An automobile EPB switch button stroke and force size test bench and a control system thereof

Abstract

The application relates to the technical field of automobile part detection equipment, and discloses an automobile EPB switch button stroke and force size test bench and a control system thereof, which comprises a numerical control platform, a detection mechanism and a control unit, the detection mechanism comprises a force sensor and three laser sensors, and the control unit establishes an absolute coordinate system; the geometric center coordinates of a to-be-detected EPB switch button are calculated according to the data collected by the laser sensors and are translated to align the force sensor; the actual surface difference size is calculated; when being pressed, the tilt angle parameters of the pressed surface are extracted and are converted into plane compensation displacement instructions to drive the numerical control platform to translate, data are recorded, and the effective stroke is calculated; when being reset, the time domain phase difference is calculated according to the moment when the internal metal contact is disconnected and the moment when the external button is lifted; and the actual surface difference size, the effective stroke and the time domain phase difference are input into a multilayer perception machine model to determine a qualified state. According to the application, lateral frictional interference during pressing is reduced, mechanical response hysteresis is quantified, and assembly defect classification and evaluation are realized.

Description

Technical Field

[0001] This invention relates to the field of automotive parts testing equipment technology, specifically to a test bench and control system for the travel and force of automotive EPB switch buttons. Background Technology

[0002] The electronic parking brake (EPB) switch in an automotive system is an interactive component that controls the vehicle's parking status. The tactile feel of the EPB switch and the reliability of its internal trigger response are related to driving safety. To ensure product quality before shipment, a test bench is needed to test the travel and pressing force of the EPB switch button before it leaves the factory. Traditional test benches use force sensors in conjunction with a pressing mechanism to apply a normal force in the vertical direction to the EPB switch button and determine whether the EPB switch is qualified by comparing it with a preset force value or displacement threshold.

[0003] In actual press testing, due to the hinge or spring structure inside the automotive EPB switch, the movement trajectory of the automotive EPB switch button after being pressed is accompanied by a certain angle of tilt offset. Traditional test benches can only press down along a fixed vertical axis and cannot adapt to the tilt offset of the automotive EPB switch button during movement. This causes the pressing mechanism to slide relative to the surface of the automotive EPB switch button and generate lateral friction force. The lateral friction force will be superimposed on the measurement value of the force sensor, resulting in errors in the acquired normal pressing force test data. At the same time, the pressing and resetting action of the automotive EPB switch includes two stages: the disconnection of the internal metal electrical components and the rebound of the external mechanical housing.

[0004] Existing testing methods only record static single-point force or stroke values, making it difficult to separate the time difference between the disconnection of internal metal components and the start of rebound of the external button. The lack of time difference data makes it impossible for existing test benches to quantitatively evaluate the mechanical response hysteresis of the internal structure of automotive EPB switches. It is also difficult to detect deep-seated defects such as abnormal wear of internal transmission components or excessive assembly clearance without damaging the physical structure of automotive EPB switches. In addition, existing quality judgments mostly rely on single threshold comparisons, making it difficult to comprehensively process test data from multiple dimensions such as actual surface differences, dynamic effective stroke, and timing response. The single threshold judgment method lacks the ability to perform feature fusion and classification evaluation of multi-source data. When faced with complex assembly defects caused by dimensional tolerances and mechanical interference, misjudgments may occur, and it is impossible to directly output the specific defect type. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a test bench and control system for measuring the travel and force of automotive EPB switch buttons. This solves the problems of existing test benches, such as errors in force measurement caused by lateral friction due to fixed axial pressing, difficulty in separating the disconnection of internal electrical components from the rebound time difference of external mechanical components to diagnose mechanical response lag, and inability to comprehensively evaluate defects by relying on a single threshold and multi-dimensional test data.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a test bench for measuring the travel and force of an automotive EPB switch button, employing the following technical solution:

[0008] A test bench for measuring the travel and force of an automotive EPB switch button includes a base, a CNC platform, a servo component, a testing mechanism, and a tooling fixture. The base provides a fixed reference surface. The tooling fixture is mounted on the base and configured to clamp and fix the EPB switch to be tested. The CNC platform is mounted on the base and includes an X-axis drive component and a Y-axis drive component. The servo component is mounted at the end of the CNC platform and configured to provide Z-axis lifting drive. The testing mechanism is fixed to the output end of the servo component.

[0009] By adopting the above technical solution, the planar positioning and axial force application actions are separated and executed independently by the CNC platform and servo components due to the independent multi-axis drive mechanism layout. This avoids the deflection deformation amplification effect caused by the serial mechanism under heavy load. The base directly supports the tooling fixture and CNC system, forming a closed-loop force circuit. Therefore, a measurement environment with stable geometric reference height is obtained when the equipment is subjected to normal reaction force.

[0010] Preferably, the detection mechanism includes a force sensor and three laser sensors; the force sensor is vertically mounted on the central Z-axis of the detection mechanism, and the lower end of the force sensor is connected to a pressure head; the three laser sensors are distributed in a 120-degree circle around the pressure head and are installed at an angle to the central Z-axis, configured such that when the pressure head at the lower end of the force sensor presses the surface of the EPB switch button to be tested, the emitted light path avoids the pressure head and irradiates the button surface outside the pressing area.

[0011] By adopting the above technical solution, the optical sensor installed at an angle is combined with the central force sensor to project the ranging spot outside the point of force application, thus solving the interference problem of the physical pressure head blocking the optical field of view in the traditional coaxial detection scheme. This structural adjustment ensures that the equipment can simultaneously acquire the three-dimensional height change data of the pressure target surface under dynamic working conditions of continuous application of normal load.

[0012] Secondly, the present invention provides a control system for a test bench for measuring the travel and force of an automotive EPB switch button, employing the following technical solution:

[0013] A control system for a test bench for measuring the travel and force of an automotive EPB switch button is provided. The control system includes a control unit connected to a CNC platform, a servo component, and a testing mechanism. The control unit is configured to execute integrated testing logic, which includes:

[0014] The CNC platform and servo components are driven to move to the mechanical zero position to establish an absolute coordinate system, and preset calibration parameters matching the EPB switch under test are loaded.

[0015] The CNC platform is controlled to move within a preset search area, the detection mechanism is controlled to collect discrete height data, the actual geometric center coordinates of the EPB switch button to be tested are calculated based on the discrete height data, and the CNC platform is driven to translate and align the central axis of the detection mechanism to the actual geometric center coordinates.

[0016] At the actual geometric center coordinates, the relative height of the EPB switch button to be tested and the relative height of the product reference surface are obtained by the testing mechanism, and the actual surface difference dimension of the product to be tested is calculated based on the height difference.

[0017] The driving detection mechanism applies a normal force to the EPB switch button under test by pressing down along the Z-axis, extracts the tilt angle parameter of the pressed surface and converts it into a plane compensation displacement command to drive the CNC platform to synchronously translate and compensate, and records the force data and absolute displacement data to obtain the effective stroke.

[0018] The drive detection mechanism reverses and resets, synchronously recording the unloading time series data of the normal force value and the rebound time series data of the center height. The processed curve extracts the moment when the internal metal contact is disconnected and the moment when the external button is lifted, and calculates the time domain phase difference.

[0019] The actual surface difference size, effective travel distance, and time-domain phase difference are combined to construct a three-dimensional feature input vector. This three-dimensional feature input vector is then input into a pre-deployed multilayer perceptron evaluation model to perform forward propagation calculations. The qualified status is determined based on the classification results output by the multilayer perceptron evaluation model.

[0020] By adopting the above technical solutions, the system reconstructs the true center of the target under test based on the measured height data, thereby correcting the initial value error caused by assembly deviation. In the force-pressing stage, the surface tilt data is extracted and converted into translation compensation action to reduce the test error caused by non-axial off-center load. At the same time, the system extracts and quantifies the degree of timing misalignment between the internal electrical signal disconnection and the external mechanical springback. This type of time domain data, along with the stroke and surface difference indicators, is input into the multilayer perceptron model for classification and calculation. This breaks through the limitations of relying on a single threshold judgment in multi-factor coupled scenarios and improves the ability to determine the accumulation of product assembly tolerances and potential failure states.

[0021] Preferably, calculating the actual geometric center coordinates of the EPB switch button under test based on discrete height data specifically includes:

[0022] The control unit controls the CNC platform to move step by step within a two-dimensional rectangular search area according to a set gridded scanning step size;

[0023] When moving to each grid node position, three laser sensors are triggered to synchronously collect local surface relative tilt distance data. The local surface relative tilt distance data is multiplied by the cosine of the physical installation tilt angle to convert it into vertical absolute height data, and the average value is taken to generate fused height data.

[0024] After traversing the two-dimensional rectangular search area, all merged height data are used to construct a discrete height data matrix according to the spatial coordinate mapping relationship of the grid nodes;

[0025] The X-axis coordinates of the actual geometric center are obtained by calculating the sum of the products of the discrete X-axis coordinates and the corresponding fused height data, and the first ratio of the sum of the fused height data. The Y-axis coordinates of the actual geometric center are obtained by calculating the sum of the products of the discrete Y-axis coordinates and the corresponding fused height data, and the second ratio of the sum of the fused height data.

[0026] By adopting the above technical solution, the calculation deviation introduced by the installation tilt angle is eliminated by using cosine trigonometric functions, and the height data of multiple points in the area is obtained by grid scanning. The coordinates are solved based on the discrete weighted centroid algorithm, replacing the traditional positioning method that relies solely on the theoretical design dimensions as the origin. This algorithm processing method can resist the calculation interference caused by local surface injection molding burrs or abnormal smoothness to a certain extent, thereby obtaining a more realistic reference coordinate of the force center.

[0027] Preferably, obtaining the relative height of the EPB switch button to be tested and the relative height of the product reference surface, and calculating the actual surface difference dimension of the product to be tested, specifically includes:

[0028] Three laser sensors are triggered to acquire the local relative tilt distance of the EPB switch button surface under test and convert it into the vertical absolute height measurement value of the three button surfaces. The arithmetic mean of the three vertical absolute height measurement values ​​of the button surfaces is calculated to obtain the relative height of the button center.

[0029] The CNC platform is driven to move the detection mechanism to a fixed reference surface above the EPB switch housing to be tested. Three laser sensors are triggered to obtain the local relative tilt distance of the reference surface and convert it into the vertical absolute height measurement value of the three reference surfaces. The arithmetic mean is calculated to obtain the relative height of the center of the reference surface.

[0030] The actual surface difference dimension is obtained by subtracting the relative height of the button center from the relative height of the reference surface center.

[0031] By adopting the above technical solution, the existing three sets of laser sensors in the system are directly applied. By performing an arithmetic mean calculation on the distance between the three collected points, the local waviness error existing in the manufacturing process of the button and the reference surface is filtered out during the processing. Furthermore, by performing a difference calculation on the same sensor reference, the deviation introduced by the installation tilt of the test fixture is eliminated, thereby obtaining the actual surface difference dimension that can objectively reflect the relative assembly accuracy of the product.

[0032] Preferably, the extraction of the tilt angle parameters of the pressure surface and its conversion into planar compensation displacement commands to drive the CNC platform for synchronous translation compensation specifically includes:

[0033] During the constant-speed downward pressure cycle, three laser sensors are triggered to continuously collect discrete relative distance data sequences. Trigonometric functions are used to convert these data into transient vertical absolute height data, which are then combined with fixed plane projection coordinates to generate the coordinates of three transient three-dimensional spatial points.

[0034] A normal vector is generated by performing an outer product operation on the first spatial vector from the first transient 3D spatial point to the second transient 3D spatial point and the second spatial vector from the first transient 3D spatial point to the third transient 3D spatial point.

[0035] Extract the three-dimensional coordinate components of the normal vector, and calculate the instantaneous tilt angle of the button surface around the X-axis and the instantaneous tilt angle around the Y-axis in real time using the arctangent trigonometric function;

[0036] The instantaneous tilt angle tangent around the X-axis is multiplied by the kinematic decoupling coefficient to obtain the planar compensation displacement in the Y-axis direction. The instantaneous tilt angle tangent around the Y-axis is multiplied by the kinematic decoupling coefficient to obtain the planar compensation displacement in the X-axis direction. The planar compensation displacement in the X-axis direction and the planar compensation displacement in the Y-axis direction are used as dynamic position commands and sent to the CNC platform in real time.

[0037] By adopting the above technical solution, the normal vector of the pressure surface is calculated and generated in each control cycle by using the principle of spatial vector cross product. The instantaneous tilt angle is calculated by extracting the normal vector component and combining it with the arctangent function. This is then converted into a horizontal displacement command and sent to the platform. This allows the pressure head to move instantly when the button is tilted due to internal force, preventing the pressure head from sliding laterally with the switch button and ensuring that the loading force is always transmitted along the true normal direction of the button surface.

[0038] Preferably, the method of recording force data and absolute displacement data to derive the effective stroke specifically includes:

[0039] The control unit monitors the changes in the normal force data in real time and locks the initial absolute displacement data when the normal force data reaches the preset contact force threshold.

[0040] When the normal force data reaches the target trigger force value, the current target absolute displacement data is locked.

[0041] The effective travel of the EPB switch button under test is obtained by subtracting the initial force absolute displacement data from the current target absolute displacement data.

[0042] By adopting the above technical solution, the true contact start point between the actuator and the product surface is determined by relying on the normal force data to reach the contact threshold. When the target trigger force value is reached, the position of the motion endpoint is extracted. By subtracting the two, the invalid stroke during the no-load descent phase of the pressure head is eliminated. The basis for determining the stroke measurement start point is changed from a simple fixed position to the actual force data, thereby reducing the zero-point drift problem of stroke testing caused by the overall thickness dimensional tolerance of the product.

[0043] Preferably, extracting the time difference between the internal metal contact disconnection moment and the external button release moment and calculating the time-domain phase difference specifically includes:

[0044] Map the unloading time series data of normal force value and the rebound time series data of center height to the same time axis for alignment.

[0045] After alignment, a first-order backward difference operation is performed to generate the normal force unloading rate of change sequence and the center height springback rate of change sequence, respectively.

[0046] The first timestamp corresponding to the moment when the number of consecutive sampling points whose rate of change of normal unloading force is continuously greater than or equal to the preset force value change judgment threshold reaches the consecutive trigger point number threshold is used as the moment when the internal metal contact is disconnected.

[0047] The first timestamp corresponding to the moment when the number of consecutive sampling points whose center height rebound rate is continuously greater than or equal to the preset height change judgment threshold reaches the consecutive trigger point number threshold is used as the external button release time.

[0048] The time-domain phase difference data is generated by calculating the difference between the moment when the internal metal contact is disconnected and the moment when the external button is released.

[0049] By adopting the above technical solution, the acquired force value and displacement curve are mapped to the same time axis for alignment, and the numerical change points of the two signals with time are captured by first-order differential operation. On this basis, high-frequency sensing noise is eliminated by using a set threshold for the number of continuous trigger points, so as to accurately extract the timing breakpoints of internal contact disconnection and external mechanism rebound. The time-domain phase difference data obtained in this way objectively quantifies the asynchronous response defects caused by aging, hysteresis or increased friction damping of the guide groove of the internal spring element of the button.

[0050] Preferably, inputting the 3D feature input vector into a pre-deployed multilayer perceptron evaluation model to perform forward propagation computation specifically includes:

[0051] Extract the preset maximum and minimum normalization boundary parameters and perform linear mapping on the 3D feature input vector to generate a standardized feature input vector with a mapping interval between 0 and 1.

[0052] The multilayer perceptron evaluation model consists of a fully connected input layer, a first hidden layer, a second hidden layer, and an output layer.

[0053] The standardized feature input vector is input into the input layer. After multiplying it with the connection weight matrix and adding the bias vector, the first hidden layer output feature matrix and the second hidden layer output feature matrix are obtained by activating the linear rectified function.

[0054] The normalized exponential function is used to calculate a 3-dimensional classification probability vector in the output layer. The classification index corresponding to the probability element with the largest value is extracted to determine the qualified status of the EPB switch under test.

[0055] By adopting the above technical solution, the input features are pre-processed with linear mapping to eliminate the computational interference caused by the different dimensions of surface difference, travel and time features. The perceptual network with multiple hidden layers and activation functions are used to comprehensively calculate the multi-dimensional features. Finally, the output layer calculates the probability of belonging to different categories, which changes the traditional method of simply relying on manually setting multiple hard judgment boundaries and improves the ability to identify defective products on the verge of being qualified.

[0056] Preferably, the multilayer perceptron evaluation model is generated offline through the following steps:

[0057] Historical measured data is extracted to construct input features for the training sample set. A real supervision label set containing qualified labels, assembly tolerance exceedance labels, and mechanical structure failure labels, annotated by manual quality inspection experts, is introduced.

[0058] The cross-entropy loss function is used as the objective function, and an adaptive moment estimation optimizer is configured to perform error backpropagation and parameter weight update iteration on the training sample set.

[0059] When the value of the cross-entropy loss function decreases by less than a preset threshold for consecutive preset iteration cycles, training is stopped and the multilayer perceptron evaluation model parameters are exported and deployed to the control unit.

[0060] By adopting the above technical solution, historical measured data labeled with specific state tags is used to drive model training. The cross-entropy loss function is used in combination with an adaptive moment estimation optimizer to update parameters. The stopping strategy set during training avoids the network parameters from getting stuck in the overfitting state of a specific batch of samples, ensuring that the exported model parameter network has good generalization and analytical ability for future unknown batches of test data, and maintaining the judgment stability of the detection equipment in long-term use.

[0061] This invention provides a test bench and control system for measuring the travel and force of an automotive EPB switch button. It offers the following advantages:

[0062] 1. This invention uses three laser sensors to continuously acquire discrete relative distance data to construct the normal vector of the pressure surface, and converts the calculated instantaneous tilt angle into a planar compensation displacement command to drive the CNC platform to translate synchronously. The above displacement compensation method enables the detection mechanism to adjust its horizontal position according to the slight tilt of the tested car EPB switch button during the pressing process, thereby reducing the lateral friction force component generated by the relative sliding between the force sensor head and the surface of the car EPB switch button, and improving the accuracy of the normal pressing pressure test data.

[0063] 2. This invention aligns the normal force unloading sequence with the center height rebound sequence on the time axis, extracts the first timestamp of the sudden change in force and height using first-order backward differential operation, and calculates the time-domain phase difference between the moment the internal metal contact is disconnected and the moment the external button is lifted. The above method of calculating the time-domain phase difference separates the disconnection time of the internal electrical components and the rebound time of the external mechanical components, which are difficult to measure directly. This allows for the quantitative reflection of the mechanical response hysteresis caused by wear of internal transmission components or assembly gaps without disassembling the automotive EPB switch.

[0064] 3. This invention combines the actual surface difference dimensions, effective travel, and temporal phase difference obtained from the test into a feature vector, and inputs it into a multilayer perceptron evaluation model with a nonlinear activation function for calculation and analysis. The above data processing method integrates test data from three dimensions: spatial dimensions, dynamic displacement, and temporal response. It overcomes the shortcomings of single threshold judgment that may miss complex defects, and can directly classify and output the judgment results of qualified products, defective products with excessive assembly tolerances, or defective products with mechanical structure failures, thereby improving the comprehensiveness of automotive EPB switch quality judgment. Attached Figure Description

[0065] Figure 1 This is a perspective view of the present invention;

[0066] Figure 2 This is a schematic diagram of the system architecture of the present invention;

[0067] Figure 3 This is a flowchart of the method of the present invention;

[0068] Figure 4 This is a schematic diagram illustrating the construction of the absolute coordinate system and parameter initialization of the test bench according to the present invention;

[0069] Figure 5 This is a schematic diagram illustrating the principle of laser array scanning and geometric center extraction of the present invention.

[0070] Figure 6This is a schematic diagram illustrating the principle of non-contact measurement of surface difference dimensions according to the present invention.

[0071] Figure 7 This is a schematic diagram illustrating the dynamic follow-up pressure control and stroke extraction principle of the present invention.

[0072] Figure 8 This is a diagnostic diagram of the time-domain phase difference response of the heterogeneous sensor of the present invention;

[0073] Figure 9 This is a topology diagram of the multidimensional quality assessment model based on a multilayer perceptron of the present invention.

[0074] Figure 10 This is a scatter plot of the three-dimensional feature distribution of the present invention;

[0075] Figure 11 This is a classification confusion matrix diagram for the present invention.

[0076] Among them, 10 is the base; 20 is the CNC platform; 30 is the servo component; 40 is the detection mechanism; 50 is the control unit; 60 is the tooling fixture; 41 is the force sensor; and 42 is the laser sensor. Detailed Implementation

[0077] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0078] See attached document Figure 1 and Figure 2 The present invention provides an automotive EPB switch test bench, comprising: a base 10, a CNC platform 20, a servo component 30, a testing mechanism 40, a control unit 50, and a tooling fixture 60;

[0079] The base 10 provides a fixed reference surface. The tooling fixture 60 is installed on the base 10 to hold and fix the EPB switch to be tested. The CNC platform 20 is installed on the base 10. The CNC platform 20 includes an X-axis drive assembly and a Y-axis drive assembly. The servo assembly 30 is installed at the end of the CNC platform 20 to provide Z-axis lifting drive. The detection mechanism 40 is fixed to the output end of the servo assembly 30. The control unit 50 is connected to the CNC platform 20, the servo assembly 30 and the detection mechanism 40 respectively.

[0080] The detection mechanism 40 includes a force sensor 41 and three laser sensors 42. The force sensor 41 is vertically mounted on the central Z-axis of the detection mechanism 40. The lower end of the force sensor 41 is connected to a pressure head. The three laser sensors 42 are distributed in a 120° circle around the pressure head and are installed at an angle to the central Z-axis. When the force sensor 41 presses the surface of the EPB switch button to be tested, the light emitted by the three laser sensors 42 avoids the pressure head and irradiates the button surface outside the pressing area.

[0081] See attached document Figure 3 This invention provides a comprehensive testing method for automotive EPB switches, the method comprising the following steps:

[0082] S10, after the EPB switch under test is fixed to the base 10 by the tooling fixture 60, the control unit 50 drives the CNC platform 20 and the servo component 30 to move to the mechanical zero position to establish an absolute coordinate system, and loads the preset calibration parameters of the corresponding EPB switch model.

[0083] S20, the control unit 50 controls the CNC platform 20 to move within the preset search area, controls the three laser sensors 42 to collect discrete height data, calculates the actual geometric center coordinates of the EPB switch button under test based on the discrete height data, and drives the CNC platform 20 to translate and align the central axis of the force sensor 41 to the actual geometric center coordinates.

[0084] S30, at the actual geometric center coordinates, the control unit 50 obtains the relative height of the EPB switch button under test and the relative height of the product reference surface through three laser sensors 42, and calculates the actual surface difference dimension of the product under test based on the height difference;

[0085] S40, the servo component 30 drives the detection mechanism 40 to press down along the Z-axis. The force sensor 41 contacts and applies a normal force to the EPB switch button under test. During the pressing, three laser sensors 42 continuously collect surface height data. The control unit 50 fits the transient plane of the pressed surface and extracts the tilt angle parameter. The tilt angle parameter is converted into a plane compensation displacement command and sent to the CNC platform 20 to drive the detection mechanism 40 to translate along the X and Y directions. The force data of the force sensor 41 and the absolute displacement data of the servo component 30 are recorded simultaneously. When the force data reaches the preset contact force threshold, the initial force absolute displacement data is locked. When the force data reaches the target trigger force value, the current absolute displacement data is locked and the difference is calculated to obtain the effective stroke.

[0086] S50, servo component 30 drives detection mechanism 40 to reverse and reset, control unit 50 synchronously records force value unloading curve of force sensor 41 and center height rebound curve obtained by three laser sensors 42, processes the two curves to extract the mechanical disconnection time of internal metal contact and the physical release time of external button, and calculates the time domain phase difference between the two times.

[0087] S60, the control unit 50 combines the actual surface difference size, effective stroke and time domain phase difference to construct a three-dimensional feature input vector, inputs the three-dimensional feature input vector into a pre-trained and deployed multilayer perceptron evaluation model to perform forward propagation calculation, determines the qualified status of the EPB switch under test based on the classification result output by the multilayer perceptron evaluation model and outputs a judgment signal.

[0088] See attached document Figure 4 The EPB switch to be tested is placed and locked inside the fixture 60. The fixture 60 is fixed to the upper surface of the base 10 and fits the outline of the EPB switch to be tested. Specifically, the mechanical locking and clamping structure of the fixture 60 can be achieved by a cylinder-driven linkage mechanism or a manual quick clamp.

[0089] The control unit 50 sends a home reset command to the CNC platform 20 and the servo component 30. The CNC platform 20 contains drive motors and displacement feedback components for the X and Y axes, and the servo component 30 contains a servo motor and linear transmission mechanism for the Z axis. The X-axis component of the CNC platform 20, the Y-axis component of the CNC platform 20, and the Z-axis component of the servo component 30 are moved under control until the limit sensors at the axial ends of the CNC platform 20 and the servo component 30 are triggered to generate a physical zeroing signal.

[0090] After receiving the physical zeroing signals of all three axes, the control unit 50 establishes a three-dimensional absolute coordinate system O−XYZ of the device with the physical origin position of the current execution component as the spatial coordinate origin. In the three-dimensional absolute coordinate system O−XYZ, the plane containing the X-axis and Y-axis is parallel to the reference horizontal plane of the base 10, and the Z-axis is perpendicular to the reference horizontal plane of the base 10 and faces directly upward. The three-dimensional spatial motion trajectory data of the detection mechanism 40 and the position feedback data of each sensor are all uniformly converted into the three-dimensional absolute coordinate system O−XYZ for subsequent mathematical calculation.

[0091] Control unit 50 retrieves and loads a preset calibration parameter matrix that matches the specific model of the EPB switch currently under test from its internal storage module. The control unit 50 pre-acquires and stores a preset calibration parameter matrix by reading the 3D digital model of the product design drawings or by pre-testing the standard EPB switch to be tested. Preset calibration parameter matrix It includes the physical control parameters and spatial reference base required for the testing process.

[0092] Preset calibration parameter matrix The specific mathematical expression is:

[0093] ;

[0094] In the formula, This is the preset calibration parameter matrix; The reference X-axis coordinate value is the theoretical geometric center of the EPB switch button surface under test in the three-dimensional absolute coordinate system O−XYZ. The value range is set between 100 and 500 mm depending on the installation position of the tooling fixture 60; The reference Y-axis coordinate value is the theoretical geometric center of the EPB switch button surface under test in the three-dimensional absolute coordinate system O−XYZ. The value range is set between 100 and 500 millimeters; The target trigger force value required for the downward pressure test. The value range is set to 15 to 30 Newtons according to the specifications of the EPB switch under test; This refers to the offset of the product's reference plane coordinates. The value range is determined based on the actual measured three-dimensional dimensions of the EPB switch housing under test and is set between 20 and 50 mm.

[0095] See attached document Figure 5 The control unit 50 extracts the reference X-axis coordinate value and reference Y-axis coordinate value from the preset calibration parameter matrix. The control unit 50 drives the CNC platform 20 to move and positions the center Z-axis of the detection mechanism 40 to the initial upper position determined by the reference X-axis coordinate value and reference Y-axis coordinate value. The control unit 50 sets the boundary length and boundary width of the two-dimensional rectangular search area according to the external dimension parameters of the EPB switch to be tested. The boundary length of the two-dimensional rectangular search area is set to a range of 10 to 30 mm, and the boundary width of the two-dimensional rectangular search area is set to a range of 10 to 30 mm. The control unit 50 sets the gridded scanning step size of the CNC platform 20 in the two-dimensional rectangular search area. The gridded scanning step size is set to a range of 0.1 to 0.5 mm.

[0096] The control unit 50 determines the total number of scanning nodes in the X-axis direction based on the ratio of the boundary length of the two-dimensional rectangular search area to the gridded scanning step size. The control unit 50 determines the total number of scanning nodes in the Y-axis direction based on the ratio of the boundary width of the two-dimensional rectangular search area to the gridded scanning step size. The control unit 50 drives the CNC platform 20 to perform line-by-line stepping motion within the two-dimensional rectangular search area according to the gridded scanning step size. When the CNC platform 20 moves to each grid node position, the control unit 50 triggers the three laser sensors 42 on the periphery of the detection mechanism 40 to synchronously emit laser beams and receive reflected light signals. The three laser sensors 42 respectively collect the corresponding local surface relative tilt distance data.

[0097] The control unit 50 extracts the physical installation tilt angles of the three laser sensors 42. The control unit 50 multiplies the relative tilt distance data of the local surface by the cosine value of the physical installation tilt angle to convert it into vertical absolute height data. The control unit 50 takes the average value of the vertical absolute height data converted by the three laser sensors 42 at the same grid node position to generate fused height data. The control unit 50 constructs a discrete height data matrix based on the spatial coordinate mapping relationship of the grid nodes after traversing the entire two-dimensional rectangular search area and all the fused height data generated.

[0098] The control unit 50 calls the discrete height data matrix and calculates the actual geometric center coordinates of the EPB switch button under test according to the discrete weighted centroid algorithm. The calculation models of the X-axis and Y-axis components of the actual geometric center coordinates are expressed by discrete weighted summation formulas. The formula for the X-axis coordinate of the actual geometric center is:

[0099] ;

[0100] In the formula, The X-axis coordinate of the actual geometric center; This is the index number of the scan node in the X-axis direction;

[0101] This represents the total number of scan nodes in the X-axis direction. This is the index number of the scan node in the Y-axis direction; This represents the total number of scan nodes in the Y-axis direction. For the first Discrete X-axis coordinates of each scan node in the three-dimensional absolute coordinate system; For the first Discrete Y-axis coordinates of each scan node in the three-dimensional absolute coordinate system; coordinates The fusion height data corresponding to the location.

[0102] The formula for the Y-axis coordinate of the actual geometric center is:

[0103] ;

[0104] In the formula, The Y-axis coordinate of the actual geometric center; This is the index number of the scan node in the X-axis direction; This represents the total number of scan nodes in the X-axis direction. This is the index number of the scan node in the Y-axis direction; This represents the total number of scan nodes in the Y-axis direction. For the first Discrete X-axis coordinates of each scan node in the three-dimensional absolute coordinate system; For the first Discrete Y-axis coordinates of each scan node in the three-dimensional absolute coordinate system; coordinates The fusion height data corresponding to the location.

[0105] After calculating and obtaining the X-axis coordinates and Y-axis coordinates of the actual geometric center, the control unit 50 performs a difference calculation between the X-axis coordinates of the actual geometric center and the reference X-axis coordinates to obtain the X-axis compensation amount. The control unit 50 also performs a difference calculation between the Y-axis coordinates of the actual geometric center and the reference Y-axis coordinates to obtain the Y-axis compensation amount. Based on the X-axis compensation amount and the Y-axis compensation amount, the control unit 50 generates a coordinate compensation vector command and sends the coordinate compensation vector command to the CNC platform 20. The CNC platform 20 drives the detection mechanism 40 to translate the corresponding compensation distance in the horizontal plane according to the coordinate compensation vector command. The control unit 50 determines that the central axis of the force sensor 41 in the detection mechanism 40 is precisely aligned with the actual geometric center coordinate position of the EPB switch button to be tested. After completing the central axis alignment action, the control unit 50 locks the current spatial coordinates of the detection mechanism 40 and updates them as the starting working reference point for subsequent vertical pressure tests.

[0106] See attached document Figure 6 The control unit 50 keeps the center Z-axis of the detection mechanism 40 at the actual geometric center coordinates. The control unit 50 triggers three laser sensors 42 to acquire the local relative tilt distance of the EPB switch button surface to be tested. The control unit 50 extracts the physical installation tilt angle of the three laser sensors 42. The value range of the physical installation tilt angle is set between 30 degrees and 60 degrees to construct an outgoing light path that avoids the pressure head in the central area. The control unit 50 uses the cosine value of the trigonometric function to convert the local relative tilt distance into the vertical absolute height measurement value of the three button surfaces.

[0107] Three laser sensors 42 are evenly distributed in a 120-degree circle around the pressure head. The projections of the three laser sensors 42 on the horizontal plane form an equilateral triangle. The geometric center of the equilateral triangle coincides with the actual geometric center coordinates. The control unit 50 uses the vertical absolute height measurement values ​​of the three button surfaces to fit and generate the initial spatial plane equation of the pressure surface of the EPB switch button to be measured in a three-dimensional absolute coordinate system. Based on the coincidence of the centroid and geometric center of the equilateral triangle and the geometric characteristics of the linear gradual change of the spatial plane height, the control unit 50 calculates the arithmetic mean of the vertical absolute height measurement values ​​of the three button surfaces to directly calculate the relative height of the button center at the actual geometric center coordinates of the initial spatial plane.

[0108] The formula for calculating the relative height of the button center is:

[0109] ;

[0110] In the formula, The relative height of the button center; For the number index of laser sensor 42; For the first The vertical absolute height of the button surface is measured and converted by a laser sensor 42.

[0111] The control unit 50 extracts the product reference surface coordinate offset based on the preset calibration parameter matrix. The product reference surface coordinate offset is determined based on the measured three-dimensional dimensions of the EPB switch housing under test and the value range is set between 20 and 50 mm. The control unit 50 drives the CNC platform 20 to move the detection mechanism 40 to above the fixed reference surface of the EPB switch housing under test based on the product reference surface coordinate offset. The control unit 50 triggers the three laser sensors 42 again above the fixed reference surface to obtain the local relative tilt distance of the fixed reference surface of the EPB switch housing under test.

[0112] The control unit 50 converts the local relative tilt distance of the fixed reference surface multiplied by the cosine of the physical installation tilt angle into the reference surface vertical absolute height measurement value according to the same trigonometric function conversion logic. The control unit 50 calculates the relative height of the reference surface center by taking the arithmetic mean of the three reference surface vertical absolute height measurement values. The control unit 50 calls the relative height of the reference surface center and the relative height of the button center to perform the difference calculation to obtain the actual surface difference dimension of the EPB switch surface to be measured.

[0113] The formula for calculating the actual surface difference is:

[0114] ;

[0115] In the formula, This refers to the actual surface difference dimensions; The relative height of the center of the reference plane. The relative height of the button's center.

[0116] Temperature drift compensation and sensor nonlinear error calibration in surface dimension measurement can be achieved using conventional lookup table compensation methods or polynomial fitting compensation algorithms.

[0117] See attached document Figure 7The control unit 50 controls the detection mechanism 40 to move to the initial safe height corresponding to the actual geometric center coordinates. The value range of the initial safe height is set to 2 to 5 mm above the surface of the EPB switch button to be tested. The control unit 50 sends a constant speed pressing command to the servo component 30. The value range of the pressing speed corresponding to the constant speed pressing command is set to 1 to 10 mm per second. The servo component 30 drives the detection mechanism 40 to move downward along the Z-axis. The pressure head at the lower end of the force sensor 41 at the center of the detection mechanism 40 makes physical contact with the surface of the EPB switch button to be tested and continuously applies a normal force to the surface of the EPB switch button to be tested.

[0118] During the dynamic downward pressure cycle in which the normal force continues to increase, the control unit 50 triggers three laser sensors 42 to continuously acquire discrete relative distance data sequences on the button surface at a preset high-frequency sampling rate. The preset high-frequency sampling rate is set to a range of 1000 to 5000 Hz. The control unit 50 uses trigonometric functions to convert the discrete relative distance data sequences into transient vertical absolute height data of three measurement points in real time. The control unit 50 extracts the fixed plane projection coordinates of the three measurement points in the three-dimensional absolute coordinate system. The control unit 50 combines the fixed plane projection coordinates with the transient vertical absolute height data to generate three transient three-dimensional spatial point coordinates.

[0119] Three transient three-dimensional spatial point coordinates are set to correspond to the first spatial point, the second spatial point, and the third spatial point, respectively. The control unit 50 constructs a first spatial vector with the first spatial point as the starting point and the second spatial point as the ending point. The control unit 50 constructs a second spatial vector with the first spatial point as the starting point and the third spatial point as the ending point. The control unit 50 generates a normal vector perpendicular to the pressure surface by performing an outer product operation on the first spatial vector and the second spatial vector. The control unit 50 extracts the X-axis component, Y-axis component, and Z-axis component of the normal vector. Based on the X-axis component, Y-axis component, and Z-axis component of the normal vector, the control unit 50 calculates the instantaneous tilt angle of the button surface around the X-axis and the instantaneous tilt angle of the button surface around the Y-axis in real time.

[0120] The formula for the instantaneous tilt angle about the X-axis is:

[0121] ;

[0122] In the formula, For a moment The instantaneous tilt angle of the button surface around the X-axis; For continuous time variables; For a moment The Y-axis component of the transient space plane normal vector; For a moment Z-axis component of the transient space plane normal vector; This is the operator for the arctangent trigonometric function.

[0123] ;

[0124] In the formula, For a moment The instantaneous tilt angle of the button surface around the Y-axis; For continuous time variables; For a moment The X-axis component of the transient space plane normal vector; For a moment Z-axis component of the transient space plane normal vector; This is the operator for the arctangent trigonometric function.

[0125] During the button press, the surface tilts and causes a shift in the normal direction. The control unit 50 retrieves the kinematic decoupling coefficient of the detection mechanism 40 stored internally. The kinematic decoupling coefficient is generated by calibrating the vertical distance from the end of the force sensor 41 pressure head to the physical rotation center of the detection mechanism 40. The range of the kinematic decoupling coefficient is set to 50 to 150 mm. The control unit 50 multiplies the tangent of the instantaneous tilt angle around the X-axis with the kinematic decoupling coefficient to obtain the planar compensation displacement in the Y-axis direction. The control unit 50 multiplies the tangent of the instantaneous tilt angle around the Y-axis with the kinematic decoupling coefficient to obtain the planar compensation displacement in the X-axis direction.

[0126] The formulas for the planar compensation displacement in the X-axis direction and the planar compensation displacement in the Y-axis direction are as follows:

[0127] ;

[0128] ;

[0129] In the formula, For a moment The planar compensation displacement in the X-axis direction; For a moment The planar compensation displacement in the Y-axis direction; For continuous time variables; The kinematic decoupling coefficient of the testing institution is 40. For a moment The instantaneous tilt angle of the button surface around the Y-axis; For a moment The instantaneous tilt angle of the button surface around the X-axis; This is the operator for the tangent trigonometric function.

[0130] The control unit 50 sends the planar compensation displacement in the X-axis direction and the planar compensation displacement in the Y-axis direction as dynamic position commands to the CNC platform 20 in real time. The CNC platform 20 drives the detection mechanism 40 to perform synchronous translation and follow-up in the horizontal plane according to the dynamic position commands. During the entire motion cycle of the servo component 30 driving the detection mechanism 40 to press down, the control unit 50 activates a high-frequency hardware timer. The control unit 50 synchronously records the normal force data output by the force sensor 41 and the absolute displacement data fed back by the encoder inside the servo component 30 according to a unified timestamp.

[0131] The control unit 50 monitors the changes in the normal force data in real time. When the normal force data reaches the preset contact force threshold, the control unit 50 locks the initial force timestamp. The contact force threshold is set to a range of 0.1 to 0.5 Newtons. When the normal force data reaches the target trigger force value in the preset calibration parameter matrix, the control unit 50 locks the current target timestamp. The control unit 50 extracts the absolute displacement data corresponding to the target timestamp and performs a difference calculation between the absolute displacement data corresponding to the target timestamp and the absolute displacement data corresponding to the initial force timestamp to obtain the effective stroke of the EPB switch button under test.

[0132] The formula for calculating the effective travel distance is:

[0133] ;

[0134] In the formula, The effective travel of the EPB switch button under test; The target timestamp is the time stamp corresponding to the normal force data reaching the target trigger force value; This is the initial force timestamp corresponding to when the normal force data reaches the contact force threshold. target timestamp The corresponding absolute displacement data; Initial force timestamp The corresponding absolute displacement data.

[0135] See attached document Figure 8 After the normal force data reaches the target trigger force value, the control unit 50 sends a reverse unloading command to the servo component 30. The servo component 30 drives the detection mechanism 40 to move upward along the Z-axis to perform the unloading action. The control unit 50 activates the dual-channel high-speed data acquisition module. The control unit 50 synchronously records the normal force value unloading time series data output by the force sensor 41 and the center height rebound time series data jointly calculated by the three laser sensors 42 based on the same system hardware clock source. The control unit 50 maps the normal force value unloading time series data and the center height rebound time series data to the same time axis for data alignment processing.

[0136] The control unit 50 performs a first-order backward difference operation on the aligned normal force unloading time series data to generate a normal force unloading rate of change sequence. The control unit 50 performs a first-order backward difference operation on the aligned center height rebound time series data to generate a center height rebound rate of change sequence. The control unit 50 extracts the internally stored force value mutation judgment threshold and height mutation judgment threshold. The force value mutation judgment threshold and height mutation judgment threshold are obtained by offline calibration based on the rebound mechanical test curve of the standard qualified EPB switch under test. The value range of the force value mutation judgment threshold is set to 10 to 50 Newtons per second, and the value range of the height mutation judgment threshold is set to 5 to 20 millimeters per second.

[0137] The control unit 50 sets a threshold for the number of consecutive trigger points. The threshold range is set to 3 to 5 discrete sampling periods. The control unit 50 compares the continuous values ​​in the normal force unloading rate of change sequence with the force value mutation judgment threshold. The control unit 50 extracts the first timestamp corresponding to the number of consecutive sampling points where the normal force unloading rate of change is continuously greater than or equal to the force value mutation judgment threshold, and uses it as the moment when the internal metal contact is disconnected. The moment when the internal metal contact is disconnected maps to the instant when the internal electrical connection of the EPB switch under test is physically disconnected.

[0138] The control unit 50 compares the continuous values ​​in the center height rebound rate sequence with the height change judgment threshold in turn. The control unit 50 extracts the first timestamp corresponding to the number of continuous sampling points where the center height rebound rate is continuously greater than or equal to the height change judgment threshold and reaches the number of continuous trigger points as the external button release time. The external button release time maps to the physical moment when the external structural component of the EPB switch under test overcomes static friction and begins to move upward. The control unit 50 calls the internal metal contact disconnection time and the external button release time to perform timing difference calculation.

[0139] The control unit 50 calculates the difference between the time when the internal metal contact is opened and the time when the external button is lifted to generate time-domain phase difference data. The time-domain phase difference data objectively reflects the physical state of response lag caused by the internal mechanical assembly tolerance of the EPB switch under test.

[0140] The formula for calculating the time-domain phase difference data is:

[0141] ;

[0142] In the formula, This is time-domain phase difference data; This is the moment when the internal metal contacts disconnect; The moment the external button is released.

[0143] For the specific algorithm implementation of first-order backward difference operation on discrete time series data, conventional numerical analysis difference algorithm or Kalman filter smoothing derivative algorithm can be used to implement the first-order backward difference operation.

[0144] See attached document Figure 9 The control unit 50 extracts the actual surface difference size, effective travel, and temporal phase difference data obtained from the aforementioned steps. The control unit 50 combines the actual surface difference size, effective travel, and temporal phase difference data to construct a three-dimensional feature input vector with a dimension of 3. The control unit 50 extracts the preset maximum and minimum normalized boundary parameters and performs linear mapping processing on the three-dimensional feature input vector to generate a standardized feature input vector.

[0145] The maximum and minimum normalized boundary parameters are determined offline by statistically analyzing the extreme values ​​of feature data from historical production batches of the EPB switch under test. The maximum and minimum normalized boundary parameters include the extreme values ​​of the actual surface difference size boundary, the extreme value of the effective travel boundary, and the extreme value of the time-domain phase difference boundary. The extreme value of the actual surface difference size boundary is set to a range of -1.0 to 1.0 mm, the extreme value of the effective travel boundary is set to a range of 0 to 5.0 mm, and the extreme value of the time-domain phase difference boundary is set to a range of 0 to 100 milliseconds. The mapping interval of the normalized feature input vector is set to 0 to 1. The control unit 50 inputs the normalized feature input vector into the multilayer perceptron evaluation model pre-deployed in the system's physical memory to perform forward propagation calculation.

[0146] The multilayer perceptron evaluation model consists of a fully connected input layer and a first hidden layer, as well as a second hidden layer and an output layer. The input layer contains 3 neurons and receives a standardized feature input vector. The first hidden layer contains 64 neurons and the second hidden layer contains 32 neurons. Both the first and second hidden layers use a linear rectified function as a nonlinear activation function. The output layer contains 3 neurons and uses a normalized exponential function as an activation function. The output layer outputs a 3-dimensional classification probability vector. The three elements in the classification probability vector sequentially map the business physical probability values ​​of the tested EPB switch being in a qualified state, an assembly tolerance exceeding the limit state, and a mechanical structure failure state.

[0147] The control unit 50 calculates the node output states of each hidden layer sequentially according to the forward propagation logic of the multilayer perceptron evaluation model. The matrix operation formula for the first hidden layer is:

[0148] ;

[0149] In the formula, This is the output feature matrix of the first hidden layer; It is a linear rectification activation function; This is the connection weight matrix from the input layer to the first hidden layer; The input vector is a standardized feature vector; This is the bias vector for the first hidden layer.

[0150] After calculating and obtaining the output feature matrix of the first hidden layer, the control unit 50 passes the output feature matrix of the first hidden layer as input to the second hidden layer for forward propagation calculation. The matrix operation formula of the second hidden layer is as follows:

[0151] ;

[0152] In the formula, This is the output feature matrix of the second hidden layer; It is a linear rectification activation function; This is the connection weight matrix from the first hidden layer to the second hidden layer; This is the output feature matrix of the first hidden layer; This is the bias vector for the second hidden layer.

[0153] After obtaining the output feature matrix of the second hidden layer, the control unit 50 performs the output layer classification calculation of the multilayer perceptron evaluation model. The output layer calculation formula of the multilayer perceptron evaluation model is as follows:

[0154] ;

[0155] In the formula, This is the classification probability vector output by the multilayer perceptron evaluation model; The normalized exponential activation function; This is the connection weight matrix from the second hidden layer to the output layer; This is the output feature matrix of the second hidden layer; This is the bias vector for the output layer.

[0156] Before deploying the multilayer perceptron evaluation model to the control unit 50, the R&D terminal extracts a pre-established historical detection database of EPB switches as a training sample set. The input features in the training sample set are derived from historical measured data records. The supervision labels in the training sample set consist of qualified labels, assembly tolerance exceedance labels, and mechanical structure failure labels marked by human quality inspection experts. The R&D terminal uses the cross-entropy loss function as the objective function for model training. The R&D terminal configures an adaptive moment estimation optimizer to perform error backpropagation update iterations on the training sample set with an initial learning rate of 0.001. The R&D terminal stops training and derives the parameter network of the multilayer perceptron evaluation model when the cross-entropy loss function value decreases by less than 0.0001 for 10 consecutive iterations.

[0157] The formula for calculating the cross-entropy loss function used during model training is as follows:

[0158] ;

[0159] In the formula, This is the scalar error value output by the cross-entropy loss function; The index number of the output layer category; For the corresponding categories in the training sample set The actual supervised label value, which takes the value of 0 or 1; The classification probability vector output by the forward propagation of the multilayer perceptron evaluation model. Medium category The corresponding predicted probability value.

[0160] The control unit 50 extracts the classification index corresponding to the probability element with the largest value in the classification probability vector. Based on the classification index, the control unit 50 determines the EPB switch under test as a qualified product, a defective product with excessive assembly tolerance, or a defective product with mechanical structure failure. The control unit 50 outputs a comprehensive inspection report with corresponding quality appraisal labels to the external host computer. For the adaptive moment estimation optimization algorithm and error backpropagation differentiation operation involved in the multilayer perceptron evaluation model, the standard function interface built into the open source deep learning framework can be called to implement it.

[0161] To aid in understanding the technical solution of this invention, the following provides a comprehensive testing application example based on a mass-produced automotive EPB switch.

[0162] This application example selects 150 automotive EPB switches from the same historical production batch as the standard test sample. Among the 150 automotive EPB switches, there are 105 qualified products identified by manual disassembly and verification, 25 defective products with assembly tolerance exceeding the limit, and 20 defective products with mechanical structure failure. The ambient temperature in the test workshop is kept constant at 23°C. The test equipment is assembled and fixed according to the aforementioned mechanical topology of the base, CNC platform, and testing mechanism, and the hardware bus is wired. The multilayer perceptron evaluation model parameter network, which is exported and compiled from an open-source deep learning framework, is pre-programmed into the memory of the control unit.

[0163] The equipment operator places 150 test samples sequentially into the tooling fixture above the base and triggers the pneumatic mechanism to lock them. The control unit drives the CNC platform and servo components to perform the origin limit reset action and establish a physical three-dimensional absolute coordinate system.

[0164] The control unit calls the preset calibration parameter matrix of the corresponding EPB switch in the internal non-volatile memory. The control unit controls the laser sensor array to perform gridded scanning in the two-dimensional rectangular search area. The control unit solves the discrete height data matrix and locates the actual geometric center coordinates of each test sample. The servo component drives the detection mechanism to perform vertical constant speed pressing and reverse reset unloading actions in sequence at the actual geometric center position.

[0165] During the pressure test cycle, the control unit synchronously records the time-series discrete sampling data returned by the force sensor and three laser sensors. The control unit extracts and calculates the actual surface difference size, effective stroke, and time-domain phase difference data of each corresponding test sample.

[0166] The control unit combines the actual surface difference dimensions, effective travel, and temporal phase difference data into a three-dimensional feature input vector. The multilayer perceptron evaluation model receives the three-dimensional feature input vector and propagates it forward to output a classification probability vector containing three floating-point numbers. The control unit outputs the corresponding predicted state identification label based on the maximum probability index of the classification probability vector. The host computer software automatically summarizes the measured three-dimensional feature data of 150 test samples and the model prediction results and generates a comprehensive experimental sampling data table.

[0167] Table 1: Sampling Data of Comprehensive Testing Experiment for Automotive EPB Switches

[0168] #003 0.47 2.13 23.6 qualified products qualified products #012 0.52 2.08 25.1 qualified products qualified products #045 -0.12 1.95 21.4 qualified products qualified products #067 1.24 2.65 41.3 Assembly tolerance exceeded Assembly tolerance exceeded #078 1.18 2.51 38.9 Assembly tolerance exceeded Assembly tolerance exceeded #092 0.41 4.31 86.5 Mechanical structure failure Mechanical structure failure #114 0.38 4.15 92.7 Mechanical structure failure Mechanical structure failure #138 1.05 2.44 28.5 Assembly tolerance exceeded qualified products #142 0.49 4.56 77.2 Mechanical structure failure Mechanical structure failure #149 0.58 2.21 24.8 qualified products qualified products

[0169] Table 2: Comparison of the effects of traditional detection methods and the multidimensional fusion detection method of the present invention

[0170] Overall state recognition accuracy 81.33% 99.33% Number of misjudged samples (qualified products mistakenly judged as defective products) 16 items 1 item Number of samples missed (defective products mistakenly judged as qualified) 12 items 0 items Average time to complete test per unit 4.2 seconds 4.5 seconds

[0171] Refer to Tables 1 and 2 and Figure 10 , Figure 11 It can be seen that traditional single-axis pressure displacement detection methods lack a three-dimensional spatial dynamic calibration mechanism and physical posture follow-up displacement compensation logic. Traditional single-axis pressure displacement detection methods are affected by the crosstalk of button switch installation eccentricity and surface tilt during the mechanical pressing cycle, resulting in defective product misjudgment and missed judgment. The multi-dimensional heterogeneous fusion detection scheme of the present invention eliminates the physical measurement noise interference caused by mechanical assembly bias. The multi-dimensional heterogeneous fusion detection scheme of the present invention realizes the isolation and status identification of the internal physical defects of the EPB switch under test through the solution of multi-dimensional heterogeneous data and multilayer perceptron evaluation model. The multi-dimensional heterogeneous fusion detection scheme of the present invention improves the accuracy of comprehensive quality identification while maintaining the cycle time of a single test device.

[0172] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A test bench for measuring the travel and force of an automotive EPB switch button, comprising a base, a CNC platform, a servo component, a testing mechanism, and tooling fixtures, characterized in that: The base provides a fixed reference surface; the tooling fixture is mounted on the base and configured to clamp and fix the EPB switch to be tested; the CNC platform is mounted on the base, and the CNC platform includes an X-axis drive assembly and a Y-axis drive assembly; the servo assembly is mounted at the end of the CNC platform and configured to provide Z-axis lifting drive; The detection mechanism is fixed to the output end of the servo component.

2. The test bench for the travel and force of automotive EPB switch buttons according to claim 1, characterized in that, The detection mechanism includes a force sensor and three laser sensors; The force sensor is vertically mounted on the central Z-axis of the detection mechanism, and the lower end of the force sensor is connected to the pressure head; the three laser sensors are distributed in a 120-degree circle around the pressure head and are installed at an angle to the central Z-axis, configured such that when the pressure head at the lower end of the force sensor presses the surface of the EPB switch button to be tested, the emitted light path avoids the pressure head and irradiates the button surface outside the pressing area.

3. A control system for a test bench for measuring the travel and force of an automotive EPB switch button, applied to the test bench described in claim 1 or 2, characterized in that, The control system includes a control unit; The control unit is connected to the CNC platform, the servo component, and the testing mechanism, respectively; the control unit is configured to execute comprehensive testing logic, which includes: The CNC platform and the servo component are driven to move to the mechanical zero position to establish an absolute coordinate system, and preset calibration parameters matching the EPB switch under test are loaded. The CNC platform is controlled to move within a preset search area, the detection mechanism is controlled to collect discrete height data, the actual geometric center coordinates of the EPB switch button to be tested are calculated based on the discrete height data, and the CNC platform is driven to translate and align the central axis of the detection mechanism to the actual geometric center coordinates. At the actual geometric center coordinates, the relative height of the EPB switch button to be tested and the relative height of the product reference surface are obtained by the detection mechanism, and the actual surface difference dimension of the product to be tested is calculated based on the height difference. The detection mechanism is driven to press down along the Z-axis to apply a normal force to the EPB switch button under test. The tilt angle parameter of the pressed surface is extracted and converted into a plane compensation displacement command to drive the CNC platform to synchronously translate and compensate. The force data and absolute displacement data are recorded to obtain the effective stroke. The detection mechanism is driven to reverse and reset, and the time series data of unloading normal force value and rebound center height are recorded synchronously. The curve is processed to extract the time when the internal metal contact is disconnected and the time when the external button is lifted, and the time domain phase difference is calculated. The actual surface difference size, the effective travel distance, and the temporal phase difference are combined to construct a three-dimensional feature input vector. The three-dimensional feature input vector is then input into a pre-deployed multilayer perceptron evaluation model to perform forward propagation calculation. The qualified status is determined based on the classification result output by the multilayer perceptron evaluation model.

4. The control system according to claim 3, characterized in that, Calculating the actual geometric center coordinates of the EPB switch button under test based on the discrete height data specifically includes: The control unit controls the CNC platform to move step by step within the two-dimensional rectangular search area according to the set gridded scanning step size; Upon moving to each grid node, three laser sensors are triggered to synchronously collect local surface relative tilt distance data. This local surface relative tilt distance data is multiplied by the cosine of the physical installation tilt angle to convert it into vertical absolute height data, and the average value is taken to generate fused height data. After traversing the two-dimensional rectangular search area, all fused height data are used to construct a discrete height data matrix according to the spatial coordinate mapping relationship of the grid nodes. The discrete weighted centroid algorithm is used to calculate the sum of the products of the X-axis discrete coordinate values ​​and the corresponding fused height data, and the first ratio of the sum of the fused height data, to obtain the X-axis coordinates of the actual geometric center. The sum of the products of the Y-axis discrete coordinate values ​​and the corresponding fused height data, and the second ratio of the sum of the fused height data, is calculated to obtain the Y-axis coordinates of the actual geometric center.

5. The control system according to claim 3, characterized in that, Obtaining the relative height of the EPB switch button under test and the relative height of the product reference surface, and calculating the actual surface difference of the product under test, specifically includes: Three laser sensors are triggered to acquire the local relative tilt distance of the EPB switch button surface under test and convert it into three absolute vertical height distance values ​​of the button surface. The arithmetic mean of the three absolute vertical height distance values ​​is calculated to obtain the relative height of the button center. The CNC platform is driven to move the detection mechanism above the fixed reference surface of the EPB switch housing under test. The three laser sensors are triggered to acquire the local relative tilt distance of the reference surface and convert it into three absolute vertical height distance values ​​of the reference surface. The arithmetic mean of the three reference surface distance values ​​is calculated to obtain the relative height of the reference surface center. The relative height of the button center is subtracted from the relative height of the reference surface center to obtain the actual surface difference dimension.

6. The control system according to claim 3, characterized in that, Extracting the tilt angle parameters of the pressure surface and converting them into planar compensation displacement commands to drive the CNC platform for synchronous translation compensation specifically includes: During a constant-speed downward pressure cycle, three laser sensors are triggered to continuously acquire discrete relative distance data sequences. These data are then converted into transient vertical absolute height data using trigonometric functions and combined with fixed-plane projection coordinates to generate three transient three-dimensional spatial point coordinates. A normal vector is generated by performing an outer product operation on the first spatial vector from the first transient three-dimensional spatial point to the second transient three-dimensional spatial point and the second spatial vector from the first transient three-dimensional spatial point to the third transient three-dimensional spatial point. The three-dimensional coordinate components of the normal vector are extracted, and the instantaneous tilt angles of the button surface around the X-axis and around the Y-axis are calculated in real time using arctangent trigonometric functions. The tangent of the instantaneous tilt angle around the X-axis is multiplied by the kinematic decoupling coefficient to obtain the planar compensation displacement in the Y-axis direction. The planar compensation displacements in the X-axis and Y-axis directions are then used as dynamic position commands and sent to the CNC platform in real time.

7. The control system according to claim 3, characterized in that, The effective stroke is derived by recording force data and absolute displacement data, specifically including: The control unit monitors the changes in the normal force data in real time, and locks the initial force absolute displacement data when the normal force data reaches the preset contact force threshold; it locks the current target absolute displacement data when the normal force data reaches the target trigger force value; and it subtracts the initial force absolute displacement data from the current target absolute displacement data to obtain the effective stroke of the EPB switch button under test.

8. The control system according to claim 3, characterized in that, Extracting the time difference between the internal metal contact disconnection moment and the external button release moment and calculating the time-domain phase difference specifically includes: The normal force unloading time series data and the center height rebound time series data are mapped to the same time axis for alignment. After alignment, a first-order backward difference operation is performed to generate the normal force unloading rate change sequence and the center height rebound rate change sequence, respectively. The first timestamp corresponding to the number of consecutive sampling points where the normal force unloading rate change is continuously greater than or equal to the preset force change threshold reaches the consecutive trigger point threshold is extracted as the internal metal contact disconnection time. The first timestamp corresponding to the number of consecutive sampling points where the center height rebound rate change is continuously greater than or equal to the preset height change threshold reaches the consecutive trigger point threshold is extracted as the external button release time. The difference between the internal metal contact disconnection time and the external button release time is calculated to generate time domain phase difference data.

9. The control system according to claim 3, characterized in that, The process of inputting the three-dimensional feature input vector into a pre-deployed multilayer perceptron evaluation model to perform forward propagation calculations specifically includes: Extract the preset maximum and minimum normalization boundary parameters and perform linear mapping processing on the three-dimensional feature input vector to generate a standardized feature input vector with a mapping interval between 0 and 1. The multilayer perceptron evaluation model comprises a fully connected input layer, a first hidden layer, a second hidden layer, and an output layer. The standardized feature input vector is input into the input layer, multiplied by the connection weight matrix, and a bias vector is added. Then, a linear rectified function is used to activate the first hidden layer output feature matrix and the second hidden layer output feature matrix. A normalized exponential function is used to calculate a 3-dimensional classification probability vector in the output layer. The classification index corresponding to the probability element with the largest value is extracted to determine the pass / fail status of the EPB switch under test.

10. The control system according to claim 9, characterized in that, The multilayer perceptron evaluation model is generated through offline training via the following steps: Historical measured data is extracted to construct input features for the training sample set. A real supervision label set containing qualified labels, assembly tolerance exceedance labels, and mechanical structure failure labels, annotated by manual quality inspection experts, is introduced. The cross-entropy loss function is used as the objective function, and an adaptive moment estimation optimizer is configured to perform error backpropagation and parameter weight update iteration on the training sample set. When the value of the cross-entropy loss function decreases by less than a preset threshold for a consecutive preset iteration period, training is stopped and the multilayer perceptron evaluation model parameter network is exported and deployed to the control unit.

Images