Method of detecting and manufacturing a balance spring for a timepiece

Abstract

The invention relates to a method for detecting a hairspring or a hairspring blank arranged as a hairspring, the hairspring having to have at least one preset oscillation frequency, the method comprising the following steps: a. applying a time-varying vibration excitation to the hairspring or the hairspring blank to cover a preset frequency range; b. determining at least one characteristic of the oscillation frequency of the hairspring or the hairspring blank, such as a resonance peak, during the vibration excitation in the preset frequency range; c. providing the oscillation frequency characteristic determined in step b to a prediction machine to determine the stiffness of the hairspring or the hairspring blank and / or to determine whether a dimensional correction of the hairspring or the hairspring blank is required to achieve the preset oscillation frequency.

Description

Technical Field

[0001] This invention relates to the field of testing and manufacturing of watch components, and specifically to a method for testing and manufacturing a watch hairspring (also known as an oscillator). Background Technology

[0002] Mechanical watch movements are regulated by a mechanical regulator consisting of an oscillator, a flexible, deformable component whose oscillations determine the watch's rate. For example, many watch regulators include a hairspring acting as an oscillator, mounted on the balance wheel's axis and oscillating via the escapement. The natural frequency of the balance wheel and hairspring pair allows the watch to be regulated, depending particularly on the stiffness of the balance wheel and hairspring.

[0003] In fact, the frequency f of the regulating mechanism, which is formed by coupling a hairspring with stiffness R and a balance wheel with inertia I, is given by the formula:

[0004] [Equation 1]

[0005]

[0006] The stiffness of the hairspring also defines its inherent vibrational characteristics, such as its natural frequency and oscillation frequency. In current applications, the natural frequency of an elastic system (a single oscillator or an oscillator-balanced pair) is the oscillation frequency of the system when it is freely evolving, i.e., without any excitation force. Furthermore, the oscillation frequency of an elastic system is a frequency at which the local maximum displacement amplitude at a given point of the elastic system can be measured. In other words, if the elastic system is excited by a variable-frequency excitation source over a period of time, then before this oscillation frequency, the displacement amplitude exhibits an upward slope and afterwards a downward slope at any point not corresponding to a vibration node. Typically, in such measurements, the displacement amplitude is recorded as a function of the excitation frequency, showing a displacement amplitude peak or resonance peak that is related to or has the characteristics of the oscillation frequency.

[0007] The stiffness of a hairspring oscillator typically depends on the material properties and its dimensions, particularly the thickness (i.e., width) of the helical coil along the outer stud. More specifically, the stiffness is given by the following formula:

[0008] [Equation 2]

[0009]

[0010] In the formula:

[0011] φ: Spring torsion angle;

[0012] M: Spiral spring return torque;

[0013] In the formula, for a constant cross-section external pile made of a specific material, M is calculated as follows:

[0014] [Equation 3]

[0015]

[0016] In the formula:

[0017] E Young's modulus of the material used for the external piles;

[0018] L : Length of the outer pile;

[0019] h Height of the outer pile;

[0020] e: Thickness or width of the outer pile.

[0021] The natural frequency of the regulating element, formed by coupling a hairspring with stiffness R and a balance wheel with inertia I, is in particular proportional to the square root of the hairspring's stiffness. The primary specification of the hairspring is its stiffness, which must be within a defined range to properly match the balance wheel, which constitutes the oscillator's inertial element. This pairing is crucial for accurately setting the frequency of the mechanical oscillator.

[0022] It is crucial that the oscillator's characteristics be as stable as possible, so that the watch's movement will also be stable. In the modern environment, the importance of magnetic fields has led watchmakers to use silicon hairsprings for some years, as they are less sensitive to magnetic interference than metal hairsprings.

[0023] It is now possible to fabricate hundreds of silicon filaments on a single wafer using microfabrication techniques. In particular, it is known to produce multiple silicon oscillators with very high precision within a silicon wafer using photolithography and machining / etching methods. The methods for producing these mechanical oscillators typically use monocrystalline silicon wafers, but silicon wafers made of other materials can also be used, such as polycrystalline or amorphous silicon, other semiconductor materials, glass, ceramics, carbon, carbon nanotubes, or composites comprising these materials. Monocrystalline silicon belongs to the m³m cubic crystal class and has an isotropic coefficient of thermal expansion (α).

[0024] Silicon's first thermoelastic modulus is a very negative value; therefore, the stiffness of a silicon oscillator, and its natural frequency, varies greatly with temperature. To at least partially compensate for this drawback, documents EP1422436, EP2215531, and WO2016128694 describe a hairspring-type mechanical oscillator made of a single-crystal silicon core (or two cores as described in WO2016128694), whose temperature variation in Young's modulus is compensated for by a layer of amorphous silicon oxide (SiO2) surrounding the core (or core), one of the few materials with a positive thermoelastic modulus.

[0025] When the hairsprings are fabricated collectively on a wafer from silicon or other materials, the final functional efficiency will be obtained by dividing the number of hairsprings whose stiffness corresponds to the mating gap by the total number of hairsprings on the wafer.

[0026] However, the microfabrication steps used in fabricating hairsprings on wafers, particularly the etching step, often result in significant geometric dispersion between hairspring dimensions on the same wafer. Therefore, even though the etched pattern is identical for each hairspring, there is a noticeable dispersion in their stiffness. The measured stiffness dispersion typically follows a Gaussian distribution. To optimize manufacturing yield, we are interested in centering the average of this Gaussian distribution onto a nominal stiffness value while reducing the standard deviation of this Gaussian distribution.

[0027] Furthermore, the stiffness dispersion is even greater between the hairsprings of two wafers etched at different times using the same method. This is shown in Figure 1 The figure shows the stiffness dispersion curves Rd1, Rd2, and Rd3 of the hairsprings on three different wafers. Generally, for each plate, the distribution of stiffness R (relative to the number of hairsprings N with that stiffness) follows a normal or Gaussian distribution, and each dispersion curve is centered on its respective average values ​​Rm1, Rm2, and Rm3.

[0028] Documents WO2015113973 and EP3181938 propose to remedy this problem by forming a hairspring with a size larger than that required to obtain a hairspring with a preset stiffness, measuring the stiffness of the formed hairspring by coupling it to a balance wheel with a preset inertia, calculating the material thickness that needs to be removed to obtain the required size of the hairspring with the preset stiffness, and removing this thickness from the hairspring. Similarly, EP3181939 proposes to remedy the same problem by forming a hairspring with a size smaller than that required to obtain a hairspring with a preset stiffness, determining the stiffness of the formed hairspring by coupling it to a balance wheel with a preset inertia, calculating the material thickness that needs to be added to obtain the required size of the hairspring with the preset stiffness, and adding this material thickness to the hairspring.

[0029] In this way, such as Figure 2 As shown, although there is an average stiffness R on a given wafer 平均 1. R 平均 Grade 2, but relative to the nominal stiffness R 公称 In this regard, stiffness dispersion curves such as Rd1 and Rd2 can be recent.

[0030] This method requires a high degree of accuracy when measuring the frequency of the hairspring to determine its stiffness. In particular, measurement errors can be caused by the preset inertia of the balance wheel or the components used. A step must then be taken to calculate the thickness to be removed in order to remove the calculated thickness with high accuracy again. Furthermore, it should be noted that coupling the hairspring to the balance wheel with its preset inertia requires meticulous operation and a significant amount of preparation time. Finally, it should also be noted that any assembly operations performed on components or blanks still existing on the wafer greatly increase the risk of contamination (e.g., the presence of fine silicon particles (fragments) generated during processing).

[0031] The purpose of this invention is to provide a method that does not have the aforementioned drawbacks, allowing for faster production processes and / or less risk of contamination, and / or more sampling, and / or more accurate measurements, thereby enabling more personalized adjustments to wafer springs. Summary of the Invention

[0032] This invention specifically relates to a method for detecting a hairspring or a hairspring blank arranged as a hairspring, wherein the hairspring must have at least one preset oscillation frequency, and the method includes the following steps:

[0033] a. Apply time-varying vibration excitation to the hairspring or hairspring blank to cover a preset frequency range;

[0034] b. During vibration excitation within a preset frequency range, determine at least one characteristic of the oscillation frequency of the hairspring or hairspring blank, such as a resonance peak;

[0035] c. Provide the oscillation frequency characteristics determined in step b to the predictor to determine the stiffness of the hairspring or hairspring blank and / or to determine whether the hairspring or hairspring blank needs to be dimensionally modified to achieve the preset oscillation frequency.

[0036] According to the above embodiments, the method includes the steps of vibrating the hairspring or hairspring blank and measuring the characteristics of the oscillation frequency, and then inferring the stiffness and / or whether dimensional correction is needed by prediction. No assembly with the balance wheel or other components is required, which saves time. Furthermore, the measurement is performed only on the hairspring or blank, which limits errors caused by other components or their assembly, as well as any contamination. Measurement accuracy is improved due to fewer sources of variation caused by other components or contamination. In other words, the hairspring or hairspring blank is tested individually. Vibration excitation is applied to a single component or blank, not associated with any balance, weights, or oscillation system. This method can test a free, single part (i.e., at least one free end, not connected to any mechanism or balance wheel), which brings advantages such as increased productivity (no need for assembly with an oscillation system), improved quality (no part contamination, no breakage, more parts can be tested within the same budget), and improved accuracy (no errors associated with other components of the oscillation system).

[0037] In one embodiment, vibrational excitation is applied to a hairspring or hairspring blank, with its free end (typically a central inner stud) and the other end fixed to an insert or clip. From a mechanical perspective, the vibrational excitation can be considered as being applied via a spring (the elastic portion of the hairspring) to a mass (located at the hairspring's center of gravity) and connected to a reference frame (a clamp for the individual hairspring, or, for example, the base or remaining portion of a blank made of silicon). The vibrational excitation causes the suspended mass to move.

[0038] It should also be noted that if it is determined that dimensional correction is required for the component under test (or all individual components connected to the same wafer, or even a single component connected to a region of the wafer that may or may not include the component under test), the correction can be made for the individual component without disassembling anything again (e.g., the silicon component can be directly oxidized at the end of the test). Therefore, it is possible to add or remove material from the unit component to change its inherent stiffness. In other words, dimensional correction of the unit component is achieved by changing its dimensions (typically the width and / or thickness of the outer stud that constitutes the elastic part of the hairspring).

[0039] Therefore, the process described above makes it possible to inspect the hairspring blank during the manufacturing process while limiting the risk of contamination or assembly errors. Dimensional corrections (section, height, and / or thickness) can then be made. This process can also be used to inspect finished hairsprings, for example, by classifying them according to stiffness increments to plan pairing with specific balances.

[0040] Of course, the frequency range of the obtained spectrum depends not only on the vibration excitation source but also on the sensor of the measuring instrument used. Therefore, the frequency range is related to both the excitation frequency range and the sensitive frequency range of the instrument (vibration meter or other) used to measure the oscillation amplitude. However, the selection of the excitation frequency range will include at least one oscillation frequency of the hairspring or billet being tested.

[0041] The preset oscillation frequency that the finished hairspring is to exhibit can be the target natural frequency, the target oscillation frequency, the target natural frequency range, or the target oscillation frequency range defined by the allowable error around the target value.

[0042] The dimensional corrections predicted by the predictor can typically be corrections to the cross-section of the flexible outer stud that forms the hairspring or hairspring blank, i.e., corrections to the height or thickness or both.

[0043] In the above method, the characteristic of the oscillation frequency is the characteristic of the oscillation response measured within a preset frequency range, including at least one oscillation frequency. This characteristic is typically determined after processing the original measurement signal (e.g., measuring the displacement amplitude, velocity, or acceleration of certain points on the hairspring or hairspring blank), a process that can include, for example, Fourier transform to determine the resonance peak, thereby determining the oscillation frequency.

[0044] It should be noted that this process can determine a stiffness in order to classify parts and / or calculate / infer the level of dimensional correction to be applied to obtain the target stiffness. However, it is also possible to consider only the determined oscillation frequency in order to directly calculate / derive the level of dimensional correction to be applied to obtain the target stiffness.

[0045] In one embodiment, in step a, the frequency range is applied simultaneously to multiple hairsprings or hairspring blanks. Speed ​​is improved because the vibrational excitation can typically be applied to a wafer supporting hundreds of hairspring blanks, which are still attached to the wafer.

[0046] In one embodiment, the frequency range is pre-set to cover at least one frequency range:

[0047] - The frequency range is centered on a preset oscillation frequency;

[0048] - At least 30% of the preset oscillation frequency, i.e. ±15% of the preset oscillation frequency. For example, if the preset oscillation frequency is 1kHz, then the frequency range will be from 850Hz to 1150Hz.

[0049] In one embodiment, the hairspring has at least two preset oscillation frequencies, and the frequency range is preset to cover at least two preset oscillation frequencies. By covering or sweeping a wide frequency range, several resonance peaks (or oscillation frequencies) can be measured, which can provide better accuracy.

[0050] In one embodiment, step a includes using a source, such as a piezoelectric power source, to induce or apply acoustic excitation on the edge of the wafer supporting the hairspring blank, or preferably on or under the hairspring or hairspring blank to be specifically excited.

[0051] In one embodiment, the sound source may be coupled to an excitation cone that selectively excites at least one hairspring or one hairspring blank. More preferably, if a wafer supports several hairspring blanks, the sound source may be connected to an excitation cone that selectively excites at least some, preferably all, of the hairspring blanks.

[0052] In one embodiment, the sound source can be selected and / or adjusted to generate time-varying vibration excitation covering a preset frequency range:

[0053] - The amplitude is sufficient to cause the hairspring or hairspring blank to vibrate with sufficient amplitude so that it can be detected by a device used to measure the displacement amplitude, velocity, or acceleration of at least one point of the hairspring or hairspring blank, and / or

[0054] - Over a sufficiently long period of time, the vibration spectrum of the hairspring or hairspring blank is deduced.

[0055] In one embodiment, step b includes using optical measurement means, such as a laser vibrometer using the Doppler effect.

[0056] In one embodiment, step b is based on time-varying measurements of the displacement amplitude, velocity, or acceleration of at least one point of the hairspring or hairspring blank, and the measurements are preferably performed at least partially in step a.

[0057] In one embodiment, step b includes:

[0058] - Sub-step: Determine the oscillation frequency of the hairspring or hairspring blank based on the running or modal deformation at at least one point of the hairspring or hairspring blank. The running or modal deformation is typically defined by the displacement amplitude or velocity or acceleration and the direction of oscillation (outside or in a specific plane), as a function of the excitation frequency.

[0059] In one embodiment, the hairspring or hairspring blank is disposed on a base surface, and step b includes the following sub-steps:

[0060] - Sub-step b': In the normal direction of the base plane, measure the displacement amplitude, velocity or acceleration of at least one point of the hairspring or hairspring blank;

[0061] - and / or, sub-step b'': measuring the displacement amplitude, velocity or acceleration of at least one point of the hairspring or hairspring blank in the direction contained by the base plane.

[0062] Measuring displacement or velocity in several directions makes it easier to determine resonance peaks and oscillation frequencies.

[0063] In one embodiment:

[0064] - For the first preset oscillation frequency, only step b' is performed, which involves measuring the displacement or velocity of at least one point of the hairspring or hairspring blank in the direction normal to the base plane.

[0065] And / or

[0066] - For the second preset oscillation frequency, only step b'' is performed to measure the displacement or velocity of at least one point of the hairspring or hairspring blank in the direction within the base plane.

[0067] Depending on the oscillation frequency, measurements can be taken in one or another direction to measure the maximum possible displacement or velocity, thereby minimizing measurement error. The vibration mode (typically the direction of vibration) in response to vibrational excitation may vary depending on the geometry of the hairspring or hairspring blank.

[0068] In one embodiment, step b includes processing the measurement signal with, for example, a Fourier transform to determine the resonant peaks of displacement amplitude or velocity or acceleration and / or phase as a function of the excitation frequency.

[0069] In one embodiment, step b includes:

[0070] - Sub-step: Determine the resonance peak of the hairspring or hairspring blank based on the displacement amplitude or velocity of at least one point of the hairspring or hairspring blank.

[0071] In one embodiment, the oscillation frequency is determined based on the width of the resonance or amplitude peak, which is half the maximum value of the amplitude resonance peak. This approach makes it possible to limit calculation errors arising solely from determining the frequency position of the peak defined by its maximum value.

[0072] In one embodiment, step c includes calculating the stiffness of the hairspring or hairspring blank. Calculating the stiffness can more accurately determine whether a dimensional correction is needed and by what value. Furthermore, it can pre-determine the dimensions or select a balance wheel to couple with after the hairspring is manufactured.

[0073] In one embodiment, if dimensional correction is required, the method includes the step of:

[0074] d. Using a predictor and based on the oscillation frequency characteristics determined in step b, calculate the dimensional corrections to be implemented, such as corrections to the cross-section, height, and / or thickness.

[0075] In one embodiment, the predictor uses a polynomial formula to predict whether a size correction is needed. For example, a linear regression model can be used.

[0076] In one embodiment, the predictor performs classification, for example by a neural network, to predict whether size correction is needed.

[0077] In one embodiment, the predictor performs classification based on k-means or k-medians partitions to predict whether size correction is needed.

[0078] In one embodiment, a hairspring blank is formed on a wafer, the wafer including a plurality of hairspring blanks distributed in a plurality of fan-shaped regions of the wafer;

[0079] Step b includes a sub-step, namely: determining at least one characteristic of the oscillation frequency of at least one hairspring blank in each sector region;

[0080] Step c includes determining the stiffness of the hairspring blank and / or determining whether the hairspring blank for each sector region requires dimensional correction. By refining the analysis of each sector of the wafer, the accuracy of dimensional corrections (section, height, and / or thickness) is improved.

[0081] In one embodiment, the detection process includes a step of using a predictor to calculate the dimensional correction of the hairspring blank to be applied to each sector region.

[0082] In one embodiment, step a includes modifying the direction of vibration excitation over time, preferably in the direction pointing towards the hairspring or hairspring blank, the oscillation frequency characteristics of which are determined in step b.

[0083] In one embodiment, the detection method includes a preliminary step of taking into account the material of the hairspring or hairspring blank and adjusting the maximum amplitude of the vibration excitation and / or the frequency range of a preset frequency range according to the material of the hairspring or hairspring blank.

[0084] In one embodiment, the obtained frequency range covers a frequency range from 0 Hz to 100 kHz, preferably from 0 Hz to 50 kHz, more preferably from 0 Hz to 40 kHz, and most preferably from 10 kHz to 35 kHz. The applicant has found that the prediction accuracy is better for peak or oscillation frequencies located in the higher frequency range. Indeed, the effect of stiffness on oscillation frequency is stronger in the higher frequency range (e.g., between 10 kHz and 35 kHz), so both sensitivity and accuracy are better within this specific range.

[0085] In one embodiment, steps a and b are repeated at least several times for the same measurement point of the hairspring or hairspring blank.

[0086] In one embodiment, steps a and b are performed simultaneously. This synchronization makes it possible to detect phase shifts, or attenuation, or coupling, which, if taken into account, can improve the accuracy of predictions or make it possible to adjust or recalibrate the vibration excitation source.

[0087] A second aspect of the present invention relates to a method for manufacturing a hairspring having at least one preset oscillation frequency, comprising the following steps:

[0088] • Form at least one hairspring or hairspring blank with the dimensions (including the preset allowable error) required to obtain the preset oscillation frequency;

[0089] • The detection method according to the first aspect of the present invention is used to detect the hairspring or hairspring blank.

[0090] In one embodiment, the manufacturing method includes a step comprising:

[0091] • Based on the calculation in step d, at least one dimension of the hairspring blank formed in step a is corrected to obtain a hairspring with a preset oscillation frequency.

[0092] These dimensions (cross-section, height, and / or thickness) can be corrected by removing or adding material.

[0093] In one embodiment, the hairspring or hairspring blank is made of silicon, glass, ceramic, metal, or carbon nanotubes. In particular, conventional metal hairsprings (steel, nickel-phosphorus alloys, etc.) can also be detected. In this case, the metal hairspring is held in place by a tool or used as a reference, which positions it opposite the emission source and the displacement measuring device.

[0094] In one embodiment, the hairspring blank is formed on a wafer together with a plurality of other hairspring blanks.

[0095] A third aspect of the present invention relates to a predictive machine learning method for implementing step c of the detection method described in the first aspect of the present invention, comprising the following steps:

[0096] (1) Forming a hairspring or a hairspring blank;

[0097] (2) Apply time-varying vibration excitation to each hairspring or each hairspring blank to cover a preset frequency range;

[0098] (3) During the vibration excitation process within the preset frequency range, determine at least one characteristic of the oscillation frequency of each hairspring or each hairspring blank;

[0099] (4') Install multiple hairsprings or hairspring blanks in an oscillation mechanism with a preset inertia in order to measure the free oscillation frequency or stiffness of each hairspring or each hairspring blank;

[0100] And / or

[0101] (4'') Using simulation tools, model multiple hairsprings or hairspring blanks in an oscillating mechanism with preset inertia in order to calculate the free oscillation frequency or stiffness of each hairspring or each hairspring blank;

[0102] (5) For each hairspring or each hairspring blank, provide the following to the predictor:

[0103] - Characteristics of the oscillation frequency determined in step (3);

[0104] - The free oscillation frequency or stiffness measured in step (4'') and / or step (4'').

[0105] It is preferable to select a measuring instrument that is sufficiently sensitive within the selected frequency range, and to ensure that the vibrational behavior of the hairspring can be utilized within that selected frequency range.

[0106] In one embodiment, step (3) includes a preliminary stage, namely: determining the reference measurement point, and:

[0107] - Measure the displacement or displacement velocity of multiple preset points on the hairspring or hairspring blank;

[0108] - Determine a node among multiple preset points, wherein the displacement amplitude of the node at at least one oscillation frequency or peak value is zero or less than the peak value of a first threshold;

[0109] - Among multiple preset points, a reference point to be measured during detection is selected. The reference point is different from the determined node, and preferably, the peak displacement amplitude of each reference point is greater than the peak value of the second threshold.

[0110] Such a reference point determination process makes it possible to eliminate points or regions of nodes (i.e., fixed points) at one or more oscillation frequencies.

[0111] In one embodiment, the hairspring shape or hairspring blank has a radius Ra defined between a free center end and an embedded peripheral end, and at least two reference points are selected, preferably four reference points, located at:

[0112] - In the first zone, less than 0.20 × Ra, or

[0113] - In the second zone, between 0.05 × Ra and 0.30 × Ra, or

[0114] - The third zone is between 0.35 × Ra and 0.65 × Ra, or

[0115] - The fourth zone is between 0.65×Ra and 0.85×Ra.

[0116] The selection of these regions ensures that the points whose displacement is being tracked have sufficient displacement amplitude for correct and accurate measurement.

[0117] Alternatively, displacement / motion / vibration can be measured at a specific point, preferably in a large and / or undeformable area of ​​the component (or with minimal deformation associated with the oscillation of the hairspring). In particular, the measurement can be targeted at a point on the hairspring stud or the hairspring blank. On the one hand, the stud has a relatively large dimension relative to bends (a bend is typically 20µm to 40µm wide, and the stud can be at least 110µm in size), making it easier to aim the measuring tool. On the other hand, the stud can be considered undeformable during vibration excitation, and all points on the stud exhibit similar displacement / motion / vibration. Therefore, it is easier to lock the measurement point (e.g., 4µm in size for a laser sensor) onto the housing, and / or small errors in locating the measurement point on the housing have little impact on the final result. Furthermore, after selecting a specific measurement point on the part, it becomes possible to determine and select a specific frequency range for stiffness prediction.

[0118] In one embodiment, in this scheme, several parts still connected to the base or mold will be detected in series, which can be specified as follows:

[0119] - Steps for capturing images of the component under test;

[0120] - Steps for analyzing images, such as determining the type of each part and / or the position and / or orientation of each part;

[0121] - Select one or more measurement points for each part, and / or select an excitation vibration spectrum for each part and / or each selected point;

[0122] - For each component under test (DUT), the step of positioning the substrate or supporting the DUT in the vibration excitation and measurement equipment. According to this embodiment, excitation and measurement can be performed automatically while the wafer still carries the spindle blank:

[0123] - Take one or more images of the wafer;

[0124] - Perform automatic image analysis to determine at least the XY position of each part (or the type or model of the part).

[0125] - Determine or select specific measurement points that have been pre-established, based on the determined location and / or orientation and / or type of part (e.g., on the housing); a specific excitation cycle may also be selected based on the type of part or a specific point;

[0126] For example, using a tool carrying a blade and including an XY stage, each hairspring blank is continuously and automatically positioned opposite the excitation source and measuring device, and inspection is performed by aiming at the correct measuring point and applying the correct excitation specifications. An automatic focusing step can be optionally performed, adjusting the relative position according to the Z-axis of the vibratory head position, thus making it possible to obtain the clearest image of the observed part. Therefore, the laser beam is accurately focused on the surface of the part, provided that the focal plane of the laser beam and the focal plane of the observation camera coincide, or their offset is known and systematically compensated for.

[0127] In one embodiment, based on a measurement point selected on the component under test (DUT) and / or based on an excitation frequency and / or based on a DUT model, a step can be provided including specifying the excitation direction and / or measurement direction. For this purpose, an excitation direction perpendicular to the DUT (or the axial direction of the excitation source) can be selected to maximize displacement perpendicular to the plane formed by the DUT. An excitation direction inclined relative to the DUT (or the axial direction of the excitation source) can be selected to maximize displacement within the plane formed by the stationary component. In terms of measurement, a measurement direction perpendicular to the DUT (or the axial direction of the laser beam from the measuring device) can be selected to maximize the measurement accuracy of displacement perpendicular to the stationary plane of the DUT. A measurement direction inclined relative to the DUT (or the axial direction of the laser beam from the measuring device) can be selected to maximize the measurement accuracy of displacement contained within the plane formed by the stationary component. When measuring along an oblique axis, a suitable receiving sensor for receiving reflected signals can be provided based on the roughness of the part: for "mirror" parts with low roughness, a receiving sensor with a large collecting cone (preferably covering at least twice the tilt angle) or an offset can be provided, while for "rough" parts, a receiving sensor combined with the light emission source can be provided.

[0128] In one embodiment, several components are attached to a substrate, such as a wafer. Sampling can be performed by separating one or more components to test them as a unit, and from this, the specific excitation frequency to be applied, and / or the specific measurement point to be used, and / or the specific range of the vibrational spectrum to be considered, in order to derive the desired oscillation frequency characteristics. In other words, this initial sampling makes it possible to test parts individually under favorable conditions (measurement errors and interferences are limited) to select optimal testing conditions for parts still attached to the substrate. In the case of exciting components attached to the substrate, the response of the substrate may be excited and measured in order to subsequently determine and exclude the spectral range of its vibrations. Attached Figure Description

[0129] Further details of the invention will become clearer upon reading the following description, which is made with reference to the accompanying drawings, wherein:

[0130] - Figure 1 The uncorrected stiffness distribution curves of the springs on three different plates are shown;

[0131] - Figure 2 This shows the mean stiffness of the plate near its nominal value;

[0132] - Figures 3A-3F It is a simplified representation of the process of fabricating a mechanical oscillator (in this case, a hairspring) on ​​a wafer;

[0133] - Figure 4 A device for evaluating hairspring torque is shown;

[0134] - Figure 5 This schematically illustrates how to evaluate the stiffness of a hairspring through vibration analysis;

[0135] - Figure 6 An example of a frequency applied to a silicon wafer supporting a hairspring blank is shown in order to apply vibrational excitation;

[0136] - Figure 7 An example is shown where the displacement amplitude at a point on a hairspring blank is measured in response to... Figure 6 The imposed frequency range;

[0137] - Figure 8 Detailed illustrations are shown in Figure 7 The resonance peak determined at a specific frequency;

[0138] - Figure 9 It shows in Figure 8 Oscillation peaks measured and superimposed at specific frequencies;

[0139] - Figure 10 It shows from Figure 9 An example of a predictive model built from data extracted from [the dataset]. Detailed Implementation

[0140] Figures 3A-3F This is a simplified representation of a method for fabricating a mechanical oscillator 100 on wafer 10. The oscillator is specifically intended for mounting on the regulating components of a watch; in this example, it takes the form of a silicon hairspring 100, which is mounted on the balance wheel of a mechanical watch movement.

[0141] exist Figure 3AIn this design, wafer 10 is described as an SOI (“silicon on insulator”) wafer, comprising a substrate or “processor” 20 carrying a sacrificial silicon oxide (SiO2) layer 30 and a monocrystalline silicon layer 40. For example, the substrate 20 may be 500 µm thick, the sacrificial layer 30 may be 2 µm thick, and the silicon layer 40 may be 120 µm thick. The monocrystalline silicon layer 40 may have any crystal orientation.

[0142] Figure 3B and 3C The photolithography process is illustrated. "Photolithography," as used here, refers to all operations that transfer an image or pattern onto or over wafer 10. (See reference...) Figure 3B In this exemplary embodiment, layer 40 is covered by a protective layer 50, for example, made of a polymerizable resin. Layer 50 is structured, typically through photolithography steps using an ultraviolet light source and, for example, a photomask (or other type of exposure mask) or a stepping and stencil system. Figure 3C As shown, this photolithographic structure forms a pattern of multiple oscillators in layer 50.

[0143] Subsequently, Figure 3D In the steps shown, the pattern is processed, particularly etched, to form multiple oscillators 100 in layer 40. Etching can be performed using deep active ion etching (DRIE). After etching, the remainder of the protective layer 50 is subsequently removed.

[0144] exist Figure 3E In this process, the oscillator is released from the substrate 20 by locally removing the sacrificial layer 30, or even by etching all or part of the silicon of the substrate or processor 20. Prior to the release step, the etched surface may also be smoothed (not shown), for example by a thermal oxidation step, followed by a deoxidation step, such as wet etching using hydrofluoric acid (HF).

[0145] exist Figure 3F In the final step of the manufacturing process, the silicon oscillator 100 with 110 turns is coated with a layer of silicon oxide (SiO2) 120, typically through a thermal oxidation step to produce a thermally compensated oscillator. The formation of this layer 120, typically 2-5 µm thick, also affects the final stiffness of the oscillator and therefore needs to be considered in the preceding steps to obtain the vibrational characteristics of the hairspring, thereby achieving the specific natural frequency of the hairspring-balance wheel pair in a given watch mechanism.

[0146] As mentioned above, in the stage prior to the production of the thermal compensation layer, the various oscillators formed in the wafer typically exhibit significant geometric dispersion among themselves, and thus significant dispersion in their stiffness, even though the steps of patterning and processing / engraving through these patterns are the same for all oscillators.

[0147] Furthermore, even when using the same method specifications, the stiffness dispersion is even greater between the hairsprings of two wafers etched at different times.

[0148] The above description relates to silicon 100 oscillators, but oscillators can also be produced using glass, ceramics, carbon nanotubes, or even metals.

[0149] To make the average stiffness of oscillators on different wafers relative to Figure 2 The nominal stiffness value shown is centered, and the oscillator obtained on the relevant wafer 10 in step 3E can be intentionally formed to a different size (e.g., larger) than the size d required to obtain the nominal or target stiffness. This makes it possible to set up a control process designed to estimate the vibrational characteristics (natural frequency and / or oscillation frequency) of the oscillator in order to infer the stiffness and / or actual size of the oscillator 100, in order to correct the size, which will result in obtaining the ideal natural frequency of the oscillator-balance pair.

[0150] The present invention proposes, in step 3E, to determine, from at least one characteristic of the oscillation frequency of the oscillator 100 sample on the wafer, whether geometric correction of the oscillator is required. If so, the present invention proposes to precisely calculate the material thickness to be modified (to be removed or added) around each inflection point to obtain dimensions that result in vibrational characteristics (natural frequency and / or oscillation frequency, and / or stiffness) of the oscillator corresponding to the target value, using a method more efficient than prior art methods.

[0151] Therefore, the present invention proposes to determine at least one characteristic of the oscillation frequency of a resonator sample by vibration measurement, and to apply a prediction method (e.g., a digital model or a classification or categorization method) to link the results of the vibration measurement with the necessary geometric corrections.

[0152] In this way, the modal characteristics of the hairspring connected to the plug can be utilized. During the learning phase, using analytical and numerical methods, a predictive machine can be built by establishing a predictive model that links dimensions (especially thickness) and / or stiffness to certain selected frequencies (natural frequencies or oscillating frequencies related to resonance peaks or half-width at half-maximum).

[0153] Once the learning phase is complete (once the mode and excitation frequency to be used are determined), it becomes possible to move into the prediction phase, by using a prediction model to examine the oscillator of the produced wafer to predict whether dimensional corrections are needed. This involves using a prediction machine to calculate or predict, if necessary, precise corrections to be made to the oscillator dimensions (e.g., by shrinking if the produced blank is larger than the final required size, or by adding material if the produced blank is smaller than the final required size).

[0154] In this way, it is possible to integrate the control process into the manufacturing method to correct the vibration characteristics (natural frequency and / or oscillation frequency, and / or stiffness) of the oscillator when necessary to obtain a specific and preset natural oscillation frequency once the oscillators are each coupled to the balance wheel of a particular watch mechanism.

[0155] Vibration excitation

[0156] By measuring the vibration response of an oscillator, it is possible to deduce at least one characteristic of the oscillation frequency, such as its value. Specifically, the wafer must first be vibrated. Several methods are available:

[0157] a. Frequency domain measurement :

[0158] 1- Excitation (continuous single-frequency excitation) is performed on or under the wafer, on or below the hairspring blank 200, using a piezoelectric power source (or any other source capable of inducing or applying acoustic excitation) at a specific frequency f0. In this variant, the excitation is held.

[0159] 2. Alternatively, a piezoelectric power source (or any other source capable of inducing or applying acoustic excitation) can be used on or below the wafer, on or below the springform blank 200, to provide a specific excitation (preferably) whose frequency varies over time, covering a preset frequency range, such as from 0 to 100 kHz, preferably from 0 to 75 kHz, preferably from 0 to 50 kHz, preferably from 5 kHz to 50 kHz, preferably from 10 to 35 kHz. The entire frequency range can be scanned or covered in time intervals ranging from fractions of a second to several seconds. For example, the frequency range can be scanned or covered in time intervals of less than 0.5 seconds, less than 1 second, or less than 1.5 seconds. In this variant, the excitation frequency varies continuously.

[0160] b. Time-domain measurement: An excitation hammer (or any other source capable of generating pulsed acoustic excitation) is used at the wafer edge, on, or under the spindle to obtain the shortest possible acoustic pulse (multi-frequency pulse excitation). In this variant, the excitation is timely and discontinuous.

[0161] Furthermore, measurements can be performed according to a specific sampling method, such as a sampling range of 4, 2, or 1 Hz. In fact, the resolution of the acquired data, processed according to the Fourier transform, depends directly on the acquisition time.

[0162] Furthermore, if the frequency range is extended to 50 kHz, for example, a signal sampling frequency of at least 100 kHz can be selected.

[0163] Generally, the excitation direction, i.e., the direction of motion applied by the source, can also be changed (vibration can be applied along one or more axes, and this or these directions can vary over time). If a wafer consisting of multiple oscillators is excited, the direction of vibration may be adjusted to point towards one or another oscillator, depending on the displacement amplitude measurement described below.

[0164] Finally, the sound source can be coupled to a diverging cone pointing towards the oscillator to be excited, and the sound source can be adjusted to emit an excitation signal with an amplitude sufficient to apply vibration excitation to the oscillator, and the amplitude is sufficient to be accurately detected and measured by a selected measuring instrument.

[0165] Measurement of amplitude, velocity, or displacement acceleration

[0166] During the excitation process, appropriate measuring equipment is used to record the amplitude and phase (relative to the excitation source) of a specific excitation spring in the X, Y (in-plane), and Z (out-of-plane) directions. Non-limiting examples of possible measurement methods are as follows:

[0167] - Optical methods using interferometers:

[0168] a. Through the three-dimensional Doppler effect (laser Doppler effect vibration meter);

[0169] b. Holographic method;

[0170] - Optical methods for stroboscopic effects;

[0171] - High temporal resolution chromatic confocal profiler;

[0172] - Optical reflector:

[0173] a. Vibration analysis using beam deflection on a multi-screen detector or camera;

[0174] b. TCSPC-type spatiotemporal analysis;

[0175] - Acoustic methods using Doppler effect ultrasound.

[0176] Figure 5 A silicon wafer 25 is schematically shown on which multiple hairspring blanks 200 are formed. A vibration excitation source 400 is connected to the silicon wafer 25, thus applying vibration excitation. Therefore, each hairspring blank 200 vibrates, and a laser vibrometer 300 (focused here on a point on the right-hand hairspring blank 200) is able to measure the amplitude of the vibration at that point over time. Displacement can be measured in the normal direction to the plane of the insert 25, but it can also be measured in one or more directions contained within the plane of the insert 25.

[0177] Once a specific point has been studied, the laser vibrometer 300 can be moved to another measuring point on the hairspring blank 200, or to another hairspring blank 200 on the insert 25. Alternatively, the hairspring blank 200 can be moved relative to the laser vibrometer.

[0178] Figure 6 An example of a time-varying vibration excitation is shown. In the example shown, the excitation frequency varies with time, between 0 Hz and 50 kHz, and continuous rising edges can be applied, each rising edge separated by a rest period without excitation. For each measurement point on the hairspring blank 200, multiple rising edges can be applied (between 2 and 60 rising edges), each rising edge lasting between 0.5 seconds and 2 seconds, for example.

[0179] Select the reference point to be measured.

[0180] Regarding the measurement of displacement amplitude, a step can be provided during the learning phase, including identifying points on the resonator where the vibration response is significant. When the hairspring is subjected to vibration, especially when the frequency varies with time, the vibration response will result in nodes on the hairspring—specific points on the hairspring where the displacement amplitude is low or zero. If displacement measurement is performed at a point on the hairspring that is a node at one or more specific frequencies, then the determination of the oscillation frequency characteristics will be negatively affected.

[0181] Therefore, a preliminary step is provided to measure displacement at multiple preset points on the hairspring, for example, at least 10 preset points, preferably at least 20 preset points, and very preferably at least 30 preset points. The preset points can be specified to be selected on an orthogonal XY reference frame arranged on the hairspring plane.

[0182] At the end of the preliminary amplitude measurement step at the preset points, the oscillation frequency of each measurement point can be determined. Then, reference points are selected, and the displacement amplitude measurements during excitation show that they are not nodes at these oscillation frequencies. In other words, the identified nodes have a displacement amplitude of zero or less than a first threshold peak at at least one oscillation frequency. These nodes are removed from the reference points for consideration in subsequent measurements. It should also be noted that the reference points are different and depend on the location of the filamentary blank 200 on wafer 25.

[0183] Typically, it can be assumed that at least two reference points will be selected, and preferably at least four reference points. In the case where the oscillator has a radius Ra and is fixed or embedded in the wafer through its outer end, it is preferable to select and position four reference points:

[0184] - In the first region, less than 0.20 × Ra (e.g., in the central inner pile), or

[0185] - The second zone is between 0.05×Ra and 0.30×Ra (e.g., at the second helical coil at the distance from the inner stake), or

[0186] - The third region between 0.35×Ra and 0.65×Ra (e.g., a spiral coil located in the middle of the hairspring), or

[0187] - The fourth region between 0.65 × Ra and 0.85 × Ra (e.g., the spiral coil located at three-quarters of the way down the hairspring).

[0188] In this way, the reference point is far from the part anchored on the plate, which naturally has a higher oscillation displacement capability, thus ensuring better accuracy of displacement measurement.

[0189] The displacement of a point on the insert body and / or a point on the excitation source can also be measured to determine or measure, for example, phase shift or vibration decay, or even resonance from vibration coupling or the insert. These additional measurements ensure that the determined peak value is indeed the peak value of the individual hairspring. The measurement of displacement amplitude can also be performed synchronously with the vibration excitation.

[0190] Determination of vibration characteristics

[0191] Depending on the selected incentive range, there are several possible scenarios:

[0192] a. Frequency domain measurement

[0193] 1- Variations of continuous excitation:

[0194] (1) The amplitude and phase of the oscillation are integrated over a sufficiently long time to obtain good spectral resolution at the excitation frequency f0;

[0195] (2) Shift the oscillation frequency by delta f, excite at frequency f0 + Δ f, and repeat integration step i;

[0196] (3) Based on the excitation frequency, reconstruct the amplitude spectrum and phase spectrum of the oscillation (which may have peaks at several frequencies).

[0197] 2- Variables that change in excitation frequency over time:

[0198] (1) During the frequency sweep process, the amplitude and phase of the oscillation are recorded at regular intervals;

[0199] (2) Repeat step i at least once, preferably at least three times;

[0200] (3) Based on the excitation frequency, reconstruct the amplitude spectrum and phase spectrum of the oscillation (which may have peaks at several frequencies).

[0201] b. Time-domain measurement :

[0202] (1) Record the time displacement of the spiral coil along the X, Y and Z directions for a sufficiently long time to obtain a sufficiently representative signal, such as a few seconds.

[0203] (2) The signal can be recorded and used as a reference signal for comparison with other signals measured on other components. Alternatively, the signal can be processed using a Fourier transform to determine the oscillation frequency in the recorded signal.

[0204] Therefore, each stimulated resonator can be identified with at least one resonance peak. It is recommended that the oscillation frequency not be determined based on the peak value of the resonance peak, i.e., the maximum amplitude, but rather on the curve region between 25% and 75% of the maximum amplitude value of the resonance peak, for example, from the width at half its height. This approach focuses on the curve portion between 25% and 75% of the maximum amplitude value of the resonance peak, making it possible to limit errors arising from the singularity of the maximum amplitude point and approximate calculations of the reconstructed peak apex. The curve region between 25% and 75% of the maximum amplitude value of the resonance peak is more accurate than the portion above 75% (typically the peak apex), providing greater precision for determining the accurate oscillation frequency. For example, the midpoint connecting the two points at the midpoint of the oscillation peak's height can be used to determine the oscillation frequency associated with the relevant peak value.

[0205] Figure 7 It shows Figure 5 An example of the vibration spectrum of a point on the midstream wire blank 200, which is composed of... Figure 6 The displacement amplitude measurements at relevant measurement points in response to vibrational excitation were reconstructed, falling between 10 kHz and 15 kHz. Note that three amplitude peaks are present around 11 kHz, 12.3 kHz, and 13.7 kHz. Although not shown, 10 to 30 amplitude peaks can typically be identified if the frequency range of the vibrational excitation is between 0 Hz and 50 kHz. Each amplitude peak has an oscillation frequency, and the maximum amplitude varies considerably.

[0206] Figure 8 The process that can be performed on an amplitude peak, for example, at 11 kHz, is illustrated in detail. The aim is to find the oscillation frequency and give it the most accurate value possible. The applicant realized that better accuracy could be achieved by determining a segment connecting the rising and falling portions of the curve, i.e., half the peak value, rather than processing based on the maximum value of the peak. The oscillation frequency is typically the value in the middle of this segment. However, interpolation can be performed on points near the resonance peak to improve accuracy; the selected points on this segment can be moved, and these points will not be the midpoints, especially if the actual location of the resonance peak has shifted, for example, due to the selected sampling frequency.

[0207] Figure 9 The diagram illustrates an example of an amplitude peak of approximately 10 kHz, constructed from about 10 blank hairsprings using a 200-count detection method. It can be seen that the frequency position of the amplitude peak differs from one hairspring blank to another (from approximately 9.8 kHz to 10.02 kHz), with the maximum displacement amplitude varying by approximately a 1:5 ratio. Since the peak values ​​of the amplitude peaks are not truly symmetrical, it is meaningful to determine the oscillation frequency based on the peak width at half-height.

[0208] for Figure 9 These detections allow us to deduce the following oscillation frequencies:

[0209]

[0210] Determine the actual stiffness of the oscillator being tested and / or the dimensions of the outer pile.

[0211] To establish a predictive model that can take vibrational characteristics (typically oscillation frequency) as input and output stiffness and / or dimensional corrections, it is necessary to provide data related to the actual stiffness and / or dimensions of the outer stud of the oscillator under test during the learning phase. For this purpose, the natural frequency of the hairspring system can be measured in an environment similar to that of a specific watch mechanism.

[0212] There are two methods that can be used. First, a pre-set balance wheel can be directly coupled to the oscillator still connected to the plate, and the natural oscillation frequency of the oscillator-balance wheel pair can be measured to compare this natural frequency with the expected natural frequency. Most importantly, the actual stiffness or actual dimensions can be calculated based on formulas 1 to 3 above. According to the second option, the oscillator under test can be completed so that it can be mounted or coupled separately to the balance, where the natural oscillation frequency of the oscillator-balance pair is measured again.

[0213] In both of the above options, the stiffness of each oscillator can be determined through an intermediate stage, followed by the actual dimensions of the tested oscillator stud. In other words, the natural frequency or oscillation frequency can be determined by analyzing the free oscillation of the hairspring coupled to the reference balance wheel, and then the stiffness or dimensions of the oscillator stud can be determined. In this method, a laser is used to align with the balance wheel arm or balance wheel spring carrier, and the movement time of the balance wheel arm or micrometer is recorded. This is used to estimate the period, then the frequency, and finally the stiffness. The collected data is essentially a point cloud over time.

[0214] Several solutions exist for evaluating the stiffness of a wire on a wafer, as described by M. Vermot et al. in *Traité deconstruction horlogère* (2011), pp. 178-179. For example, a dynamic evaluation can be performed by connecting the wire to a reference balance with known inertia. By measuring the assembly frequency, the wire stiffness can be accurately inferred. This evaluation can be performed on the wafer or by separating the wire from the wafer. The references and prior art given above provide details of this method.

[0215] Similarly, stiffness can be inferred by measuring the response torque at the inner pile using a rheometer. The obtained signal represents the relationship between torque change and amplitude. By analyzing the slope of this curve at low amplitude (the linear portion), the stiffness can be inferred, and then the dimensions of the oscillator outer pile can be deduced. The dimensions of the spiral spring outer pile can then be determined.

[0216] For each oscillator tested on the wafer, the natural frequency and / or oscillation frequency and / or stiffness can also be estimated through simulation. To do this, the dimensions of each tested oscillator can be measured so that the oscillator can be reconstructed through numerical modeling, thereby simulating its vibrational response to the imposed spectrum through numerical calculation, and the stiffness of the oscillator can also be found.

[0217] High-resolution three-dimensional X-ray tomography will make it possible to extract point clouds, provide the three-dimensional material density of the hairspring, and, using appropriate image reconstruction, derive a map of the hairspring's cross-section. These different types of data can be used to infer the dimensions of the outer pile and to estimate the hairspring's stiffness using geometric methods.

[0218] Another approach involves analyzing the forced oscillations of the hairspring on a reference balance wheel with an escapement. As mentioned above, laser measurements of the balance wheel arm's transit time (point cloud) can be used to measure frequency and infer stiffness. Alternatively, an acoustic acquisition system (Witschi type microphone) can be used to record the impacts of the exhaust / anchoring system at various stages of operation. The measured data are either the point cloud of the balance arm's transit moment or the time evolution of the sound pressure level. These types of experimental data can be used to infer the period, then the frequency, then the stiffness, and finally the dimensions of the oscillator's outer stud.

[0219] Back to above Figure 9 The test discussed earlier, which involves coupling each hairspring blank 200 to a reference balance wheel and measuring its stiffness, allows us to derive the following stiffness:

[0220]

[0221] Building a prediction model

[0222] During the learning phase, the oscillation amplitude of the physical oscillator is measured, and the oscillation frequency is determined. In order to subsequently correlate the oscillation frequency measured on the oscillator with the stiffness and / or dimensional (thickness) corrections to be made, a correlation phase is required, in which a predictive model is built.

[0223] The aforementioned operations (vibration measurement, determination of resonance peaks, bandwidth at half-height and its median or correction value, determination of the stiffness and / or size of the outer pile) make it possible to provide a database that correlates the position of the hairspring on the wafer, the spectrum or period of vibration, or the bandwidth at half-height and its median or correction value with the effective stiffness and / or size of the outer pile of the hairspring. As mentioned above, this database can be constructed from numerical simulations of the finite element model of the hairspring. These simulations can be used to generate a reference oscillation spectrum or period related to stiffness. This database can also be supplemented by experimental measurements, by measuring the vibration spectrum, oscillation period, and position of the hairspring on the wafer, as well as their associated stiffness. One advantage of this approach is that the learning database is enriched as the testing progresses. This makes it possible to have a model that is adaptive to both the pad and the spring, and helps to reduce the standard deviation of the pad stiffness.

[0224] This database can be used to build a predictive model, and several solutions are available.

[0225] A numerical model, such as a polynomial model, can be established as a function of the oscillation frequency to calculate the actual thickness, dimensional correction, or actual stiffness.

[0226] Alternatively, classification can be achieved by dividing the input data (results of vibration measurements, typically the frequency of the resonance peak) and the output data (stiffness and / or dimensions of the oscillator's outer pile) into k-means and establishing a correspondence between them.

[0227] Neural networks (such as perceptrons) can also be used to process images of resonance peaks, classifying them according to stiffness or external pile size, with the category defined by the numerical increment.

[0228] In summary, the learning phase includes a detection phase (exciting the oscillator, measuring vibration characteristics to reconstruct the vibration spectrum and determine the oscillation frequency). The stiffness and / or dimensions of the oscillator's outer pile are also measured. Once the input data (oscillation frequency) and output data (stiffness and / or outer pile dimensions) of the important samples are available, the predictive model building phase can proceed.

[0229] Returning to the discussion and description Figure 9 The relevant examples and the collected data are as follows:

[0230]

[0231] Linear regression modeling was performed on the data from the first six lines above, and the following relationships were established:

[0232] R = 0.0015 F – 10.894

[0233] In the formula:

[0234] R is the stiffness (unit: 10). -7 N.mm)

[0235] F is the oscillation frequency, measured in Hz.

[0236] Therefore, stiffness can be predicted and compared with the actual measured stiffness, as shown in the table below. The first six rows are the data used to build or train the linear regression, and the last four rows are only the predicted values:

[0237]

[0238] The maximum measured error was 4.40%. Figure 10 The linear regression line for the values ​​in the first six rows is shown.

[0239] It should be noted that, further, the established prediction model should be examined to determine if it possesses good sensitivity, i.e., that the model will produce two different output values ​​for two different input values. The applicant has acknowledged that the sensitivity of the prediction model is not uniform for all resonance peaks. In particular, if the established prediction formula is referenced and... Figure 10 The guidance coefficient is shown to be 0.0015 10. -7 N.mm / Hz. On the one hand, the applicant found that for higher oscillation frequencies, the guiding coefficient can be larger, which provides better predictive sensitivity for predicting significant stiffness or size corrections, even from close oscillation frequency values. Further, during the learning phase, a step is provided to compare predictive sensitivities to check / confirm that certain high-frequency (e.g., above 5 kHz) resonant peaks are best considered and selected to subsequently predict stiffness and / or size corrections as a function of the measured vibration response as accurately as possible.

[0240] On the other hand, the applicant also recognizes that even at similar oscillation frequencies, resonance modes (especially deformation and / or displacement modes of the resonator) can differ significantly, which can also affect the sensitivity of stiffness and / or size correction predictions. Further, during the learning phase, a step is provided to compare prediction sensitivities so that one oscillation frequency is subsequently considered over another in order to predict stiffness and / or size corrections as a function of vibration response as accurately as possible.

[0241] Building upon the aforementioned research on predictive sensitivity, it is conceivable that during the learning phase, the identified resonance peaks would be categorized based on the sensitivity of predicted stiffness and / or size corrections. An excitation frequency range (to be applied in the pure prediction phase) can then be defined to include at least one or more resonance peaks or frequencies that provide optimal sensitivity. Therefore, applying a variable vibration excitation within a preset frequency range will ensure accurate prediction of identified resonance peaks, or prediction of each identified resonance peak that overlaps or supports each other.

[0242] Generally, during the learning phase, high-frequency resonant peaks and / or resonant peaks corresponding to a specific resonant mode can be selected to make accurate and reliable numerical predictions. The frequency range will be preset to include at least one resonant peak, preferably several, so that a single prediction can be made as accurately as possible, or several predictions can be made (one for each resonant peak of interest) for cross-checking, averaging, or recalibrating of the predicted values.

[0243] For example, several stiffness or size correction values ​​can be predicted from several resonance peaks or frequencies, and then the final value can be calculated using the predicted values. A weighted average is generated by assigning weights to each predicted value, with each weight determined as a function of the sensitivity of each corresponding resonance peak or frequency.

[0244] Alternatively, and more preferably, a single model can be provided that accepts all input resonant peaks or frequencies and returns stiffness or size corrections, with the model learning phase used to calculate the weights of the input resonant peaks or frequencies.

[0245] Prediction phase

[0246] Once the learning phase is complete, it may be possible to proceed to the prediction phase, such as during oscillator control. The inspection process can typically be performed on the hairspring blank produced on the wafer and still attached to that wafer, in order to estimate the stiffness and / or dimensions of the hairspring stud on the sample to determine if dimensional corrections are needed.

[0247] Once the model training is complete, the following check procedure applies:

[0248] 1) Locate the position of the hairspring on the wafer and measure the vibration spectrum or oscillation period (as described above);

[0249] 2) Predict the stiffness and / or dimensions of the outer pile by applying a predictive model;

[0250] 3) Determine whether dimensional corrections are needed to achieve the target natural frequency or stiffness.

[0251] During the control process, the precise corrections to be made can also be quantified. Therefore, in addition to the control methods described above, the manufacturing method may also include:

[0252] 1) Based on the model and target stiffness and / or target outer pile dimensions, know the estimated effective hairspring stiffness: apply the necessary correction dose.

[0253] Repeat steps 1) and 2) of the inspection process to check the spring stiffness / size and confirm that it has reached the target value within the allowable error range, or repeat these steps and size corrections until the stiffness / size predicted by the model reaches the target value.

[0254] sampling

[0255] As is well known, hundreds of hairsprings are fabricated on a single wafer, and the dimensions of the hairspring studs can vary depending on the region of the wafer. Although individual hairsprings can be evaluated, in practice, samples of hairsprings distributed across the wafer are evaluated.

[0256] Based on the assessments performed, corrections can be applied uniformly across the entire wafer, or by region if the results differ from one hairspring to another. This approach reduces the standard deviation of stiffness dispersion. Furthermore, if the stiffness of all hairsprings is known from the application of the model, it becomes possible to determine the optimal correction to minimize overall dispersion.

[0257] It is even possible to evaluate all the springs on the insert, especially through vibration evaluation, because doing so is very fast.

[0258] Although the above examples are primarily based on the manufacture of a hairspring with an initial outer pile size larger than the target outer pile size, it is also possible to manufacture a hairspring with an initial outer pile size smaller than the target outer pile size. The corrective steps then involve adding material, as described in the aforementioned document EP3181939.

[0259] This method involves determining the oscillation frequency simply by applying vibration excitation to the hairspring blank, making it possible to quickly obtain measurement data without having to assemble the balance wheel. For example, it also limits measurement errors because only the hairspring blank is detected (there are no errors that may be related to the balance wheel, such as its mass, assembly position, etc.).

Claims

1. A method for detecting a hairspring or a hairspring blank arranged as a hairspring, wherein the hairspring needs to have at least one preset oscillation frequency, the detection method comprising the following steps: a. Apply time-varying vibration excitation to the hairspring or hairspring blank to cover a preset frequency range; b. During vibration excitation within a preset frequency range, determine at least one characteristic of the oscillation frequency of the hairspring or hairspring blank; c. Provide the oscillation frequency characteristics determined in step b to the predictor to determine the stiffness of the hairspring or hairspring blank and / or to determine whether dimensional corrections to the hairspring or hairspring blank are needed to achieve the preset oscillation frequency. The predictor is established by using analytical and numerical methods to build a predictive model that correlates size and / or stiffness with certain selected frequencies.

2. The detection method according to claim 1, characterized in that, The frequency range is preset to cover at least one frequency range: - The frequency range is centered on a preset oscillation frequency; - And the frequency range covers ±15% of the preset oscillation frequency.

3. The detection method according to any one of claims 1 to 2, wherein the hairspring has at least two preset oscillation frequencies, characterized in that, The frequency range is preset to cover at least two preset oscillation frequencies.

4. The detection method according to any one of claims 1 to 2, characterized in that, Step b is based on time-varying measurements of the displacement amplitude, velocity, or acceleration of at least one point on the hairspring or hairspring blank.

5. The detection method according to any one of claims 1 to 2, wherein the hairspring or hairspring blank is disposed on a base surface, characterized in that, Step b includes the following sub-steps: - Sub-step b': In the normal direction of the base plane, measure the displacement amplitude, velocity or acceleration of at least one point of the hairspring or hairspring blank; - and / or, sub-step b'': measuring the displacement amplitude, velocity or acceleration of at least one point of the hairspring or hairspring blank in the direction contained by the base plane.

6. The detection method according to claim 4, characterized in that, Step b includes: - Sub-step: Determine the resonance peak of the hairspring or hairspring blank based on the displacement amplitude or velocity of at least one point of the hairspring or hairspring blank.

7. The detection method according to claim 6, characterized in that, The oscillation frequency in step b is determined by taking the midpoint of the line segment connecting two points on the resonance peak at 50% of the maximum value of the resonance peak.

8. The detection method according to any one of claims 1 to 2, characterized in that, When it is determined in step c that a dimensional correction is needed, the method includes a step: d. Using a predictor and based on the oscillation characteristics determined in step b, calculate the size corrections that need to be implemented.

9. The detection method according to any one of claims 1 to 2, characterized in that, The predictor uses a polynomial formula to predict whether a size correction is needed.

10. The detection method according to any one of claims 1 to 2, characterized in that, The predictor performs classification by a neural network to predict whether size correction is needed.

11. The detection method according to any one of claims 1 to 2, wherein the hairspring blank is formed on a wafer, and the wafer includes a plurality of hairspring blanks distributed in a plurality of fan-shaped regions of the wafer; Its features are, Step b includes a sub-step: determining at least one characteristic of the oscillation frequency of at least one hairspring blank in each sector region; Furthermore, step c includes a sub-step: determining the stiffness of the hairspring blank in each sector and / or whether dimensional correction is required.

12. The detection method according to any one of claims 1 to 2, characterized in that, The method includes a preliminary step of taking into account the material of the hairspring or hairspring blank and adjusting the maximum amplitude of the vibration excitation and / or the frequency range of the preset frequency range according to the material of the hairspring or hairspring blank.

13. The detection method according to any one of claims 1 to 2, characterized in that, The frequency range covers a frequency range from 0 Hz to 100 kHz.

14. The detection method according to claim 1, characterized in that, The at least one feature includes a resonance peak.

15. The detection method according to claim 4, characterized in that, The measurement is performed at least partially in step a.

16. The detection method according to claim 13, characterized in that, The frequency range covers a frequency range from 0 Hz to 50 kHz.

17. The detection method according to claim 16, characterized in that, The frequency range covers a frequency range from 0 Hz to 40 kHz.

18. The detection method according to claim 17, characterized in that, The frequency range covers a range from 10 kHz to 35 kHz.

19. A method for manufacturing a hairspring having at least one preset oscillation frequency, comprising the following steps: • Form at least one hairspring or hairspring blank having the dimensions required to obtain a preset oscillation frequency, the dimensions including a preset allowable error; • The detection method according to claim 1 is used to detect the hairspring or hairspring blank.

20. The manufacturing method according to claim 19, characterized in that, The method includes a step: • According to the calculation of step d as described in claim 8, at least one dimension of the hairspring blank formed in step a is corrected to obtain a hairspring with a preset oscillation frequency.

21. The manufacturing method according to any one of claims 19 to 20, characterized in that, The hairspring blank is formed together with a plurality of other hairspring blanks on a single wafer.

22. A predictive machine learning method for implementing step c of the detection method according to any one of claims 1 to 18, comprising the following steps: (1) Forming a hairspring or a hairspring blank; (2) Apply time-varying vibration excitation to each hairspring or each hairspring blank to cover a preset frequency range; (3) When applying a preset frequency range, determine at least one characteristic of the oscillation frequency of each hairspring or each hairspring blank; (4') Install multiple hairsprings or hairspring blanks in an oscillation mechanism with a preset inertia in order to measure the free oscillation frequency or stiffness of each hairspring or hairspring blank; And / or (4'') Using simulation tools, model multiple hairsprings or hairspring blanks in an oscillating mechanism with preset inertia in order to calculate the free oscillation frequency or stiffness of each hairspring or hairspring blank; (5) For each hairspring or hairspring blank, provide the following to the predictor: - The characteristics of the oscillation frequency determined in step (3); - The free oscillation frequency or stiffness measured in step (4') and / or step (4'').

23. The learning method according to claim 22, characterized in that, Step (3) includes a preliminary stage, namely: determining the benchmark measurement point, and: - Measure the displacement of multiple preset points on the hairspring or hairspring blank; - Determine a node among multiple preset points, wherein the displacement amplitude of the node at at least one oscillation frequency or peak value is zero or less than the peak value of a first threshold; - Among multiple preset points, select the reference point to be measured during detection, which is different from the determined node.

24. The learning method according to claim 23, wherein, The peak displacement amplitude of each reference point is greater than the peak value of the second threshold.

Images