength of the gold-sprayed particles; Step 4: Spatiotemporally align the above manufacturing process data with the real-time current ripple spectrum, voltage surge pulse sequence, and temperature rise spatial gradient. Step 5: Capture the spatiotemporal coordinates of self-healing discharge based on the acoustic emission array, and invert the film shrinkage stress field by combining the thermal setting process parameters; Step 6: Reconstruct the electrical tree propagation path through a multiphysics coupling engine to dynamically correct the pulse load spectrum of the accelerated aging test; finally, output a lifetime dial with process traceability coding.

[0008] As a preferred embodiment of the intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in this invention, the deployment and monitoring of the distributed optical fiber sensor array are specifically implemented according to the following steps: Before the winding process begins, polyimide-coated optical fibers with a thickness of ≤3μm are selected and pre-laid in the gap between the guide rollers of the unwinding mechanism, and are embedded between the film layers synchronously as the dielectric film is released. The optical fibers are arranged in a spiral path with a pitch of 8mm to ensure that one optical fiber is embedded for every 8 layers of thin film wound, and the end is connected to the fiber Bragg grating demodulator through a slip ring. During the winding process, the demodulator captures the Bragg wavelength offset Δλ of each optical fiber in real time. When the difference in Δλ between adjacent optical fibers on the same circumference exceeds ±0.3nm, the misalignment calculation module is automatically triggered. According to the calibration formula Δd=0.15·Δλ, the unit is mm, and the transverse misalignment value of the film is output, where the unit of Δλ is nm. At the same time, by calculating the strain difference Δε between two adjacent optical fibers, when |Δε|>50με, it is determined that the tension gradient is abnormal. At this time, the winding machine control system immediately dynamically reduces the linear speed from the reference value of 5500mm / s to 5000mm / s and starts the tension roller fine adjustment mechanism to bring the gradient difference back to the safe threshold. After winding is completed, the fiber array is permanently retained inside the core, and its pigtail is drawn out axially from the core rod for stress state monitoring during subsequent operation.

[0009] As a preferred embodiment of the intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in this invention, the terahertz wave scanning process is precisely executed according to the following steps: A terahertz scanning unit is installed before the slitting and winding process to pass the slit metallized film through the scanning area at a constant speed of 600 mm / s; A pulse emission source with a center frequency of 1.5THz is used to vertically illuminate the edge of the membrane, and a 64-element focal plane detector array is symmetrically deployed on both sides of the membrane, with the detector spacing precisely calibrated to 0.5mm. When the edge of the membrane passes through the scanning area, the detector collects the distribution of scattered field intensity at a rate of 2000 frames per second, and the three-dimensional shape is reconstructed by the finite difference time-domain algorithm for each frame of data. The specific reconstruction process is as follows: First, wavelet denoising is performed on the original signal to extract the peak time t of the time-domain pulse envelope. peak Based on the refractive index of the medium n=1.8, the burr height H is calculated as H=0.5×c×Δt. peak / n, where c is the speed of light, Δt peak The peak time difference between adjacent detection units; A height value H(x) is generated every 0.1 mm along the membrane length. When the H value of three consecutive points exceeds 30% of the nominal membrane thickness, the coordinate interval is automatically marked as a red warning zone. The warning signal is transmitted to the slitting machine control system in real time, triggering the following actions: immediately reducing the slitting speed from 600mm / s to 200mm / s, and simultaneously starting the diamond grinding wheel dressing device to grind the slitting blade online until the warning area disappears in subsequent scan data before restoring the original speed; All 3D data of burrs are overlaid with the coating thickness distribution map to generate a composite defect map, which is then transmitted to the central database via an industrial bus to be associated with the process batch number of the roll film.

[0010] As a preferred embodiment of the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data as described in this invention, the construction and control process of the gold-sprayed monitoring layer is specifically implemented according to the following steps: Before the gold spraying process begins, the capacitor core is fixed in a six-axis robotic arm fixture and transferred to the magnetron sputtering chamber. Under vacuum conditions of ≤5×10⁻³Pa, pulsed bias sputtering technology is used to deposit a continuous thin film with a thickness of 200±10nm onto the sensor end face. The target purity is ≥99.99%, and the deposition rate is controlled at 3.2nm / s. After deposition is complete, the core is loaded onto the rotary table of the gold spraying machine. Each gold spraying station is equipped with a high-sensitivity charge amplifier with a gain of 10. 9 V / C, whose input terminal is connected to the edge contact of the piezoelectric layer via a nano-silver wire; During the gold spraying process, the transient piezoelectric current I in the action area of ​​each spray gun is collected in real time. p It is converted into a charge Q=∫I through an integrating circuit. p dt, integration time window width 50ms; When the system detects that the charge per unit area Q / A in a local area is less than 0.8 nC / mm², where A is the area of ​​the area scanned by the spray gun, it is determined that the adhesion is insufficient and the distance adjustment mechanism is immediately triggered: the spray gun bracket is driven by a servo motor to gradually reduce the distance between the spray gun nozzle and the end face of the core from the initial value of 10 mm to the optimal adhesion range of 5~8 mm with an accuracy of 0.1 mm / step, while the wire feeding speed of the zinc-tin alloy wire is increased from the reference value of 4.5 m / min to 5.2 m / min; After each adjustment, the area is rescanned until Q / A ≥ 0.8nC / mm² before proceeding with the subsequent gold spraying process; The charge distribution data of all regions are used to generate a two-dimensional heat map, which is then bound to the core ID database along with the gold spraying process parameters.

[0011] As a preferred embodiment of the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data as described in this invention, the execution process of the thermal shaping stress field inversion algorithm is specifically implemented according to the following steps: The completed capacitor core is placed on the heat-setting equipment platform; The dual-wavelength laser speckle interferometer is activated, with its 532nm and 635nm laser beams irradiating the core surface at an incident angle of 30°. The displacement field changes during the heat-drying process are acquired by a high-speed CMOS camera at 500 frames per second. After the heat setting begins, the equipment heats up according to the preset program: the round core adopts a two-stage heating curve, which heats up from room temperature to 95℃ in 0~10 minutes and maintains 95±2℃ in 10~30 minutes; The flat core adopts a three-stage heating curve: from room temperature to 100℃ in 0~5 minutes, to 120℃ in 5~15 minutes, and maintained at 120±3℃ in 15~30 minutes. The real-time acquired speckle images are used to calculate the three-dimensional displacement field u(x,y,t) using a digital image correlation algorithm. Substituting this into the stress field inversion equation, and based on the linear elastic constitutive relation, the displacement gradient tensor ∇u is input into the finite element solver to calculate the stress tensor σ. ij = C ijkl ·ε kl C ijkl Let ε be the stiffness matrix. kl =0.5(∂u k / ∂x l + ∂u l / ∂x k ) represents the strain tensor; Simultaneously, a thermal expansion effect correction term is introduced, and based on the measured shrinkage rate α of the thin film material and the real-time temperature T(t), the thermal stress component σ is superimposed in the stress field. thermal = E·α·ΔT / (1-ν), where E is the elastic modulus and ν is Poisson's ratio; When the inversion results show that the equivalent stress in a local area exceeds 45% of the material's yield strength, the temperature control program is automatically adjusted: for a circular core, the heating rate is reduced to 70% of the original rate at the axial position corresponding to the stress concentration area; For flat cores, graphite thermal pads are added to the contact surface of the press to even out the temperature distribution; All stress field data generate a 3D cloud map every 5 minutes, marking the coordinates of stress concentration areas and out-of-limit values, and transmits it to the central database via industrial Ethernet to be associated with the core process number.

[0012] As a preferred embodiment of the intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in this invention, the workflow of the multi-physics coupling engine is specifically implemented according to the following steps: Real-time datasets are retrieved from the central database, including interlayer stress distribution maps, three-dimensional topology of slit burrs, thermal maps of gold spray adhesion, and thermal shaping inversion stress fields. During the capacitor operation phase, the current ripple spectrum is collected in real time by an embedded fiber optic sensor, while an infrared thermal imager is used to monitor the surface temperature distribution of the casing at a rate of 30 frames per second. After inputting the above data into the multi-field coupling engine, a fracture propagation model of the metal coating is first constructed based on the principles of fracture mechanics: the local stress concentration factor K is extracted based on the burr location coordinates. t Combined with operating current and Joule heat power P heat =I rms ²·R contact , where I rmsR is the root mean square value of the current, and R is the contact resistance. contact The thermal stress increment in the temperature gradient field is calculated by inverting the Q value of the gold spray adhesion. The crack propagation rate is calculated using the Paris formula, where the stress intensity factor amplitude ΔK = 1.12·Δσ·(π·a), Δσ is the sum of mechanical stress and thermal stress, and a is the initial crack length, which is obtained by converting the burr height H through a shape factor. When a specific location satisfies ΔK ≥ 0.7K IC K IC For the fracture toughness of the coating material, when the zinc-aluminum alloy is 18 MPa·m¹ / ², this point is marked as a high-risk failure zone; Customized accelerated aging solutions are developed for high-risk areas: based on the actual ripple current fundamental frequency f. base Set the pulse load rise time t r =1 / (6f base ), descent time t d =1 / (20f base Peak current I peak =2.5I rms +3σ, where σ is the standard deviation of current fluctuation; During the aging test, after every 1000 pulse cycles, the crack propagation Δa in the high-risk area is detected by an acoustic emission array. If Δa > 5 μm, the pulse amplitude is automatically increased by 10% until failure is triggered. The final output is a lifetime decay cloud map with spatial coordinates, indicating the percentage of remaining lifetime in each region.

[0013] As a preferred embodiment of the intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in this invention, the implementation process of the dynamic correction strategy for the pulse load spectrum is specifically carried out according to the following steps: The current waveform of the thin-film capacitor during operation is captured in real time by a high-speed data acquisition card, with a sampling rate of not less than 1MS / s, and 10 power frequency cycles are continuously recorded. Wavelet packet decomposition is performed on the original data to extract the fundamental frequency f. min and the highest effective harmonic frequency f max Typically, harmonics with an energy percentage greater than 1% are selected, and the current fluctuation rate δ is calculated simultaneously. I / I0, where I0 is the DC component, δ I The peak value of the ripple wave; Based on the above parameters, an asymmetric triangular wave test current is generated: the rise time t is set. r =1 / (6f min (Falling edge time t) d =1 / (20f max For example, in photovoltaic inverter application scenarios, fmin =4kHz t r =41.7μs, f max =100kHz t d =0.5μs; The peak current of the waveform is according to I peak =I0[1+2.5(δ I The calculation is performed as follows: I0)]+3σ, where σ is the standard deviation of the measured current, and the valley current remains constant at I0. The generated load spectrum is input into a programmable current source and applied to the capacitor terminals via a four-wire connection. At the same time, an infrared thermal imager is activated to monitor the surface temperature rise distribution. After every 500 pulse cycles, the acoustic emission array data is retrieved: if the count of acoustic emission events in the 300-500kHz range in the high-risk area exceeds 50% of the baseline value, then I is automatically switched to the next higher level. peak Increase the current value by 10%; If the thermal imager shows a local temperature rise ΔT > 15K and the reference temperature is 40℃, then extend the pulse interval time to 150% of the original value. This process continues until the capacitor capacitance decays by more than 5% or the loss tangent increases by more than 20%, and the final failure cycle number N is recorded. f According to formula L rem =L0·(N curr / N f Calculate the remaining lifespan, where L0 is the rated lifespan and N... curr This represents the number of loops that have been run.

[0014] As a preferred embodiment of the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data as described in this invention, the deployment of the acoustic emission array and the location of the self-healing point are specifically implemented according to the following steps: Before the capacitor potting process, 12 acoustic emission sensors with a resonant frequency of 150kHz are fixed to the inner wall of the metal shell with high-temperature epoxy resin. The sensors are symmetrically distributed in an icosahedral pattern: one at the top center and one at the bottom center, and 10 equally spaced on the equatorial plane. The included angle between adjacent sensors is precisely calibrated to 36°. Each sensor is connected to a 128-channel acoustic emission acquisition instrument via a shielded coaxial cable, with the sampling rate set to 5MS / s and the preamplifier gain set to 40dB. During capacitor operation, when an event is detected where the signal amplitude exceeds the threshold of 45dB and the duration is greater than 2μs, the time difference positioning algorithm is activated: first, wavelet denoising is performed on the original signal to extract the arrival time t of the direct wave. i (i=1~12), according to the sound speed model v(T)=4100 6.5(T 25) m / s, where T is the real-time temperature, calculate the three-dimensional coordinates; The positioning formula is minΣ[(t) i -t j )-(d i -d j ) / v]², where d i Given the distance from sensor i to the predicted point, the optimal coordinate point is iteratively solved using the gradient descent method; When the event characteristic frequency is in the range of 300~500kHz, which is the typical frequency band of self-healing discharge, the infrared thermal imager data is retrieved simultaneously: if the surface of the shell corresponding to the coordinate point has a temperature step ΔT>8K before and after the event, the reference temperature is 40℃, and the duration of the temperature rise is less than 100ms, then it is determined to be an effective self-healing point. The location coordinates and energy integral value ∫A²dt of all valid events, where A is the signal amplitude, and the associated temperature rise data are packaged into a data packet. The data packet is matched with the high-risk area coordinates of the multi-field coupling engine through the timestamp. When the match is successful, the point is marked as an activated self-healing zone.

[0015] As a preferred embodiment of the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data as described in this invention, the generation and updating process of the lifespan scale is specifically performed according to the following steps: 3,000 concentric annular microgrooves were pre-machined on the surface of the capacitor shell, with a groove depth of 50 μm, a groove width of 20 μm, and a spacing of 40 μm between adjacent grooves, covering 80% of the axial height of the shell. When a valid self-healing event is determined to have occurred, the discharge energy integral value E is used as the basis for determining the event. discharge =0.5C(U pre ²-U post ²), where C is the real-time capacity, U pre U is the voltage before discharge. post Given the voltage after discharge, calculate the equivalent lifetime loss Δt caused by this event. eq =η·E discharge / (U N ·C0), where η is the damage coefficient, and the zinc-aluminum alloy coating is taken as 1.8×10⁻³, U N C0 is the rated voltage, and C0 is the initial capacity. For every 0.1% of lifetime equivalent consumed, the femtosecond laser ablation unit is triggered: a laser beam with a wavelength of 1064nm and a pulse energy of 20mJ is used to position the corresponding annular groove through a galvanometer system, and burns through the 100nm thick aluminum film remaining at the bottom of the microgroove within 200μs to form a light-transmitting hole with a diameter of 5μm. After ablation is completed, the transmittance is verified using an integrated optical sensor: if the transmittance of the 635nm probe light is detected to be >90%, the ablation is deemed effective, and the remaining lifetime display value L is updated. rem =L rem 0.1%; Simultaneously, the values ​​are displayed on the LED digital tube on the top of the casing; when the cumulative number of burned holes reaches 2500 and the remaining lifespan is 25%, an early warning signal is automatically activated, and the capacitor serial number is uploaded to the cloud maintenance system; The coordinates of each ablation event are bound to the associated self-healing event data and stored to generate a spatiotemporal distribution map of lifespan loss.

[0016] As a preferred embodiment of the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data as described in this invention, the process of constructing and recording the process traceability code is specifically performed according to the following steps: After the capacitor completes the potting process, the calculation period for the winding tension fluctuation variance σ² is taken as the entire winding process and the range of burr height H during slitting. max H min That is, the statistical value of a single roll of film and the adhesion strength of gold spraying Q. avg That is, the four sets of data, namely the global average value of the end face, the integral value of the heat setting temperature ∫Tdt of the heat drying process, are normalized to the interval [0,1] to form the process feature vector [V1,V2,V3,V4]; Input a three-layer convolutional neural network with a kernel size of 3×3, a stride of 1, and 64-128-256 channels. The first layer expands the vector into an 8×8 matrix, and after two convolution-pooling operations, it outputs a 128-dimensional feature tensor. The binary code is generated by binarization using the Sigmoid activation function, with a bit "1" corresponding to an output value ≥ 0.5. On a 40mm×40mm anodized aluminum plate with a reserved QR code on the capacitor shell, an ultraviolet laser engraving machine is used to etch the coding pattern with a dot diameter of 20μm and an engraving depth accuracy of 30%: the bit "1" area has a micro pit array with an etch depth of 50μm and a pit spacing of 40μm, while the bit "0" area retains the original surface. After the recording is completed, the QR code area is scanned with a white light interferometer to verify that the micro-pit morphology qualification rate is >99.9%; at the same time, the binary code is converted to Base32 and uploaded to the cloud database and bound to the capacitor serial number. When a user reads the code using a scanning device, the three-dimensional shape recognition bit value of the micro-pit is reconstructed first, and then the complete process file in the cloud is called, including the winding speed curve, the distribution map of the slitting burrs, the thermal map of gold spraying, and the thermal setting stress cloud map, so as to realize the full life cycle data traceability.

[0017] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent prediction method for the lifespan of a thin-film capacitor that integrates multi-dimensional data as described in the first aspect of the present invention.

[0018] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent prediction method for the lifespan of a thin-film capacitor that integrates multi-dimensional data as described in the first aspect of the present invention.

[0019] The beneficial effects of this invention are: This invention achieves a revolutionary breakthrough in the field of thin-film capacitor lifetime prediction through multi-dimensional embedded sensing and physical field reconstruction technology, creatively solving the industry pain point that traditional methods rely on statistical models and are detached from actual failure mechanisms. Its core beneficial effect is primarily reflected in a substantial improvement in prediction accuracy: based on native manufacturing process data such as fiber optic monitoring of interlayer stress, terahertz burr 3D reconstruction, and piezoelectric feedback of gold sputtering adhesion, combined with the spatial coordinates of self-healing discharge events captured by an acoustic emission array, a precise mapping chain of process defects, operational degradation, and failure location is constructed, reducing lifetime prediction error from over 30% in existing technologies to less than 8%. For example, in photovoltaic inverter applications, the pulse load spectrum (t) is dynamically corrected through a multi-physics engine. r =41.7μs precise matching of 4kHz ripple fundamental frequency), which improves the equivalence between accelerated testing and real-world operating conditions by more than 3 times.

[0020] Secondly, this solution pioneers a new proactive intervention model for failure prevention. Real-time detection of localized stress exceeding limits (>18MPa) during thermal stress field inversion automatically triggers temperature control adjustments (e.g., adding graphite thermal pads to the flat core) to suppress film wrinkling at its source. The pulsed load spectrum dynamically increases the Ipeak intensity based on acoustic emission event counts (>50% of baseline value), enabling early verification of coating crack propagation. This closed-loop "monitoring-prediction-intervention" system allows high-risk areas to be identified as early as 20% of the capacitor's lifespan, more than 80% earlier than traditional capacity decay alarms, providing at least a 48-hour maintenance window for power electronic systems.

[0021] Third, the whole life cycle management paradigm has undergone a major upgrade. Process traceability coding will incorporate winding tension variance σ² (e.g., fluctuation of 432±72g) and slitting burr range (H... max H minKey parameters such as the ≤1.44μm threshold and the gold plating adhesion Qavg (≥0.8nC / mm²) are compressed into 128-bit laser micro-pit codes using CNN. Combined with complete cloud-based process archives, manufacturing defects in each capacitor can be traced back to specific processes. For example, when early failure occurs in capacitors for new energy vehicles, scanning the code can retrieve the distribution map of cutting burrs and the thermal map of gold plating, accurately attributing it to the risk of peeling of the zinc-aluminum alloy coating at the burrs, guiding the direction of process optimization. At the same time, the mechanical life scale dial maintains visibility even in high-temperature and high-humidity environments by ablating 3000 annular micro-grooves with femtosecond laser (each hole corresponds to 0.1% life loss), avoiding maintenance blind spots caused by the failure of electronic display devices. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data in Example 1. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0027] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for intelligent prediction of the lifetime of thin-film capacitors by fusing multi-dimensional data, including the following steps: Step 1: Embed a distributed fiber optic sensor array during the capacitor core winding process to collect the winding tension gradient, film misalignment, and interlayer stress distribution in real time; Step 2: Perform terahertz wave scanning on the slit metallized film to construct a three-dimensional topological map of the edge burrs and correlate it with the coating thickness distribution; Step 3: Deposit a nanoscale piezoelectric sensing layer on the gold-sprayed end face to simultaneously monitor the impact energy distribution and adhesion strength of the gold-sprayed particles; Step 4: Spatiotemporally align the above manufacturing process data with the real-time current ripple spectrum, voltage surge pulse sequence, and temperature rise spatial gradient. Step 5: Capture the spatiotemporal coordinates of self-healing discharge based on the acoustic emission array, and invert the film shrinkage stress field by combining the thermal setting process parameters; Step 6: Reconstruct the electrical tree propagation path through a multiphysics coupling engine to dynamically correct the pulse load spectrum of the accelerated aging test; finally, output a lifetime dial with process traceability coding.

[0028] The deployment and monitoring of distributed fiber optic sensor arrays are implemented according to the following steps: Before the winding process begins, polyimide-coated optical fibers with a thickness of ≤5μm are selected and pre-laid in the gap between the guide rollers of the unwinding mechanism, and are embedded between the film layers synchronously as the dielectric film is released. The optical fibers are arranged in a spiral path with a pitch of 8mm to ensure that one optical fiber is embedded for every 8 layers of thin film wound, and the end is connected to the fiber Bragg grating demodulator through a slip ring. During the winding process, the demodulator captures the Bragg wavelength offset Δλ of each optical fiber in real time. When the difference in Δλ between adjacent optical fibers on the same circumference exceeds ±0.3nm, the misalignment calculation module is automatically triggered. According to the calibration formula Δd=0.15·Δλ, the unit is mm, and the transverse misalignment value of the film is output, where the unit of Δλ is nm. At the same time, by calculating the strain difference Δε between two adjacent optical fibers, when |Δε|>50με, it is determined that the tension gradient is abnormal. At this time, the winding machine control system immediately dynamically reduces the linear speed from the reference value of 5500mm / s to 5000mm / s and starts the tension roller fine adjustment mechanism to bring the gradient difference back to the safe threshold. After winding is completed, the fiber array is permanently retained inside the core, and its pigtail is drawn out axially from the core rod for stress state monitoring during subsequent operation.

[0029] The terahertz wave scanning process is executed precisely according to the following steps: A terahertz scanning unit is installed before the slitting and winding process to pass the slit metallized film through the scanning area at a constant speed of 600 mm / s; A pulse emission source with a center frequency of 1.5THz is used to vertically illuminate the edge of the membrane, and a 64-element focal plane detector array is symmetrically deployed on both sides of the membrane, with the detector spacing precisely calibrated to 0.5mm. When the edge of the membrane passes through the scanning area, the detector collects the distribution of scattered field intensity at a rate of 2000 frames per second, and the three-dimensional shape is reconstructed by the finite difference time-domain algorithm for each frame of data. The specific reconstruction process is as follows: First, wavelet denoising is performed on the original signal to extract the peak time t of the time-domain pulse envelope. peak Based on the refractive index of the medium n=1.8, the burr height H is calculated as H=0.5×c×Δt. peak / n, where c is the speed of light, Δt peak The peak time difference between adjacent detection units; A height value H(x) is generated every 0.1 mm along the membrane length. When the H value of three consecutive points exceeds 30% of the nominal membrane thickness, the coordinate interval is automatically marked as a red warning zone. The warning signal is transmitted to the slitting machine control system in real time, triggering the following actions: immediately reducing the slitting speed from 600mm / s to 200mm / s, and simultaneously starting the diamond grinding wheel dressing device to grind the slitting blade online until the warning area disappears in subsequent scan data before restoring the original speed; All 3D data of burrs are overlaid with the coating thickness distribution map to generate a composite defect map, which is then transmitted to the central database via an industrial bus to be associated with the process batch number of the roll film.

[0030] The construction and control process of the gold-sprayed monitoring layer is implemented in the following steps: Before the gold spraying process begins, the capacitor core is fixed in a six-axis robotic arm fixture and transferred to the magnetron sputtering chamber. Under vacuum conditions of ≤5×10⁻³Pa, pulsed bias sputtering technology is used to deposit a continuous thin film with a thickness of 200±10nm onto the sensor end face. The target purity is ≥99.99%, and the deposition rate is controlled at 3.2nm / s. After deposition is complete, the core is loaded onto the rotary table of the gold spraying machine. Each gold spraying station is equipped with a high-sensitivity charge amplifier with a gain of 10. 9 V / C, whose input terminal is connected to the edge contact of the piezoelectric layer via a nano-silver wire; During the gold spraying process, the transient piezoelectric current I in the action area of ​​each spray gun is collected in real time. p It is converted into a charge Q=∫I through an integrating circuit. p dt, integration time window width 50ms; When the system detects that the charge per unit area Q / A in a local area is less than 0.8 nC / mm², where A is the area of ​​the area scanned by the spray gun, it is determined that the adhesion is insufficient and the distance adjustment mechanism is immediately triggered: the spray gun bracket is driven by a servo motor to gradually reduce the distance between the spray gun nozzle and the end face of the core from the initial value of 10 mm to the optimal adhesion range of 5~8 mm with an accuracy of 0.1 mm / step, while the wire feeding speed of the zinc-tin alloy wire is increased from the reference value of 4.5 m / min to 5.2 m / min; After each adjustment, the area is rescanned until Q / A ≥ 0.8nC / mm² before proceeding with the subsequent gold spraying process; The charge distribution data of all regions are used to generate a two-dimensional heat map, which is then bound to the core ID database along with the gold spraying process parameters.

[0031] The execution process of the thermally shaped stress field inversion algorithm is implemented in the following steps: The completed capacitor core is placed on the heat-setting equipment platform; The dual-wavelength laser speckle interferometer is activated, with its 532nm and 635nm laser beams irradiating the core surface at an incident angle of 30°. The displacement field changes during the heat-drying process are acquired by a high-speed CMOS camera at 500 frames per second. After the heat setting begins, the equipment heats up according to the preset program: the round core adopts a two-stage heating curve, which heats up from room temperature to 95℃ in 0~10 minutes and maintains 95±2℃ in 10~30 minutes; The flat core adopts a three-stage heating curve: from room temperature to 100℃ in 0~5 minutes, to 120℃ in 5~15 minutes, and maintained at 120±3℃ in 15~30 minutes. The real-time acquired speckle images are used to calculate the three-dimensional displacement field u(x,y,t) using a digital image correlation algorithm. Substituting this into the stress field inversion equation, and based on the linear elastic constitutive relation, the displacement gradient tensor ∇u is input into the finite element solver to calculate the stress tensor σ. ij = C ijkl ·ε kl C ijkl Let ε be the stiffness matrix. kl =0.5(∂u k / ∂x l + ∂u l / ∂x k ) represents the strain tensor; Simultaneously, a thermal expansion effect correction term is introduced, and based on the measured shrinkage rate α of the thin film material and the real-time temperature T(t), the thermal stress component σ is superimposed in the stress field. thermal = E·α·ΔT / (1-ν), where E is the elastic modulus and ν is Poisson's ratio; When the inversion results show that the equivalent stress in a local area exceeds 45% of the material's yield strength, the temperature control program is automatically adjusted: for a circular core, the heating rate is reduced to 70% of the original rate at the axial position corresponding to the stress concentration area; For flat cores, graphite thermal pads are added to the contact surface of the press to even out the temperature distribution; All stress field data generate a 3D cloud map every 5 minutes, marking the coordinates of stress concentration areas and out-of-limit values, and transmits it to the central database via industrial Ethernet to be associated with the core process number.

[0032] The workflow of the multiphysics coupling engine is implemented in the following steps: Real-time datasets are retrieved from the central database, including interlayer stress distribution maps, three-dimensional topology of slit burrs, thermal maps of gold spray adhesion, and thermal shaping inversion stress fields. During the capacitor operation phase, the current ripple spectrum is collected in real time by an embedded fiber optic sensor, while an infrared thermal imager is used to monitor the surface temperature distribution of the casing at a rate of 30 frames per second. After inputting the above data into the multi-field coupling engine, a fracture propagation model of the metal coating is first constructed based on the principles of fracture mechanics: the local stress concentration factor K is extracted based on the burr location coordinates. t Combined with operating current and Joule heat power P heat =I rms ²·R contact , where I rms R is the root mean square value of the current, and R is the contact resistance. contact The thermal stress increment in the temperature gradient field is calculated by inverting the Q value of the gold spray adhesion. The crack propagation rate is calculated using the Paris formula, where the stress intensity factor amplitude ΔK = 1.12·Δσ·(π·a), Δσ is the sum of mechanical stress and thermal stress, and a is the initial crack length, which is obtained by converting the burr height H through a shape factor. When a specific location satisfies ΔK ≥ 0.7K IC K IC For the fracture toughness of the coating material, when the zinc-aluminum alloy is 18 MPa·m¹ / ², this point is marked as a high-risk failure zone; Customized accelerated aging solutions are developed for high-risk areas: based on the actual ripple current fundamental frequency f. base Set the pulse load rise time t r =1 / (6f base ), descent time t d =1 / (20f base Peak current I peak =2.5I rms +3σ, where σ is the standard deviation of current fluctuation; During the aging test, after every 1000 pulse cycles, the crack propagation Δa in the high-risk area is detected by an acoustic emission array. If Δa > 5 μm, the pulse amplitude is automatically increased by 10% until failure is triggered. The final output is a lifetime decay cloud map with spatial coordinates, indicating the percentage of remaining lifetime in each region.

[0033] The implementation process of the dynamic correction strategy for pulse load spectrum is carried out in the following steps: The current waveform of the thin-film capacitor during operation is captured in real time by a high-speed data acquisition card, with a sampling rate of not less than 1MS / s, and 10 power frequency cycles are continuously recorded. Wavelet packet decomposition is performed on the original data to extract the fundamental frequency f. min and the highest effective harmonic frequency f max Typically, harmonics with an energy percentage greater than 1% are selected, and the current fluctuation rate δ is calculated simultaneously. I / I0, where I0 is the DC component, δ I The peak value of the ripple wave; Based on the above parameters, an asymmetric triangular wave test current is generated: the rise time t is set. r =1 / (6f min (Falling edge time t) d =1 / (20f max For example, in photovoltaic inverter application scenarios, f min =4kHz t r =41.7μs, f max =100kHz t d =0.5μs; The peak current of the waveform is according to I peak =I0[1+2.5(δ I The calculation is performed as follows: I0)]+3σ, where σ is the standard deviation of the measured current, and the valley current remains constant at I0. The generated load spectrum is input into a programmable current source and applied to the capacitor terminals via a four-wire connection. At the same time, an infrared thermal imager is activated to monitor the surface temperature rise distribution. After every 500 pulse cycles, the acoustic emission array data is retrieved: if the count of acoustic emission events in the 300-500kHz range in the high-risk area exceeds 50% of the baseline value, then I is automatically switched to the next higher level. peak Increase the current value by 10%; If the thermal imager shows a local temperature rise ΔT > 15K and the reference temperature is 40℃, then extend the pulse interval time to 150% of the original value. This process continues until the capacitor capacitance decays by more than 5% or the loss tangent increases by more than 20%, and the final failure cycle number N is recorded. f According to formula L rem =L0·(N curr / N f Calculate the remaining lifespan, where L0 is the rated lifespan and N... curr This represents the number of loops that have been run.

[0034] The deployment and self-healing point location of the acoustic emission array are implemented in the following steps: Before the capacitor potting process, 12 acoustic emission sensors with a resonant frequency of 150kHz are fixed to the inner wall of the metal shell with high-temperature epoxy resin. The sensors are symmetrically distributed in an icosahedral pattern: one at the top center and one at the bottom center, and 10 equally spaced on the equatorial plane. The included angle between adjacent sensors is precisely calibrated to 36°. Each sensor is connected to a 128-channel acoustic emission acquisition instrument via a shielded coaxial cable, with the sampling rate set to 5MS / s and the preamplifier gain set to 40dB. During capacitor operation, when an event is detected where the signal amplitude exceeds the threshold of 45dB and the duration is greater than 2μs, the time difference positioning algorithm is activated: first, wavelet denoising is performed on the original signal to extract the arrival time t of the direct wave. i (i=1~12), according to the sound speed model v(T)=4100 6.5(T 25) m / s, where T is the real-time temperature, calculate the three-dimensional coordinates; The positioning formula is minΣ[(t) i -t j )-(d i -d j ) / v]², where d i Given the distance from sensor i to the predicted point, the optimal coordinate point is iteratively solved using the gradient descent method; When the event characteristic frequency is in the range of 300~500kHz, which is the typical frequency band of self-healing discharge, the infrared thermal imager data is retrieved simultaneously: if the surface of the shell corresponding to the coordinate point has a temperature step ΔT>8K before and after the event, the reference temperature is 40℃, and the duration of the temperature rise is less than 100ms, then it is determined to be an effective self-healing point. The location coordinates and energy integral value ∫A²dt of all valid events, where A is the signal amplitude, and the associated temperature rise data are packaged into a data packet. The data packet is matched with the high-risk area coordinates of the multi-field coupling engine through the timestamp. When the match is successful, the point is marked as an activated self-healing zone.

[0035] The process of generating and updating the lifespan dial is carried out in the following steps: 3,000 concentric annular microgrooves were pre-machined on the surface of the capacitor shell, with a groove depth of 50 μm, a groove width of 20 μm, and a spacing of 40 μm between adjacent grooves, covering 80% of the axial height of the shell. When a valid self-healing event is determined to have occurred, the discharge energy integral value E is used as the basis for determining the event. discharge =0.5C(U pre ²-U post ²), where C is the real-time capacity, U pre U is the voltage before discharge. post Given the voltage after discharge, calculate the equivalent lifetime loss Δt caused by this event. eq =η·E discharge / (U N ·C0), where η is the damage coefficient, and the zinc-aluminum alloy coating is taken as 1.8×10⁻³, U N C0 is the rated voltage, and C0 is the initial capacity. For every 0.1% of lifetime equivalent consumed, the femtosecond laser ablation unit is triggered: a laser beam with a wavelength of 1064nm and a pulse energy of 20mJ is used to position the corresponding annular groove through a galvanometer system, and burns through the 100nm thick aluminum film remaining at the bottom of the microgroove within 200μs to form a light-transmitting hole with a diameter of 5μm. After ablation is completed, the transmittance is verified using an integrated optical sensor: if the transmittance of the 635nm probe light is detected to be >90%, the ablation is deemed effective, and the remaining lifetime display value L is updated. rem =L rem 0.1%; Simultaneously, the values ​​are displayed on the LED digital tube at the top of the casing; When the cumulative number of burned holes reaches 2,500 and the remaining lifespan is 25%, an early warning signal will be automatically activated, and the capacitor serial number will be uploaded to the cloud maintenance system. The coordinates of each ablation event are bound to the associated self-healing event data and stored to generate a spatiotemporal distribution map of lifespan loss.

[0036] The process of constructing and recording the process traceability code is carried out in the following steps: After the capacitor completes the potting process, the calculation period for the winding tension fluctuation variance σ² is taken as the entire winding process and the range of burr height H during slitting. max H min That is, the statistical value of a single roll of film and the adhesion strength of gold spraying Q. avg That is, the four sets of data, namely the global average value of the end face, the integral value of the heat setting temperature ∫Tdt of the heat drying process, are normalized to the interval [0,1] to form the process feature vector [V1,V2,V3,V4]; Input a three-layer convolutional neural network with a kernel size of 3×3, a stride of 1, and 64-128-256 channels. The first layer expands the vector into an 8×8 matrix, and after two convolution-pooling operations, it outputs a 128-dimensional feature tensor. The binary code is generated by binarization using the Sigmoid activation function, with a bit "1" corresponding to an output value ≥ 0.5. On a 40mm×40mm anodized aluminum plate with a reserved QR code on the capacitor shell, an ultraviolet laser engraving machine is used to etch the coding pattern with a dot diameter of 20μm and an engraving depth accuracy of 30%: the bit "1" area has a micro pit array with an etch depth of 50μm and a pit spacing of 40μm, while the bit "0" area retains the original surface. After the recording is completed, the QR code area is scanned with a white light interferometer to verify that the micro-pit morphology qualification rate is >99.9%; at the same time, the binary code is converted to Base32 and uploaded to the cloud database and bound to the capacitor serial number. When a user reads the code using a scanning device, the three-dimensional shape recognition bit value of the micro-pit is reconstructed first, and then the complete process file in the cloud is called, including the winding speed curve, the distribution map of the slitting burrs, the thermal map of gold spraying, and the thermal setting stress cloud map, so as to realize the full life cycle data traceability.

[0037] Example 2, refer to Figure 1 This is the second embodiment of the present invention. The workflow of the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data in this embodiment is as follows: First, an optical fiber array is embedded in the winding process to capture the tension gradient and misalignment in real time. After slitting, a burr-coating correlation model is constructed by terahertz scanning. During gold spraying, a piezoelectric layer is deposited simultaneously to monitor adhesion. During operation, the above process data is spatiotemporally aligned with the current ripple spectrum and temperature rise field (timestamp accuracy ±1ms, spatial grid 0.5mm³). The coordinates of the self-healing point are located based on the acoustic emission array, and the thermal shaping shrinkage stress field is inverted. The electric tree propagation path is reconstructed through a multiphysics engine (e.g., the fracture toughness of zinc-aluminum alloy coating is 18MPa·m¹ / ²), and an asymmetric triangular wave pulse load spectrum is dynamically generated (rising edge 41.7μs matching fundamental frequency). Finally, a lifespan dial is ablated on the shell surface and the process code is engraved to achieve full-cycle lifespan management from manufacturing to scrapping. Detailed implementation method: Step 1 Execution Details: Install a distributed sensing system on the BL850 automatic winding machine. Polyimide optical fiber (5μm diameter) is pre-laid at the unwinding roller guide wheel and wound synchronously with a 4.8μm thick PP film. One fiber is embedded every 8 layers of film, with a pitch of 8.0±0.1mm. During winding, the FBG demodulator monitors wavelength drift in real time. When it detects an adjacent fiber Δλ>0.3nm (corresponding to misalignment Δd>0.045mm) or strain difference Δε>50με (tension fluctuation>15%), the control system reduces the linear speed from 5500mm / s to 5000mm / s and adjusts the tension roller to bring the gradient value back within ±40με. After winding, the pigtail is led out through the center hole of the Φ9mm mandrel and connected to an external optical switch.

[0039] Step 2 Execution Details: The slitting process uses a 920mm wide galvanized aluminum film, and a terahertz scanning unit is installed at the slitting machine exit. The film passes through a 1.5THz emission source at a speed of 600mm / s, and a 64-element detector collects the edge scattering field. The burr height H(x) is reconstructed using a finite-difference time-domain algorithm. When the H value is >1.44μm (30% of a 4.8μm film thickness) at three consecutive points, the slitting machine automatically reduces its speed to 200mm / s and starts diamond wheel grinding of the cutters for 10 seconds. The burr 3D coordinates are mapped to the coating thickness distribution map (the area with sheet resistance of 20~100Ω is monitored in detail), and the data is bound to the film roll ID.

[0040] Step 3 Execution Details: Before gold spraying, the core is placed in the magnetron sputtering chamber, and a 200nm aluminum nitride piezoelectric layer is deposited on the end face (sputtering rate 3.2nm / s, vacuum degree 5×10⁻³Pa). Each station of the gold spraying machine is equipped with a charge amplifier to collect the charge Q of the spray gun area in real time. When the local Q / A < 0.8nC / mm², the servo mechanism adjusts the spray gun distance from 10mm to 6.5mm and the zinc-tin alloy wire feeding speed from 4.5m / min to 5.2m / min. After adjustment, the Q is re-measured until it meets the standard, and an adhesion thermal map is generated.

[0041] Step 4 Execution Details: The capacitor is installed into the DC support stage of the photovoltaic inverter. Current ripple is collected via a fiber optic sensor, and an infrared thermal imager monitors the casing temperature rise at 30fps. The central server integrates the data using timestamp alignment: winding tension variance σ², burr range H... max H min Gold-plated Q avg Thermal setting ∫Tdt and real-time current spectrum (fundamental frequency 4kHz±5%), temperature rise gradient ΔT max (Threshold 15K). Spatial alignment is achieved through a finite element mesh, with each cell storing process-run coupled data.

[0042] Step 5 execution details: Twelve 150kHz acoustic emission sensors (spaced 36° apart) on the inner wall of the casing capture events in the 300-500kHz range. A time-difference positioning algorithm is used in conjunction with temperature compensation to determine the velocity of sound v(T) = 4100. 6.5(T 25) Calculate coordinates in m / s. Simultaneously retrieve heat-setting parameters: circular core, 95℃ / 30min, invert stress field σ based on displacement field u(x,y,t). ij =E·ε ij / (1+ν)+βΔT(β=1.2×10⁻ 4 K⁻¹). When the coordinates of the self-healing point overlap with the stress concentration zone (σ>18MPa), it is marked as a high-risk point.

[0043] Step 6 Execution Details: The multi-field coupling engine retrieves high-risk data: burr height H converted to crack length a (a=0.7H), stress field Δσ superimposed with Joule heating ΔT joule =I²R t / C p The crack propagation rate da / dt = C(ΔK) 4 (ΔK=1.12Δσ(πa), C=3×10⁻¹ 0 When ΔK > 12.6 MPa·m¹ / ² (0.7 K IC When t is generated, a pulse load spectrum is generated: r =1 / (6×4000)=41.7μs,t d=1 / (20×100000)=0.5μs, I peak =1.5I0 (I0=200A). Acoustic emission is used to verify the crack propagation Δa after every 1000 pulse cycles. If Δa>5μm, then I0 is increased. peak 10%.

[0044] Lifespan scale implementation: 3000 annular microgrooves (groove depth 50μm) are pre-fabricated on the shell surface. Self-healing event discharge energy E discharge =0.5×750×10⁻ 6 ×(800² 650²) = 35J, life loss Δt eq =1.8×10⁻³×35 / (800×750×10⁻ 6 =0.105%. Femtosecond laser ablation of the corresponding micro-grooves (one groove per 0.1%), updating the LED display value when the light transmittance is >90%. A cumulative ablation of 750 grooves (25%) triggers a cloud-based alert.

[0045] Process coding implementation: winding σ²=0.021, burr H max H min =1.38μm, gold-plated Q avg =0.92nC / mm², heat-set ∫Tdt=2850℃·min, normalized and input into CNN, output 128-bit encoding (e.g., 1011...0110). Ultraviolet laser etched a micro-pit array on a 40×40mm aluminum plate ("1" pit depth 50μm), and after passing white light interferometry verification, it was bound to the cloud process file.

[0046] This embodiment was verified on photovoltaic inverter capacitors, with a predicted lifespan error of <7.5%, a high-risk area location accuracy of 92%, and a yield improvement of 11.8%, fully achieving the invention objectives.

[0047] This embodiment also provides a computer device applicable to the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent prediction method for the lifespan of thin-film capacitors that integrates multi-dimensional data as proposed in the above embodiment.

[0048] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0049] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data, as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0050] In summary, this invention achieves a revolutionary breakthrough in the field of thin-film capacitor lifetime prediction through multi-dimensional embedded sensing and physical field reconstruction technology, creatively solving the industry pain point that traditional methods rely on statistical models and are detached from actual failure mechanisms. Its core beneficial effect is primarily reflected in the substantial improvement in prediction accuracy: based on native manufacturing process data such as fiber optic monitoring of interlayer stress, terahertz burr 3D reconstruction, and piezoelectric feedback of gold sputtering adhesion, combined with the spatial coordinates of self-healing discharge events captured by an acoustic emission array, a precise mapping chain of process defects, operational degradation, and failure location is constructed, reducing lifetime prediction error from over 30% in existing technologies to less than 8%. For example, in photovoltaic inverter applications, the pulse load spectrum (t) is dynamically corrected through a multi-physics engine. r =41.7μs precise matching of 4kHz ripple fundamental frequency), which improves the equivalence between accelerated testing and real-world operating conditions by more than 3 times.

[0051] Secondly, this solution pioneers a new proactive intervention model for failure prevention. Real-time detection of localized stress exceeding limits (>18MPa) during thermal stress field inversion automatically triggers temperature control adjustments (e.g., adding graphite thermal pads to the flat core) to suppress film wrinkling at its source; the pulse load spectrum dynamically increases I based on the acoustic emission event count (>50% of baseline value). peak This strength enables early verification of coating crack propagation. This "monitoring-prediction-intervention" closed loop allows high-risk areas to be located when the capacitor reaches 20% of its lifespan degradation, more than 80% earlier than traditional capacity degradation alarm points, reserving at least a 48-hour maintenance window for power electronic systems.

[0052] Third, the whole life cycle management paradigm has undergone a major upgrade. Process traceability coding will incorporate winding tension variance σ² (e.g., fluctuation of 432±72g) and slitting burr range (H... max H min ≤1.44μm threshold), gold spray adhesion Q avg Key parameters such as (≥0.8nC / mm²) are compressed into 128-bit laser micro-pit codes using CNN. Combined with complete cloud-based process archives, manufacturing defects in each capacitor can be traced back to specific processes. For example, when early failure occurs in capacitors for new energy vehicles, scanning the code can retrieve the distribution map of cutting burrs and the thermal map of gold spraying, accurately attributing it to the risk of peeling of the zinc-aluminum alloy coating at the burrs, guiding the direction of process optimization. At the same time, the mechanical life scale dial maintains visibility even in high-temperature and high-humidity environments by ablating 3000 annular micro-grooves with femtosecond laser (each hole corresponds to 0.1% life loss), avoiding maintenance blind spots caused by the failure of electronic display devices.

[0053] Ultimately, this technology brings significant economic benefits: online burr removal during manufacturing (speed reduced to 200mm / s + diamond wheel trimming) improves yield by 12%; early replacement of high-risk capacitors using crack propagation models reduces photovoltaic inverter system failure rate by 40%; and at the recycling end, process coding identifies repairable components (such as products with insufficient gold plating adhesion), increasing the recycling rate of high-end materials by 35%. These breakthroughs lay the foundation for a technological paradigm shift in thin-film capacitors from "statistical lifetime prediction" to "physical mechanism-driven precise prediction."

[0054] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent prediction of the lifespan of thin-film capacitors by integrating multi-dimensional data, characterized in that, Includes the following steps: Step 1: Embed a distributed fiber optic sensor array during the capacitor core winding process to collect the winding tension gradient, film misalignment, and interlayer stress distribution in real time; Step 2: Perform terahertz wave scanning on the slit metallized film to construct a three-dimensional topological map of the edge burrs and correlate it with the coating thickness distribution; Step 3: Deposit a nanoscale piezoelectric sensing layer on the gold-sprayed end face to simultaneously monitor the impact energy distribution and adhesion strength of the gold-sprayed particles; Step 4: The manufacturing process data collected in Steps 1 to 3 are spatiotemporally aligned with the real-time current ripple spectrum, voltage surge pulse sequence, and temperature rise spatial gradient. The central server integrates the data using timestamp alignment, and spatial alignment is achieved through finite element meshes. Each cell stores the coupled data of the process operation. Step 5: Based on the spatiotemporal coordinates of self-healing discharge captured by the emission sensors in the acoustic emission array, the real-time acquired speckle images are used to calculate the three-dimensional displacement field u(x,y,t) using a digital image correlation algorithm. This displacement field is then combined with the heat-setting process parameters to invert the film shrinkage stress field; among which, the stress field σ is inverted based on the displacement field u(x,y,t). ij =E·ε ij / (1+ν)+βΔT(β= ), where ν is Poisson's ratio; When the coordinates of the self-healing point overlap with the stress concentration area, it is marked as a high-risk point, and the stress concentration area satisfies σ>18MPa; Step 6: Reconstruct the electrical tree propagation path through a multiphysics coupling engine and dynamically correct the pulse load spectrum of the accelerated aging test; finally, output a lifetime dial with process traceability coding, in which the multiphysics coupling engine retrieves high-risk point data, converts burr height into crack length, superimposes stress field and Joule heat, and reconstructs the electrical tree propagation path according to the crack propagation rate.

2. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 1, characterized in that: The deployment and monitoring of the distributed fiber optic sensor array are implemented according to the following steps: Before the winding process begins, polyimide-coated optical fibers with a thickness of ≤3μm are selected and pre-laid in the gap between the guide rollers of the unwinding mechanism, and are embedded between the film layers synchronously as the dielectric film is released. The optical fibers are arranged in a spiral path with a pitch of 8mm to ensure that one optical fiber is embedded for every 8 layers of thin film wound, and the end is connected to the fiber Bragg grating demodulator through a slip ring. During the winding process, the demodulator captures the Bragg wavelength offset Δλ of each optical fiber in real time. When the difference in Δλ between adjacent optical fibers on the same circumference exceeds ±0.3nm, the misalignment calculation module is automatically triggered. According to the calibration formula Δd=0.15·Δλ, the unit is mm, and the transverse misalignment value of the film is output, where the unit of Δλ is nm. At the same time, by calculating the strain difference Δε between two adjacent optical fibers, when |Δε|>50με, it is determined that the tension gradient is abnormal. At this time, the winding machine control system immediately dynamically reduces the linear speed from the reference value of 5500mm / s to 5000mm / s and starts the tension roller fine adjustment mechanism to bring the gradient difference back to the safe threshold. After winding is completed, the fiber array is permanently retained inside the core, and its pigtail is drawn out axially from the core rod for stress state monitoring during subsequent operation.

3. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 2, characterized in that: The terahertz wave scanning process is executed precisely according to the following steps: A terahertz scanning unit is installed before the slitting and winding process to pass the slit metallized film through the scanning area at a constant speed of 600 mm / s; A pulse emission source with a center frequency of 1.5THz is used to vertically illuminate the edge of the membrane, and a 64-element focal plane detector array is symmetrically deployed on both sides of the membrane, with the detector spacing precisely calibrated to 0.5mm. When the edge of the membrane passes through the scanning area, the detector collects the distribution of scattered field intensity at a rate of 2000 frames per second, and the three-dimensional shape is reconstructed by the finite difference time-domain algorithm for each frame of data. The specific reconstruction process is as follows: First, wavelet denoising is performed on the original signal to extract the peak time t of the time-domain pulse envelope. peak Based on the refractive index of the medium n=1.8, the burr height H is calculated as H=0.5×c×Δt. peak / n, where c is the speed of light, Δt peak The peak time difference between adjacent detection units; A height value H(x) is generated every 0.1 mm along the membrane length. When the H value of three consecutive points exceeds 30% of the nominal membrane thickness, the coordinate interval is automatically marked as a red warning zone. The warning signal is transmitted to the slitting machine control system in real time, triggering the following actions: immediately reducing the slitting speed from 600mm / s to 200mm / s, and simultaneously starting the diamond grinding wheel dressing device to grind the slitting blade online until the warning area disappears in subsequent scan data before restoring the original speed; All burr 3D data and coating thickness distribution map are overlaid to generate a composite defect map, which is transmitted to the central database via industrial bus and associated with the process batch number of the slit metallized film.

4. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 3, characterized in that: The process of simultaneously monitoring the impact energy distribution and adhesion strength of gold-plated particles and constructing and controlling these parameters is implemented in the following steps: Before the gold spraying process begins, the capacitor core is fixed in a six-axis robotic arm fixture and transferred to the magnetron sputtering chamber. Under vacuum conditions of ≤5×10⁻³Pa, pulsed bias sputtering technology is used to deposit a continuous thin film with a thickness of 200±10nm onto the sensor end face. The target purity is ≥99.99%, and the deposition rate is controlled at 3.2nm / s. After deposition is complete, the core is loaded onto the rotary table of the gold spraying machine. Each gold spraying station is equipped with a high-sensitivity charge amplifier with a gain of 10. 9 V / C, whose input terminal is connected to the edge contact of the piezoelectric layer via a nano-silver wire; During the gold spraying process, the transient piezoelectric current I in the action area of ​​each spray gun is collected in real time. p It is converted into a charge Q=∫I through an integrating circuit. p dt, integration time window width 50ms; When the system detects that the charge per unit area Q / A in a local area is less than 0.8 nC / mm², where A is the area of ​​the area scanned by the spray gun, it is determined that the adhesion is insufficient and the distance adjustment mechanism is immediately triggered: the spray gun bracket is driven by a servo motor to gradually reduce the distance between the spray gun nozzle and the end face of the core from the initial value of 10 mm to the optimal adhesion range of 5~8 mm with an accuracy of 0.1 mm / step, while the wire feeding speed of the zinc-tin alloy wire is increased from the reference value of 4.5 m / min to 5.2 m / min; After each adjustment, the area is rescanned until Q / A ≥ 0.8 nC / mm² before proceeding with the subsequent gold spraying process; The charge distribution data of all regions are used to generate a two-dimensional heat map, which is then bound to the core ID database along with the gold spraying process parameters.

5. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 4, characterized in that: The execution process of the thermally shaped stress field inversion algorithm is implemented in the following steps: The completed capacitor core is placed on the heat-setting equipment platform; The dual-wavelength laser speckle interferometer is activated, with its 532nm and 635nm laser beams irradiating the core surface at an incident angle of 30°. The displacement field changes during the heat-drying process are acquired by a high-speed CMOS camera at 500 frames per second. After the heat setting begins, the equipment heats up according to the preset program: the round core adopts a two-stage heating curve, which heats up from room temperature to 95℃ in 0~10 minutes and maintains 95±2℃ in 10~30 minutes; The flat core adopts a three-stage heating curve: from room temperature to 100℃ in 0~5 minutes, to 120℃ in 5~15 minutes, and maintained at 120±3℃ in 15~30 minutes. The real-time acquired speckle images are used to calculate the three-dimensional displacement field u(x,y,t) using a digital image correlation algorithm. Substituting this into the stress field inversion equation, and based on the linear elastic constitutive relation, the displacement gradient tensor ∇u is input into the finite element solver to calculate the stress tensor σ. ij = C ijkl ·ε kl C ijkl Let ε be the stiffness matrix. kl =0.5(∂u k / ∂x l + ∂u l / ∂x k Let x be the strain tensor; where k and l are coordinate direction indices, and x is the strain tensor. k and x l U represents the k-th and l-th coordinate directions, respectively. k and u l Let u represent the components of the displacement field u in the k-th and l-th coordinate directions, respectively. Simultaneously, a thermal expansion effect correction term is introduced, and based on the measured shrinkage rate α of the thin film material and the real-time temperature T(t), the thermal stress component σ is superimposed in the stress field. thermal = E·α·ΔT / (1-ν), where E is the elastic modulus, ν is Poisson's ratio, α is the measured shrinkage rate of the film material, and ΔT is the temperature increment; When the inversion results show that the equivalent stress in a local area exceeds 45% of the material's yield strength, the temperature control program is automatically adjusted: for a circular core, the heating rate is reduced to 70% of the original rate at the axial position corresponding to the stress concentration area; For flat cores, graphite thermal pads are added to the contact surface of the press to even out the temperature distribution; All stress field data generate a 3D cloud map every 5 minutes, marking the coordinates of stress concentration areas and out-of-limit values, and transmits it to the central database via industrial Ethernet to be associated with the core process number.

6. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 5, characterized in that: The workflow of the multiphysics coupling engine is implemented in the following steps: Real-time datasets are retrieved from the central database, including interlayer stress distribution maps, three-dimensional topology of slit burrs, thermal maps of gold spray adhesion, and thermal shaping inversion stress fields. During the capacitor operation phase, the current ripple spectrum is collected in real time by an embedded fiber optic sensor, while an infrared thermal imager is used to monitor the surface temperature distribution of the casing at a rate of 30 frames per second. After inputting the above data into the multi-field coupling engine, a fracture propagation model of the metal coating is first constructed based on the principles of fracture mechanics: the local stress concentration factor K is extracted based on the burr location coordinates. t Combined with operating current and Joule heat power P heat =I rms ²·R contact , where I rms R is the root mean square value of the current, and R is the contact resistance. contact The thermal stress increment in the temperature gradient field is calculated by inverting the Q value of the gold spray adhesion. The crack propagation rate is calculated using the Paris formula, where the stress intensity factor amplitude ΔK = 1.12·Δσ·(π·a), Δσ is the sum of mechanical stress and thermal stress, and a is the initial crack length, which is obtained by converting the burr height H through a shape factor. When a specific location satisfies ΔK ≥ 0.7K IC K IC For the fracture toughness of the coating material, when the zinc-aluminum alloy is 18 MPa·m¹ / ², this point is marked as a high-risk failure zone; Customized accelerated aging solutions are developed for high-risk areas: based on the actual ripple current fundamental frequency f. base Set the pulse load rise time t r =1 / (6f base ), descent time t d =1 / (20f base Peak current I peak =2.5I rms +3σ, where σ is the standard deviation of current fluctuation; During the aging test, after every 1000 pulse cycles, the crack propagation Δa in the high-risk area is detected by an acoustic emission array. If Δa > 5 μm, the pulse amplitude is automatically increased by 10% until failure is triggered. The final output is a lifetime decay cloud map with spatial coordinates, indicating the percentage of remaining lifetime in each region.

7. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 6, characterized in that: The implementation process of the dynamic correction strategy for the pulse load spectrum is carried out in the following steps: The current waveform of the thin-film capacitor during operation is captured in real time by a high-speed data acquisition card, with a sampling rate of not less than 1MS / s, and 10 power frequency cycles are continuously recorded. Wavelet packet decomposition is performed on the original data to extract the fundamental frequency f. min and the highest effective harmonic frequency f max Typically, harmonics with an energy percentage greater than 1% are selected, and the current fluctuation rate δ is calculated simultaneously. I / I0, where I0 is the DC component, δ I The peak value of the ripple wave; Based on the above parameters, an asymmetric triangular wave test current is generated: the rise time t is set. r =1 / (6f min Falling edge time t d =1 / (20f max ); The peak current of the waveform is according to I peak =I0[1+2.5(δ I The calculation is performed as follows: I0)]+3σ, where σ is the standard deviation of the measured current, and the valley current remains constant at I0. The generated load spectrum is input into a programmable current source and applied to the capacitor terminals via a four-wire connection. At the same time, an infrared thermal imager is activated to monitor the surface temperature rise distribution. After every 500 pulse cycles, the acoustic emission array data is retrieved: if the count of acoustic emission events in the 300-500kHz range in the high-risk area exceeds 50% of the baseline value, then I is automatically switched to the next higher level. peak Increase the current value by 10%; If the thermal imager shows a local temperature rise ΔT > 15K and the reference temperature is 40℃, then extend the pulse interval time to 150% of the original value. This process continues until the capacitor capacitance decays by more than 5% or the loss tangent increases by more than 20%, and the final failure cycle number N is recorded. f According to formula L rem =L0·(N curr / N f Calculate the remaining lifespan, where L0 is the rated lifespan and N... curr This represents the number of loops that have been run.

8. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 7, characterized in that: The deployment and self-healing point location of the acoustic emission array are implemented in the following steps: Before the capacitor potting process, 12 acoustic emission sensors with a resonant frequency of 150kHz are fixed to the inner wall of the metal shell with high-temperature epoxy resin. The sensors are symmetrically distributed in an icosahedral pattern: one at the top center and one at the bottom center, and 10 equally spaced on the equatorial plane. The included angle between adjacent sensors is precisely calibrated to 36°. Each sensor is connected to a 128-channel acoustic emission acquisition instrument via a shielded coaxial cable, with the sampling rate set to 5MS / s and the preamplifier gain set to 40dB. During capacitor operation, when an event is detected where the signal amplitude exceeds the threshold of 45 dB and the duration is greater than 2 μs, the time difference positioning algorithm is activated: first, wavelet denoising is performed on the original signal to extract the arrival time t of the direct wave. i (i=1~12), according to the sound speed model v(T)=4100 6.5(T 25) m / s, where T is the real-time temperature, calculate the three-dimensional coordinates; The positioning formula is minΣ[(t) i -t j )-(d i -d j ) / v]², where d i Let t be the distance from sensor i to the predicted point. The optimal coordinate point is solved iteratively using the gradient descent method, where t is the distance from sensor i to the predicted point. i and t j d represents the arrival time of the direct wave received by the i-th and j-th sensors, respectively. i and d j , , are the distances from the i-th and j-th sensors to the estimated point, respectively, and v is the speed of sound corresponding to the real-time temperature T; When the event characteristic frequency is in the range of 300~500kHz, which is the typical frequency band of self-healing discharge, the infrared thermal imager data is retrieved simultaneously: if the surface of the shell corresponding to the coordinate point has a temperature step ΔT>8K before and after the event, the reference temperature is 40℃, and the duration of the temperature rise is less than 100ms, then it is determined to be an effective self-healing point. The location coordinates and energy integral value ∫A²dt of all valid events, where A is the signal amplitude, and the associated temperature rise data are packaged into a data packet. The data packet is matched with the high-risk area coordinates of the multi-field coupling engine through the timestamp. When the match is successful, the point is marked as an activated self-healing zone.

9. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 8, characterized in that: The generation and updating process of the lifespan dial is carried out in the following steps: 3,000 concentric annular microgrooves were pre-machined on the surface of the capacitor shell, with a groove depth of 50 μm, a groove width of 20 μm, and a spacing of 40 μm between adjacent grooves, covering 80% of the axial height of the shell. When a valid self-healing event is determined to have occurred, the discharge energy integral value E is used as the basis for determining the event. discharge =0.5C(U pre ²-U post ²), where C is the real-time capacity, U pre U is the voltage before discharge. post Given the voltage after discharge, calculate the equivalent lifetime loss Δt caused by this event. eq =η·E discharge / (U N ·C0), where η is the damage coefficient, and the zinc-aluminum alloy coating is taken as 1.8×10⁻³, U N C0 is the rated voltage, and C0 is the initial capacity. For every 0.1% of lifetime equivalent consumed, the femtosecond laser ablation unit is triggered: a laser beam with a wavelength of 1064nm and a pulse energy of 20mJ is used to position the corresponding annular microgroove through a galvanometer system, and burns through the 100nm thick aluminum film remaining at the bottom of the microgroove within 200μs to form a light-transmitting hole with a diameter of 5μm. After ablation is completed, the transmittance is verified using an integrated optical sensor: if the transmittance of the 635nm probe light is detected to be >90%, the ablation is deemed effective, and the remaining lifetime display value L is updated. rem =L rem 0.1%; Simultaneously, the values ​​are displayed on the LED digital tube at the top of the casing; When the cumulative number of burned holes reaches 2,500 and the remaining lifespan is 25%, an early warning signal will be automatically activated, and the capacitor serial number will be uploaded to the cloud maintenance system. The coordinates of each ablation event are bound to the associated self-healing event data and stored to generate a spatiotemporal distribution map of lifespan loss.

10. The intelligent prediction method for the lifetime of thin-film capacitors that integrates multi-dimensional data as described in claim 9, characterized in that: The process of constructing and recording the process traceability code is carried out in the following steps: After the capacitor completes the potting process, the calculation period for the winding tension fluctuation variance σ² is taken as the entire winding process and the range of burr height H during slitting. max H min That is, the statistical value of a single roll of film and the adhesion strength of gold spraying Q. avg That is, the four sets of data, namely the global average value of the end face, the integral value of the heat setting temperature ∫Tdt of the heat drying process, are normalized to the interval [0,1] to form the process feature vector [V1,V2,V3,V4]; Input a three-layer convolutional neural network with a kernel size of 3×3, a stride of 1, and 64-128-256 channels. The first layer expands the vector into an 8×8 matrix, and after two convolution-pooling operations, it outputs a 128-dimensional feature tensor. The sigmoid activation function is used to binarize the data, generating a 128-bit binary code. A bit "1" corresponds to an output value ≥ 0.

5. On a 40mm×40mm anodized aluminum plate with a reserved QR code on the capacitor shell, an ultraviolet laser engraving machine is used to etch the coding pattern with a dot diameter of 20μm and an engraving depth accuracy of 30%: the bit "1" area has a micro pit array with an etch depth of 50μm and a pit spacing of 40μm, while the bit "0" area retains the original surface. After the recording is completed, the QR code area is scanned with a white light interferometer to verify that the micro-pit morphology qualification rate is >99.9%; at the same time, the binary code is converted to Base32 and uploaded to the cloud database and bound to the capacitor serial number. When a user reads the code using a scanning device, the three-dimensional shape recognition bit value of the micro-pit is reconstructed first, and then the complete process file in the cloud is called, including the winding speed curve, the distribution map of the slitting burrs, the thermal map of gold spraying, and the thermal setting stress cloud map, so as to realize the full life cycle data traceability.

Images