Automatic fiber placement machine soft probe calibration method based on time sequence signal analysis

Abstract

The application discloses a soft probe calibration method of automatic fiber placement equipment based on timing signal analysis and relates to the technical field of industrial equipment calibration. The method comprises the following steps: controlling a fiber placement head to touch reference feature points in multiple postures, collecting cylinder displacement and the position of each shaft at the moment of touching, and obtaining a multiple-posture data set; based on forward kinematics, a constraint equation set containing a tool center point, a floating direction and a workpiece coordinate system is established, and the data set is used for joint solution to obtain a calibration parameter group; the theoretical position of the reference point is inversely calculated, and the difference with the known coordinates is obtained to obtain a residual sequence; fault-sensitive features are extracted, a fault mode library is matched, and a diagnosis conclusion is output; when healthy, the workpiece coordinate system offset is written into a numerical control system for updating; and when faulty, updating is prohibited and a maintenance suggestion is output. The technical problem that, in the prior art, a fiber placement head containing a passive floating mechanism cannot realize on-machine self-calibration which does not depend on external measuring instruments and eliminates parameter coupling through multiple-posture touching is solved.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment calibration technology, specifically to a soft probe calibration method for automated fiber placement equipment based on time-series signal analysis. Background Technology

[0002] Automated fiber placement is a core process in the manufacturing of composite material components. It utilizes multi-axis CNC equipment to lay prepreg fibers onto the mold surface along a planned path. The spatial calibration accuracy of the fiber placement head tool's center point relative to the workpiece coordinate system directly determines the accuracy of the fiber placement point. The system needs to accurately control the nominal position vector of the tool's center point, the floating direction vector of the compaction cylinder, and the transformation parameters from the workpiece coordinate system to the equipment's base coordinate system.

[0003] Currently, calibration mainly relies on external precision measuring instruments such as laser trackers. These methods involve expensive equipment and have high operational barriers, making it difficult to meet the high-frequency, rapid verification needs after routine mold changes or wire placement head adjustments. To reduce reliance on external instruments, in-machine calibration is a potential direction. However, traditional in-machine probe calibration schemes for CNC machine tools are based on the assumption of rigid fixed length, which is unsuitable for wire placement heads: the end of the wire placement head contains a continuously floating compaction cylinder, which elastically retracts upon contact with the reference point. Furthermore, the retraction direction is affected by assembly tolerances and is not strictly aligned with the flange coordinate system, making it impossible for traditional models to solve for the end parameters of such floating mechanisms. In addition, when touching a feature point in a single vertical posture, the axial component of the tool center point is highly collinear with the axial offset of the workpiece coordinate system in the equations, making it impossible to obtain a unique solution. Summary of the Invention

[0004] This invention provides a soft probe calibration method for automated filament placement equipment based on time-series signal analysis, which solves the technical problem in the prior art that it is impossible to achieve on-machine self-calibration without relying on external measuring instruments and by dissolving parameter coupling through multi-posture touch for filament placement heads with passive floating mechanisms.

[0005] This invention provides a soft probe calibration method for automated fiber placement equipment based on time-series signal analysis, the method comprising: The filament-laying head is controlled to touch a set of reference feature points with known coordinates in multiple postures. The cylinder displacement and the position of each axis of the equipment at the moment of contact are collected to obtain a multi-posture contact dataset. Based on forward kinematics, a set of constraint equations is established, including the tool center point vector, the floating direction vector, and the workpiece coordinate system transformation matrix. The set of constraint equations is then solved jointly using the multi-pose touch dataset to obtain the calibrated parameter set. The theoretical positions of each reference feature point are calculated back based on the calibrated parameter set, and the residual sequence is obtained by subtracting it from the known coordinates. Output the calibrated parameter set and the residual sequence.

[0006] One or more technical solutions provided in this invention have at least the following technical effects or advantages: By controlling the wire-laying head to touch the same reference feature point in at least two different end postures, independent constraints in orthogonal directions are provided for the constraint equations, thus resolving the collinear coupling problem between the axial component of the tool center point and the axial offset of the workpiece coordinate system under single-posture contact. Simultaneously, a constraint equation system including the tool center point vector, the floating direction vector, and the workpiece coordinate system transformation matrix is ​​established using forward kinematics. A nonlinear least squares algorithm is then used to jointly solve the multi-posture contact dataset, obtaining all calibration parameters at once and avoiding the error cascading caused by the step-by-step calibration of traditional methods.

[0007] The compaction cylinder of the wire-laying head is set to a low-pressure floating state and reused as a soft probe with contact detection capability. By synchronously acquiring the displacement timing signal of the LVDT (Linear Variable Differential Transformer) and the position timing signal of the encoders on each axis of the equipment, a first-order differential operation is performed on the displacement timing signal to identify the physical contact moment. This timing signal analysis method can eliminate the inertial overtravel error caused by machine tool interpolation delay and achieve high contact positioning accuracy without adding additional hardware.

[0008] In summary, this invention achieves on-machine self-calibration of multiple coupled parameters, such as the center point vector, floating direction vector, and workpiece coordinate system transformation matrix of a wire-laying head with a floating mechanism, without relying on external measuring instruments by reusing the compaction cylinder as a soft probe, dissolving parameter coupling through multi-pose contact, and nonlinear joint solution. Furthermore, it enables fault diagnosis based on the calibration residuals. This solves the technical problem in the prior art that it is impossible to achieve on-machine self-calibration without relying on external measuring instruments and dissolving parameter coupling through multi-pose contact for wire-laying heads with passive floating mechanisms. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the automatic fiber placement equipment soft probe calibration method based on time-series signal analysis provided in an embodiment of the present invention; Figure 2 This is a logic diagram of the automatic fiber placement equipment soft probe calibration method based on time-series signal analysis provided in this embodiment of the invention; Figure 3 This is an example diagram of a multi-pose touch dataset for an automated filament placement equipment soft probe calibration method based on time-series signal analysis provided in an embodiment of the present invention. Detailed Implementation

[0011] like Figure 1 As shown, this invention provides a flowchart illustrating a soft probe calibration method for automated fiber placement equipment based on time-series signal analysis; as... Figure 2 As shown, this invention provides a scheme logic diagram for a soft probe calibration method for automated fiber placement equipment based on time-series signal analysis. The method includes: S100: Controls the filament placement head to touch a set of reference feature points with known coordinates in multiple postures, collects the cylinder displacement and the position of each axis of the equipment at the moment of contact, and obtains a multi-posture contact dataset; In daily production of automated filament placement equipment, die changes, filament placement head maintenance, or changes in ambient temperature can all cause shifts in the tool center point position and workpiece coordinate system. Traditional calibration methods rely on external precision instruments such as laser trackers, which have high operational barriers and are time-consuming, making it difficult to meet the needs of daily high-frequency and rapid verification. In addition, the passive floating compaction cylinder at the end of the filament placement head makes the traditional rigid body fixed-length on-machine probe calibration scheme no longer applicable.

[0012] Step S100 in the method provided in this embodiment of the invention includes: controlling the fiber placement head to contact a set of reference feature points with known coordinates in multiple postures, including: For the same reference feature point, touch it with at least two different device end postures, wherein the angle of the posture difference between the two touches is not less than a preset posture difference threshold.

[0013] The specific implementation method is as follows: First, a set of reference feature points with known coordinates are determined. These reference feature points are hardened reference points pre-fixed on the edge of the mold or on the tooling platform, such as precision tapered recesses, high-gloss ball stages, or positioning planes. The coordinate values ​​of each reference feature point in the workpiece coordinate system have been pre-acquired and stored in the CNC system using precision measurement methods such as a coordinate measuring machine.

[0014] At the start of the calibration operation, the CNC system automatically generates a probe path based on a pre-configured array of reference feature point coordinates. To resolve collinear coupling of parameters, at least two touches with different end-effector postures are performed for each reference feature point. When the wire-laying head touches a reference feature point with a single vertical posture, the axial component of the tool center point and the axial offset of the workpiece coordinate system are highly collinear in the constraint equations, and a unique solution cannot be obtained using only a single posture. By changing the end-effector posture of the wire-laying head flange, the tool center point vector and the floating direction vector produce different projections in different directions, providing independent constraints in orthogonal directions for the constraint equation set, thereby resolving collinear coupling.

[0015] The end-effector's posture is achieved by adjusting the angle of the device's rotation axis. The angle difference between the two touches is not less than a preset posture difference threshold, which can be set according to the device's kinematic characteristics and accuracy requirements, for example, 30 degrees. For example, the first touch approaches the reference feature point vertically with the flange in a level posture (0 degrees on the C-axis). The second touch rotates the flange 45 degrees around the floating direction and then touches the same reference feature point again in an inclined posture. The 45-degree difference between the two postures, greater than the preset posture difference threshold of 30 degrees, effectively provides orthogonal constraints. Performing the above-mentioned differential posture touches on each reference feature point forms a multi-posture touch path sequence.

[0016] Step S100 in the method provided in this embodiment of the invention further includes: acquiring the cylinder displacement and the position of each axis of the device at the moment of contact, including: The compaction cylinder of the wire-laying head is set to a low-pressure floating state. During the process of the wire-laying head approaching the reference feature point, the displacement timing signal of the linear variable differential transformer and the position timing signal of the encoder of each axis of the equipment are synchronously acquired at a preset sampling frequency. The first-order differential operation is performed on the displacement timing signal, and the moment when the first-order differential value changes abruptly is identified as the physical contact moment. The corresponding cylinder displacement and the encoder position of each axis of the equipment are extracted by backtracking from this physical contact moment.

[0017] The specific implementation method is as follows: In soft probe mode, the compaction cylinder of the filament placement head is first set to a low-pressure floating state. By reducing the cylinder back pressure to a preset low pressure value at the control end, the cylinder is no longer rigidly locked, but has elastic retraction capability. At this time, when the pressure roller contacts the rigid reference feature point, the cylinder push rod will automatically compress and retract. Combined with the existing displacement sensor, the pressure roller can be reused as a soft probe that can detect contact. The air pressure value in the low-pressure floating state can be set according to the cylinder specifications and contact requirements, for example, set to 0.1 MPa to 0.2 MPa, corresponding to a contact force of approximately 10 Newtons to 20 Newtons.

[0018] As the wire-laying head approaches the reference feature point at low speed along the floating direction, two sets of timing signals are simultaneously acquired at a preset sampling frequency. The first set is the displacement timing signal output by the linear variable differential transformer, reflecting the real-time compression change of the cylinder; the second set is the position timing signal fed back by the encoders of each axis of the equipment, recording the real-time position readings of each moving axis. The sampling frequency is set according to the equipment interpolation cycle and contact detection accuracy requirements; for example, it can be set to 1000 Hz, corresponding to a 1-millisecond sampling period. In typical low-speed contact scenarios, the contact event lasts for approximately 10 to 50 milliseconds, and a 1000 Hz sampling frequency can capture 10 to 50 sampling points within this window, meeting the requirements for contact moment detection.

[0019] The equipment drives the pressure roller at low speed to approach the reference feature point. Once the pressure roller contacts the reference feature point, if the movement does not stop, the machine tool will continue to advance several millimeters due to the interpolation delay. During this period, the cylinder is forced to retract to absorb the excess displacement. The synchronously acquired displacement timing signal fully records the above process.

[0020] The acquired displacement timing signal is subjected to first-order differential operation. The reason for using first-order differential operation to detect the moment of physical contact is as follows: Before contact, the pressure roller is in a free-suspension state, the cylinder push rod is not subjected to external reaction force, the displacement sensor reading remains relatively stable or only drifts slowly due to the movement of the equipment, and the rate of displacement change is close to zero. When the pressure roller contacts the surface of the rigid reference feature point, the equipment continues to advance, the reference surface generates a reaction force on the pressure roller, the cylinder push rod begins to be compressed, and the rate of displacement change undergoes a significant abrupt change before and after contact, that is, from a slow change close to zero to a rapid compression determined by the feed rate of the equipment. If the absolute value of displacement is directly used for threshold judgment, the absolute position of the reference surface needs to be accurately calibrated as a reference, and in low-speed contact scenarios, the absolute value of displacement change may be very small in the early stage of contact, which is easily drowned out by sensor noise. However, first-order differential is sensitive to the rate of displacement change, which can amplify the abrupt change in rate at the moment of contact, transforming the judgment of "whether the displacement exceeds the threshold" into the judgment of "whether the rate of displacement change changes abruptly", thus detecting the event in the very early stage of contact.

[0021] The first-order differential mathematically reflects the instantaneous rate of change of a function at a point. For displacement time-series signals, the first-order differential value represents the rate of change of cylinder displacement with time, i.e., the instantaneous movement speed of the cylinder push rod. In discrete signal processing, displacement time-series signals are acquired with a fixed sampling period to obtain a displacement sequence. The first-order differential is approximated by calculating the difference between adjacent sampling points, i.e., the displacement value of the later sampling point minus the displacement value of the earlier sampling point, and then divided by the sampling period. To suppress single-point misjudgments caused by sensor noise, the original displacement signal can first be filtered by a five-point moving average, and then the difference between adjacent sampling points can be calculated as the first-order differential value. At the instant of contact, the cylinder push rod is subjected to the reaction force of the rigid reference surface, and the rate of displacement change changes abruptly, resulting in a significant peak in the differential value curve.

[0022] The moment when the first-order derivative value changes abruptly is identified as the moment of physical contact. The determination of this abrupt change relies on a preset derivative threshold and a continuous confirmation mechanism. For example, the derivative threshold can be set to 0.5 millimeters per millisecond; when the rate of displacement change exceeds this threshold, it is considered a sign of contact. To avoid false triggering caused by vibration, a continuous multi-point confirmation strategy can be adopted, for example, contact is confirmed only when three consecutive sampling points exceed the derivative threshold. The moment of physical contact is then traced back to the moment when the first of these three sampling points exceeds the threshold.

[0023] Based on the identified physical contact moment, the system backwards to eliminate inertial overtravel segments caused by interpolation delays, extracts the cylinder displacement reading corresponding to that moment, and records it as the cylinder displacement at the instant of contact. Simultaneously, it extracts the encoder position readings for each axis at that moment and records them as the positions of each axis of the device at the instant of contact. The data collected in a single contact are combined into a single contact record. After traversing all poses of contact at all reference feature points, a multi-pose contact dataset is obtained.

[0024] For example, there are 6 conical recesses on the flange skirt of the flat mold as reference feature points. The 6 feature points are numbered P_w1 to P_w6 in order from left to right and from front to back. Taking one of the feature points P_w1 (coordinates 200 mm, 150 mm, 50 mm) as an example, the two posture contact data are as follows: The first contact is performed in a vertical posture with the C-axis at 0 degrees. At the moment of contact, the cylinder displacement is 0.75 mm, and the flange posture is X=198.5 mm, Y=148.2 mm, Z=128.6 mm, A=0 degrees, B=0 degrees, C=0 degrees; The second contact is performed in a tilted posture with the C-axis rotated to 45 degrees. At the moment of contact, the cylinder displacement is 1.12 mm, and the flange posture is X=197.8 mm, Y=149.1 mm, Z=127.4 mm, A=0 degrees, B=0 degrees, C=45 degrees. The remaining 5 feature points are touched according to the same dual-pose rule, forming a multi-pose touch dataset with a total of 12 sets of touch data. Figure 3 This is an example diagram of a multi-pose touch dataset for an automated filament placement equipment soft probe calibration method based on time-series signal analysis provided in an embodiment of the present invention.

[0025] The following technical effects were achieved through this step: First, by reusing the compaction cylinder of the wire placement head as a soft probe, without adding any extra hardware, the contact detection function is realized using the existing cylinder and displacement sensor of the equipment, which reduces calibration costs and system complexity.

[0026] Second, the collinear coupling problem between the axial component of the tool center point and the axial offset of the workpiece coordinate system is resolved by the multi-pose touch strategy, providing independent constraints in orthogonal directions for the joint solution of the subsequent constraint equations.

[0027] Third, by performing first-order differential analysis on the displacement time sequence signal, the physical contact moment is accurately identified and the data at the moment of contact is extracted back, eliminating the inertial overtravel error caused by the machine tool interpolation delay, and ensuring that the calibration data truly reflects the spatial relationship of the contact point.

[0028] S200: Based on forward kinematics, establish a set of constraint equations including the tool center point vector, the floating direction vector, and the workpiece coordinate system transformation matrix. Use the multi-pose touch dataset to jointly solve the set of constraint equations to obtain the calibrated parameter set. The above steps yielded a multi-pose touch dataset containing multiple sets of touch data. Each set of data records the end-effector posture, cylinder displacement, and encoder position for each axis at the moment of contact. Traditional rigid body kinematics models assume the tool is a fixed-length rigid body, while the end of the wire-laying head contains a passive floating cylinder, resulting in elastic retraction at the moment of contact. Furthermore, the retraction direction is not perfectly parallel to the flange Z-axis due to assembly tolerances. If tool center point calibration and workpiece coordinate system alignment are separated into two independent processes, their respective measurement errors will be cascaded and superimposed, and a single vertical posture touch cannot eliminate the collinear coupling of parameters.

[0029] Step S200 in the method provided in this embodiment of the invention includes: jointly solving the constraint equations using the multi-pose touch dataset, including: Using the design nominal value or the previous calibration value of the tool center point vector, the floating direction vector, and the workpiece coordinate system transformation matrix as initial values, construct the objective function as the sum of squares of the differences between the known reference feature point coordinates and the theoretical positions of all contact points. Use a nonlinear least squares algorithm for iterative optimization until the change in the objective function between two adjacent iterations is less than the preset convergence tolerance. Use the parameter value at convergence as the calibration result to obtain the calibrated parameter set. The specific implementation method is as follows: First, initial values ​​are assigned to the parameters to be solved. The tool center point vector T0 is taken from the nominal tool length value on the design drawing or the result of the last calibration. For example, the last calibration value [0, -45.1, 80] mm is used as the initial value. The floating direction vector V̂ is taken as the ideal assumption value [0, 0, -1], indicating that the cylinder is perpendicular to the flange mounting surface by default. The workpiece coordinate system transformation matrix T_W_B is taken as the workpiece coordinate system offset value currently set by the CNC system, that is, the value written into the workpiece coordinate system parameters such as G54 / G55 after the last calibration. Initial values ​​close to the true values ​​help the algorithm converge quickly, but do not affect the accuracy of the final solution.

[0030] Further, an objective function is constructed. For each contact, the theoretical coordinates of the known reference feature point in the workpiece coordinate system and the encoder positions of each axis recorded in step S100 at the moment of contact are used. The pose matrix T_B_F of the flange coordinate system in the equipment base coordinate system is calculated from the encoder positions of each axis through forward kinematics, and is a known quantity. The position of the pressure roller center in the equipment base coordinate system at the moment of contact is represented as T_B_F×[T0+ΔL×V̂]. After mapping by the workpiece coordinate system transformation matrix T_W_B, its theoretical position should be equal to the known coordinates of the reference feature point in the workpiece coordinate system. The sum of the squares of the differences between the known coordinates and the theoretical positions of all contact points is used as the objective function. The smaller the objective function value, the closer the calibration parameters are to the true values.

[0031] Furthermore, a nonlinear least squares algorithm is used to iteratively optimize the objective function. The nonlinear least squares problem can be formulated as follows: given a dataset containing multiple touch points, each touch point provides a constraint equation regarding the parameter vector to be solved. The objective is to find the optimal parameter vector that minimizes the sum of squared residuals between the known coordinates of all touch points and the calculated values ​​from the model. Since the constraint equations are nonlinear functions of the parameters to be solved, analytical solutions cannot be obtained directly; iterative optimization methods are needed to gradually approximate the optimal value.

[0032] The Levenberg-Marquardt algorithm was chosen as the iterative optimization method. This algorithm is a standard algorithm for solving nonlinear least squares problems, and its core idea is to perform adaptive interpolation between gradient descent and the Gauss-Newton method in each iteration. The specific iterative steps are as follows: First, parameter initialization. Using the design nominal value or the previous calibration value of the parameter vector to be solved as the starting point of the iteration, initial values ​​are set for each component in the parameter vector, as explained above.

[0033] Second, construct the Jacobian matrix. At the current parameter vector estimate, calculate the partial derivatives of the constraint equations for each contact point with respect to each parameter, assembling them into an n x m Jacobian matrix, where n is the number of contact points and m is the number of parameters to be solved. The element in the i-th row and j-th column of the Jacobian matrix represents the local sensitivity of the theoretical position of the i-th contact point to the j-th parameter. Taking two different poses of the reference feature point P_w1 as an example, the flange rotation angles of the two contacts are different, causing the components of T0 and T_W_B to produce differentiated partial derivative values ​​in the Jacobian matrix, eliminating column collinearity under a single pose and ensuring that the Jacobian matrix has a good condition number.

[0034] Third, calculate the current residual. Substitute the current parameter vector estimate into the constraint equation to calculate the theoretical position of each contact point, and use the difference between this theoretical position and the known coordinates as the residual vector. The residual vector is an n-dimensional vector, with each component corresponding to the position deviation of a contact point.

[0035] Fourth, construct and solve the incremental equation. The incremental equation is the transpose of the Jacobian matrix multiplied by the Jacobian matrix, plus the damping factor multiplied by the identity matrix, and the resulting matrix multiplied by the parameter increment vector, which is equal to the transpose of the negative Jacobian matrix multiplied by the residual vector. The solution process for this equation is as follows: first, transform the equation into an equivalent system of linear equations, and then directly solve for the parameter increment vector using matrix decomposition methods, such as Cholesky decomposition or QR decomposition.

[0036] The damping factor is the core parameter controlling the iterative behavior. When the damping factor is large, the incremental equation degenerates into the incremental form of the gradient descent method, the parameter update direction is close to the negative gradient direction, the step size is small but robust, and it is suitable for situations where the current parameter estimate is far from the optimal solution. When the damping factor is small, the incremental equation degenerates into the incremental form of the Gauss-Newton method, the parameter update direction is close to the optimal direction of the second approximation, the step size is large and the convergence speed is fast, and it is suitable for situations where the current parameter estimate is close to the optimal solution.

[0037] Fifth, adaptively adjust the damping factor. Calculate the ratio of the actual decrease in the objective function after parameter updates to the decrease predicted by the quadratic model. If this ratio is close to 1, it indicates that the quadratic model and the objective function are well-matched at the current point, and the damping factor can be reduced to make the next iteration closer to the Gauss-Newton method to accelerate convergence. If this ratio is small or even negative, it indicates that the actual decrease is much smaller than expected or the objective function has actually increased, and the damping factor needs to be increased to make the next iteration closer to the gradient descent method to ensure robustness. Through the above adaptive adjustment strategy, the damping factor is automatically adjusted according to the actual changing trend of the objective function during the iteration process, without the need for manually setting a fixed value.

[0038] Sixth, update the parameters and check convergence. Add the obtained parameter increment vector to the current parameter vector estimate to obtain the updated parameter vector estimate. During the iteration process, continuously calculate the change in the objective function between adjacent iterations. When the change is less than the preset convergence tolerance, the optimization is considered to have converged, and the iteration stops. The convergence tolerance can be set according to the calibration accuracy requirements, for example, 10 to the power of -5 millimeters, corresponding to the sub-millimeter level calibration requirement for wire placement accuracy. Use the parameter values ​​at convergence as the calibration result to obtain the calibrated parameter set containing the tool center point vector T0, the floating direction vector V̂, and the workpiece coordinate system transformation matrix T_W_B. At the same time, record the root mean square residual of the final objective function value as a calibration quality evaluation index.

[0039] Taking the two touch data of the reference feature point P_w1 (coordinates 200 mm, 150 mm, 50 mm) in step S100 as an example: the first touch is at 0 degrees on the C-axis, the cylinder displacement is 0.75 mm, and the flange pose is (198.5, 148.2, 128.6, 0°, 0°, 0°); the second touch is at 45 degrees on the C-axis, the cylinder displacement is 1.12 mm, and the flange pose is (197.8, 149.1, 127.4, 0°, 0°, 45°). The two touch data with different poses contain different flange rotation angles, which makes the components of the tool center point vector T0 and the workpiece coordinate system transformation matrix T_W_B project differently in the constraint equations, eliminating the collinear coupling under a single pose. The 12 sets of touch data constitute an over-constrained equation system for 11 unknown parameters. The LM algorithm finds the optimal solution that satisfies the least squares meaning under the over-constraint conditions.

[0040] For example, after convergence of the LM algorithm iterative solution, the output calibrated tool center point vector is [0.1, -45.16, 79.8] mm, the floating direction vector V̂ is [-0.013, 0, -0.99], the workpiece coordinate system transformation matrix T_W_B corresponds to the updated workpiece coordinate system offset values, and the calibration root mean square residual is approximately 0.08 mm. The components of the tool center point vector are corrected from their initial values, and the floating direction vector is corrected from an idealized value to an actual value including a small installation tilt angle.

[0041] The following technical effects were achieved through this step: This method incorporates three types of coupled parameters—the tool center point vector, the floating direction vector, and the workpiece coordinate system transformation matrix—into a single set of constraint equations. Orthogonal direction-independent constraints are provided using multi-pose touch data, and a nonlinear least squares algorithm is employed to solve all calibration parameters simultaneously. Compared to the traditional method of performing tool center point geometric calibration followed by workpiece coordinate system alignment in separate steps, this method eliminates the cascading and superposition of errors from the separate calibration steps and improves the robustness of the calibration results through constraint-based solutions.

[0042] Step S300: Calculate the theoretical position of each reference feature point based on the calibrated parameter set, and obtain the residual sequence by subtracting it from the known coordinates; Step S200 obtains the calibrated parameter set through joint solution. This parameter set includes the estimated tool center point vector, floating direction vector, and workpiece coordinate system transformation matrix for the current calibration cycle. If the calibration parameter set is consistent with the actual geometric state of the equipment, the theoretical positions of each reference feature point calculated using this parameter set should closely match the known coordinates. Conversely, if there is a significant deviation between the calculated positions and the known coordinates, it indicates that there are abnormal factors in the calibration process, such as damage to the reference feature points, backlash in the transmission system, or loosening of the wire placement head structure.

[0043] The specific implementation method is as follows: First, extract the calibrated parameter set obtained in step S200, including the calibrated tool center point vector T0_cal, the calibrated floating direction vector V̂_cal, and the calibrated workpiece coordinate system transformation matrix T_W_B_cal.

[0044] Secondly, for each touch recorded in the multi-pose touch dataset in step S100, the known reference feature point coordinates P_w_i, the flange pose matrix T_B_F_i at the moment of contact, and the cylinder displacement ΔL_i corresponding to that touch are extracted. Taking two touches of the reference feature point P_w1 (coordinates 200 mm, 150 mm, 50 mm) as an example: In the first touch record, the flange pose is (198.5, 148.2, 128.6, 0°, 0°, 0°), and the cylinder displacement is 0.75 mm; in the second touch record, the flange pose is (197.8, 149.1, 127.4, 0°, 0°, 45°), and the cylinder displacement is 1.12 mm.

[0045] Furthermore, based on the forward kinematics model, the position of the pressure roller contact center in the equipment base coordinate system corresponding to this contact is calculated using T_B_F_i, T0_cal, ΔL_i, and V̂_cal. This position is then mapped to the workpiece coordinate system via T_W_B_cal to obtain the theoretical inverse position P_w_i_calc for this contact. The theoretical inverse position P_w_i_calc is then subtracted from the known coordinates P_w_i in the X, Y, and Z directions dimension by dimension to obtain the three-dimensional residual vector for this contact.

[0046] For example, for the first touch of P_w1, the theoretical position is calculated using the calibrated parameter set as (199.98, 150.03, 49.99) mm, and the residual vector between this position and the known coordinates (200, 150, 50) mm is (-0.02, 0.03, -0.01) mm. For the second touch of P_w1, the theoretical position is calculated as (200.01, 149.97, 50.02) mm, and the residual vector is (0.01, -0.03, 0.02) mm.

[0047] Traverse all touch data of all reference feature points in step S100, perform the above inverse calculation and subtraction operations on each touch record, and combine the residual vectors of all touches in the execution order to form a residual sequence. Each residual vector in this residual sequence reflects the degree of agreement between the calibration parameter set and the known coordinates at the corresponding touch position.

[0048] This step uses the calibrated parameter set to reverse verify the spatial position of each benchmark feature point, quantifying the accuracy of the calibration parameters into a three-dimensional residual sequence of each touch position, providing a quantitative basis for extracting fault-sensitive features and evaluating calibration quality in subsequent steps.

[0049] Step S400: Extract fault-sensitive features from the residual sequence, match them with the pre-stored fault mode library, and output diagnostic conclusions; Step S300 uses the calibrated parameter set to back-calculate the theoretical positions of each reference feature point, and subtracts them from the known coordinates to obtain the residual sequence. The residual sequence not only contains random measurement noise but may also implicitly contain systematic information about abnormal equipment geometry. If calibration quality is judged solely based on the magnitude of the root mean square value of the residuals, it can distinguish whether the calibration is acceptable, but it cannot pinpoint the specific cause of the calibration anomaly. For example, damage to a single reference feature point and overall offset of the workpiece coordinate system can both lead to an abnormal increase in residuals, but the handling methods for the two are drastically different.

[0050] Step S400 in the method provided in this embodiment of the invention includes: outputting the calibrated parameter set and the residual sequence, extracting fault-sensitive features from the residual sequence, matching them with a pre-stored fault mode library, and outputting a diagnostic conclusion; Extracting fault-sensitive features from the residual sequence includes: The median and standard deviation of the residuals at each benchmark feature point are extracted as the first feature; the correlation coefficient of the residuals with the contact posture angle is extracted as the second feature; the periodic frequency component of the residuals in the machine tool space is extracted as the third feature; the average magnitude of all residual vectors is extracted as the fourth feature; and the rate of change of the residuals with the cylinder displacement is extracted as the fifth feature.

[0051] The specific implementation method is as follows: The median and standard deviation of the residuals at each benchmark feature point are extracted as the first feature. For each benchmark feature point, the magnitudes of the three-dimensional residual vectors from each touch are aggregated into a set, and the median and standard deviation of this set of magnitude data are calculated. The median reflects the level of deviation of the residuals from the center at that feature point, and the standard deviation reflects the degree of dispersion of the residuals at that feature point. When a feature point is locally damaged, the median of the residuals at that point will be systematically higher than that at other normal points, and if the damage causes irregularities in the contact surface, the volatility of the residuals will also increase.

[0052] The correlation coefficient between the residual and the touch attitude angle is extracted as the second feature. The end-effector rotation axis angle corresponding to each touch, such as the C-axis angle, is used as the independent variable, and the residual magnitude of that touch is used as the dependent variable. The Pearson correlation coefficient between the two is calculated. The higher the absolute value of this coefficient, the stronger the linear correlation between the residual magnitude and the end-effector attitude. Its physical meaning is that there is a deviation between the calibration result of the floating direction vector and the actual installation direction, leading to a fixed direction deviation pattern under different attitudes.

[0053] The periodic frequency components of the residual in the machine tool space are extracted as the third feature. The residual sequence is arranged in order of the corresponding spatial coordinates of the contact with the machine tool, such as the X-axis or Z-axis. A Fast Fourier Transform is performed on the one-dimensional sequence to obtain the spectral distribution of the residual. If a peak value exceeding a preset amplitude threshold appears at the spatial frequency position corresponding to the lead screw pitch, it indicates that the residual has periodic fluctuations corresponding to the lead screw pitch, which may be due to local wear or backlash of the lead screw.

[0054] The average magnitude of all residual vectors is extracted as the fourth feature. The average residual vector is obtained by summing all the three-dimensional residual vectors of all touches and dividing by the total number of touches. Its magnitude is then calculated. This feature reflects the degree of systematic deviation of the calibration parameters throughout the workspace. When the workpiece coordinate system shifts overall, the residual directions at all touch points tend to be consistent, and the magnitude of the average residual vector will increase significantly.

[0055] The rate of change of residual with cylinder displacement is extracted as the fifth feature. Using the cylinder displacement ΔL recorded for each contact as the independent variable and the residual modulus of that contact as the dependent variable, a univariate linear regression is performed, and the regression slope is calculated. The physical meaning of this slope is: the degree of influence of the change in contact force between the wire-laying head and the reference feature point, i.e., the change in cylinder compression, on the residual. When the regression slope exceeds the normal range, it indicates that the connection rigidity between the wire-laying head flange and the cylinder may have decreased, leading to an unexpected offset of the end position when the contact force changes.

[0056] The fault mode library mentioned above includes at least one of the following modes: The first mode is characterized by the median residual of a single reference feature point exceeding a preset single-point deviation threshold, and the difference between the median residual of the reference feature point and the average of the median residuals of all other reference feature points exceeding a preset relative deviation threshold, which is then diagnosed as a damaged reference feature point. The second mode is characterized by the absolute value of the correlation coefficient between the residuals of all points and the rotation axis angle exceeding the preset correlation coefficient threshold, which is diagnosed as an unstable floating direction vector. The third mode is characterized by a peak value exceeding the preset amplitude threshold appearing at the frequency corresponding to the lead screw after the fast Fourier transform, which is diagnosed as backlash or lead screw wear in the transmission system. The fourth mode is characterized by the average residual vector having a magnitude exceeding a preset systematic deviation threshold and all residuals having the same direction, which is diagnosed as an overall offset of the workpiece coordinate system. The fifth mode is characterized by the regression slope of the residual with cylinder displacement exceeding the preset force sensitivity threshold, which is diagnosed as a decrease in the rigidity of the connection between the wire-laying head flange and the cylinder.

[0057] The specific implementation method is as follows: The first to fifth features extracted from the residual sequence in the above steps are matched one by one with each mode in the fault mode library.

[0058] For the first mode, the median of the residuals of each reference feature point in the first feature is checked. If the median of the residuals of a reference feature point exceeds a preset single-point deviation threshold, and the difference between the median of the residuals of that point and the average of the medians of the residuals of all other reference feature points exceeds a preset relative deviation threshold, then the reference feature point is determined to have local damage, and is diagnosed as damaged. For example, the preset single-point deviation threshold can be set to 0.3 mm, and the preset relative deviation threshold can be set to 0.2 mm.

[0059] For the second mode, the absolute value of the correlation coefficient between the residual modulus and the rotation axis angle in the second feature is checked. If the absolute value of this correlation coefficient exceeds a preset correlation coefficient threshold, the calibration value of the floating direction vector is determined to be unstable, and the floating direction vector is diagnosed as unstable. For example, the preset correlation coefficient threshold can be set to 0.7.

[0060] For the third mode, examine the spectrum after the Fast Fourier Transform in the third feature. If a peak appears at the spatial frequency corresponding to the lead screw lead, and the amplitude of the peak exceeds the preset amplitude threshold, it is determined that there is a periodic error source related to the lead screw in the transmission system, and the diagnosis is that there is backlash or lead screw wear in the transmission system.

[0061] For the fourth mode, the magnitude of the average residual vector in the fourth feature is checked. If the magnitude exceeds a preset systematic deviation threshold, and the residual directions at each touch point are statistically consistent, then it is determined that there is an overall offset in the workpiece coordinate system, and this is diagnosed as an overall offset in the workpiece coordinate system. For example, the preset systematic deviation threshold can be set to 0.5 mm.

[0062] For the fifth mode, examine the regression slope of the residual in the fifth feature as a function of cylinder displacement. If the absolute value of the regression slope exceeds the preset force sensitivity threshold, it is determined that the connection rigidity of the end floating mechanism may have decreased, and the diagnosis is a decrease in the connection rigidity between the wire-laying head flange and the cylinder.

[0063] The preset thresholds for each fault mode can be set based on the equipment's accuracy level, historical calibration data statistics, and engineering experience. The matched fault mode and its corresponding characteristic value are output as the diagnostic conclusion. If none of the above five modes are triggered, the diagnostic conclusion is a healthy state.

[0064] When the diagnostic conclusion is a healthy state, the calibrated parameter set is written into the CNC system to update the workpiece coordinate system offset; when the diagnostic conclusion is a fault state, updating the workpiece coordinate system offset is prohibited, and maintenance suggestions are output. The conditions for prohibiting updating the workpiece coordinate system offset also include: When the root mean square of the residual Euclidean distance of all touch points in the residual sequence is greater than the preset first root mean square threshold, regardless of the diagnostic conclusion, updating the workpiece coordinate system offset is prohibited and a calibration quality abnormality alarm is output.

[0065] The specific implementation method is as follows: After the appeal step outputs the diagnostic conclusion, the root mean square (RMS) value of the Euclidean distance of the residuals at all touch points in the residual sequence is first calculated. This value reflects the overall dispersion of the calibration data. This value is then compared with a preset first RMS threshold. The first RMS threshold is a global safety threshold independent of the fault mode library, and its set value is higher than the healthy RMS threshold.

[0066] When the root mean square (RMS) of the residuals exceeds the first RMS threshold, it indicates that the overall quality of the calibration data is severely out of tolerance. Regardless of the fault mode matching result, the calibration result is deemed unreliable, updating the workpiece coordinate system offset parameters is prohibited, and a calibration quality anomaly alarm is output to the operator, suggesting possible global factors such as: deformation of the wire-laying head due to equipment collision, large-area sensor failure, or overall displacement of reference feature points. When the RMS of the residuals is less than or equal to the first RMS threshold, this mandatory prohibition condition is not triggered, and subsequent operations continue according to the diagnostic conclusion. For example, the first RMS threshold can be set to 0.8 mm, corresponding to the global failure threshold of the calibration data.

[0067] Step S400 in the method provided in this embodiment of the invention further includes: after each calibration is completed, if the diagnosis conclusion is a healthy state, the statistical characteristics of the residual sequence in this calibration are stored as a new health benchmark for comparative analysis in subsequent calibrations.

[0068] The specific implementation method is as follows: When the output diagnostic conclusion is a healthy state, the statistical characteristics of the residual sequence of this calibration are extracted, including at least the root mean square value of the residuals, the median of the residuals at each benchmark feature point, and the standard deviation. These statistical characteristics are then associated with the timestamp and execution conditions of this calibration and stored as a new health benchmark in the historical calibration database of the CNC system.

[0069] In each subsequent calibration, the statistical characteristics of the current residual sequence can be compared with historical health benchmarks to calculate the deviation. If the root mean square of the residuals from each calibration shows a continuous upward trend, even if the residual value of a single calibration is still within the health threshold range, it can indicate that the mechanical condition of the wire-laying head is at risk of long-term degradation, providing an early warning basis for preventive maintenance.

[0070] Step S400 in the method provided in this embodiment of the invention further includes: the condition for determining the diagnosis conclusion as a healthy state is: the root mean square distance of the residual Euclidean distance of all touch points in the residual sequence is less than or equal to a preset root mean square threshold for health; and each of the fault-sensitive features does not trigger any fault mode in the fault mode library.

[0071] The specific implementation method is as follows: The root mean square (RMS) value of the residual Euclidean distance between all touch points in the residual sequence is used as the global quality indicator, and the root mean square threshold for health is used as the overall accuracy requirement for qualified calibration, set according to the typical motion accuracy of AFP equipment, for example, 0.3 mm. Each feature in the fault-sensitive features is used as a local diagnostic indicator, extracted by the above steps and compared one by one with the preset threshold of each mode in the fault mode library. Only when both conditions are met simultaneously, i.e., the root mean square value of the residual is less than or equal to the preset root mean square threshold for health, and all fault-sensitive features have not triggered any fault mode, is the diagnostic conclusion of this calibration determined to be a healthy state.

[0072] For example, in step S300, the residual vector of the first contact of the reference feature point P_w1 is (-0.02, 0.03, -0.01) mm, with a magnitude of approximately 0.037 mm, and the residual vector of the second contact is (0.01, -0.03, 0.02) mm, with a magnitude of approximately 0.037 mm. The root mean square distance of the residual Euclidean distance for all 12 contacts is approximately 0.08 mm, which is less than the healthy root mean square threshold of 0.3 mm. Furthermore, none of the five features triggered the corresponding fault mode. Therefore, the diagnostic conclusion is a healthy state.

[0073] If the root mean square value of the residuals exceeds the healthy root mean square threshold, it will not be considered healthy even if no specific fault mode is matched, and it is recommended to check the fault mode library for further investigation. If a certain feature triggers the corresponding mode in the fault mode library, then regardless of whether the root mean square value of the residuals meets the standard, the corresponding fault is determined to exist.

[0074] The health determination criteria consist of the two conditions mentioned above, and both conditions must be met simultaneously in order to recognize the diagnostic conclusion of this determination as a healthy state.

[0075] Step S400 in the method provided in this embodiment of the invention further includes: after outputting the diagnostic conclusion, it further includes: When the diagnosis is a fault condition, a maintenance work order containing the fault location, fault type and maintenance steps is automatically generated based on the matched fault mode and sent to the equipment maintenance terminal.

[0076] The specific implementation method is as follows: When the output diagnostic conclusion is a fault status, a preset work order template is filled according to the matched fault mode type. The work order content should include at least: fault location, such as the reference feature point number, lead screw shaft name or cylinder connection flange location, fault type, such as the diagnostic conclusion in the fault mode library, and repair steps, such as replacing the reference feature point, retightening the flange connection bolts, adjusting the lead screw backlash, or contacting the supplier. The generated repair work order is sent to the maintenance personnel's terminal via the equipment network for rapid response.

[0077] For example, if the matched fault mode is mode 5, and the diagnostic conclusion is a decrease in the rigidity of the connection between the wire-laying head flange and the cylinder, then the generated maintenance work order content is as follows: Fault location: Connection between the wire-laying head flange and the cylinder; Fault type: Decreased connection rigidity, residual-displacement regression slope exceeds the standard; Maintenance steps: After stopping the machine, use a torque wrench to retighten the flange connection bolts to the specified torque, and re-measure the calibrated residual. If the problem still exists, replace the flange elastic connector.

[0078] The following technical effects were achieved through this step: First, a dual safety decision-making mechanism based on residual root mean square (RMS) and fault mode matching was established. Before fault mode matching, a first RMS threshold is used as a global safety threshold. If the threshold is exceeded, parameter updates are prohibited regardless of the diagnostic conclusion, thus forming a fallback protection at the output end and effectively preventing erroneous updates caused by missed fault mode detections or unknown modes in the fault mode library.

[0079] Secondly, a closed-loop decision-making process has been established, comprising four branches: parameter update, benchmark iteration, fault alarm, and maintenance guidance. In a healthy state, the workpiece coordinate system offset is automatically updated, and the statistical characteristics of this operation are recorded as a reference for long-term health monitoring. In a fault state, the writing of erroneous parameters is blocked, and a maintenance work order is automatically generated and pushed to the maintenance terminal, shortening fault response time and improving equipment maintainability.

Claims

1. A soft probe calibration method for automated fiber placement equipment based on time-series signal analysis, characterized in that, include: The filament-laying head is controlled to touch a set of reference feature points with known coordinates in multiple postures. The cylinder displacement and the position of each axis of the equipment at the moment of contact are collected to obtain a multi-posture contact dataset. Based on forward kinematics, a set of constraint equations is established, including the tool center point vector, the floating direction vector, and the workpiece coordinate system transformation matrix. The set of constraint equations is then solved jointly using the multi-pose touch dataset to obtain the calibrated parameter set. The theoretical positions of each reference feature point are calculated back based on the calibrated parameter set, and the residual sequence is obtained by subtracting it from the known coordinates. Output the calibrated parameter set and the residual sequence.

2. The method for calibrating a soft probe in an automated fiber placement equipment based on time-series signal analysis as described in claim 1, characterized in that, Controlling the fiber-laying head to contact a set of reference feature points with known coordinates in multiple postures includes: For the same reference feature point, touch it with at least two different device end postures, wherein the angle of the posture difference between the two touches is not less than a preset posture difference threshold.

3. The method for calibrating a soft probe in an automated fiber placement equipment based on time-series signal analysis as described in claim 1, characterized in that, Collect the cylinder displacement and the position of each axis of the equipment at the moment of contact, including: The compaction cylinder of the wire-laying head is set to a low-pressure floating state. During the process of the wire-laying head approaching the reference feature point, the displacement timing signal of the linear variable differential transformer and the position timing signal of the encoder of each axis of the equipment are synchronously acquired at a preset sampling frequency. The first-order differential operation is performed on the displacement timing signal, and the moment when the first-order differential value changes abruptly is identified as the physical contact moment. The corresponding cylinder displacement and the encoder position of each axis of the equipment are extracted by backtracking from this physical contact moment.

4. The method for calibrating a soft probe in an automated fiber placement equipment based on time-series signal analysis as described in claim 1, characterized in that, The constraint equations are solved jointly using the multi-pose touch dataset, including: Using the design nominal value or the previous calibration value of the tool center point vector, the floating direction vector, and the workpiece coordinate system transformation matrix as initial values, the objective function is constructed as the sum of squares of the differences between the known reference feature point coordinates and the theoretical positions of all contact points. The nonlinear least squares algorithm is used for iterative optimization until the change in the objective function between two adjacent iterations is less than the preset convergence tolerance. The parameter value at the time of convergence is used as the calibration result to obtain the calibrated parameter set.

5. The method for calibrating a soft probe in an automated fiber placement equipment based on time-series signal analysis as described in claim 1, characterized in that, It also includes extracting fault-sensitive features from the residual sequence, matching them with a pre-stored fault mode library, and outputting diagnostic conclusions; Extracting fault-sensitive features from the residual sequence includes: The median and standard deviation of the residuals at each benchmark feature point are extracted as the first feature; the correlation coefficient of the residuals with the contact posture angle is extracted as the second feature; the periodic frequency component of the residuals in the machine tool space is extracted as the third feature; the average magnitude of all residual vectors is extracted as the fourth feature; and the rate of change of the residuals with the cylinder displacement is extracted as the fifth feature.

6. The method as described in claim 5, characterized in that, The fault mode library contains at least one of the following modes: The first mode is characterized by the median residual of a single reference feature point exceeding a preset single-point deviation threshold, and the difference between the median residual of the reference feature point and the average of the median residuals of all other reference feature points exceeding a preset relative deviation threshold, which is then diagnosed as a damaged reference feature point. The second mode is characterized by the absolute value of the correlation coefficient between the residuals of all points and the rotation axis angle exceeding the preset correlation coefficient threshold, which is diagnosed as an unstable floating direction vector. The third mode is characterized by a peak value exceeding the preset amplitude threshold appearing at the frequency corresponding to the lead screw after the fast Fourier transform, which is diagnosed as backlash or lead screw wear in the transmission system. The fourth mode is characterized by the average residual vector having a magnitude exceeding a preset systematic deviation threshold and all residuals having the same direction, which is diagnosed as an overall offset of the workpiece coordinate system. The fifth mode is characterized by the regression slope of the residual with cylinder displacement exceeding the preset force sensitivity threshold, which is diagnosed as a decrease in the rigidity of the connection between the wire-laying head flange and the cylinder.

7. The method as described in claim 5, characterized in that, When the diagnosis conclusion is a healthy state, the calibrated parameter set is written into the CNC system to update the workpiece coordinate system offset; When the diagnostic conclusion is a fault condition, updating the workpiece coordinate system offset and outputting maintenance suggestions are prohibited. The conditions under which updating the workpiece coordinate system offset is prohibited also include: When the root mean square of the residual Euclidean distance of all touch points in the residual sequence is greater than the preset first root mean square threshold, regardless of the diagnostic conclusion, updating the workpiece coordinate system offset is prohibited and a calibration quality abnormality alarm is output.

8. The method as described in claim 5, characterized in that, Also includes: After each calibration, if the diagnosis is a healthy state, the statistical characteristics of the residual sequence are stored as a new health benchmark for comparative analysis in subsequent calibrations.

9. The method as described in claim 5, characterized in that, The diagnostic conclusion is determined as a healthy state under the following conditions: the root mean square distance of the residual Euclidean distance of all touch points in the residual sequence is less than or equal to the preset root mean square threshold for health; and each of the fault-sensitive features does not trigger any fault mode in the fault mode library.

10. The method as described in claim 5, characterized in that, After outputting the diagnostic conclusion, it also includes: When the diagnosis is a fault condition, a maintenance work order containing the fault location, fault type and maintenance steps is automatically generated based on the matched fault mode and sent to the equipment maintenance terminal.

Images