Method for predicting service life of nib spring

Abstract

The invention relates to the field of mechanical engineering, in particular to a method for predicting the service life of a nib spring, and aims to solve the problem that in the prior art, although a traditional digital simulation technology can provide part of prediction capacity, model construction of the traditional digital simulation technology is often limited to macroscopic structure parameters and cannot be brought into microcosmic geometric defect factors generated in the manufacturing process. By acquiring the key defect parameters and combining the digital twinborn model and the lightweight time sequence AI model for prediction, the method has the advantage of efficiently and accurately predicting the fatigue life of the nib spring.

Description

A method for predicting the lifespan of a pen tip spring. Technical Field

[0001] This application relates to the field of mechanical engineering, and more specifically, to a method for predicting the lifespan of a pen tip spring. Background Technology

[0002] In the production and quality control of pen tip springs, accurate prediction of fatigue life is crucial to ensuring the reliability of writing instruments. Currently, the industry generally uses bench testing methods for life assessment. This method requires repeated simulation of a large number of writing actions, resulting in a lengthy testing cycle. This cannot meet the timeliness requirements of rapid quality testing for mass production batches, and the resource consumption and cost burden of a single test significantly increase production pressure.

[0003] While traditional digital simulation technology can provide some predictive capabilities, its model construction is often limited to macroscopic structural parameters and fails to incorporate microscopic geometric defects generated during the manufacturing process, such as surface scratches and dimensional deviations. These defects have a key impact on the fatigue performance of springs, resulting in a large deviation between simulation results and actual lifespan.

[0004] In addition, some high-precision prediction solutions rely on expensive micro-nano detection equipment and highly customized artificial intelligence architectures. Such equipment is not only expensive to purchase and maintain, but also has a high technical implementation threshold, making it difficult for small and medium-sized enterprises with limited resources to apply it effectively. Ultimately, this has led to a long-term dilemma in the industry where it is difficult to balance prediction accuracy and cost control. Summary of the Invention

[0005] (a) Technical problem to be solved The purpose of this application is to provide a method and calculation device for predicting the life of pen tip springs, which has the advantages of high efficiency and accurate prediction of the fatigue life of pen tip springs, significantly shortening the test cycle and reducing resource consumption.

[0006] (II) Technical Solution This application provides a method for predicting the lifespan of a pen tip spring. The technical solution is as follows: S1, obtain at least one key defect parameter of the pen tip spring to be tested. The key defect parameter is a microscopic geometric defect parameter that is related to the manufacturing process of the pen tip spring and affects its fatigue life; S2, based on the key defect parameter, simulate the mechanical response of the pen tip spring under writing conditions by using a simplified digital twin model to obtain its time-series stress data; S3, input the key defect parameter and time-series stress data into a trained lightweight time-series AI model. The lightweight time-series AI model includes a spatial feature extraction module for extracting the spatial distribution characteristics of stress and a time-series feature extraction module for extracting the time-series change characteristics of stress; S4, obtain the predicted lifespan value of the pen tip spring based on the output of the lightweight time-series AI model.

[0007] Furthermore, this application also proposes that key defect parameters include any one or more of the following: winding gap tolerance, surface scratch depth, end burr size, spring wire diameter tolerance, and winding turn deviation.

[0008] Furthermore, this application also proposes that, in step S1, key defect parameters be obtained by using a laser confocal microscope, a high-magnification metallurgical microscope, or a high-precision optical profilometer.

[0009] Furthermore, this application proposes that, in step S2, the simplified construction and simulation of the digital twin model includes: S21, establishing a macroscopic structural finite element model based on the CAD model of the pen tip spring, with a mesh size of 0.05mm-0.2mm; S22, embedding key defect parameters into the macroscopic structural finite element model by modifying geometric features or material properties; S23, applying boundary conditions and loads to the model to simulate writing actions, including releasing the axial rotational degree of freedom of the spring end and constraining its radial displacement, and loads including simulating at least one of light pressure writing, heavy pressure writing, or tilted writing; S24, performing transient dynamic simulation and outputting the time-series stress data of key parts of the pen tip spring.

[0010] Furthermore, this application also proposes that the tools for simplifying the construction of digital twin models are ANSYS, ABAQUS, Code_Aster, or HyperMesh.

[0011] Furthermore, this application proposes that the spatial feature extraction module is a convolutional neural network, and the temporal feature extraction module is a long short-term memory network, a gated recurrent unit, or a bidirectional long short-term memory network.

[0012] Furthermore, this application also proposes that, before step S3, a step of preprocessing the time-series stress data is included, the preprocessing including: cutting the continuous time-series stress data into fixed-length time windows, and normalizing the cut data.

[0013] Furthermore, this application also proposes a training method for a lightweight temporal AI model, which includes: acquiring multiple sets of training samples, each set of training samples including: key defect parameters of a sample pen tip spring, temporal stress data obtained through a simplified digital twin model, and actual fatigue life data of the sample pen tip spring obtained through bench testing; using the key defect parameters and temporal stress data as inputs and the actual fatigue life data as the expected output, supervising the training of the lightweight temporal AI model.

[0014] Furthermore, this application also proposes a method for rapid quality inspection of mass-produced pen tip spring batches, specifically including: sampling 3%-10% of a batch of pen tip springs, performing steps S1 to S4 on each sample; if the life prediction value of all samples is not lower than the preset life threshold, the batch is deemed qualified; if the life prediction value of any sample is lower than the preset life threshold, the main failure cause is located based on the intermediate features output by the lightweight time-series AI model.

[0015] Furthermore, this application also proposes a computing device, comprising: a processor; a memory storing a computer program; and when the computer program is executed by the processor, implementing the above-mentioned method for predicting the lifespan of a pen tip spring.

[0016] (III) Beneficial Effects Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention obtains key defect parameters and combines them with digital twin models and lightweight time-series AI models for prediction, which has the advantages of efficient and accurate prediction of pen tip spring fatigue life. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0018] Figure 1 is a schematic diagram of the overall logical structure of the pen tip spring life prediction method. Detailed Implementation

[0019] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Example 1

[0022] In pen tip spring life prediction technology, existing methods suffer from problems such as extended prediction cycles, increased prediction costs, and discrepancies between predicted results and actual lifespan. Specifically, bench testing methods require simulating numerous writing cycles, making the prediction cycle unsuitable for the rapid quality inspection requirements of mass production environments; traditional digital simulation technologies only model macroscopic structures, ignoring microscopic geometric defects generated during manufacturing, which have a decisive impact on fatigue life; and existing high-precision prediction solutions rely on high-end micro-nano inspection equipment and complex artificial intelligence architectures, resulting in equipment procurement and maintenance costs exceeding the implementation capabilities of small and medium-sized enterprises, making the technical threshold too high. These problems collectively restrict improvements in prediction efficiency, cost control, and accuracy.

[0023] For example, in the mass production quality inspection process of pen tip spring manufacturers, when predicting the lifespan of new batches of products, the long-term operation of bench testing methods leads to production interruptions, and the quality inspection results cannot be fed back to the production stage in a timely manner. If traditional digital simulation is used, the predicted fatigue life is inconsistent with the early failure phenomena that actually occur in the writing conditions of the actual product because it does not take into account the micro-geometric defect parameters such as micro-cracks and surface roughness generated in the stamping process. High-precision prediction solutions cannot be deployed in resource-limited environments due to equipment costs and technical complexity, resulting in a lack of effective lifespan assessment basis in the quality inspection stage, which increases the possibility of unexpected failures of products in actual use.

[0024] If the above problems are not solved, the prediction of pen tip spring life will be limited by inefficient methods for a long time, resulting in low efficiency of mass production quality inspection and inability to identify batch defects in a timely manner; the deviation of the prediction results will reduce product reliability and affect user satisfaction; the inaccessibility of high-cost solutions will hinder small and medium-sized enterprises from adopting advanced technologies, exacerbate the technological imbalance in the industry, and be detrimental to the continuous improvement of product quality stability and market competitiveness.

[0025] As shown in Figure 1, this application proposes a method for predicting the lifespan of a pen tip spring, comprising the following steps: S1, obtaining at least one key defect parameter of the pen tip spring to be tested, wherein the key defect parameter is a microscopic geometric defect parameter related to the manufacturing process of the pen tip spring and affecting its fatigue life; S2, based on the key defect parameter, simulating the mechanical response of the pen tip spring under writing conditions using a simplified digital twin model to obtain its temporal stress data; S3, inputting the key defect parameter and temporal stress data into a trained lightweight temporal AI model, wherein the lightweight temporal AI model includes a spatial feature extraction module for extracting stress spatial distribution features and a temporal feature extraction module for extracting stress temporal change features; S4, obtaining the predicted lifespan value of the pen tip spring based on the output of the lightweight temporal AI model.

[0026] For ease of understanding, the following explains some key terms in this embodiment: Key defect parameters refer to microscopic geometric defect parameters that are related to the manufacturing process of the pen tip spring and affect its fatigue life. These parameters are usually minute structural features that are difficult to detect with the naked eye, such as surface scratches, uneven winding gaps, or end burrs, which have a direct impact on the fatigue failure of the spring during long-term use.

[0027] A simplified digital twin model is a computational model designed to simulate the mechanical behavior of a physical pen tip spring in an efficient manner. This model reduces computational complexity by abstracting and simplifying key features of the physical entity, particularly when considering the effects of microscopic defects, while still capturing the mechanical response crucial to fatigue life.

[0028] Writing conditions refer to the various mechanical loads and boundary conditions that the pen tip spring experiences during actual writing. These include axial pressure, radial displacement, tilt angle, and the resulting dynamic stress changes, all of which collectively determine the spring's actual working environment.

[0029] Time-series stress data refers to the sequential data showing the change of stress values ​​at key components of the pen tip spring over time under simulated writing conditions. This data reflects the transient mechanical response of the spring under dynamic loads and is crucial for analyzing fatigue damage accumulation.

[0030] Lightweight time-series AI models are trained artificial intelligence models designed to have low computational resource requirements while maintaining prediction accuracy. These models can process data with time-series characteristics and extract complex patterns related to fatigue life.

[0031] The spatial feature extraction module is a component of the lightweight temporal AI model. Its function is to analyze and identify the distribution patterns of stress in the physical space of the pen tip spring. This module can capture stress concentration areas, stress gradients, and other spatial stress features related to geometry and defects.

[0032] The temporal feature extraction module is another component of the lightweight temporal AI model. Its function is to analyze and identify dynamic patterns of stress changes over time. This module can capture the frequency, amplitude, phase, and other temporal features related to dynamic loads of stress fluctuations.

[0033] The life prediction value is a quantitative estimate of the expected fatigue life of the pen tip spring obtained through the above method. This value is usually expressed as the number of cycles or the service time and is used to evaluate the reliability and durability of the spring.

[0034] This method is typically implemented on a computing device that includes a processor and a memory containing a computer program. When the computer program is executed by the processor, the method described above is implemented.

[0035] The main features of the above technical solution are explained in more detail below: First, this method includes acquiring at least one key defect parameter of the pen tip spring under test. This key defect parameter is a microscopic geometric defect parameter related to the pen tip spring manufacturing process and affecting its fatigue life. The purpose of this step is to capture microscopic manufacturing defects that are easily overlooked in traditional macroscopic simulations but have a significant impact on fatigue life. For example, these parameters can be acquired by combining manual visual inspection with standard measuring tools. Specifically, the operator can use a magnifying glass or low-power microscope to observe the spring surface, identify possible scratches, pits, or burrs, and use vernier calipers or micrometers to perform multi-point measurements on the spring's winding gap, wire diameter, etc., and record the deviation values. In addition, local images of the spring can also be taken, and then the images can be manually analyzed to identify and quantify the size and location of microscopic defects. These methods aim to provide quantitative information on the microscopic geometric characteristics of the spring, providing input for subsequent mechanical response simulation.

[0036] Furthermore, based on the acquired key defect parameters, a simplified digital twin model is used to simulate the mechanical response of the pen tip spring under writing conditions, thereby obtaining its temporal stress data. This step aims to establish a virtual model that reflects the impact of real defects and simulate actual usage conditions on this model to obtain dynamic stress information. Specifically, a macroscopic geometric model of the pen tip spring can be constructed using general-purpose finite element analysis (FEA) software. Subsequently, the acquired key defect parameters are embedded into this macroscopic model through geometric feature corrections, such as manually adding a tiny groove corresponding to the actual scratch size and location, or adjusting the local mesh to reflect the deviation of the winding gap. Boundary conditions and loads simulating writing actions are applied to the model; for example, a periodic axial compressive force is applied to simulate the pressing action of the pen tip, and a radial constraint is applied to simulate the guidance of the pen tip within the pen barrel. By performing transient dynamic simulations, the time-varying sequence data of the stress at key parts of the pen tip spring can be output.

[0037] Subsequently, key defect parameters and time-series stress data are input into a trained lightweight time-series AI model. This lightweight time-series AI model includes a spatial feature extraction module for extracting spatial distribution characteristics of stress and a time-series feature extraction module for extracting time-series variation characteristics of stress. The purpose of this step is to utilize artificial intelligence technology to efficiently learn and identify complex patterns related to fatigue life from multi-source data. Specifically, key defect parameters can be directly used as numerical inputs to the model. Time-series stress data is then input into the model's feature extraction module. The spatial feature extraction module can employ a network structure capable of processing two-dimensional or three-dimensional mesh data; for example, by analyzing stress cloud maps in simulation results, it can identify spatial distribution characteristics such as the shape, size, and location of stress concentration areas. The time-series feature extraction module can employ a network structure capable of processing sequential data; for example, by analyzing stress-time curves, it can capture time-series characteristics such as stress peak values, stress ranges, loading frequencies, and stress variation trends. The outputs of these two modules are further integrated and comprehensively analyzed by the lightweight time-series AI model to establish a mapping relationship between defects, dynamic stress, and fatigue life.

[0038] Finally, based on the output of the lightweight temporal AI model, the predicted lifespan of the pen tip spring is obtained. The purpose of this step is to provide a quantitative assessment of fatigue life. Specifically, the final output layer of the lightweight temporal AI model is designed to directly output a numerical value representing the expected fatigue life of the pen tip spring under given defects and writing conditions, for example, in tens of thousands of writing cycles. This prediction can be directly used for quality assessment, product selection, or design improvement.

[0039] The following example provides a more detailed explanation of the above technical solution: Suppose a pen tip spring manufacturer, "Manufacturer A," faces a challenge: it needs to conduct a rapid and accurate life assessment of its mass-produced batches of pen tip springs. Traditional bench testing takes several days, which cannot meet its need for rapid delivery, while conventional macroscopic simulations suffer from insufficient prediction accuracy due to neglecting microscopic defects.

[0040] In response, "Manufacturer A" decided to adopt the method proposed in this application. First, in step S1, for a batch of pen tip springs to be tested, technicians randomly selected samples from the batch. For each sample, the technicians used a regular optical microscope equipped with a high-resolution camera to scan the surface of the spring and manually identified any possible micro-scratches or burrs. Using the microscope's built-in measurement software, the technicians manually measured the length, width, and depth of these defects. Simultaneously, the technicians used a high-precision digital caliper to perform multi-point measurements of the spring's winding gap and recorded the maximum deviation from the design standard. These measurement data, such as "surface scratch depth is 3 micrometers" and "maximum deviation of winding gap is 0.015 mm," were recorded as key defect parameters.

[0041] Next, in step S2, the technicians used a workstation equipped with general-purpose finite element analysis (FEA) software to construct a macroscopic geometric model of the pen tip spring. Based on the key defect parameters obtained in S1, the technicians manually introduced corresponding geometric corrections into the FEA model. For example, a tiny groove 3 micrometers deep was manually created at the location of the scratch on the model surface; simultaneously, the geometric parameters of the local winding were adjusted to reflect a gap deviation of 0.015 mm. Subsequently, the technicians applied boundary conditions and loads simulating writing motions to the FEA model. Specifically, one end of the spring was fixed, and the other end was subjected to a periodic axial load simulating light writing (e.g., 0.5 N) to heavy writing (e.g., 1.5 N), while allowing it to freely expand and contract in the axial direction and constraining its radial displacement. The FEA software performed transient dynamic simulations, calculating detailed data on the stress changes over time at key stress points (e.g., the scratch, the inside of the winding) of the spring during the simulated writing cycle, i.e., sequential stress data.

[0042] Subsequently, in step S3, the key defect parameters obtained in S1 (such as scratch depth and gap deviation) and the time-series stress data obtained in S2 are input into a pre-trained lightweight time-series AI model. This AI model is deployed on the industrial computer of "Manufacturer A". The spatial feature extraction module inside the model analyzes the stress distribution of the time-series stress data at each time step, identifying the local stress concentration areas caused by scratches or gap deviations and their morphological characteristics. At the same time, the time-series feature extraction module analyzes the changes in stress values ​​of these stress concentration areas over time, capturing the frequency and amplitude of stress fluctuations and the cumulative effect of fatigue loading. These two modules work together to deeply fuse microscopic defect information with dynamic mechanical response information.

[0043] Finally, in step S4, the lightweight temporal AI model outputs a value, such as "180,000 cycles," based on its complex internal calculation logic. This value is the predicted lifespan of the pen tip spring for that sample. By performing steps S1 to S4 on multiple samples in the batch and comprehensively analyzing their predicted lifespans, manufacturer A can quickly determine whether the overall quality of the batch of pen tip springs meets the requirements.

[0044] The above-mentioned technical solutions, as illustrated in the example of "Manufacturer A", demonstrate a significant technological contribution to the field of pen tip spring life prediction.

[0045] Compared to traditional bench testing methods that take several days, this method can significantly shorten the life prediction cycle of pen tip springs to just a few hours. In the example of "Manufacturer A", the entire process, from obtaining microscopic defect parameters to finally obtaining the life prediction value, can be completed quickly. This allows the manufacturer to conduct rapid quality inspections on mass production batches, significantly improving production efficiency and response speed.

[0046] Furthermore, this method effectively addresses the problem of large prediction errors caused by traditional digital simulations neglecting microscopic defects by explicitly acquiring and utilizing key defect parameters and embedding them into a simplified digital twin model. In the example, the measurement of surface scratch depth and winding gap deviation, and their representation in the FEA model, ensure that the simulation results more realistically reflect the impact of defects on the spring's mechanical response, thereby improving the accuracy of life prediction. This contrasts with the limitations of traditional simulations, which only focus on macroscopic structures and cannot capture the influence of microscopic details.

[0047] Furthermore, this method employs a lightweight temporal AI model, which effectively integrates key defect parameters, stress spatial distribution characteristics, and stress temporal variation characteristics to achieve high-precision lifetime prediction while avoiding reliance on expensive micro-nano inspection equipment and complex AI architectures. In the example, by using a common optical microscope and general-purpose FEA software, combined with a lightweight AI model deployed on an industrial computer, Manufacturer A was able to achieve high-precision prediction within a controllable cost range. This contrasts sharply with the few existing high-precision solutions that require huge investments in high-end equipment and the development of complex AI systems, enabling advanced lifetime prediction technology to be adopted by a wider range of SMEs and effectively solving the industry's dilemma of "either low accuracy or high cost."

[0048] In summary, this method forms a collaborative and mutually reinforcing technical system through low-cost acquisition of key defect parameters, construction of a simplified defect-embedded digital twin model, and a lightweight AI model that fuses spatial and temporal features. This integrated approach not only significantly improves prediction speed and accuracy but also drastically reduces implementation costs and technical barriers, providing a substantial and significantly advanced technical means for the quality control and product development of pen tip springs.

[0049] In some of the solutions described above in this application, key defect parameters are proposed to predict the life of pen tip springs. However, in this process, it is necessary to specifically identify and quantify which micro-geometric defect parameters are related to the manufacturing process in order to avoid the prediction error from increasing or the detection operation from becoming infeasible due to the ambiguity of parameter definition, thereby affecting the accuracy of life prediction and the efficiency of practical application.

[0050] In this regard, this application further proposes key defect parameters including any one or more of the following: winding gap tolerance, surface scratch depth, end burr size, spring wire diameter tolerance, and winding turn deviation.

[0051] The winding gap tolerance refers to the allowable deviation range of the distance between adjacent windings of the pen tip spring. It directly reflects the precision and consistency of the winding process during spring manufacturing. Excessive or insufficient gaps can lead to localized stress concentration, accelerating fatigue failure. This parameter can be measured at multiple points using non-contact optical measurement equipment, such as laser displacement sensors or high-precision CCD cameras combined with image processing technology, to calculate the average value and deviation; or it can be precisely measured using contact probes, such as coordinate measuring machines. The surface scratch depth refers to the vertical depth of dents or damage on the pen tip spring surface caused by improper processing, handling, or storage. Surface scratches are a major source of fatigue crack initiation. Under cyclic loading, stress concentration occurs at the scratch tip, leading to rapid crack propagation and significantly reducing the spring's fatigue life. This parameter can be measured using a laser confocal microscope for three-dimensional morphological scanning; or with high precision using an atomic force microscope (AFM); or estimated using a high-magnification metallographic microscope combined with image analysis software. End burr size refers to the tiny protrusions or sharp edges remaining at the end of the pen tip spring after cutting or chamfering, exceeding the specified flatness. End burrs can cause high local stress concentration, especially when the spring end is in contact with the pen tip or other components, becoming a weak point for fatigue failure. This parameter can be measured non-contactly using a high-precision optical profilometer to measure the height and width of the burrs; or through high-resolution imaging and dimensional analysis using a scanning electron microscope (SEM). Spring wire diameter tolerance refers to the allowable deviation range of the wire diameter used to manufacture the pen tip spring. Wire diameter tolerance directly affects the spring's stiffness, load-bearing capacity, and stress distribution. Uneven wire diameter leads to inconsistent mechanical properties in different parts of the spring, thus affecting the overall fatigue life. This parameter can be continuously measured non-contactly using an online laser diameter gauge; or through multi-point sampling measurement using a high-precision micrometer or micrometer screw gauge. Winding count deviation refers to the difference between the actual number of turns of the pen tip spring and the design-specified number of turns. Deviations in the number of winding turns alter the effective number of turns in a spring, affecting its overall stiffness, deformation, and stress distribution. This leads to discrepancies between the actual mechanical response and design expectations, impacting the accuracy of lifespan predictions. This parameter can be confirmed manually by visually counting the turns with a magnifying glass; alternatively, an automated image recognition system can process the spring image to automatically identify and count the number of winding turns.

[0052] In this method for predicting the lifespan of pen tip springs, the specific types of key defect parameters are clearly defined for the step of obtaining these parameters. These include winding gap tolerance, surface scratch depth, end burr size, spring wire diameter tolerance, and winding count deviation. These parameters are not arbitrarily selected but are microscopic geometric defects strongly correlated with fatigue life, screened based on a deep understanding of the pen tip spring manufacturing process, structural characteristics, and fatigue failure mechanisms. By accurately measuring these specific parameters, information on microscopic defects that may occur during the pen tip spring manufacturing process and significantly affect fatigue life can be comprehensively and quantitatively captured. For example, winding gap tolerance directly reflects the uniformity of spring winding, surface scratch depth quantifies potential crack initiation sources, end burr size reveals the risk of local stress concentration, spring wire diameter tolerance ensures the uniformity of material mechanical properties, and winding count deviation ensures the accuracy of the overall structural response. These specific and quantified key defect parameters are then input into a simplified digital twin model. Digital twin models can more accurately simulate the mechanical response of pen tip springs under writing conditions based on these realistic microscopic defect information, generating time-series stress data that includes the effects of defects. Compared to traditional simulations that only consider macroscopic structures, introducing these specific microscopic defect parameters makes the simulation results closer to reality, significantly improving the realism and reliability of the time-series stress data. Then, these key defect parameters containing microscopic defect information, along with the more accurate time-series stress data, are input into a trained lightweight time-series AI model. The AI ​​model can learn the complex nonlinear relationship between defect features and fatigue life from this data. Due to the explicitness and accuracy of the input parameters, the AI ​​model can more effectively extract the spatial distribution characteristics and temporal variation characteristics of stress, thereby outputting a more accurate prediction of pen tip spring life. In this way, this solution avoids the problems of increased prediction errors and infeasibility of detection operations caused by ambiguous definitions of key defect parameters. Explicit parameter definitions make the detection process standardized and quantifiable, ensuring the quality of input data, and thus improving the accuracy and practical application efficiency of the entire life prediction method. This refined definition and quantification of key defect parameters is a crucial step in achieving high-precision, low-cost, and rapid lifetime prediction. It makes the entire prediction process, from source data collection to final result output, more reliable and practical.

[0053] As a specific implementation method, when obtaining the key defect parameters of the pen tip spring to be tested, the following parameter combinations can be selected for measurement and analysis: First, for the winding gap tolerance of the pen tip spring, a high-precision optical measurement system can be used to scan the distance between adjacent coils of the spring at multiple points, calculate its average gap and maximum deviation, and evaluate the uniformity of the winding process.

[0054] For surface scratch depth, a laser confocal microscope can be used to perform a three-dimensional morphological scan of the spring surface, accurately measuring the depth and length of any visible scratches to identify potential crack initiation points. Furthermore, for end burr dimensions, a high-resolution industrial camera combined with image processing software can be used to magnify and inspect the ends of the spring, quantifying the height and width of the burrs to assess the end processing quality. In addition, spring wire diameter tolerances can be monitored in real-time during production using an online diameter gauge, or multi-point sampling measurements can be performed on finished springs using a high-precision micrometer to ensure the uniformity of the wire diameter.

[0055] The deviation in the number of winding turns can be detected by analyzing the overall image of the spring using an automated vision inspection system. This system automatically identifies and counts the actual number of winding turns and compares it with the design value to verify the consistency of the structure. Obtaining these specific parameters allows the subsequent simplified digital twin model to more realistically reflect the actual physical state of the pen tip spring, thus providing high-quality input data for the lightweight time-series AI model and ultimately achieving accurate lifespan prediction.

[0056] The aforementioned technical solution clearly defines the specific types of key defect parameters, including winding gap tolerance, surface scratch depth, end burr size, spring wire diameter tolerance, and winding turn deviation. These parameters are microscopic geometric defects that significantly affect fatigue life, identified through screening based on the failure mechanism and manufacturing process characteristics of pen tip springs. This clear parameter definition solves the problem of unclear key defect parameters in life prediction methods, making the acquisition of key defect parameters more specific and operable. Inspectors can perform standardized measurements based on these clear parameters, avoiding inspection errors and operational difficulties caused by unclear parameter definitions. Furthermore, since these parameters can be quantified using conventional inspection equipment, such as laser confocal microscopes, high-magnification metallurgical microscopes, or high-precision optical profilometers, the inspection cost and technical threshold are reduced, making this life prediction method easier to promote and apply in actual production environments, especially suitable for small and medium-sized enterprises. When combined with basic prediction methods, these specific and quantified key defect parameters, as inputs to a simplified digital twin model, can more realistically reflect the actual state of the pen tip spring, making the time-series stress data simulated by the model more accurate. Furthermore, the lightweight temporal AI model can learn a more accurate defect-lifetime mapping relationship from these high-quality input data, significantly improving the accuracy of the final lifetime prediction. Therefore, this approach not only improves the accuracy of lifetime prediction but also enhances the practicality and operability of the method, effectively avoiding the problems of increased prediction errors or infeasibility of detection operations caused by ambiguous parameter definitions, thereby improving the overall efficiency and reliability of pen tip spring lifetime prediction.

[0057] In some of the solutions described above in this application, key defect parameters are proposed to accurately predict the life of the pen tip spring. However, in this process, using unsuitable testing equipment may lead to insufficient measurement accuracy or excessive cost, affecting the feasibility and economy of the prediction. This is especially true for small and medium-sized enterprises, which find it difficult to afford the procurement and maintenance of high-end equipment. Therefore, this application further proposes that in step S1, the key defect parameters be obtained using a laser confocal microscope, a high-magnification metallurgical microscope, or a high-precision optical profilometer.

[0058] A laser confocal microscope is a microscopic device that achieves high-resolution three-dimensional imaging using laser scanning and confocal principles. Its working principle involves point scanning and pinhole filtering to effectively suppress scattered light from non-focal planes, thereby obtaining clear optical slice images. It can also construct the three-dimensional morphology of the sample through Z-axis scanning. This device provides extremely high vertical and lateral resolution, making it particularly suitable for precise measurements of microscopic surface morphology, roughness, scratch depth, and minute geometric dimensions. For example, a white light interferometric confocal microscope or a laser scanning confocal microscope can be used, with software controlling the scanning path and depth, to perform non-contact measurements of surface defects in a pen tip spring.

[0059] Different objectives and detectors can be configured to suit the needs of different materials and measurement ranges. A high-magnification metallurgical microscope is a microscopic device specifically designed for observing and analyzing the microstructure of metallic materials. It is typically equipped with high-magnification objectives and eyepieces and uses reflected light illumination, enabling clear visualization of grains, phase composition, inclusions, cracks, and various geometric defects in metallic materials. This equipment is widely used in industrial applications for material quality control and failure analysis, and can accurately measure dimensional deviations. For example, upright or inverted metallurgical microscopes can be used, along with image analysis software, to quantitatively analyze the cross-sectional or surface features of a pen tip spring, such as measuring wire diameter and gaps. Furthermore, observation modes such as polarized light and differential interferometry can enhance the identification of specific microscopic features. A high-precision optical profilometer is a device that uses optical principles to perform non-contact measurements of the surface morphology of an object. It typically employs techniques such as white light interferometry, phase-shifting methods, or fringe projection, reconstructing the three-dimensional morphology data of the object's surface by analyzing changes in reflected light or projected fringes. This equipment boasts advantages such as high measurement speed, high accuracy, and non-destructive testing of samples, making it suitable for comprehensive inspection of the dimensions and morphology of complex curved surfaces, microstructures, and precision parts. For example, an optical profilometer based on the principle of white light interference or the principle of fringe projection can be used to perform rapid and high-precision scanning and data acquisition of the overall three-dimensional shape, end features, or winding structure of a pen tip spring.

[0060] Different measurement probes or sensors can be configured to accommodate pen tip springs of varying sizes and surface characteristics. Obtaining the key defect parameters refers to measuring the pen tip spring using the aforementioned testing equipment to acquire microscopic geometric defect parameters related to the pen tip spring manufacturing process and affecting its fatigue life. This process aims to provide accurate and reliable input data for subsequent life prediction. For example, by setting the equipment's measurement program, specific areas of the pen tip spring can be automatically scanned, and quantitative values ​​of specific parameters such as surface scratch depth, winding gap tolerance, end burr size, spring wire diameter tolerance, or winding turn deviation can be extracted using the equipment's built-in analysis software or third-party data processing tools. Alternatively, manual assistance in locating key areas can be combined with automated measurement programs to improve measurement efficiency and accuracy.

[0061] The solution in this application introduces specific detection equipment in step S1, namely a laser confocal microscope, a high-magnification metallurgical microscope, or a high-precision optical profilometer, to obtain the key defect parameters of the pen tip spring. The selection of these devices is not arbitrary, but based on a comprehensive consideration of the microscopic characteristics of the key defect parameters, the required measurement accuracy, and the cost affordability of small and medium-sized enterprises. The laser confocal microscope, with its high-resolution three-dimensional imaging capability, can accurately capture quantitative data of microscopic geometric defects such as surface scratch depth and end burr size, ensuring the measurement accuracy of the three-dimensional morphology parameters.

[0062] High-magnification metallurgical microscopes focus on the microstructure analysis of metallic materials, effectively identifying and quantifying microscopic dimensional fluctuations related to wire processing and winding processes, such as spring wire diameter tolerance and winding turn deviation, ensuring the authenticity of material and process-related parameters. High-precision optical profilometers employ non-contact measurement methods, enabling rapid and non-destructive acquisition of the overall three-dimensional morphology data of the pen tip spring, particularly suitable for defect parameters requiring overall morphology analysis, such as winding gap tolerance. These devices ensure sufficient accuracy and reliability of the acquired key defect parameters, providing accurate geometric and material property inputs for the simplified digital twin model simulation of the pen tip spring's mechanical response under writing conditions in subsequent step S2, avoiding simulation deviations caused by data errors. Furthermore, these devices are all mature, commonly used industrial equipment, with procurement and maintenance costs far lower than high-end micro / nano detection equipment. This allows the entire life prediction method to maintain high accuracy while possessing good economic efficiency and operability, effectively resolving the contradiction between "insufficient accuracy" and "excessive cost" in traditional methods, providing a reliable and economical data foundation for pen tip spring life prediction.

[0063] As a specific implementation method, when predicting the lifespan of a batch of pen tip springs, it is first necessary to obtain their key defect parameters. For example, for a batch of pen tip springs, a sample can be randomly selected. To obtain the surface scratch depth and end burr size of this sample, a laser confocal microscope can be used to scan its surface. The operator fixes the pen tip spring on the microscope stage, sets the scanning area and depth through software, and the laser beam scans the surface of the pen tip spring point by point, collecting the reflected light using the confocal principle, ultimately generating a high-resolution three-dimensional topographic image. Through image analysis software, the depth and width of the surface scratches, as well as the height and shape of the end burrs, can be accurately quantified.

[0064] Simultaneously, to obtain the spring wire diameter tolerance and winding count deviation of this sample, a high-magnification metallurgical microscope can be used for observation and measurement. The operator places the pen tip spring under the metallurgical microscope, adjusts the magnification, and observes the cross-section or winding structure of the spring wire through the eyepiece or a connected monitor. Using the microscope's built-in measurement function or image analysis software, the actual diameter of the spring wire is measured and compared with the design value to obtain the tolerance. At the same time, the number of winding turns is counted and compared with a standard value. Furthermore, to obtain the winding gap tolerance, a high-precision optical profilometer can be used to perform a non-contact scan of the overall winding structure of the pen tip spring. The profilometer quickly constructs three-dimensional point cloud data of the pen tip spring by emitting a beam and receiving reflected light. Then, specialized software is used to analyze the gap between adjacent winding turns and calculate its tolerance. Through the above operations, various key defect parameters of the pen tip spring can be obtained comprehensively and accurately, providing reliable input data for subsequent life prediction.

[0065] Through the above technical solution, this application effectively solves the problems of insufficient measurement accuracy or excessive cost caused by improper selection of testing equipment in traditional pen tip spring life prediction methods. Specifically, by selecting three types of industrially common and mature testing equipment—laser confocal microscope, high-magnification metallurgical microscope, and high-precision optical profilometer—this application can significantly reduce equipment procurement and maintenance costs while ensuring high-precision quantification of microscopic geometric defect parameters of pen tip springs. The laser confocal microscope ensures accurate measurement of three-dimensional microscopic defects such as surface scratch depth and end burr size; the high-magnification metallurgical microscope ensures the authenticity of manufacturing process-related parameters such as spring wire diameter tolerance and winding count deviation; and the high-precision optical profilometer enables rapid and non-destructive acquisition of overall morphological parameters such as winding gap tolerance. This equipment combination not only meets the measurement accuracy requirements of key defect parameters but also makes it economical for small and medium-sized enterprises, thereby lowering the technical threshold and improving the feasibility and scalability of the life prediction method. Furthermore, these devices are relatively easy to operate, which helps improve testing efficiency and adapts to the rapid quality inspection needs of mass production batches. Ultimately, this solution provides high-precision, low-cost, and easy-to-operate methods for acquiring key defect parameters, thus providing high-quality input data for subsequent simplified digital twin models and lightweight time-series AI models, thereby ensuring the accuracy and practicality of the entire pen tip spring life prediction method.

[0066] In some of the solutions mentioned above in this application, a simplified digital twin model is proposed to simulate the mechanical response of the pen tip spring under writing conditions and obtain time-series stress data. However, in its implementation, how to accurately embed microscopic defect parameters and efficiently perform simulations to obtain reliable data, while avoiding high-cost and complex models, is a problem that needs to be solved.

[0067] In response, this application further proposes the above-mentioned simplified digital twin model construction and simulation method, which includes the following steps: First, a macroscopic structural finite element model is established based on the CAD model of the pen tip spring, and the mesh size is set to 0.05mm-0.2mm; then, key defect parameters are embedded into the macroscopic structural finite element model by means of geometric features or material properties correction; subsequently, boundary conditions and loads simulating writing actions are applied to the model, wherein the boundary conditions include releasing the axial rotational degree of freedom of the spring end and constraining its radial displacement, and the loads include simulating at least one of light pressure writing, heavy pressure writing, or tilted writing; finally, transient dynamic simulation is performed to output the time-series stress data of key parts of the pen tip spring.

[0068] A macroscopic structural finite element model was established based on the CAD model of the pen tip spring, with a mesh size of 0.05mm-0.2mm. The CAD model, a computer-aided design model, defines the geometry, dimensions, and topology of the pen tip spring and forms the basis for subsequent finite element analysis. Establishing the macroscopic structural finite element model involves discretizing the continuous physical model into a set of finite elements and nodes to solve its mechanical response using numerical methods. This model can be established using various finite element analysis software, such as ANSYS, ABAQUS, Code_Aster, or HyperMesh. The mesh size was chosen to balance computational accuracy and efficiency. Smaller mesh sizes can more accurately capture local stress concentrations but significantly increase computational load; larger mesh sizes improve computational efficiency but may sacrifice local accuracy. The mesh size range of 0.05mm-0.2mm ensures sufficient capture of stress distribution in key parts of the pen tip spring while keeping the simulation computational load within an acceptable range, avoiding the need for high-end computing resources.

[0069] Key defect parameters are embedded into the macroscopic structural finite element model by modifying geometric features or material properties. Key defect parameters refer to microscopic geometric defect parameters related to the pen tip spring manufacturing process and affecting its fatigue life, such as winding gap tolerance, surface scratch depth, end burr size, spring wire diameter tolerance, and winding turn deviation. These parameters can be obtained using equipment such as laser confocal microscopes, high-magnification metallographic microscopes, or high-precision optical profilometers. Embedding by geometric feature modification involves directly modifying or adding geometric elements to the macroscopic structural finite element model to represent defects, such as modeling surface scratches as tiny grooves on the model surface, or adjusting the geometric dimensions of the winding gap. Embedding by material property modification involves modifying local material properties of the model in the defect area or the area affected by the defect, such as reducing the local elastic modulus or yield strength, to equivalently simulate the impact of the defect on the material's mechanical properties. This embedding method avoids complex full-size modeling of microscopic defects, thus significantly simplifying the modeling process and computational complexity, while effectively reflecting the impact of defects on the overall mechanical response.

[0070] Boundary conditions and loads simulating writing actions are applied to the model. Boundary conditions are constraints imposed on the model's edges or specific regions in finite element analysis to simulate the confined state of the pen tip spring in a real-world working environment. Releasing the axial rotational freedom of the spring end while constraining its radial displacement aims to accurately simulate the pen tip spring's installation state inside the pen barrel; that is, the spring can freely extend, contract, and rotate axially, but is restricted radially by the inner wall of the pen barrel. Loads are external forces applied to the model to simulate the pressure exerted on the pen tip spring during writing. Simulating at least one of the following conditions—light pressure writing, heavy pressure writing, or tilted writing—aims to cover the most common stress conditions experienced by the pen tip spring in actual use, ensuring that the simulation results reflect the mechanical response in real-world writing scenarios. These boundary conditions and load settings enable the digital twin model to more realistically reflect the stress state of the pen tip spring under writing conditions.

[0071] Perform transient dynamics simulations to output time-series stress data for key components of the pen tip spring. Transient dynamics simulation is an analytical method used to calculate the dynamic response of a structure under time-varying loads, including displacement, velocity, acceleration, and stress. Unlike static analysis, transient dynamics simulation can capture the dynamic process of stress changes over time, which is crucial for fatigue life prediction. Key components of the pen tip spring typically refer to areas of stress concentration or prone to fatigue failure, such as the spring's ends, winding corners, or areas with microscopic defects. By focusing on these key components to output time-series stress data, the amount of data can be effectively reduced, data processing efficiency improved, and high-quality, targeted input provided for subsequent lightweight time-series AI models.

[0072] The proposed solution constructs and simulates a simplified digital twin model through the aforementioned steps to simulate the mechanical response of a pen tip spring under writing conditions. First, a macroscopic structural finite element model is established based on a CAD model, with an appropriate mesh size set, laying an efficient and sufficiently accurate foundation for subsequent simulations. Building upon this, key defect parameters obtained from step S1 are innovatively embedded into the macroscopic model using geometric features or material properties corrections. This allows the model to realistically reflect the impact of microscopic defects on mechanical behavior without significantly increasing complexity. Subsequently, by applying precise boundary conditions simulating writing actions and diverse load conditions, the high realism of the simulation scenario is ensured. Finally, transient dynamic simulation is performed, focusing on key components to output time-series stress data. This not only captures the dynamic changes in stress but also ensures the validity and conciseness of the data, providing high-quality and efficient data support for the input of the lightweight time-series AI model in the subsequent step S3. This method cleverly balances simulation accuracy, computational efficiency, and cost, enabling reliable mechanical response data to be obtained even with limited computing resources. This effectively solves the problems of neglecting microscopic defects and low efficiency in traditional simulations, and provides a solid foundation for subsequent lifetime prediction.

[0073] As a specific implementation method, a CAD model of the pen tip spring can be created using SolidWorks software and imported into ANSYS Workbench for finite element analysis. In ANSYS Workbench, a macroscopic structural finite element model is built based on the imported CAD model, and the automatic mesh generation tool is used to set the mesh size to approximately 0.1 mm. For embedding key defect parameters, for example, if a scratch with a depth of 0.02 mm is detected on the surface of the pen tip spring, a V-groove with a depth of 0.02 mm and a width of 0.05 mm can be added to the corresponding location in the CAD model to simulate the geometric characteristics of the scratch; if a local tolerance in the spring wire diameter is detected, the elastic modulus of the material in that local area can be equivalently adjusted to 95% of its original value in the finite element model to simulate the weakening of its mechanical properties. When applying boundary conditions, one end of the spring can be fixed, allowing it to rotate freely along the axial direction, but restricting any radial displacement. Regarding loads, a heavy-pressure writing condition can be simulated, for example, applying a periodically varying axial pressure to the other end of the spring, with a peak pressure of 2 N, a frequency of 5 Hz, and a duration of 10 seconds. Subsequently, the transient dynamics solver of ANSYS is executed, and the solution is set at key locations such as the end of the spring, the winding corner, and the scratched area. Equivalent stress data is output every 0.001 seconds to form a time-series stress curve.

[0074] Through the aforementioned technical solution, this application can accurately integrate the microscopic defect parameters of the pen tip spring into a simplified digital twin model, effectively solving the problem of insufficient prediction accuracy caused by ignoring microscopic defects in traditional simulations. Simultaneously, by employing a macroscopic structural finite element model, a moderate mesh size, defect embedding methods that correct geometric or material properties, and data output focusing on key components, the efficiency of the simulation is significantly improved, keeping the time consumed in a single simulation at a low level. This significantly reduces the demand for high-end computing resources, making it affordable for small and medium-sized enterprises, thus resolving the dilemma of "either low accuracy or high cost" in existing technologies. The acquired high-quality time-series stress data provides a reliable and efficient input for subsequent lightweight time-series AI models, thereby improving the overall accuracy and practicality of pen tip spring life prediction.

[0075] In some of the solutions described above in this application, simplified digital twin models are proposed for constructing and simulating the mechanical response of pen tip springs. However, in the implementation process, the selection of construction tools may involve high-end or self-developed software, resulting in excessively high procurement and maintenance costs and increased technical barriers, making it difficult for small and medium-sized enterprises to bear the implementation burden. Therefore, this application further proposes simplified digital twin model construction tools as ANSYS, ABAQUS, Code_Aster, or HyperMesh.

[0076] The simplified digital twin model building tool refers to a software platform used to create and simulate digital twin models of pen tip springs. The core function of this tool is to provide an integrated environment that allows users to define the geometry, material properties, applied boundary conditions, and loads of the pen tip spring, and to perform mechanical simulations to obtain its dynamic response data. This can be achieved through commercial computer-aided engineering (CAE) software suites, open-source simulation platforms, or customized tools developed specifically for particular applications.

[0077] ANSYS is a commercial finite element analysis (FEA) software widely used in engineering simulation. It offers powerful simulation capabilities for various physical fields, including structural mechanics, fluid dynamics, and electromagnetics. Particularly in structural analysis, it supports complex geometric modeling, material definition, mesh generation, and transient dynamic analysis. Its modular design and user-friendly interface make it a commonly used simulation tool in industry.

[0078] ABAQUS is another leading commercial finite element analysis software, renowned for its powerful capabilities in nonlinear analysis, material behavior modeling, and handling complex contact problems. It can accurately simulate the response of various materials under different loading conditions, including fatigue and fracture, making it ideal for detailed structural mechanics simulations, especially excelling in handling large deformations and complex material models.

[0079] Code_Aster is an open-source finite element analysis software developed and maintained by EDF (Électricité de France). It provides comprehensive structural mechanics analysis capabilities, including statics, dynamics, and fatigue analysis, and can handle complex geometric and material models. As open-source software, it offers high flexibility and customizability; users can obtain it for free and modify and extend it according to their needs, while also enjoying active community support.

[0080] HyperMesh is a leading commercial finite element preprocessor primarily used for geometry cleanup, mesh generation, and model setup. It supports importing various CAD data and generates high-quality finite element meshes, preparing the model for subsequent finite element analyses (such as those performed in ANSYS or ABAQUS). HyperMesh's advantages lie in its efficient meshing capabilities and adaptability to complex geometries, significantly improving modeling efficiency and simulation accuracy.

[0081] This application specifies ANSYS, ABAQUS, Code_Aster, or HyperMesh as the tools for constructing simplified digital twin models, enabling efficient and economical implementation of the simplified digital twin model construction and simulation step S2 in the entire pen tip spring life prediction method. Specifically, when establishing the macroscopic structural finite element model S21 based on the CAD model of the pen tip spring, these tools provide powerful geometric processing and mesh generation functions, ensuring that the mesh size is within the range of 0.05mm-0.2mm, meeting the simulation accuracy requirements. When embedding key defect parameters into the macroscopic structural finite element model S22 by modifying them through geometric features or material properties, these tools support flexible model modification and property definition, accurately incorporating the influence of microscopic defects into the macroscopic model. When applying boundary conditions and loads S23 to the model to simulate writing actions, these tools provide rich types of boundary conditions and load application methods, accurately simulating the axial rotational degree of freedom of the spring end and constraining its radial displacement, as well as simulating various working conditions such as light pressure writing, heavy pressure writing, or tilted writing. Finally, when performing transient dynamic simulations and outputting the time-series stress data S24 for key components of the pen tip spring, tools such as ANSYS, ABAQUS, and Code_Aster, as mature solvers, were able to stably and efficiently complete the dynamic analysis and extract the required time-series stress data. By selecting these comprehensive, technologically mature, and widely available tools, this application ensures that the simplified construction process of the digital twin model meets technical requirements while effectively controlling costs and technical barriers, thereby making the entire pen tip spring life prediction method more operable and scalable.

[0082] As a specific implementation method, the construction of a simplified digital twin model can be achieved using the ANSYS Workbench platform. First, the CAD model of the pen tip spring is imported into the ANSYS SpaceClaim module for geometric cleanup and simplification. Then, in the ANSYS Mechanical module, the model is meshed using finite element methods, ensuring the mesh size is controlled to approximately 0.1 mm. If the critical defect parameter is the depth of a surface scratch, a corresponding groove can be created on the spring surface at the scratch location using geometric modification functions to simulate the defect. If the critical defect parameter is a localized material defect, it can be simulated by defining the material properties of the local area, such as reducing the elastic modulus or yield strength of that area. Next, boundary conditions and loads simulating writing motions are applied to the model. For example, one end of the spring is fixed and its radial displacement is constrained while allowing axial rotation; a time-varying force load is applied to the other end to simulate the dynamic stress under heavy writing conditions. Finally, transient structural analysis is configured, and the simulation is executed to extract time-series stress data from the stress concentration or defect areas of the pen tip spring.

[0083] As another specific implementation method, the open-source software Code_Aster can be used to simplify the construction of the digital twin model. First, the CAD model of the pen tip spring is imported using pre-processing software such as Salome, and meshed to generate a finite element mesh that meets the requirements. If the key defect parameter is the size of the end burr, the geometric characteristics of the burr can be accurately represented at the spring end through local mesh refinement or geometric correction. If the key defect parameter is the winding gap tolerance, it can be reflected by adjusting the geometric gap between adjacent spring coils. Then, boundary conditions are defined in the Code_Aster input file. For example, the axial rotational degree of freedom of the spring end is released while its radial displacement is constrained, while a composite load simulating a tilted writing condition is applied. This load includes axial and radial force components and varies over time. Finally, the Code_Aster solver is run to perform transient dynamic simulation, and the time-series stress data of key parts of the pen tip spring (such as defects or high-stress areas) is extracted from the simulation results.

[0084] In traditional pen tip spring life prediction technologies, inappropriate selection of tools for building simplified digital twin models can lead to excessively high procurement and maintenance costs, increased technical barriers, and make implementation difficult for small and medium-sized enterprises (SMEs). This application effectively solves these problems by limiting the tools for building simplified digital twin models to ANSYS, ABAQUS, Code_Aster, or HyperMesh. These tools have a broad application base and mature technical support in industry, natively supporting core functions such as macroscopic structural finite element modeling, geometric feature or material property correction to embed defects, transient dynamic simulation, and output of time-series stress data for key components, without requiring additional development or complex customization. This not only ensures the smoothness of the simulation process and the reliability of the data but also significantly reduces the procurement cost and technical barriers of the tools. For example, Code_Aster, as an open-source tool, can be obtained at zero cost, while simplified versions or rental models of ANSYS and ABAQUS provide economically feasible options, and HyperMesh can efficiently complete pre-processing. This tool selection strategy enables the pen tip spring life prediction method to be implemented efficiently within the budget and technical capabilities of small and medium-sized enterprises, thereby breaking the industry's inherent perception that "simulation tools equal high costs" and greatly promoting the application and popularization of this advanced prediction technology in a wider range, improving the efficiency and accuracy of mass production batch quality inspection.

[0085] In some of the solutions mentioned above in this application, a lightweight temporal AI model is proposed to extract spatial distribution features and temporal variation features of stress and predict lifetime. However, in its implementation, the specific implementation methods of the spatial feature extraction module and the temporal feature extraction module are not clear, which may lead to low computational efficiency or insufficient feature extraction when the model processes high-dimensional stress data, affecting prediction accuracy and speed, especially in mass production rapid quality inspection scenarios where it is difficult to meet real-time requirements.

[0086] In this regard, this application further proposes that the spatial feature extraction module is a convolutional neural network, and the temporal feature extraction module is a long short-term memory network, a gated recurrent unit, or a bidirectional long short-term memory network.

[0087] Convolutional neural networks (CNNs) are deep learning models that effectively extract local features from input data and abstract and combine these features through local receptive fields and weight sharing mechanisms. In this application, they are used to extract spatial distribution features, such as stress concentration areas and stress gradient changes, from the temporal stress data of a pen tip spring. These features are closely related to the microscopic geometric defects of the pen tip spring. Specifically, a one-dimensional CNN can be used to extract features from the stress distribution sequence along the key parts of the pen tip spring, or a two-dimensional CNN can be used to map the stress data of the key parts of the pen tip spring into a two-dimensional image for feature extraction.

[0088] Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network that uses gating mechanisms to control the flow of information, thereby effectively capturing and memorizing long-term dependencies in long sequences. In this application, it is used to extract temporal variation features, i.e., the dynamic changes in stress over time, from the temporal stress data of a pen tip spring. Single-layer or multi-layer stacked LSM networks can be used, or LSM networks with attention mechanisms can be employed.

[0089] Gated recurrent units (GRUs) are a simplified variant of long short-term memory (LSM) networks. While maintaining similar performance, they reduce the number of parameters, thereby lowering computational complexity and accelerating training. Similar to LSMs, GRUs are used to extract temporal variation features from time-series stress data, but their structure is more lightweight. GRU networks can be constructed using single-layer or multi-layer stacked structures, or by incorporating residual connections to build deeper GRU networks.

[0090] The bidirectional long short-term memory (LSTM) network consists of two independent LSTM networks, one processing forward sequences and the other processing backward sequences. The outputs from both directions are then merged, allowing the model to utilize both past and future contextual information to understand the data at the current moment. In this application, it can more comprehensively capture the dynamic changes in the temporal stress data of the pen tip spring. For example, during writing, the stress change of the pen tip spring may be affected by both previous pressing and subsequent rebound. A bidirectional LSTM network consisting of two independent LSTM layers can be used, or a multi-layer stacked bidirectional LSTM network can be employed.

[0091] The solution presented in this application achieves effective extraction of spatial and temporal features from a pen tip spring life prediction model through the aforementioned technical means. Specifically, after obtaining temporal stress data by simulating the pen tip spring under writing conditions using a simplified digital twin model, this data is input into a lightweight temporal AI model. The convolutional neural network, acting as a spatial feature extraction module, leverages its local receptive field and weight-sharing mechanism to efficiently identify and extract local stress concentration patterns and spatial distribution characteristics caused by microscopic defects (such as scratches and burrs) from the temporal stress data. This targeted feature extraction method effectively compensates for the shortcomings of traditional digital simulations that neglect the influence of microscopic defects, ensuring accurate capture of stress anomalies caused by defects. Simultaneously, long short-term memory networks, gated recurrent units, or bidirectional long short-term memory networks, acting as temporal feature extraction modules, focus on processing the sequential information of stress data dynamically changing over time. These networks, through their unique gating mechanisms, can effectively memorize and learn the long-term dependence and dynamic evolution of stress changes in the pen tip spring under continuous writing actions (such as light pressure, heavy pressure, and tilted writing), thereby accurately reflecting the fatigue accumulation process. The combination of these two modules is not a simple superposition, but a collaborative design targeting the entire data logic of the pen tip spring, encompassing "key defect parameters - spatial stress concentration - temporal stress fluctuations." The spatial features extracted by the convolutional neural network complement the key defect parameters, revealing the impact of defects on local stress distribution; while the temporal feature extraction module captures the evolution of stress under dynamic operating conditions. Their synergistic effect deeply integrates the three-dimensional features of "defect influence - spatial distribution - temporal change," providing comprehensive and refined input for subsequent lifespan prediction. This design not only ensures the sufficiency of feature extraction but also avoids the high computational costs and technical barriers associated with complex self-developed algorithms by employing a mature and optimized network architecture from the industry. This allows the model to maintain high accuracy while achieving lightweight and high efficiency, thus meeting the real-time requirements of rapid quality inspection in mass production batches of pen tip springs.

[0092] The following is a specific example. As a concrete implementation, after acquiring the temporal stress data of the pen tip spring under writing conditions, this data is first fed into a spatial feature extraction module. This spatial feature extraction module can be specifically a convolutional neural network containing multiple one-dimensional convolutional layers. For example, the first convolutional layer can use 16 convolutional kernels, each with a size of 3 and a stride of 1, and use the ReLU activation function. This is followed by a max-pooling layer with a pooling window size of 2 and a stride of 2, used to reduce the feature dimensionality and enhance the robustness of the features. Multiple such convolutional-pooling structures can be stacked to extract different levels of local stress concentration patterns and stress gradient information from the spatial dimension of the temporal stress data. After processing by the spatial feature extraction module, the spatial dimensionality information of the data is effectively compressed and transformed into a series of representative spatial feature vectors. Subsequently, these spatial feature vectors, along with the original key defect parameters, are input into the temporal feature extraction module. This temporal feature extraction module can be specifically a gated recurrent unit (GRU) network. For example, a single-layer gated recurrent unit (GRU) network containing 128 gated recurrent units can be used. This network receives the output from the spatial feature extraction module and key defect parameters as input, processing these sequential data step-by-step. Through its update and reset gate mechanisms, the GRU network can effectively learn and memorize long-term dependencies and dynamic changes in the stress sequence, thereby capturing the fatigue accumulation process of the pen tip spring throughout the writing cycle. Finally, the output of the GRU network will contain a fusion of the spatial distribution and temporal variation characteristics of the pen tip spring stress data. This information is then fed into subsequent prediction layers to output a lifetime prediction.

[0093] Through the above technical solutions, the specific implementation methods of the spatial feature extraction module and the temporal feature extraction module in the lightweight temporal AI model were clarified, significantly improving the model's ability to process complex stress data. Convolutional neural networks can efficiently capture the local stress concentration patterns and spatial correlations of key parts of the pen tip spring. These patterns are often closely related to microscopic defects, thus ensuring accurate identification of the impact of defects. Meanwhile, long short-term memory networks, gated recurrent units, or bidirectional long short-term memory networks can effectively model the dynamic evolution characteristics and long-term dependencies of stress sequences, accurately reflecting the fatigue accumulation process of the pen tip spring under writing conditions. This combined design enables the lightweight temporal AI model to achieve deep extraction of spatial and temporal features while maintaining its "lightweight" nature, avoiding the problems of insufficient feature extraction or low computational efficiency in traditional methods. By adopting a mature and optimized network architecture from the industry, the complexity of model development and deployment is reduced, and the model's prediction accuracy and running speed are significantly improved, enabling it to meet the stringent real-time requirements of rapid quality inspection in mass production batches of pen tip springs, effectively solving the dilemma of "either low accuracy or high cost" in existing technologies.

[0094] In some of the solutions described above in this application, time-series stress data is input into a lightweight time-series AI model for lifetime prediction. However, in this process, the original time-series stress data has inconsistent continuity and scale, making it difficult for the model to extract features efficiently, which affects training stability and prediction accuracy.

[0095] In this regard, this application further proposes a step of preprocessing the time-series stress data before step S3. The preprocessing includes: cutting the continuous time-series stress data into fixed-length time windows and normalizing the cut data.

[0096] Preprocessing refers to a series of operations performed on raw data before it is input into subsequent processing modules (such as machine learning models) to optimize its quality, format, and characteristics, making it more suitable for the model's requirements. Its role is to eliminate noise and redundant information in the data, standardize data format and scale, thereby reducing the learning difficulty of the model and improving its training efficiency, stability, and predictive accuracy. Besides the segmentation and normalization processes mentioned in this application, preprocessing can also include data smoothing, feature extraction, dimensionality reduction, missing value imputation, or outlier handling, to adapt to different data types and model requirements. Segmenting continuous time-series stress data into fixed-length time windows aims to divide long-term continuous time-series data into multiple short sequence segments of specific lengths. This segmentation method allows each segment to focus on information within a specific time range, facilitating the subsequent model to capture local temporal features. Simultaneously, by generating multiple short sequence segments, the number of training samples is effectively expanded. Implementation methods can include: one is the sliding window method, which slides across continuous data with a fixed step size smaller than the window length to generate overlapping short sequences; another is the non-overlapping window method, which uses the window length as the step size to divide continuous data into non-overlapping short sequences; alternatively, the window length can be set according to the occurrence cycle of specific events in the data to ensure that each window contains one or more complete event cycles. Normalizing the segmented data is the process of converting data of different scales or dimensions to a uniform numerical range. Its main purpose is to eliminate the influence of different dimensions and the problems caused by excessive numerical differences between data, and to avoid some features with larger values ​​masking features with smaller values ​​during model training, thus leading to unstable model training, slow convergence, or prediction bias. Common normalization methods include: one is Min-Max normalization, which linearly scales the data to a specified interval, such as [0, 1] or [-1, 1], and its formula is... Another method is Z-score standardization, which transforms the data into a distribution with a mean of 0 and a standard deviation of 1. The formula is: In addition, Decimal Scaling normalization can be used to scale the data to the range of [-1, 1] by moving the decimal point.

[0097] This application's solution, after acquiring the temporal stress data of the pen tip spring, introduces a preprocessing step before inputting it into a lightweight temporal AI model. This preprocessing step first segments the continuous temporal stress data into fixed-length time windows. The mechanical response of the pen tip spring under writing conditions typically exhibits periodic changes, such as the cycle of pressing and rebounding. By setting a fixed-length time window, this window can precisely cover one or more complete writing action cycles, ensuring that each segmented data fragment has a clear physical meaning and complete temporal characteristics. This segmentation method not only decomposes complex long-sequence data into short sequences that are easy for the model to process but also effectively increases the number of training samples, which is particularly important in scenarios with limited sample size. Based on this, the segmented data is normalized. The numerical range of the pen tip spring's temporal stress data may vary significantly under different writing conditions or when different types of micro-defects are present. Directly inputting this data with varying scales into the lightweight temporal AI model can easily lead to unstable model training; high-value features may mask low-value features, affecting the model's sensitivity to subtle stress changes caused by micro-defects. By normalizing the data, all data are mapped to a uniform numerical range, eliminating scale differences between data points. This allows the model to treat all features more fairly during training, accelerating convergence and improving generalization ability. This preprocessing step is closely integrated with the overall lifetime prediction method. Preliminary step S2 simulates the mechanical response of a pen tip spring under writing conditions using a simplified digital twin model, outputting continuous temporal stress data. After being segmented and normalized in this preprocessing step, this data forms a standardized input with a unified structure and consistent scale, seamlessly adapting to the lightweight temporal AI model in subsequent step S3. This lightweight temporal AI model includes a spatial feature extraction module for extracting stress spatial distribution characteristics and a temporal feature extraction module for extracting stress temporal variation characteristics. The preprocessed data is better analyzed by these modules. For example, the temporal feature extraction module can efficiently capture the internal temporal variation patterns of segmented data of fixed length, while normalization ensures the stability of the spatial feature extraction module when handling different stress amplitudes. This synergy significantly improves the learning efficiency and prediction accuracy of lightweight temporal AI models with a limited number of parameters, providing a reliable data foundation for predicting the lifespan of pen tip springs.

[0098] As a specific implementation method, after obtaining the continuous temporal stress data of the pen tip spring under writing conditions, the following preprocessing can be performed. First, for the continuous temporal stress data, a sliding window method can be used for segmentation. For example, the length of the time window can be set to 150 time points, with a step size of 50 time points. This means that starting from the beginning of the continuous data sequence, the first 150 time points are extracted to form the first data segment; then, 50 time points are slid forward, and another 150 time points are extracted to form the second data segment, and so on, until the end of the continuous data sequence. In this way, a series of temporal stress data segments, each with a length of 150 time points, can be generated. Subsequently, these segmented data segments are normalized. For example, the Min-Max normalization method can be used to uniformly scale all stress values ​​in each data segment to the range of [0, 1]. Specifically, for each data segment, its maximum stress value and minimum stress value are found, and then each stress value in the segment is subtracted from the minimum stress value, and then divided by the difference between the maximum stress value and the minimum stress value. After this processing, the stress values ​​of all data segments will fall within the range of [0, 1], thereby eliminating the differences in numerical scale between the original data.

[0099] Through the above technical solution, this application effectively solves the problem that the strong continuity and inconsistent scale of the original temporal stress data make it difficult for lightweight temporal AI models to efficiently extract features, affecting training stability and prediction accuracy. Specifically, continuous temporal stress data is segmented into fixed-length time windows, so that each data segment has a clear physical meaning. This allows for the accurate capture of the stress fluctuation pattern of the pen tip spring within a single writing cycle, greatly improving the lightweight temporal AI model's ability to capture local temporal features. Simultaneously, this segmentation method effectively expands the number of training samples, reducing reliance on massive amounts of real-world data, and is particularly suitable for the limited sample size of small and medium-sized enterprises. Normalization of the segmented data completely eliminates the differences in stress numerical ranges caused by different writing conditions or defect types, avoiding the problem of high-value features masking low-value features. This enables the lightweight temporal AI model to converge quickly with a limited number of parameters, significantly improving the model's training stability and prediction accuracy. More importantly, this preprocessing step is highly compatible with the time-series data characteristics of the simplified digital twin model output and the architectural characteristics of the lightweight time-series AI model, achieving seamless data flow and efficient processing. This ensures that the prediction results of different batches and different test samples are highly comparable, providing a solid data foundation and consistency guarantee for rapid quality testing of pen tip springs in mass production batches.

[0100] In some of the embodiments described above in this application, a lightweight temporal AI model is proposed to predict the lifespan of a pen tip spring. However, when training this model, it is necessary to obtain actual fatigue life data, which usually relies on time-consuming and expensive bench tests, resulting in inefficient and costly training processes.

[0101] In response, this application further proposes a training method for a lightweight temporal AI model. The method includes: acquiring multiple sets of training samples, each set of training samples including: key defect parameters of a sample pen tip spring, temporal stress data obtained through the simplified digital twin model, and actual fatigue life data of the sample pen tip spring obtained through bench testing; using the key defect parameters and the temporal stress data as inputs, and the actual fatigue life data as the expected output, to perform supervised training on the lightweight temporal AI model.

[0102] Training samples are the datasets used for machine learning model learning and generalization. Obtaining multiple sets of training samples aims to provide the model with sufficient diversity and quantity of data to enable it to identify complex patterns and correlations in pen tip spring life prediction. These samples typically contain input features and corresponding expected outputs (labels), forming the basis for model learning. Key defect parameters are microscopic geometric defect parameters related to the pen tip spring manufacturing process and affecting its fatigue life. These parameters characterize the intrinsic quality and potential failure risk of the pen tip spring. For example, they may include winding gap tolerances, surface scratch depths, end burr sizes, spring wire diameter tolerances, and winding count deviations. These parameters can be obtained using high-precision measuring equipment, such as laser confocal microscopes, high-magnification metallographic microscopes, or high-precision optical profilometers. Time-series stress data refers to the stress response data of key components of the pen tip spring over time under simulated writing conditions. Simplified digital twin models can establish macroscopic structural finite element models based on the CAD model of a pen tip spring, embedding key defect parameters as geometric features or material properties modifications. Boundary conditions and loads simulating writing actions are applied, and transient dynamic simulations are performed, outputting time-series stress data. This data reflects the impact of microscopic defects on macroscopic mechanical behavior. Actual fatigue life data refers to the real service life of the pen tip spring under specific writing conditions, obtained through physical bench testing. Bench testing typically simulates repeated loads experienced by the pen tip spring in actual use until its failure, recording the number of cycles or time. This data serves as the "truth" or "label" for supervised learning, used to calibrate and validate the model's predictive ability. Using key defect parameters and time-series stress data as input to the model means that the model will learn how to extract information from these features characterizing the microstructure and dynamic mechanical response of the pen tip spring. Key defect parameters provide static, structural information, while time-series stress data provides dynamic, condition-dependent mechanical response information. This combined input aims to provide the model with comprehensive and multi-dimensional information to capture various factors affecting fatigue life. Using actual fatigue life data as the desired output is the core of supervised learning. The model aims to learn a mapping function that accurately maps key input defect parameters and time-series stress data to corresponding actual fatigue life values. By comparing the model's predicted output with the desired output, the model can continuously adjust its internal parameters to improve prediction accuracy. Supervised training of lightweight time-series AI models is a machine learning paradigm where the model optimizes its performance by learning from labeled input data (desired output). For lightweight time-series AI models, the supervised training process involves iteratively adjusting the model's weights and biases using optimization algorithms (such as gradient descent) to minimize the error between predicted and actual fatigue life. This training method enables the model to learn the intrinsic patterns in pen tip spring life prediction from historical data.

[0103] This training method addresses the challenges of high data acquisition costs and low efficiency in AI model training by strategically constructing training samples. First, for each sample, the microscopic geometric defect parameters of the pen tip spring are acquired. Simultaneously, a simplified digital twin model, embedded with these defect parameters, is used to simulate the mechanical response of the pen tip spring under writing conditions, thereby obtaining temporal stress data. Crucially, only a small number of pen tip springs undergo physical bench testing to obtain their actual fatigue life data. Subsequently, these key defect parameters, simulated temporal stress data, and actual fatigue life data are integrated into a complete training sample. The defect parameters and simulated temporal stress data serve as input features for a lightweight temporal AI model, while the actual fatigue life data serves as the model's expected output. During supervised training, the AI ​​model continuously optimizes its internal parameters by learning the complex mapping relationship between input features and expected output. By comparing the model's predictions with the actual fatigue life data, the model can gradually improve its prediction accuracy. This data strategy fully leverages the advantages of digital twin models in rapidly generating physically realistic stress data, while calibrating the model with a small amount of actual bench test data ensures the reliability of the prediction results. This balanced approach to data acquisition and training significantly reduces the overall cost and time required to build a high-precision pen tip spring life prediction model.

[0104] As a specific implementation method, for a batch of pen tip springs, such as 1000, 50 samples can be selected for comprehensive bench testing to obtain actual fatigue life data. For these 50 samples, and an additional 200 samples (for which only defect parameters and simulated stress data need to be obtained), the following steps can be performed: First, the surface of each pen tip spring is scanned using a high-precision optical profilometer to measure the surface scratch depth and end burr size, and more detailed 3D morphology data is obtained using a laser confocal microscope to evaluate the winding clearance tolerance. Second, for each spring, a finite element model is built in ANSYS Mechanical software based on its CAD model, and the measured defect parameters (e.g., simulating scratches or burrs by modifying local geometry, or adjusting material properties to reflect tolerance changes) are embedded in the model, with the mesh size set to 0.1 mm. Then, boundary conditions and loads simulating heavy writing conditions are applied. For example, the radial displacement of one end of the spring is constrained and the axial rotational degree of freedom is released. A cyclic load simulating writing force is applied, and a transient dynamic simulation of 500 cycles is performed. The time-series von Mises stress data of key stress concentration areas (such as the spring end or the tightest coil) is output. Simultaneously, the 50 selected pen tip springs are mounted on a dedicated fatigue testing bench to simulate specific writing forces (e.g., 1 N) and frequencies (e.g., 2 Hz) until spring failure (e.g., fracture or significant deformation). The number of cycles before failure is recorded as the actual fatigue life. During the model training phase, the collected defect parameters and time-series stress data are preprocessed. For example, continuous time-series stress data are cut into fixed-length time windows (e.g., each window contains 100 data points) and normalized. Subsequently, the preprocessed defect parameters and time-series stress data are fed into the lightweight time-series AI model. This model employs a convolutional neural network as a spatial feature extraction module to process defect parameters and extract spatial features from temporal stress data, and a long short-term memory network as a temporal feature extraction module to capture the temporal variation characteristics of stress data. Using actual fatigue life data as the desired output, the model is trained under supervision using the Adam optimizer and mean squared error loss function until its performance on the validation set converges.

[0105] Through the above technical solutions, the training method significantly reduces the dependence on time-consuming and expensive physical bench tests, thereby greatly reducing the cost and time of lightweight time-series AI model training. By combining easily obtainable microscopic defect parameters and time-series stress data generated by the digital twin model with a small amount of actual fatigue life data, this method can provide comprehensive and physically realistic training samples for the model while ensuring the accuracy and reliability of the model prediction results. This enables small and medium-sized enterprises with limited resources to efficiently construct and deploy high-precision pen tip spring life prediction models, effectively solving the pain points of low efficiency and high cost of traditional training methods, and providing a solid foundation for the rapid quality inspection of pen tip springs.

[0106] In some of the above solutions of this application, a method for predicting the life of a pen tip spring is proposed to predict the life of a single spring. However, in the quality inspection of mass production batches, if the complete prediction method is executed for each spring, it will consume a large amount of time and resources, unable to meet the requirements of rapid quality inspection, and the cost is high, which is not suitable for small and medium-sized enterprises to apply. In response to this, this application proposes a solution for applying the method for predicting the life of a pen tip spring to the rapid quality inspection of mass production batches of pen tip springs, specifically including: extracting samples from a batch of pen tip springs at a ratio of 3%-10%, and performing steps S1 to S4 on each sample; if the life prediction values of all samples are not lower than the preset life threshold, it is determined that the batch is qualified; if the life prediction value of a sample is lower than the preset life threshold, the main failure cause is located according to the intermediate features output by the lightweight time-series AI model.

[0107] This solution aims to effectively apply the high-precision pen tip spring life prediction technology to the quality control link of mass production. Among them, "the method is applied to the rapid quality inspection of mass production batches of pen tip springs" clarifies the core application scenario of this solution, that is, to perform efficient and economical quality inspection on mass-produced pen tip springs, thereby expanding the laboratory-level prediction ability to the production line and solving the efficiency and cost bottlenecks in the mass production link.

[0108] "Extracting samples from a batch of pen tip springs at a ratio of 3%-10%" is one of the key strategies of this solution, and its role is to represent the quality level of the entire batch through a limited sample size, significantly reducing the detection workload and resource consumption. The selection of this sampling ratio is not arbitrary, but based on a comprehensive consideration of the manufacturing process characteristics of pen tip springs and the high precision of the previous prediction method. For example, statistical methods such as random sampling, systematic sampling (such as extracting one every fixed number), or stratified sampling can be used to ensure the representativeness of the samples.

[0109] "Performing steps S1 to S4 for each sample" means applying the aforementioned pen tip spring life prediction method to each extracted pen tip spring sample in its entirety. This means that each sample will undergo the following processes: acquiring key defect parameters (S1), simulating the mechanical response under writing conditions based on a simplified digital twin model and acquiring time-series stress data (S2), inputting the key defect parameters and time-series stress data into a trained lightweight time-series AI model for processing (S3), and finally obtaining the life prediction value for that sample (S4). This process can be integrated through an automated detection system to ensure the accuracy and consistency of the prediction results.

[0110] The statement "If the predicted lifetime of all samples is not lower than the preset lifetime threshold, then the batch is deemed qualified" provides a clear criterion for batch quality. The preset lifetime threshold can be flexibly set according to product design requirements, customer quality standards, or industry specifications, for example, set to 30,000 or 50,000 writing cycles. By comparing the predicted lifetime of all samples with this threshold, the qualification of the entire batch can be quickly and quantitatively determined, simplifying the quality decision-making process.

[0111] The innovation of this solution lies in its approach of "locating the main cause of failure based on the intermediate features output by the lightweight temporal AI model if the predicted lifetime value of any sample is lower than the preset lifetime threshold." This goes beyond simple pass / fail judgment and achieves precise tracing of the root cause of failure. During training, the lightweight temporal AI model deeply learns the complex correlation between key defect parameters, stress spatial distribution, stress temporal changes, and fatigue life. Therefore, its internal intermediate features (e.g., stress concentration region features extracted by the convolutional neural network layer and stress fluctuation pattern features captured by the long short-term memory network layer) can reflect the specific defect types that lead to reduced lifetime. By analyzing these intermediate features, the dominant factors of batch failure can be inferred, such as excessive winding gap due to precision drift of the winding equipment, surface scratches due to defects in the surface treatment process, or excessively large burr size due to end processing problems.

[0112] The overall operational logic of this solution is as follows: First, by sampling a small proportion of mass-production batches, testing costs and time are effectively controlled. Second, a high-precision and rapid pen tip spring life prediction method (i.e., steps S1 to S4) is applied to each sample to ensure the reliability of the sample life prediction. Because the preceding prediction method is highly accurate and fast (e.g., a single sample prediction can be completed within 1 hour, with costs controlled within 50 yuan), the total time and cost of small-scale sampling (e.g., a batch of several thousand pieces can be tested within only 2 hours) are perfectly suited to the mass production rhythm of small and medium-sized enterprises, avoiding a disconnect between sampling inspection and production schedules. Subsequently, based on a preset life threshold, the prediction results of all samples are quickly judged to arrive at the overall batch's pass / fail conclusion. More importantly, when a batch fails, this solution can utilize the intermediate features within a lightweight temporal AI model to accurately locate the specific defect type and cause of the failure, without requiring additional testing equipment or steps. This closed-loop management model of "detection-judgment-optimization" enables the production line to adjust process parameters in a timely manner, fundamentally solve quality problems, and avoid the recurrence of similar defects.

[0113] For example, a certain pen tip spring manufacturer produces 10,000 pen tip springs every day. To achieve rapid quality inspection, the manufacturer can set a sampling ratio of 5%, that is, 500 springs are sampled for inspection every day. These samples can be selected at fixed intervals by an automated robotic arm on the production line. For each sampled spring, its surface and end are first automatically scanned by a high-precision optical profiler to obtain key defect parameters such as the depth of surface scratches, the size of end burrs, and the winding gap tolerance (S1). Subsequently, these parameters are input into a pre-established simplified digital twin model (for example, a finite element model built based on ANSYS), which simulates the stress condition of the pen tip under the condition of light pressure writing and outputs the time-series stress data of the key parts of the spring (S2). Then, the key defect parameters and the time-series stress data are fed into a trained lightweight time-series AI model (for example, a model containing a convolutional neural network and a long short-term memory network), which quickly processes the data and outputs the predicted life value of each spring (S3, S4). Suppose the preset life threshold is 40,000 writing times. If the predicted life of all 500 samples is higher than 40,000 times, this batch of springs is judged to be qualified. However, if the predicted life of 15 of these samples is lower than 30,000 times, the system will further analyze the intermediate features output by the lightweight time-series AI model of these unqualified samples. For example, the model may indicate that the convolutional neural network feature maps of these samples show an abnormally high stress concentration pattern in the end area of the spring, and this pattern is highly correlated with the larger end burr size defect parameter. Based on this, the quality engineer can quickly locate the main reason for the batch unqualified as a problem in the deburring process link and guide the production department to check and adjust the deburring equipment, thus effectively avoiding similar defects in subsequent batches.

[0114] Through the above technical solution, this application effectively solves the problems existing in the mass production quality inspection of traditional pen tip springs, such as long cycle, high cost, and inability to trace the root cause of failure. This solution realizes rapid and low-cost batch quality inspection, significantly shortens the inspection cycle, reduces the inspection cost, and makes it fully adapt to the mass production needs of small and medium-sized enterprises. At the same time, relying on the analysis ability of the lightweight time-series AI model for intermediate features, this solution can accurately locate the main reason for batch failure, upgrade the traditional "qualified / unqualified" judgment to a closed-loop management of "inspection - judgment - optimization", provide a clear direction for the improvement of production processes, and thus fundamentally improve product quality and production efficiency, reducing after-sales repair and scrap losses. This design that deeply integrates high-precision prediction methods with the mass production quality inspection scenario breaks through the limitations of the disconnection between detection and optimization and the imbalance between accuracy and efficiency in the existing technology, constructs a complete quality control link, and has significant practical value and innovation.

[0115] Traditional pen tip spring life prediction technologies suffer from long testing cycles, high costs, and low prediction accuracy in practical applications. Specifically, bench testing requires simulating tens of thousands of writing cycles, with a cycle of 7-15 days, which cannot meet the rapid quality inspection needs of mass production batches, and the cost per test is high. Traditional digital simulation only focuses on the macroscopic structure, ignoring the impact of microscopic defects generated during the production process on life, resulting in prediction errors exceeding 40%. In addition, existing high-precision prediction solutions rely on high-end micro-nano detection equipment (such as focused ion beam scanners) and complex self-developed AI architectures, with equipment procurement and maintenance costs exceeding 5 million yuan. The high technical threshold makes it difficult for small and medium-sized enterprises to afford, leading to a common dilemma in the industry: "either low accuracy or high cost."

[0116] Example 2

[0117] To address this issue, this application proposes a computing device including a processor and a memory. The memory stores a computer program, which, when executed by the processor, implements a method for predicting the lifespan of pen tip springs. The core innovation of this embodiment lies in its collaborative approach of combining low-cost acquisition of key defect parameters with defect-embedded simulation of simplified digital twin models, and introducing a lightweight temporal AI model that fuses spatial and temporal features. This effectively solves the problems of long testing cycles, high costs, and low prediction accuracy in pen tip spring lifespan prediction without requiring high-end micro-nano inspection equipment and complex AI architectures, achieving the core advantage of enabling small and medium-sized enterprises to apply efficient prediction solutions.

[0118] For ease of understanding, the following explains some key terms in this embodiment: Key defect parameters refer to microscopic geometric defect parameters related to the pen tip spring manufacturing process and affecting its fatigue life, such as surface scratches, uneven winding gaps, or end burrs; Simplified digital twin model refers to a computational model that abstracts and simplifies the key features of the physical pen tip spring to efficiently simulate its mechanical behavior, especially when considering the influence of microscopic defects; Writing conditions refer to the various mechanical loads and boundary conditions experienced by the pen tip spring during actual writing; Temporal stress data refers to the sequence data of stress values ​​at key parts of the pen tip spring changing over time under simulated writing conditions; Lightweight temporal AI model refers to a trained artificial intelligence model designed to have low computational resource requirements while ensuring prediction accuracy, capable of processing temporal data and extracting spatial and temporal features; Spatial feature extraction module is a component of the lightweight temporal AI model used to analyze the spatial distribution characteristics of stress; Temporal feature extraction module is a component of the lightweight temporal AI model used to analyze the temporal variation characteristics of stress; Life prediction value refers to the quantitative estimate of the expected fatigue life of the pen tip spring obtained through the above methods.

[0119] The workflow of this computing device includes the following steps: First, acquiring the key defect parameters of the pen tip spring under test. This can be achieved through manual visual inspection combined with standard measuring tools. For example, operators use a magnifying glass or low-power microscope to observe the spring surface, identify possible scratches, dents, or burrs, and use calipers or micrometers to measure the spring's winding gap, wire diameter, etc., at multiple points, recording the deviation values. Alternatively, local images of the spring can be captured and manually analyzed to quantify the size and location of microscopic defects. Second, based on the acquired key defect parameters, a simplified digital twin model is used to simulate the mechanical response of the pen tip spring under writing conditions, thereby obtaining time-series stress data. Specifically, a macroscopic geometric model of the pen tip spring is constructed using general-purpose finite element analysis software. The key defect parameters are embedded into the model through geometric feature corrections, such as creating tiny grooves at the corresponding scratch locations on the model surface or adjusting local winding geometry parameters. Subsequently, boundary conditions and loads simulating writing actions, such as periodic axial compressive force and radial constraints, are applied to perform transient dynamic simulations and output stress time-series data for key components. Subsequently, key defect parameters and time-series stress data are input into a trained lightweight time-series AI model. This model analyzes the spatial distribution characteristics of stress (such as the shape and location of stress concentration areas) through a spatial feature extraction module and analyzes the temporal variation characteristics of stress (such as the frequency and amplitude of stress fluctuations) through a time-series feature extraction module, achieving deep fusion of multi-source data. Finally, based on the output of the lightweight time-series AI model, the predicted lifespan of the pen tip spring is obtained, expressed in terms of cycle count or usage time, used to quantitatively evaluate the reliability and durability of the spring.

[0120] The following is a concrete example: Suppose a pen tip spring manufacturer needs to conduct a rapid life assessment of a mass-produced batch. Technicians randomly select samples from the batch and scan the spring surface using a standard optical microscope equipped with a high-resolution camera. They manually identify and measure surface scratches with a depth of 3 micrometers and a maximum winding gap deviation of 0.015 mm as key defect parameters. Next, a macroscopic geometric model is built in general-purpose finite element analysis software, embedding the aforementioned defect parameters (such as creating a 3-micrometer-deep micro-groove and adjusting the winding gap). Periodic axial loads of 0.5N to 1.5N are applied to perform transient dynamic simulations, obtaining time-series stress data for key stress-bearing areas. Subsequently, the key defect parameters and time-series stress data are input into a lightweight time-series AI model deployed on an industrial computer. The model's spatial feature extraction module identifies local stress concentration areas at the scratches, and the time-series feature extraction module captures the cumulative effect of stress fluctuations, ultimately outputting a life prediction value of 180,000 cycles. By performing the above process on multiple samples, the manufacturer can complete the batch quality assessment within hours.

[0121] Through the above technical solutions, this application achieves rapid, low-cost, and high-precision prediction of pen tip spring lifespan. Compared to traditional bench testing, the prediction cycle is shortened from 7-15 days to several hours, significantly improving the quality inspection efficiency of mass production batches. By explicitly acquiring and utilizing key defect parameters, combined with defect-embedded simulation of a simplified digital twin model, the problem of large prediction errors caused by neglecting microscopic defects in traditional digital simulation is effectively solved, resulting in a substantial improvement in prediction accuracy. Simultaneously, the use of a lightweight temporal AI model to replace high-end micro-nano testing equipment and complex AI architecture keeps equipment costs within the affordability range of small and medium-sized enterprises, thus breaking through the industry's technical bottleneck of "either low accuracy or high cost." Overall, this application, through low-cost acquisition of key defect parameters, construction of a simplified defect-embedded digital twin model, and a lightweight AI model integrating spatial-temporal features, forms a collaborative and mutually reinforcing technical system, providing a technically advanced and substantial means for the quality control and product development of pen tip springs.

[0122] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting the lifespan of a pen tip spring, characterized in that, Includes the following steps: S1, Obtain at least one key defect parameter of the pen tip spring to be tested. The key defect parameter is a microscopic geometric defect parameter related to the manufacturing process of the pen tip spring and affecting its fatigue life. S2, Based on the key defect parameter, simulate the mechanical response of the pen tip spring under writing conditions using a simplified digital twin model to obtain its time-series stress data. S3, Input the key defect parameter and the time-series stress data into a trained lightweight time-series AI model. The lightweight time-series AI model includes a spatial feature extraction module for extracting stress spatial distribution features and a time-series feature extraction module for extracting stress time-series change features. S4. Based on the output of the lightweight time-series AI model, the predicted lifespan of the pen tip spring is obtained.

2. The method according to claim 1, characterized in that, The key defect parameters include any one or more of the following: winding gap tolerance, surface scratch depth, end burr size, spring wire diameter tolerance, and winding turn deviation.

3. The method according to claim 1 or 2, characterized in that, In step S1, the key defect parameters are obtained by using a laser confocal microscope, a high-magnification metallurgical microscope, or a high-precision optical profilometer.

4. The method according to claim 1, characterized in that, In step S2, the construction and simulation of the simplified digital twin model includes: S21, establishing a macroscopic structural finite element model based on the CAD model of the pen tip spring, with a mesh size of 0.05mm-0.2mm; S22, embedding the key defect parameters into the macroscopic structural finite element model by modifying geometric features or material properties; S23, applying boundary conditions and loads to the model to simulate writing actions, wherein the boundary conditions include releasing the axial rotational degree of freedom of the spring end and constraining its radial displacement, and the loads include simulating at least one of light pressure writing, heavy pressure writing, or tilted writing; S24, performing transient dynamic simulation and outputting the time-series stress data of the key parts of the pen tip spring.

5. The method according to claim 4, characterized in that, The tools for building the simplified digital twin model are ANSYS, ABAQUS, Code_Aster, or HyperMesh.

6. The method according to claim 1, characterized in that, The spatial feature extraction module is a convolutional neural network, and the temporal feature extraction module is a long short-term memory network, a gated recurrent unit, or a bidirectional long short-term memory network.

7. The method according to claim 1, characterized in that, Before step S3, the process further includes a preprocessing step for the time-series stress data. The preprocessing includes: cutting the continuous time-series stress data into fixed-length time windows and normalizing the cut data.

8. The method according to claim 1, characterized in that, The training method of the lightweight temporal AI model includes: acquiring multiple sets of training samples, each set of training samples including: key defect parameters of a sample pen tip spring, temporal stress data obtained through the simplified digital twin model, and actual fatigue life data of the sample pen tip spring obtained through bench testing; using the key defect parameters and temporal stress data as inputs and the actual fatigue life data as the expected output, the lightweight temporal AI model is trained under supervision.

9. The method according to claim 1, characterized in that, The method is applied to rapid quality inspection of mass-produced pen tip springs, specifically including: sampling 3%-10% of a batch of pen tip springs, and performing steps S1 to S4 on each sample; if the predicted lifespan of all samples is not lower than a preset lifespan threshold, the batch is deemed qualified; if the predicted lifespan of any sample is lower than the preset lifespan threshold, the main cause of failure is located based on the intermediate features output by the lightweight time-series AI model.

10. A computing device, comprising: processor; A memory storing a computer program; when the computer program is executed by the processor, it implements the method for predicting the life of a pen tip spring as described in any one of claims 1 to 9.

Images