A method for real-time separation and compensation of multi-source errors in torque measurement

Abstract

The embodiment of the application provides a multi-source error real-time separation and compensation method in torque measurement, relates to the technical field of mechanical measurement and precision instruments, and comprises the following steps: a multi-source error mathematical model is established, which comprises a temperature compensation submodel, a friction loss submodel, a nonlinear correction submodel and a misalignment correction submodel; temperature sensors, rotation speed sensors, acceleration sensors and torque sensors are arranged to obtain real-time measurement data; an adaptive Kalman filtering algorithm is used to perform online recursive estimation on error characteristic parameters, so that real-time separation of each error component is realized; and the original torque measurement value is compensated according to the separated error components and the correction result is output. The application has the effect of significantly improving the dynamic precision of torque measurement under variable working conditions.

Description

Technical Field

[0001] This invention relates to the field of mechanical measurement and precision instrument technology, and in particular to a method for real-time separation and compensation of multi-source errors in torque measurement. Background Technology

[0002] Torque is a core parameter for power transmission in rotating machinery, fastener assembly quality, and evaluation of material mechanical properties. High-precision torque testers are widely used in metrology calibration, automotive manufacturing, aerospace, and new energy equipment testing. Existing torque measurement systems face interference from multiple error sources in actual working environments. The first type of error originates from temperature changes; the second type originates from frictional resistance in mechanical transmission links; the third type originates from the nonlinearity of the sensor itself; and the fourth type originates from installation alignment deviations.

[0003] Currently, error compensation methods in the industry mainly employ offline, single-factor processing. For temperature compensation, static temperature compensation is typically performed via hardware circuitry at the sensor's factory, or correction is made in the measurement software based on a single temperature coefficient. This approach cannot address dynamic errors caused by temperature gradient changes and thermal conduction delays in actual operating conditions. Friction compensation largely relies on operator experience for data smoothing, lacking quantitative model support. Nonlinear correction generally employs piecewise linearization or multi-point calibration fitting, with correction accuracy limited by the density of calibration points. Alignment error is primarily controlled by the machining accuracy of mechanical tooling and repeated manual adjustments; residual alignment deviations and the resulting measurement fluctuations are difficult to eliminate at the signal level.

[0004] To address the aforementioned issues, there is an urgent need for a method that can uniformly model, separate in real time, and jointly compensate for the multi-source errors during the measurement process. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method for real-time separation and compensation of multi-source errors in torque measurement.

[0006] This application provides a method for real-time separation and compensation of multi-source errors in torque measurement, which adopts the following technical solution: A method for real-time separation and compensation of multi-source errors in torque measurement includes the following steps: S1. Establish a multi-source error mathematical model that includes a temperature compensation sub-model, a friction loss sub-model, a nonlinear correction sub-model, and a misalignment correction sub-model. The functional relationship of the temperature compensation sub-model is as follows: ΔT1 = k1·ΔT + k2·ΔT², where ΔT = T - T_ref, T is the current measured temperature, T_ref is the reference temperature, the function relationship of the friction loss sub-model is ΔT2 = μ·ω + T0, where ω is the current rotational angular velocity, the function relationship of the nonlinear correction sub-model is ΔT3 = α·T_raw² + β·T_raw, where T_raw is the original torque measurement value, the function relationship of the misalignment correction sub-model is ΔT4 = γ·sin(ω·t + φ), where t is time, and k1, k2, μ, T0, α, β, γ and φ are the error characteristic parameters to be identified. S2. In the torque testing system, a temperature sensor is arranged to obtain the real-time temperature value T, a speed sensor is arranged to obtain the real-time rotational angular velocity ω, an acceleration sensor is arranged to obtain the misalignment vibration signal, and the raw torque measurement value T_raw is provided by the torque sensor itself. S3. An adaptive Kalman filter algorithm is used to recursively estimate the error characteristic parameters online. The error characteristic parameters are used as the state vector X=[k1,k2,μ,T0,α,β,γ,φ]ᵀ. The sum of the outputs of the four sub-models, ΔT_total=ΔT1+ΔT2+ΔT3+ΔT4, is used as the prediction part of the observation equation. In each filtering cycle, the data of each sensor is read, the total error prediction value ΔT_total is calculated, the observation residual e=T_raw-ΔT_total is calculated, and the state vector is corrected and updated by the Kalman gain matrix through the residual to achieve real-time separation of various error components. S4. Based on the separated error components, the compensated torque measurement result is obtained and output according to the corrected torque calculation formula T_corrected=T_raw-ΔT1-ΔT2-ΔT3-ΔT4.

[0007] Optionally, the initial values ​​of the state vector in the adaptive Kalman filter algorithm are determined by offline calibration experiments. Specifically, k1 and k2 are obtained by performing multi-point temperature cycling experiments on the sensor in a constant temperature chamber and fitting a quadratic curve of ΔT versus output drift. μ and T0 are obtained by running the sensor under no-load at different speeds and recording the output value after stabilization. α and β are obtained by applying a series of known standard torque values ​​and fitting a quadratic polynomial of the relationship between T_raw and output.

[0008] Optionally, the vibration signal collected by the accelerometer is preprocessed before being fed into the filtering algorithm. The pulse signal of the speed sensor is used as a reference for order tracking sampling. The sampled data is subjected to fast Fourier transform to extract the amplitude A and phase φ of the fundamental frequency component of the rotation frequency, which are used to drive the update of parameters γ and φ in the filtering algorithm.

[0009] Optionally, the adaptive Kalman filter algorithm introduces a fading factor λ. When a sudden change in the measurement condition is detected, the filter gain is automatically increased by monitoring the statistical characteristics of the residual sequence e, so as to accelerate the response speed of the error characteristic parameters to the change in the condition.

[0010] Optionally, the method further includes a self-checking step: when the torque testing system is in a stable no-load operating state, a sequence of corrected torque output values ​​over a period of time is collected and its mean μ_check and standard deviation σ_check are calculated. If |μ_check| exceeds a preset threshold or σ_check exceeds a preset fluctuation threshold, a prompt signal is triggered.

[0011] In summary, this application includes at least one of the following beneficial technical effects: This invention, for the first time, incorporates four types of measurement errors with different sources and mechanisms—temperature drift, friction loss, nonlinear distortion, and misalignment fluctuation—into a unified mathematical modeling and filtering estimation framework, achieving online joint identification and separation of multi-source errors. It employs an adaptive Kalman filter algorithm, which has advantages such as recursive operation, low computational load, and suitability for real-time operation in embedded systems. By introducing a fading factor and a residual monitoring mechanism, the algorithm maintains the accuracy of error estimation under non-steady-state conditions such as sudden torque load changes, rapid speed changes, and ambient temperature fluctuations, effectively solving the problem of accuracy degradation under dynamic conditions in traditional fixed-parameter compensation methods. Attached Figure Description

[0012] Figure 1 This is the overall flowchart of this method. Detailed Implementation

[0013] The terminology used in the following embodiments is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to also include expressions such as “one or more,” unless the context clearly indicates otherwise. It should also be understood that in the following embodiments of this application, “at least one” and “one or more” refer to one, two, or more than two. The term “and / or” is used to describe the relationship between related objects, indicating that three relationships may exist; for example, A and / or B can indicate: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character “ / ” generally indicates that the preceding and following related objects are in an “or” relationship.

[0014] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0015] The following is in conjunction with the appendix Figure 1 The present invention will be described in further detail below.

[0016] This application discloses a method for real-time separation and compensation of multi-source errors in torque measurement. The overall process includes four stages: error modeling, multi-sensor signal acquisition, adaptive filtering estimation, and error compensation output.

[0017] In the error modeling stage, a multi-source error mathematical model containing four sub-models is established: The first branch is the temperature compensation sub-model. It takes as input the difference ΔT = T - T_ref between the current temperature sensor reading T and the reference temperature T_ref, and outputs the torque error ΔT1 caused by temperature. The mathematical expression for this sub-model is: ΔT1 = k1·ΔT + k2·ΔT 2 In this embodiment, the reference temperature T_ref is taken as the ambient temperature of 20 degrees Celsius during sensor calibration; The second branch is the friction loss sub-model. The current rotational angular velocity ω is input, and the torque loss ΔT2 caused by friction is output. The mathematical expression of this sub-model is ΔT2=μ·ω+T0. The third branch is the nonlinear correction sub-model, which takes the original measured value T_raw from the torque sensor as input and outputs the nonlinear deviation ΔT3. The mathematical expression of this sub-model is ΔT3=α·T_raw²+β·T_raw. The fourth branch is the misalignment correction sub-model. It takes the rotational angular velocity ω and time t as inputs and outputs a periodic fluctuation error ΔT4. The mathematical expression of this sub-model is ΔT4=γ·sin(ω·t+φ).

[0018] The outputs of the four sub-models are summed at the summation node to form the total error estimate ΔT_total = ΔT1 + ΔT2 + ΔT3 + ΔT4.

[0019] In the multi-sensor signal acquisition stage, the hardware system includes the following components: a torque sensor is installed between the rotating part being measured and the drive or load device, with strain gauges attached inside to form a Wheatstone bridge, outputting a voltage signal proportional to the torque, which is then amplified and converted to an analog-to-digital converter before entering the digital signal processor; a DS18B20 single-bus digital temperature sensor is attached to the surface of the elastic housing of the torque sensor, transmitting temperature data to the digital signal processor via a single-bus interface; an incremental photoelectric encoder is coaxially mounted on the rotating spindle, outputting 1024 pulses per revolution, which are converted into a rotational speed value ω by a pulse counting circuit and sent to the digital signal processor; a piezoelectric accelerometer is fixed to the side of the torque sensor housing via a magnetic mount, and its output signal is amplified by a charge amplifier and an anti-aliasing filter before entering the second channel of the analog-to-digital converter; and a TMS320F28335 digital signal processor is used as the central processing unit, responsible for executing filtering algorithms and data fusion calculations.

[0020] In the adaptive filtering estimation stage, the adaptive Kalman filter algorithm is used to recursively estimate the error characteristic parameters online. The single-step recursive calculation process of the filtering algorithm is as follows: First, a state-space model is established, defining a state vector X = [k1,k1,μ,T0,α,β,γ,φ]ᵀ, with eight parameters to be identified. The observation equation is defined as Z_k = H_k·X_k + V_k, where Z_k is the observed value at the current moment, taking the original torque measurement value T_raw_k, and H_k is the observation matrix, the specific form of which is: H_k=[ΔT,ΔT²,ω,1,T_raw_k²,T_raw_k,sin(ω·t+φ),γ·cos(ω·t+φ)·t], V_k represents the observation noise, and the state equation adopts the random walk model X_k=X_{k-1}+W_{k-1}, assuming that the parameters change slowly in a short time, and W_{k-1} represents the process noise. At time k, sensor data is first read, including temperature T_k, rotational speed ω_k, and raw torque T_raw_k. The temperature deviation ΔT = T_k - T_ref is calculated. For the vibration signal collected by the accelerometer, order tracking sampling is performed based on the encoder pulse signal, i.e., 64 fixed points are collected per revolution to eliminate the influence of rotational speed fluctuations on the spectrum analysis. A fast Fourier transform is performed on the collected data block of one revolution to extract the amplitude A_k and phase φ_k of the fundamental frequency component of the rotational frequency, i.e., the first-order component. The extracted phase information φ_k is used to update the φ parameter correlation term in the observation matrix.

[0021] Then, the Kalman filtering prediction step is performed, calculating the prior state estimate X̂_k⁻=X̂_{k-1} and the prior error covariance P_k⁻=P_{k-1}+Q, where Q is the process noise covariance matrix. In this embodiment, a diagonal matrix is ​​used, with diagonal elements having a value of 10. -4 Magnitude.

[0022] Calculate the output values ​​of the four sub-models in the current state: ΔT1_k = k1·ΔT + k2·ΔT², ΔT2_k = μ· ω_k+T0, ΔT3_k=α·T_raw_k²+β·T_raw_k, ΔT4_k=γ·sin(ω_k·t_k+φ), Summing up the total error prediction value ΔT_total_k=ΔT1_k+ΔT2_k+ΔT3_k+ΔT4_k, we can calculate the observation residual e_k=T_raw_k-ΔT_total_k; Next, the Kalman filter update step is executed; the innovation covariance S_k = H_k·P_k⁻·H_kᵀ + R is calculated, where R is the observation noise covariance; in this embodiment, the value is taken as 0.1². Calculate the Kalman gain matrix K_k = P_k⁻·H_kᵀ·S_k -1 Update the posterior state estimate X̂_k=X̂_k⁻+K_k·e_k, and update the posterior error covariance P_k=(I-K_k·H_k)·P_k⁻, where I is the identity matrix.

[0023] To improve the algorithm's ability to track sudden changes in operating conditions, this embodiment introduces a fading factor λ. In each filtering cycle, the statistical characteristics of the residual sequence e_k are monitored, and the mean square value of the sliding window of the residual is calculated. When the mean square value exceeds a preset threshold, it is determined that a sudden change in operating conditions has occurred. At this time, the prediction error covariance P_k⁻ is multiplied by the fading factor λ, where λ is between 1.01 and 1.1, thereby increasing the Kalman gain and enabling the state estimation to respond quickly to changes in operating conditions.

[0024] During the error compensation output stage, based on the eight error feature parameters stored in the updated state vector X̂_k, the output values ​​of the four sub-models at the current moment are recalculated, and compensation is performed according to the formula T_corrected=T_raw-ΔT1-ΔT2-ΔT3-ΔT4 to obtain the corrected torque value T_corrected. This value is sent to the host computer for display and recording via the RS485 digital communication interface.

[0025] The method described in this embodiment also includes a self-checking function. When the torque testing system is in a stable no-load operating state, it collects the corrected torque output values ​​of one hundred consecutive sampling points and calculates their mean μ_check and standard deviation σ_check. If |μ_check| exceeds two ten-thousandths of the full scale or σ_check exceeds one ten-thousandth of the full scale, it is determined that the current error compensation effect of the system is not good or there is a sensor abnormality. A high-level trigger prompt signal is output through a general-purpose input / output pin of the digital signal processor to remind the operator to perform maintenance or recalibration.

[0026] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or variations made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A method for real-time separation and compensation of multi-source errors in torque measurement, characterized in that, Includes the following steps: S1. Establish a multi-source error mathematical model that includes a temperature compensation sub-model, a friction loss sub-model, a nonlinear correction sub-model, and a misalignment correction sub-model. The functional relationship of the temperature compensation sub-model is as follows: ΔT1 = k1·ΔT + k2·ΔT², where ΔT = T - T_ref, T is the current measured temperature, T_ref is the reference temperature, the function relationship of the friction loss sub-model is ΔT2 = μ·ω + T0, where ω is the current rotational angular velocity, the function relationship of the nonlinear correction sub-model is ΔT3 = α·T_raw² + β·T_raw, where T_raw is the original torque measurement value, the function relationship of the misalignment correction sub-model is ΔT4 = γ·sin(ω·t + φ), where t is time, and k1, k2, μ, T0, α, β, γ and φ are the error characteristic parameters to be identified. S2. In the torque testing system, a temperature sensor is arranged to obtain the real-time temperature value T, a speed sensor is arranged to obtain the real-time rotational angular velocity ω, an acceleration sensor is arranged to obtain the misalignment vibration signal, and the raw torque measurement value T_raw is provided by the torque sensor itself. S3. An adaptive Kalman filter algorithm is used to recursively estimate the error characteristic parameters online. The error characteristic parameters are used as the state vector X=[k1,k2,μ,T0,α,β,γ,φ]ᵀ. The sum of the outputs of the four sub-models, ΔT_total=ΔT1+ΔT2+ΔT3+ΔT4, is used as the prediction part of the observation equation. In each filtering cycle, the data of each sensor is read, the total error prediction value ΔT_total is calculated, the observation residual e=T_raw-ΔT_total is calculated, and the state vector is corrected and updated by the Kalman gain matrix through the residual to achieve real-time separation of various error components. S4. Based on the separated error components, the compensated torque measurement result is obtained and output according to the corrected torque calculation formula T_corrected=T_raw-ΔT1-ΔT2-ΔT3-ΔT4.

2. The method for real-time separation and compensation of multi-source errors in torque measurement according to claim 1, characterized in that: The initial values ​​of the state vector in the adaptive Kalman filter algorithm are determined by offline calibration experiments. k1 and k2 are obtained by performing multi-point temperature cycling experiments on the sensor in a constant temperature chamber and fitting a quadratic curve of ΔT and output drift. μ and T0 are obtained by running the sensor under no-load at different speeds and recording the output value after stabilization. α and β are obtained by applying a series of known standard torque values ​​and fitting a quadratic polynomial of the relationship between T_raw and output.

3. The method for real-time separation and compensation of multi-source errors in torque measurement according to claim 1, characterized in that: The vibration signal collected by the accelerometer is preprocessed before being fed into the filtering algorithm. The pulse signal of the speed sensor is used as a reference for order tracking sampling. The sampled data is subjected to fast Fourier transform to extract the amplitude A and phase φ of the fundamental frequency component of the rotation frequency, which are used to drive the update of parameters γ and φ in the filtering algorithm.

4. The method for real-time separation and compensation of multi-source errors in torque measurement according to claim 1, characterized in that: The adaptive Kalman filter algorithm introduces a fading factor λ. When a sudden change in the measurement condition is detected, the filter gain is automatically increased by monitoring the statistical characteristics of the residual sequence e, so as to accelerate the response speed of the error characteristic parameters to the change in the condition.

5. The method for real-time separation and compensation of multi-source errors in torque measurement according to claim 1, characterized in that: The method also includes a self-checking step. When the torque testing system is in a stable no-load operation state, the corrected torque output value sequence over a period of time is collected and its mean μ_check and standard deviation σ_check are calculated. If |μ_check| exceeds a preset threshold or σ_check exceeds a preset fluctuation threshold, a prompt signal is triggered.

Images