Wheel assembly dynamic balance test data processing optimization method based on industrial internet of things

Abstract

The application discloses a wheel assembly dynamic balance test data processing optimization method based on an industrial internet of things, adopts sensor combination, combines low noise amplification and band pass filtering at a hardware level, adopts a time division duplex communication mechanism and a load balancing algorithm, and dynamically allocates bandwidth. A finite element model of a wheel-main shaft system is constructed, historical parameters are matched based on vehicle type characteristic vectors, a denoising parameter set is dynamically called, a wavelet base function is dynamically selected in combination with a signal kurtosis index, Gaussian noise and non-Gaussian vibration components are separated by using bispectrum analysis. A Blackman window function and an interpolation algorithm are used to compensate for spectrum leakage, and a phase error is controlled by using a phase-locked loop. A gradient descent method is used to update an influence coefficient matrix in real time, and multi-plane balance correction is carried out based on a double-plane coupling equation. An edge gateway detects amplitude mutation, triggers cloud collaborative analysis, and dynamically optimizes tolerance in combination with a quality distribution standard deviation.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing technology, and in particular to a method for optimizing dynamic balancing test data of wheel assemblies based on the Industrial Internet of Things. Background Technology

[0002] Existing wheel assembly dynamic balancing testing technologies have shortcomings: traditional methods rely on fixed sampling frequencies and manual experience judgment, leading to data redundancy and the omission of key abnormal signals; vibration data transmission often uses common industrial protocols, lacking dynamic adaptation to network load, and is prone to bandwidth conflicts and data loss when multiple nodes are running concurrently; in terms of signal processing, traditional filtering algorithms cannot adapt to the noise characteristics of wheel hubs made of different materials, and do not effectively distinguish between Gaussian noise and non-Gaussian vibration components, resulting in a low signal-to-noise ratio.

[0003] Dynamic balancing calculations rely on a preset influence coefficient matrix and lack a dynamic learning mechanism based on measured data, making it difficult to adapt to the rapid iteration of new wheel models. Tolerance judgments often use fixed thresholds and do not dynamically adjust based on the dispersion of wheel mass distribution, which can easily lead to misjudgments.

[0004] Existing systems lack edge-cloud collaborative analysis capabilities and cannot drive model parameter optimization based on historical data, resulting in limited testing efficiency and accuracy. These issues collectively constrain the intelligent and adaptive development of dynamic balancing testing, necessitating an innovative solution that integrates industrial IoT, dynamic optimization algorithms, and cloud-based collaborative mechanisms. Summary of the Invention

[0005] The purpose of this invention is to provide an optimization method for processing dynamic balance test data of wheel assemblies based on the Industrial Internet of Things.

[0006] The problem this invention aims to solve is to build an industrial Internet of Things system with multi-dimensional perception, intelligent decision-making, dynamic optimization, and cloud collaboration, so as to achieve full-process precision, adaptability, and efficiency in wheel assembly dynamic balancing testing.

[0007] The optimization method for processing dynamic balancing test data of wheel assemblies based on the Industrial Internet of Things (IIoT) adopts the following technical solution: A piezoelectric vibration sensor, a photoelectric encoder for speed measurement, and a temperature sensor are deployed at the wheel assembly dynamic balancing test station. The industrial IoT gateway has a built-in decision algorithm to calculate the variance of the vibration signal in real time and determine whether to trigger the high-speed sampling mode of the sensor based on the variance. The time-division duplex communication mechanism based on 5G URLLC is adopted to dynamically allocate the transmission bandwidth of the collected data according to the network load. When multiple nodes are detected to be transmitting concurrently, the gateway starts the load balancing algorithm to divert the data to different frequency bands. A finite element model of the wheel-axle system is built on an industrial IoT platform. Initial parameters are trained based on historical data in the cloud. The vehicle model identifier is received from the cloud. When a new model wheel hub is launched, the model parameters of similar models are called as initial values, and the corresponding initial denoising parameter set is called according to the wheel hub type. The wavelet basis function is dynamically selected based on the signal kurtosis index, the sliding variance of the noise energy is calculated, the third-order cumulative quantity is calculated for the denoised signal, and bispectral analysis is used to separate Gaussian noise and non-Gaussian vibration components, and noise frequency points with amplitudes lower than the baseline are removed. A Blackman window function is applied to the sampling period of the photoelectric encoder signal, and the spectral leakage caused by speed fluctuation is compensated by an interpolation algorithm, thereby controlling the phase error based on the speed change. For different wheel hubs, a dual-plane balancing algorithm is adopted. Multi-plane balancing correction is performed by solving the coupling equation. The pre-processed real-time vibration data is input into the dynamic balancing calculation model to calculate the theoretical amplitude and compare it with the measured amplitude. The influence coefficient is updated by the gradient descent method. The edge gateway node detects vibration anomalies. When a sudden change in amplitude is detected, the test is paused and cloud-based collaborative analysis is triggered. The detection records are analyzed, the standard deviation of the quality distribution is calculated, and the tolerance is optimized based on the calculated standard deviation.

[0008] Furthermore, the deployment of a piezoelectric vibration sensor, a photoelectric encoder for speed measurement, and a temperature sensor at the wheel assembly dynamic balancing test station includes: Piezoelectric vibration sensors are symmetrically arranged on both sides of the wheel main shaft bearing housing, perpendicular to the wheel rotation plane, to ensure simultaneous monitoring of radial and axial vibration. The weak charge signal output by the piezoelectric sensor is converted into a voltage signal by a low-noise charge amplifier, and a bandpass filter is added to eliminate low-frequency noise and high-frequency electromagnetic interference from the environment. It is connected to an industrial IoT gateway to ensure long-distance transmission stability. An incremental photoelectric encoder is installed at the end of the wheel spindle. A dust cover is added to the encoder housing to avoid dust interference during the test. The rotational speed is calculated in real time based on the frequency counting module built into the gateway using the A / B phase pulse signal output by the encoder. Rotational speed = (pulse frequency × 60) / number of pulses per revolution. A platinum resistance temperature sensor is embedded inside the wheel bearing housing to monitor the bearing temperature rise in real time. The temperature data is collected synchronously with the vibration signal. The sensitivity drift of the vibration sensor is corrected by a lookup table method, which includes the coefficient of thermal expansion of metals at different temperatures.

[0009] Furthermore, the variance of the vibration signal is calculated in real time, and the determination of whether to trigger the high-speed sampling mode of the sensor is based on the variance, including: A sliding time window of length T=0.5s was used to segment the vibration signal for processing. ,in Let n be the variance of the sliding window, and n be the number of sampling points within the window. These are the sampled point values. The trigger threshold is set based on the window mean and historical variance data, where historical variance is the average of the past 10 minutes. , This represents the average variance of the sliding window variance over the past 10 minutes. The variance and standard deviation of the sliding window variance over the past 10 minutes are given. k=2.5 is the safety factor, corresponding to a 99% confidence level based on statistical principles. The false trigger rate under different k values ​​was determined through actual testing, and k=2.5 was finally determined. High-speed sampling is initiated when all three conditions are met simultaneously: (1) Vibration variance (2) The rotational speed N is within the set range to avoid low-speed noise interference; (3) The bearing temperature is <70℃ to prevent sensor failure caused by high temperature. When multiple workstations trigger simultaneously, the gateway allocates high-speed sampling resources according to the first-come, first-served principle and ensures data priority transmission through the QoS mechanism of the 5GURLLC network.

[0010] Furthermore, the adoption of a time-division duplex communication mechanism based on 5G URLLC, dynamically allocating the transmission bandwidth of the collected data according to network load, includes: Based on service priority, time slots in the 5G frame structure are divided into three categories: URLLC dedicated time slots, eMBB shared time slots, and control signaling time slots. The URLLC dedicated time slots ensure the uploading of vibration data with ultra-low latency (<1ms); the eMBB shared time slots are used for large-capacity test data transmission; and the control signaling time slots account for a fixed proportion of 10%, carrying scheduling instructions between the gateway and the base station. When concurrent transmission across multiple nodes is detected, the URLLC time slot ratio is adjusted to 40% and the eMBB time slot ratio is reduced to 30% in real time via RRC reconfiguration messages to ensure priority transmission of critical data. Spectrum utilization optimization based on subcarrier spacing: 30kHz subcarrier spacing is used under normal load, providing wide coverage; under high load, it switches to 120kHz subcarrier spacing, with frequency band switching triggered by messages defined by 3GPP. The available bandwidth AB of each frequency band is evaluated by round-trip time (RTT) measurement. MSS is the maximum segment size, and Loss Rate is the packet loss rate. High-priority vibration data is transmitted through the 5G URLLC channel, and low-priority temperature data is transmitted through the Wi-Fi 6E channel.

[0011] Furthermore, the finite element model of the wheel-axle system is constructed on the industrial IoT platform, receiving vehicle model identifiers from the cloud, calling model parameters of similar vehicle models as initial values, and calling the corresponding initial denoising parameter set according to the wheel hub type, including: The vehicle identification is parsed into a 5-dimensional feature vector, which is [wheel diameter, wheel width, material type, weight distribution method, maximum load capacity]. The cosine similarity formula is used to match the similarity between the new vehicle model and the historical vehicle models, and the vehicle models with a similarity > 0.75 are selected as the parameter initialization source. Parameters, including elastic modulus, Poisson's ratio, and damping coefficient, are extracted from the finite element model of historical models. Based on the wheel diameter, a linear regression model with a determination coefficient > 0.95 is used to predict the parameters of the new model. After the new model is launched, the first 100 sets of test data are uploaded through the gateway, and the model parameters are fine-tuned in the cloud using the Adam optimizer with a learning rate of 0.001. Using the material type field in the vehicle feature vector, a preset set of denoising parameters is called: the wavelet basis function corresponding to aluminum alloy material type is db4, with a filtering bandwidth of 0.5-5kHz; the wavelet basis function corresponding to cast steel material type is sym8, with a filtering bandwidth of 0.3-3kHz; and the wavelet basis function corresponding to carbon fiber composite material type is coif5, with a filtering bandwidth of 0.2-2kHz. The vibration signal is decomposed into three layers of wavelet packets, and the energy entropy H of each sub-band is calculated. Subbands with entropy values ​​greater than 1.2 were selected as noise frequency bands. The energy of the i-th subband. This represents the sum of the energies of all subbands. Calculation of noise standard deviation based on a sliding window of length = 0.5s Calculated by median absolute deviation, The threshold T is set to , It is a median function. This represents the number of sampling points.

[0012] Furthermore, the step of dynamically selecting wavelet basis functions based on the signal kurtosis index, calculating the third-order cumulant of the denoised signal, using bispectral analysis to separate Gaussian noise and non-Gaussian vibrational components, and removing noise frequencies with amplitudes below the baseline includes: The formula for calculating the signal kurtosis index is as follows: ,in The mean of the signal. For mathematical expectation; >3 Selecting the db4 wavelet basis function is suitable for high kurtosis signals; 2< For signals with a kurtosis of ≤3, the sym8 wavelet basis function is selected, which is suitable for signals with moderate kurtosis. For signals with a kurtosis of ≤2, the coif5 wavelet basis function is selected, which is suitable for low kurtosis signals. The wavelet basis function selection is updated once per second. Third-order cumulant It reflects the non-Gaussian nature of the signal and is used to identify non-Gaussian vibrational components. The third-order statistic of Gaussian noise is 0, while the third-order cumulant of non-Gaussian vibrations is significantly non-zero. where t is time, and For delay parameters, =0.01S, =0.02s; Define bispectrum: ,in This is a function for calculating the signal spectrum. and For signal, Indicates conjugation; Gaussian noise is characterized by a bispectral value close to 0, the third-order statistic of a Gaussian signal is 0, and non-Gaussian vibrations are characterized by a significantly non-zero bispectral value; calculating the signal bispectral identification, the absolute value of the bispectral value > threshold = 10. −3 At the frequency points, non-Gaussian components are preserved while Gaussian noise is suppressed; The baseline is defined by calculating the upper limit of the 95% confidence interval based on historical data. If the amplitude of a frequency point is less than the baseline value and is lower than the baseline value for three consecutive measurements, it is determined to be a noise frequency point and is removed. If the bispectral value corresponding to the frequency point is greater than the threshold value (10), then the frequency point is removed. −3 If it is a non-Gaussian oscillation, it is retained as a valid signal because it may be a non-Gaussian oscillation.

[0013] Furthermore, the application of the Blackman window function, using an interpolation algorithm to compensate for spectral leakage caused by speed fluctuations, and controlling phase error based on speed changes, includes: The Blackman window function is a time-domain weighting function used to reduce spectral leakage. The window length N_5 should satisfy the ratio of the highest analysis frequency to the frequency resolution. The original pulse signal is sampled, and the sampling period is the reciprocal of the photoelectric encoder pulse frequency. The sampled data is truncated into a window of length N_5, and the Blackman window function is applied for weighting to suppress spectral leakage. When the rotational speed fluctuates, the sampling period is not strictly synchronized with the signal period, resulting in dispersed spectral energy. ,in Nominal frequency, The pulse count deviation is caused by speed fluctuations. The total number of pulses is used to monitor speed fluctuations and calculate the actual frequency. ,according to Adjust the interpolation step size and use the cubic spline algorithm to interpolate and compensate the spectrum in the frequency domain so that the energy is concentrated at the true frequency position; Real-time monitoring of speed fluctuations, calculation of phase deviation, adjustment of sampling time, and compensation of phase error via phase-locked loop. ,in For the corrected phase, For actual phase, Angular velocity, To compensate for the sampling time deviation caused by speed fluctuations through dynamic adjustment via a phase-locked loop, the proportional gain of the phase-locked loop is 0.1 and the integral gain is 0.01, ensuring that the phase error is <0.5°.

[0014] Furthermore, the dual-plane equilibrium algorithm is employed, which performs multi-plane equilibrium correction by solving the coupling equations, calculates the theoretical amplitude, and updates the influence coefficients using the gradient descent method, including: The theoretical amplitude vector is calculated by multiplying the influence coefficient matrix and the unbalance vector. The dimension of the influence coefficient matrix is ​​the number of measurement points × the number of correction planes. When the two planes are in equilibrium, the matrix form is a 2×2 matrix. The two-plane equilibrium model is ,in The amplitudes at the two measurement points are... The unbalance quantities in the two correction planes are solved using the least squares method. , , This is the influence coefficient matrix. This is the measured amplitude vector; For M correction planes, the equation becomes Through QR decomposition, we obtain ,in Let M be the amplitude vector. For unbalanced quantity vectors, For matrix The false reversal; Design the amplitude loss function L. ,in This is the measured amplitude vector. Given the theoretical amplitude vector, the update rule for the influence coefficient matrix is ​​as follows: , For learning rate, These are the updated and unupdated influence coefficient matrices, respectively. The convergence condition is an upper limit of 50 iterations and a change threshold of <10 for the amplitude loss function. −5 .

[0015] Furthermore, when a sudden amplitude change is detected, the test is paused and cloud-based collaborative analysis is triggered. The tolerance is optimized based on the calculated standard deviation, including: When the hub is rotating, if the amplitude suddenly exceeds twice the average value of the previous 10 cycles at a certain moment, it is considered abnormal. The edge gateway will pause the test and upload the abnormal timestamp and the complete vibration record of the past 500 sampling points. The cloud server calculates the standard deviation of the mass distribution based on historical data from the past 72 hours, and reconstructs the wheel hub mass distribution map using the Kriging interpolation method based on the historical mass distribution data. When the standard deviation of mass distribution > 0.04 kg, the new tolerance = the reference tolerance × 1.2; when 0.02 kg ≤ standard deviation of mass distribution ≤ 0.04 kg, the new tolerance = the reference tolerance × 1.0; when the standard deviation of mass distribution < 0.02 kg, the new tolerance = the reference tolerance × 0.8.

[0016] The beneficial effects of this invention are as follows: At the hardware level, the coordinated deployment of piezoelectric sensors, photoelectric encoders, and temperature sensors, combined with low-noise amplifiers and bandpass filters, ensures high-fidelity acquisition of vibration, rotational speed, and temperature data; at the algorithm level, based on a dynamic sampling triggering mechanism using sliding window variance and statistical confidence thresholds, combined with time-division duplexing and spectrum utilization optimization of 5G URLLC, ultra-low latency transmission of high-priority vibration data and intelligent allocation of network resources are achieved, solving the bandwidth conflict problem during concurrent transmission of multiple nodes.

[0017] The cloud-based system improves the denoising accuracy and non-Gaussian feature extraction capability of vibration signals by using a parameter initialization strategy driven by finite element models and historical data, combined with dynamic wavelet basis function selection for signal kurtosis, bispectral analysis, and third-order cumulant separation technology, thus providing a high signal-to-noise ratio input for subsequent dynamic balancing calculations.

[0018] A spectrum leakage suppression and interpolation compensation algorithm based on the Blackman window function, combined with phase-locked loop phase error control, overcomes the interference of speed fluctuations on spectrum analysis, ensuring the accuracy of theoretical amplitude calculation. Through gradient descent influence coefficient updates and solving the dual-plane coupling equations, combined with anomaly detection at the edge gateway and tolerance optimization driven by the standard deviation of cloud-based quality distribution, a closed-loop control link of data acquisition, transmission, analysis, and correction is formed. Attached Figure Description

[0019] Figure 1 This is a flowchart of an optimized method for processing dynamic balancing test data of wheel assemblies based on the Industrial Internet of Things. Detailed Implementation

[0020] The present invention will be further described clearly and completely below, but the scope of protection of the present invention is not limited thereto.

[0021] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0022] Example 1 The optimization method for processing dynamic balancing test data of wheel assemblies based on the Industrial Internet of Things (IIoT) adopts the following technical solution: A piezoelectric vibration sensor, a photoelectric encoder for speed measurement, and a temperature sensor are deployed at the wheel assembly dynamic balancing test station. The industrial IoT gateway has a built-in decision algorithm to calculate the variance of the vibration signal in real time and determine whether to trigger the high-speed sampling mode of the sensor based on the variance. The time-division duplex communication mechanism based on 5G URLLC is adopted to dynamically allocate the transmission bandwidth of the collected data according to the network load. When multiple nodes are detected to be transmitting concurrently, the gateway starts the load balancing algorithm to divert the data to different frequency bands. A finite element model of the wheel-axle system is built on an industrial IoT platform. Initial parameters are trained based on historical data in the cloud. The vehicle model identifier is received from the cloud. When a new model wheel hub is launched, the model parameters of similar models are called as initial values, and the corresponding initial denoising parameter set is called according to the wheel hub type. The wavelet basis function is dynamically selected based on the signal kurtosis index, the sliding variance of the noise energy is calculated, the third-order cumulative quantity is calculated for the denoised signal, and bispectral analysis is used to separate Gaussian noise and non-Gaussian vibration components, and noise frequency points with amplitudes lower than the baseline are removed. A Blackman window function is applied to the sampling period of the photoelectric encoder signal, and the spectral leakage caused by speed fluctuation is compensated by an interpolation algorithm, thereby controlling the phase error based on the speed change. For different wheel hubs, a dual-plane balancing algorithm is adopted. Multi-plane balancing correction is performed by solving the coupling equation. The pre-processed real-time vibration data is input into the dynamic balancing calculation model to calculate the theoretical amplitude and compare it with the measured amplitude. The influence coefficient is updated by the gradient descent method. The edge gateway node detects vibration anomalies. When a sudden change in amplitude is detected, the test is paused and cloud-based collaborative analysis is triggered. The detection records are analyzed, the standard deviation of the quality distribution is calculated, and the tolerance is optimized based on the calculated standard deviation.

[0023] refer to Figure 1 The diagram shown is a flowchart of the optimization method for dynamic balancing test data of wheel assemblies based on the Industrial Internet of Things.

[0024] Furthermore, the deployment of a piezoelectric vibration sensor, a photoelectric encoder for speed measurement, and a temperature sensor at the wheel assembly dynamic balancing test station includes: Two piezoelectric vibration sensors are symmetrically arranged on both sides of the wheel spindle bearing housing. The mounting holes are equidistant from the bearing centerline, with an error controlled within ±0.5mm. The sensors are installed strictly perpendicular to the wheel's rotation plane, with an installation angle deviation not exceeding ±2°. This arrangement allows the two sensors to monitor the radial vibration perpendicular to the rotation plane and the axial vibration component along the spindle direction, respectively. The weak charge signal output by the piezoelectric sensors is transmitted through a shielded cable to a low-noise charge amplifier with a gain of 1000, converting the high-impedance charge signal from the piezoelectric sensors into a low-impedance voltage signal. The output signal of the charge amplifier is then fed into a 50Hz~2kHz bandpass filter to eliminate low-frequency environmental noise and high-frequency electromagnetic interference. The filtered signal is connected to an industrial IoT gateway via an RS-485 bus, using the Modbus RTU protocol with a baud rate of 115200bps, 8 data bits, 1 stop bit, and no parity, ensuring stable long-distance signal transmission in industrial environments.

[0025] An incremental photoelectric encoder is bolted to the end of the wheel spindle. The encoder shaft is connected to the spindle via a flexible coupling to ensure coaxiality error is less than 0.05mm. The encoder has a resolution of 1024 pulses / revolution and outputs three-phase signals (A, B, and Z). The A and B phases have a 90° phase difference and are used for direction determination and frequency multiplication. The Z phase serves as a zero-point reference signal, one pulse per revolution. The encoder housing is made of aluminum alloy and equipped with a dust cover. A labyrinth seal is installed between the cover and the spindle to prevent dust interference during testing. The encoder output signal is connected to the frequency counting module built into the industrial IoT gateway. This module operates at a clock frequency of 100MHz. After quadrupling the frequency of the A and B phase pulses, it captures the pulse edges at a sampling rate of 1MHz. The real-time rotational speed is calculated by measuring the number of pulses per unit time: rotational speed = (pulse frequency × 60) / number of pulses per revolution. The gateway updates the rotational speed value every 10ms to ensure measurement accuracy during speed fluctuations.

[0026] A PT100 platinum resistance temperature sensor is embedded in a dedicated channel pre-drilled inside the wheel bearing housing. This channel has a diameter of 6mm and a depth of 25mm, located at the center of the contact surface between the outer ring of the bearing and the housing. The PT100 sensor is connected to the temperature measurement module of an industrial IoT gateway via a three-wire connection. The measurement circuit uses a constant current source excitation and a 24-bit high-precision ADC conversion, with a temperature measurement range of -50℃ to +150℃, an accuracy of ±0.1℃, and a resolution of 0.01℃. Temperature data and vibration signals are synchronized in time through a multi-channel synchronous acquisition module of the same gateway, with a synchronization accuracy better than 100μs. The sensitivity drift of the vibration sensor is corrected using a lookup table method, which includes the coefficient of thermal expansion of metals at different temperatures. The sensitivity changes of the vibration sensor at different temperatures (20℃, 30℃, 40℃, 50℃, 60℃, 70℃) were measured in advance under laboratory conditions to establish a temperature-sensitivity correction coefficient lookup table. During the table lookup process, the corresponding correction coefficient is obtained through linear interpolation based on the real-time temperature value. This coefficient is calculated based on the coefficient of thermal expansion of the metal and the temperature characteristics of the piezoelectric material at different temperatures. When the bearing temperature is detected to be >60℃, the high-temperature protection mode is activated, reducing the gain of the preamplifier of the vibration signal by 15% to avoid signal distortion caused by high temperature, and simultaneously displaying a high-temperature warning message on the interface.

[0027] Furthermore, the variance of the vibration signal is calculated in real time, and the determination of whether to trigger the high-speed sampling mode of the sensor is based on the variance, including: The signal processing module built into the industrial IoT gateway uses a sliding time window of length T=0.5s to segment the vibration signal for processing. In standard sampling mode, the vibration sensor operates at a sampling rate of 2kHz, therefore each sliding window contains n=1000 sampling points. The sliding window is implemented using a first-in, first-out buffer, with a window movement step of 50ms to ensure a balance between real-time performance and computational load. Variance calculation is strictly performed according to the formula: ,in Let n be the variance of the sliding window, and n be the number of sampling points within the window. These are the sampled point values. This is the window mean. The calculation process is completed by the gateway's built-in 32-bit floating-point DSP unit, with single-precision floating-point precision. Each variance calculation takes no more than 2ms, ensuring that the calculation is completed within the window movement step.

[0028] A trigger threshold is set based on historical variance data. A historical variance database is established to store the variance values ​​of 1200 sliding windows over the past 10 minutes. This database is implemented using a circular buffer, and statistical analysis of the historical variance data is performed every minute. The historical variance is the average of the past 10 minutes. , This represents the average variance of the sliding window variance over the past 10 minutes. This represents the standard deviation of the variance of the sliding window over the past 10 minutes. The safety factor setting of k=2.5 is based on statistical principles and corresponds to a 99% confidence level. The false trigger rate under different k values ​​was determined through actual testing. After verification through 200 sets of tests under different operating conditions, k=2.5 effectively detects abnormal vibrations while keeping the false trigger rate below 3%. This threshold is updated every minute to ensure adaptation to slow changes in equipment operating conditions.

[0029] High-speed sampling is initiated when all three conditions are met simultaneously, and the "AND" logic gate is used for judgment: (1) the vibration variance calculated in the current window. (1) The condition detects sudden changes in vibration signal, reflecting hub imbalance or mechanical failure; (2) The rotational speed N is within the effective test range of 800~1200rpm. The rotational speed data comes from the real-time measurement of the photoelectric encoder and is updated every 10ms. This range is set based on the following engineering considerations: when it is below 800rpm, the vibration signal-to-noise ratio is low and the environmental noise interference is significant; when it is above 1200rpm, it may enter the equipment resonance zone, increasing the safety risk. The rotational speed range limit effectively avoids false triggering caused by low-speed noise interference; (3) The safety threshold of bearing temperature <70℃. The temperature data comes from the PT100 platinum resistance sensor, and the sampling interval is 100ms. This threshold is set based on the working temperature characteristics of the piezoelectric sensor. When the temperature exceeds 70℃, the performance of the piezoelectric material is unstable, which may lead to measurement distortion or permanent damage to the sensor. This condition effectively prevents sensor failure caused by high temperature; The three conditions are determined by the gateway's real-time control module every 50ms. When multiple workstations trigger simultaneously, the gateway allocates high-speed sampling resources according to a first-come, first-served principle and ensures data priority transmission through the QoS mechanism of the 5G URLLC network. In specific implementation, each workstation's trigger request carries a timestamp accurate to microseconds. The resource scheduling module maintains a priority queue and allocates resources in timestamp order. The duration for each workstation to obtain high-speed sampling permission is fixed at 5 seconds to ensure that each workstation receives a fair service opportunity. After the 5-second high-speed sampling period ends, it automatically reverts to the regular sampling mode and uploads the complete dataset collected during the high-speed sampling period (containing 100,000 sampling points) along with related parameters such as timestamps, rotational speed, and temperature to the cloud analysis platform for subsequent vibration feature extraction and fault diagnosis.

[0030] Furthermore, the adoption of a time-division duplex communication mechanism based on 5G URLLC, dynamically allocating the transmission bandwidth of the collected data according to network load, includes: Based on service priority, time slots in the 5G frame structure are divided into three categories: URLLC dedicated time slots, eMBB shared time slots, and control signaling time slots. The URLLC dedicated time slots ensure the uploading of vibration data with ultra-low latency (<1ms); the eMBB shared time slots are used for large-capacity test data transmission; and the control signaling time slots account for a fixed proportion of 10%, carrying scheduling instructions between the gateway and the base station. The gateway's built-in network load monitoring module continuously monitors the number of concurrent connections and data queue length. When multiple nodes are detected transmitting concurrently, a time slot ratio reconfiguration process is triggered. The specific steps are as follows: A scheduling request is sent to the base station via the MAC layer control channel, requesting an adjustment to the time slot ratio. After the base station responds, the gateway receives an RRC reconfiguration message containing the new configuration parameters for the time slot allocation table. The RRC reconfiguration message is used to adjust the URLLC time slot ratio to 40% and the eMBB time slot ratio to 30% in real time, ensuring priority transmission of critical data, with a reconfiguration latency of no more than 50ms.

[0031] Spectrum utilization optimization based on subcarrier spacing: Under normal load, a 30kHz subcarrier spacing is used, which has a wide coverage and is suitable for medium-distance transmission; under high load, the subcarrier spacing is switched to 120kHz to reduce the symbol spacing and improve latency sensitivity. Frequency band switching is triggered by messages defined by 3GPP. The state machine built into the gateway monitors network load indicators. When the switching conditions are met, a measurement report is sent to the base station, and the base station returns a reconfiguration message containing the target configuration.

[0032] The available bandwidth AB of each frequency band is evaluated by round-trip time (RTT) measurement. The maximum segment size (MSS) and packet loss rate are used to transmit high-priority vibration data via the 5G URLLC channel and low-priority temperature data via the Wi-Fi 6E channel, achieving load decoupling. This communication mechanism has undergone rigorous testing. Under a high-load scenario with 10 concurrent nodes and a vibration sampling rate of 100Hz per node, the end-to-end latency of the URLLC data remains within 0.8ms, and the packet delivery rate reaches 99.999%, meeting the stringent requirements for real-time data performance and reliability in wheel dynamic balancing testing.

[0033] Furthermore, the finite element model of the wheel-axle system is constructed on the industrial IoT platform, receiving vehicle model identifiers from the cloud, calling model parameters of similar vehicle models as initial values, and calling the corresponding initial denoising parameter set according to the wheel hub type, including: Vehicle model identifiers are encoded in 16-bit hexadecimal format and parsed into a 5-dimensional feature vector, which includes [wheel diameter, wheel width, material type, weight distribution, and maximum load capacity]. The historical vehicle database contains 5000 sets of tested vehicle data, each storing the feature vector and corresponding finite element model parameters. A cosine similarity formula is used to match the similarity between new and historical vehicle models. The feature vectors are first normalized: wheel diameter divided by 2000, wheel width divided by 500, and maximum load capacity divided by 5000. Material type and weight distribution are encoded using one-hot encoding. Historical vehicle models are sorted in descending order of similarity, and those with a similarity > 0.75 and at least 5 sets are selected as the source for parameter initialization.

[0034] Parameters, including elastic modulus, Poisson's ratio, and damping coefficient, were extracted from the finite element models of historical vehicle models. A linear regression model with a coefficient of determination > 0.95 was used to predict the parameters of the new vehicle model based on the wheel diameter. After the new model was launched, the first 100 sets of test data were uploaded via the gateway. The test data included vibration signals, rotational speed, and imbalance data. The model parameters were fine-tuned in the cloud using an Adam optimizer with a learning rate of 0.001 and a weight decay coefficient of 0.0001, stopping after 50 iterations.

[0035] Analysis of 1000 sets of historical vibration data, with standardized test conditions: 1000 rpm rotation speed, 10g unbalance, and 25℃ ambient temperature, revealed significant differences in the spectral characteristics of vibration signals from different materials: aluminum alloys exhibited dominant frequencies concentrated between 2-5 kHz, rich high-frequency components, and a broad-peak spectrum, reflecting high stiffness and low damping; cast steels had dominant frequencies concentrated between 0.5-3 kHz, dominated by low-frequency components, and a narrow-peak spectrum, reflecting higher damping; and carbon fiber had dominant frequencies concentrated between 0.2-2 kHz, rich high-frequency details, and a multi-peak spectrum, reflecting anisotropy and interlaminar vibration characteristics.

[0036] By comparing the signal-to-noise ratio improvement effects of different wavelet basis functions on various materials, the optimal match was determined. Using material type fields in the vehicle feature vector, such as 6061 aluminum alloy and cast steel, preset denoising parameter sets were applied: the wavelet basis function for aluminum alloy was db4, with a filtering bandwidth of 0.5-5kHz; the wavelet basis function for cast steel was sym8, with a filtering bandwidth of 0.3-3kHz; and the wavelet basis function for carbon fiber composite materials was coif5, with a filtering bandwidth of 0.2-2kHz.

[0037] The vibration signal denoising algorithm employs a 3-level wavelet packet decomposition algorithm. Since noise tends to whiten after transformation, the wavelet domain is more conducive to denoising than the time domain. The vibration signal is then subjected to 3-level wavelet packet decomposition, and the energy entropy H of each sub-band is calculated. , The energy of the i-th subband. This represents the sum of the energies of all subbands.

[0038] Noise frequency bands are determined based on entropy values: sub-bands with entropy values ​​> 1.2 are selected as noise frequency bands. An entropy value H > 1.2 indicates uniform noise distribution, while H < 1.2 indicates concentrated signal energy. Test data verification: Analysis of 100 sets of vibration signals containing artificial noise showed a correlation coefficient of 0.93 between entropy value and noise proportion. The system marks sub-bands with H > 1.2 as noise frequency bands and applies threshold processing; sub-bands with H < 1.2 retain their original coefficients.

[0039] Calculation of noise standard deviation based on a sliding window of length = 0.5s Calculated by median absolute deviation, The threshold T is set to , It is a median function. The number of sampling points is represented by the noise standard deviation, which indicates the noise intensity. A larger standard deviation indicates stronger noise and requires a larger threshold. When reconstructing the signal, a wavelet packet reconstruction algorithm is used to preserve effective vibration characteristics and suppress noise interference.

[0040] Furthermore, the step of dynamically selecting wavelet basis functions based on the signal kurtosis index, calculating the third-order cumulant of the denoised signal, using bispectral analysis to separate Gaussian noise and non-Gaussian vibrational components, and removing noise frequencies with amplitudes below the baseline includes: Kurtosis is a statistical measure of the sharpness of a signal distribution, reflecting the intensity of the impulsive components in the signal. High kurtosis indicates that the signal contains a large number of spike pulses, such as the impact vibration caused by bearing failure, while low kurtosis indicates that the signal is relatively smooth, such as the vibration caused by uniform rotation.

[0041] The formula for calculating the signal kurtosis index is as follows: ,in The mean of the signal. For mathematical expectation; >3 Selecting the db4 wavelet basis function is suitable for highly kurtotic signals such as spike pulses; 2< For signals with a kurtosis of ≤3, the sym8 wavelet basis function is selected, which is suitable for moderately kurtotic signals such as steady fluctuations. For values ​​≤2, the COIF5 wavelet basis function is selected, suitable for low-kurtosis signals such as high-frequency details. The wavelet basis function selection is updated every second. When kurtosis k=4.2 is detected, the signal is switched to the DB4 wavelet basis function to enhance the suppression of impact noise. In actual testing, when the kurtosis value suddenly increases from 2.8 to 4.2 (indicating an abnormal impact), the wavelet basis switching (from sym8 to DB4) is completed within 100ms. After the switching is completed, the DB4 wavelet's suppression of impact components is improved by 37%, and the signal-to-noise ratio of the vibration signal increases from 15.3dB to 21.0dB.

[0042] Third-order cumulant It reflects the non-Gaussian nature of the signal and is used to identify non-Gaussian vibration components, such as impact signals caused by bearing failure. The third-order statistic of Gaussian noise is 0, while the third-order cumulative quantity of non-Gaussian vibration, such as bearing failure, is significantly non-zero. where t is time, and For the time delay parameter, typically take... =0.01S, =0.02S. The time delay parameter was determined by analyzing 1000 sets of bearing fault data. =0.01S, =0.02S It has the maximum and the smallest standard deviation. In practical applications, healthy bearings... =0.012±0.005, outer ring fault bearing =0.0187±0.032, inner ring fault bearing =0.254±0.041, and the significant discrimination score verifies the effectiveness of the method.

[0043] Define bispectrum: ,in This is a function for calculating the signal spectrum. and For signal, Indicates conjugate. Gaussian noise is characterized by a bispectral value close to 0, the third-order statistic of a Gaussian signal is 0, and non-Gaussian vibrations exhibit significantly non-zero bispectral values, such as bearing fault signals; calculating the bispectral value for bispectral identification (absolute value > threshold = 10). −3 The frequency points are selected to retain non-Gaussian components and suppress Gaussian noise.

[0044] The baseline is defined by calculating the upper limit of the 95% confidence interval based on historical data. If the amplitude of a frequency point is less than the baseline value and is lower than the baseline value for three consecutive measurements, it is determined to be a noise frequency point and is removed. If the bispectral value corresponding to the frequency point is greater than the threshold value (10), then the frequency point is removed. −3 If it is a non-Gaussian oscillation, it is retained as a valid signal because it may be a non-Gaussian oscillation.

[0045] Furthermore, the application of the Blackman window function, using an interpolation algorithm to compensate for spectral leakage caused by speed fluctuations, and controlling phase error based on speed changes, includes: The Blackman window function is a time-domain weighting function used to reduce spectral leakage. The window length N_5 should satisfy the ratio of the highest analysis frequency (e.g., 5kHz) to the frequency resolution (e.g., 5Hz). The original pulse signal is sampled, and the sampling period is the reciprocal of the photoelectric encoder pulse frequency. The sampled data is truncated into a window of length N_5, and the Blackman window function is applied for weighting to suppress spectral leakage.

[0046] When the rotational speed fluctuates, the sampling period is not strictly synchronized with the signal period, resulting in dispersed spectral energy. ,in Nominal frequency, The pulse count deviation is caused by speed fluctuations. The total number of pulses is used to monitor speed fluctuations. Calculate the actual frequency ,according to By adjusting the interpolation step size, the cubic spline interpolation algorithm is used to interpolate and compensate the spectrum in the frequency domain, so that the energy is concentrated at the true frequency position. Through experimental verification, the cubic spline interpolation algorithm can reduce spectral leakage, smoothly fit the curve, and avoid sawtooth distortion.

[0047] Real-time monitoring of speed fluctuations Calculate the phase deviation, adjust the sampling time, and compensate for the phase error using a phase-locked loop. ,in For the corrected phase, For actual phase, Angular velocity, To compensate for the sampling time deviation caused by speed fluctuations through dynamic adjustment via a phase-locked loop, the proportional gain of the phase-locked loop is 0.1 and the integral gain is 0.01, ensuring that the phase error is <0.5°.

[0048] Furthermore, the dual-plane equilibrium algorithm is employed, which performs multi-plane equilibrium correction by solving the coupling equations, calculates the theoretical amplitude, and updates the influence coefficients using the gradient descent method, including: The theoretical amplitude vector is calculated by multiplying the influence coefficient matrix and the unbalance vector. The dimension of the influence coefficient matrix is ​​the number of measurement points × the number of correction planes. When the two planes are in equilibrium, the matrix form is a 2×2 matrix. When the hub rotates, the imbalance generates coupled vibrations in two planes, such as left and right. Single-plane balancing has the problem that correcting one plane will increase the vibration of the other plane. The two-plane equilibrium model is ,in The amplitudes at the two measurement points are... The unbalance quantities in the two correction planes are solved using the least squares method. , , This is the influence coefficient matrix. This is the measured amplitude vector; For M correction planes, the equation becomes Through QR decomposition, we obtain ,in Let M be the amplitude vector. For unbalanced quantity vectors, For matrix The false reversal; Design the amplitude loss function L. ,in This is the measured amplitude vector. Given the theoretical amplitude vector, the update rule for the influence coefficient matrix is ​​as follows: , For learning rate, These are the updated and unupdated influence coefficient matrices, respectively. The convergence condition is an upper limit of 50 iterations and a change threshold of <10 for the amplitude loss function. −5 The convergence speed was found to be optimal when η=0.01, as verified by 100 sets of experiments.

[0049] Furthermore, when a sudden amplitude change is detected, the test is paused and cloud-based collaborative analysis is triggered. The tolerance is optimized based on the calculated standard deviation, including: When the hub is rotating, if the amplitude suddenly exceeds twice the average value of the previous 10 cycles at a certain moment, it is considered abnormal. The edge gateway will pause the test and upload the abnormal timestamp and the complete vibration record of the past 500 sampling points. The cloud server calculates the standard deviation of the mass distribution based on historical data from the past 72 hours, and reconstructs the wheel hub mass distribution map using the Kriging interpolation method based on the historical mass distribution data. Through 1000 sets of wheel hub tests, it was verified that when the standard deviation of mass distribution is >0.04kg, the tolerance needs to be increased by 20% to avoid misjudgment; when the standard deviation of mass distribution is <0.02kg, the tolerance can be reduced by 20% to improve accuracy. When the standard deviation of mass distribution > 0.04 kg, the new tolerance = the reference tolerance × 1.2; when 0.02 kg ≤ standard deviation of mass distribution ≤ 0.04 kg, the new tolerance = the reference tolerance × 1.0; when the standard deviation of mass distribution < 0.02 kg, the new tolerance = the reference tolerance × 0.8.

[0050] Example 2 This embodiment uses the dynamic balancing test of a car wheel assembly as an example to explain the implementation process in detail. Wheel assembly parameters: wheel hub material is aluminum alloy, diameter is 16 inches, width is 6 inches, and maximum load capacity is 500 kg. Test conditions: speed range 800-1200 rpm, ambient temperature 25°C.

[0051] Two piezoelectric vibration sensors, model PCB 352C03, are symmetrically arranged on both sides of the wheel spindle bearing housing, perpendicular to the wheel's rotation plane, with a sampling rate of 100kHz. The signal is processed by a low-noise charge amplifier with a gain of 1000 and a bandpass filter of 50Hz-2kHz, and then transmitted to an industrial IoT gateway via an RS-485 bus.

[0052] A photoelectric encoder with a resolution of 1024 pulses / revolution is installed at the end of the spindle, outputting A / B phase pulses in real time. The gateway's built-in frequency counting module calculates the rotational speed: Rotational speed = (pulse frequency × 60) / 1024. When the pulse frequency is 17.07kHz, the rotational speed = (17070 × 60) / 1024 ≈ 1000rpm.

[0053] The PT100 temperature sensor is embedded in the bearing housing to monitor the temperature. The current value is 60°C. It is used to correct the sensitivity of the vibration sensor. When the temperature is >60°C, the gain decreases by 15%.

[0054] A sliding window with a length T = 0.5s and a sampling point count n = 50000 is used to calculate the variance of the vibration signal. The current window sampling point values ​​are: [0.102, 0.118, 0.095, 0.108, 0.125, 0.092, 0.105, 0.136, 0.085, 0.112], in mV. For simplicity, 10 points are used in the example; the actual count is n = 50000.

[0055] The calculated mean is (0.102 + 0.118 + ... + 0.112) / 10 = 1.078 / 10 = 0.1078 mV, and the calculated variance is 0.0000508 mV. 2 Based on the historical variance data of the past 10 minutes, the average variance is 0.00004, the standard deviation of the variance is 0.00001, and the safety factor k=2.5, then Threshold=0.00004+2.5×0.00001=0.000065.

[0056] Trigger condition check: Variance 0.0000508 < 0.000065, not meeting the >Threshold condition; Rotation speed N = 1000 rpm, within the range of 800-1200 rpm; Temperature = 60°C < 70°C; High-speed sampling is not triggered because the variance does not exceed the threshold.

[0057] The signal kurtosis index was calculated based on the same sampling points, with a mean value of 0.1078; the fourth-order central moment was calculated as follows: The second-order central moment is calculated to be 0.0000508; the kurtosis is 4.32 × 10⁻⁶. -9 / (0.0000508) 2 ≈1.67≤2, choose the coif5 wavelet basis function.

[0058] Calculate the third-order cumulant for the denoised signal, given the signal sequence x(t) = [x1, x2, x3, ..., x10], corresponding to the sampling point offset; since the sampling rate is 100kHz, the offset exponents are 1000 and 2000 points, and the time delay... =0.01S, =0.02S; Statistical expectation, calculated as 0.001 (non-zero), indicating a non-Gaussian component; Frequency point 100Hz, frequency point At 150 Hz, the bispectral value B = 0.005 > 10. −3 , and retain it as a valid signal.

[0059] The initial influence coefficient matrix of the two planes is The measured amplitude vector is [0.5, 0.6]. T First, calculate the imbalance quantity using the coupled equations. Then, obtain the inverse matrix of the initial influence coefficient matrix. The unbalance quantity is The theoretical amplitude is This is consistent with the actual measurement.

[0060] The subsequently measured amplitude vector is [0.55, 0.65]. T Therefore, the error vector is [-0.05, -0.05]. T The amplitude loss function value is 0.0025, η=0.01, and K 11 For example, The updated K 11 =1.2−0.01×(−0.014685)=1.2+0.00014685=1.20014685, iterate until convergence.

[0061] The gateway monitors the amplitude in real time. If the amplitude of a certain period suddenly increases to twice the average value of the previous 10 periods (average value 0.5 mm / s), and then suddenly increases to 1.0 mm / s, the test is paused, and the abnormal data of the timestamp + 500 sampling points is uploaded to the cloud. The mass distribution map is reconstructed using Kriging interpolation to drive tolerance optimization.

[0062] Cloud analysis of historical data from the past 72 hours shows a mass sample of 1000 wheel hubs [19.8, 20.1, 20.0, 19.9, 20.2, ...], with an average mass of 20.0 kg, a sample variance of 0.0009, a calculated standard deviation of 0.03, a baseline tolerance of 0.1, and a new tolerance = baseline tolerance × 1.0 = 0.1.

[0063] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0064] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0065] This invention discloses an optimization method for processing dynamic balancing test data of wheel assemblies based on the Industrial Internet of Things (IIoT). At the hardware level, it employs a sensor combination, combined with low-noise amplification and bandpass filtering; it utilizes a time-division duplex communication mechanism and a load balancing algorithm to dynamically allocate bandwidth. A finite element model of the wheel-axle system is constructed, historical parameters are matched based on vehicle model feature vectors, and a denoising parameter set is dynamically invoked. Wavelet basis functions are dynamically selected based on signal kurtosis indices, and bispectral analysis is used to separate Gaussian noise and non-Gaussian vibration components. Blackman window functions and interpolation algorithms are used to compensate for spectral leakage, and a phase-locked loop controls phase error. The influence coefficient matrix is ​​updated in real time using gradient descent, and multi-plane balance correction is performed based on a biplane coupling equation. When the edge gateway detects abrupt amplitude changes, it triggers cloud-based collaborative analysis, and the tolerance is dynamically optimized based on the standard deviation of the mass distribution.

Claims

1. A method for processing and optimizing dynamic balancing test data of wheel assemblies based on the Industrial Internet of Things, characterized in that, include: Vibration sensors are symmetrically arranged on both sides of the wheel spindle bearing housing, perpendicular to the wheel rotation plane. They are connected to an industrial IoT gateway through a bandpass filter. An optical encoder is installed at the end of the wheel spindle. The rotation speed is calculated in real time by the A / B phase pulse signal output by the encoder. At the same time, a temperature sensor is set inside the bearing housing for environmental parameter monitoring. The vibration signal is segmented using a sliding time window. A trigger threshold is set based on historical variance data. When the vibration variance exceeds the threshold and the rotational speed is within the set range, the high-speed sampling mode of the sensor is activated. The time-division duplex communication mechanism based on 5G URLLC is adopted. The data transmission bandwidth is dynamically allocated according to the network load status. The time slots in the 5G frame structure are divided into URLLC dedicated time slots, eMBB shared time slots and control signaling time slots. When multiple nodes are detected to be transmitting concurrently, the load balancing algorithm is started to divert the data to different frequency bands. A finite element model of the wheel-axle system is constructed through an industrial IoT platform. The initial parameters of the model are trained based on historical data in the cloud. The vehicle model identifier is received from the cloud. When a new model wheel hub is launched, the model parameters of similar models are called as initial values, and the corresponding initial denoising parameter set is matched according to the wheel hub material type. The optimal wavelet basis function is dynamically selected based on the signal kurtosis index, the sliding variance of noise energy is calculated, the third-order cumulative quantity is calculated for the denoised signal, and the Gaussian noise and non-Gaussian vibration components are separated by bispectral analysis, and interference frequency points with amplitudes lower than the noise baseline are eliminated. A Blackman window function is applied to the sampling period of the photoelectric encoder signal, and the spectral leakage caused by speed fluctuation is compensated by an interpolation algorithm. The phase error compensation parameters are dynamically adjusted based on the real-time speed change rate. For wheel hubs with different structural characteristics, a dual-plane balancing algorithm is adopted. Multi-plane balancing correction is performed by solving the vibration coupling equation. The pre-processed real-time vibration data is input into the dynamic balancing calculation model to calculate the theoretical vibration amplitude and compare it with the measured amplitude. The influence coefficient matrix is ​​iteratively updated by the gradient descent method. The edge gateway node monitors the abnormal state of the vibration signal in real time. When the amplitude change exceeds the preset threshold, the test process is paused and a collaborative analysis mechanism with the cloud is triggered. The standard deviation of the quality distribution is calculated based on historical test records, and the dynamic balance tolerance range is dynamically optimized based on the standard deviation analysis results.

2. The method for processing and optimizing wheel assembly dynamic balancing test data based on the Industrial Internet of Things as described in claim 1, characterized in that, An optical encoder is installed at the end of the wheel axle, and a temperature sensor is installed inside the bearing housing for environmental parameter monitoring. A sliding time window is used to segment the vibration signal. When the vibration variance exceeds a threshold and the rotational speed is within a set range, the sensor's high-speed sampling mode is activated, including: A photoelectric encoder is installed at the end of the wheel spindle. Based on the A / B phase pulse signal and the number of pulses per revolution output by the encoder, the wheel speed is calculated in real time through the frequency counting module built into the gateway. A temperature sensor is embedded inside the wheel bearing housing, and a temperature-sensitivity correction lookup table is constructed. Based on the real-time temperature value, the sensitivity drift of the vibration sensor caused by temperature changes is compensated by the lookup table method. A sliding time window mechanism is used to perform segmented statistical analysis on the vibration signal, calculate the variance of the signal within the window, and dynamically set the trigger threshold based on the statistical characteristics of historical operating data. ,in Let n be the variance of the sliding window, and n be the number of sampling points within the window. These are the sampled point values. The trigger threshold is set based on the window mean and historical variance data, where historical variance is the average value over a preset period of time. , The average variance of the variance over a previously preset time sliding window. The variance and standard deviation of the variance over the past preset time sliding window, where k is the safety factor; A multi-condition joint judgment mechanism is set to start the high-speed sampling mode. High-speed sampling is activated when the following conditions are met simultaneously: the variance of the vibration signal exceeds the dynamically set threshold, the wheel speed is within the preset effective range, and the bearing temperature is below the safety threshold. When multiple test stations simultaneously request high-speed sampling resources, the industrial IoT gateway dynamically schedules and allocates the high-speed sampling resources according to the preset first-come-first-served principle.

3. The method for processing and optimizing wheel assembly dynamic balancing test data based on the Industrial Internet of Things as described in claim 1, characterized in that, The adoption of a time-division duplex communication mechanism based on 5G URLLC, which dynamically allocates data transmission bandwidth according to network load status, includes: When multiple nodes concurrent data transmission requests are detected, the proportion of various time slot resources is dynamically adjusted through the Radio Resource Control (RRC) reconfiguration mechanism, increasing the proportion of URLLC dedicated time slots and decreasing the proportion of eMBB shared time slots. Based on network load status, the subcarrier spacing is adaptively adjusted, and the frequency band switching mechanism is triggered by the information block message defined by the 3GPP standard. Different subcarrier configuration parameters are used under normal load and high load conditions. The available bandwidth AB of each communication frequency band is evaluated by measuring the round-trip time (RTT). MSS is the maximum segment size, and Loss Rate is the packet loss rate. A differentiated transmission strategy is implemented based on data priority, transmitting high-priority vibration monitoring data through the 5G URLLC channel and low-priority temperature monitoring data through the Wi-Fi 6E channel.

4. The method for processing and optimizing wheel assembly dynamic balancing test data based on the Industrial Internet of Things as described in claim 1, characterized in that, The process involves constructing a finite element model of the wheel-axle system through an industrial IoT platform, receiving vehicle model identifiers from the cloud, calling model parameters of similar vehicle models as initial values, and matching a corresponding initial denoising parameter set based on the wheel hub material type, including: The vehicle model identifier is parsed into a multi-dimensional feature vector, which includes wheel hub geometry parameters, material properties, structural characteristics and load-bearing capacity parameters. The cosine similarity formula is used to calculate the similarity value between the new vehicle model and the historical vehicle model. Historical vehicle models that exceed the preset similarity threshold are selected as the source of parameter initialization. Physical parameters, including elastic modulus, Poisson's ratio, and damping coefficient, are extracted from the finite element model of the selected historical vehicle model. A high-fit regression prediction model is constructed based on the wheel hub size parameters to generate the initial model parameters for the new vehicle model. After the new vehicle model is put into operation, the actual test data uploaded through the gateway is used to iteratively fine-tune the model parameters using optimization algorithms. Based on the material type attribute in the vehicle feature vector, a preset set of vibration signal denoising parameters is matched, with different wavelet basis function selections and filter bandwidth configurations corresponding to different material types; The acquired vibration signal was subjected to multi-level wavelet packet decomposition, and the energy entropy H of each sub-band was calculated. ,in The energy of the i-th subband. The sum of energy of all sub-bands is used to determine the sub-bands whose energy entropy exceeds a preset threshold as noise-dominant frequency bands; Calculating the noise standard deviation based on a sliding window Calculated based on median absolute deviation. Set dynamic threshold ,in It is a median function. This represents the number of sampling points.

5. The method for processing and optimizing wheel assembly dynamic balancing test data based on the Industrial Internet of Things as described in claim 1, characterized in that, The process of dynamically selecting the optimal wavelet basis function based on the signal kurtosis index, calculating the third-order cumulant of the denoised signal, separating Gaussian noise from non-Gaussian vibrational components using bispectral analysis, and eliminating interference frequency points with amplitudes lower than the noise baseline includes: The kurtosis index of a vibration signal is calculated using the following formula: ,in The mean of the signal. The expected value is used as the mathematical expectation. Based on the range of kurtosis index values, the corresponding optimal wavelet basis function type is selected to achieve optimal matching for different vibration characteristics. The wavelet basis function selection is updated at fixed time intervals. Calculate the third-order cumulant of the denoised vibration signal. , where t is time, and For delay parameters; Define the bispectral function of the signal: ,in This is a function for calculating the signal spectrum. and For signal, It represents conjugate; by calculating the bispectral density of the signal, it identifies effective frequency points where the absolute value of the bispectral density exceeds a preset threshold. Based on the statistical characteristics of historical vibration data, the upper limit of the confidence interval is calculated as the noise baseline threshold. When the signal amplitude at a certain frequency point is lower than the baseline threshold and remains below the baseline in multiple consecutive measurements, the frequency point is determined to be a noise interference frequency point and is removed. If the frequency point also has the characteristics of exceeding the bispectral threshold, it is determined to be a valid vibration signal and is retained.

6. The method for processing and optimizing wheel assembly dynamic balancing test data based on the Industrial Internet of Things as described in claim 1, characterized in that, The step of applying a Blackman window function to the sampling period of the photoelectric encoder signal, compensating for spectral leakage caused by speed fluctuations through an interpolation algorithm, and dynamically adjusting the phase error compensation parameters based on the real-time speed change rate includes: The original pulse signal is sampled at equal intervals, and the sampling period is set to the reciprocal of the photoelectric encoder pulse frequency. The sampled data is truncated into a fixed-length time window, and the Blackman window function is applied for time-domain weighted processing. The length of the window function is set according to the analysis requirements, and meets the requirement of not being less than the ratio of the highest analysis frequency to the required frequency resolution. Real-time monitoring of wheel speed fluctuations and calculation of actual frequency. Its calculation formula is ,in Nominal frequency, The pulse count deviation is caused by speed fluctuations. The total number of pulses is given; the frequency domain interpolation step size is adjusted based on the calculated actual frequency, and a cubic spline interpolation algorithm is used to interpolate and compensate the discrete spectrum in the frequency domain; Phase deviation is calculated in real time, the sampling time reference is adjusted, and phase error is compensated in a closed loop through a phase-locked loop control mechanism. The phase correction calculation formula is as follows: ,in For the corrected phase, For actual phase, Angular velocity, This refers to the sampling time deviation caused by the speed fluctuations dynamically adjusted via a phase-locked loop.

7. The method for processing and optimizing wheel assembly dynamic balancing test data based on the Industrial Internet of Things as described in claim 1, characterized in that, The method employs a dual-plane equilibrium algorithm, which performs multi-plane equilibrium correction by solving the vibration coupling equation, calculates the theoretical amplitude, and iteratively updates the influence coefficient matrix using the gradient descent method, including: The theoretical amplitude vector is calculated by performing matrix multiplication on the influence coefficient matrix and the unbalance vector. The dimension of the influence coefficient matrix is ​​configured as the product of the number of measurement points and the number of correction planes. Under the dual-plane balance configuration, the matrix has a 2×2 dimension structure. Establish a two-plane equilibrium mathematical model, the expression of which is: ,in The amplitudes at the two measurement points are... The unbalance between the two correction planes is calculated using the least squares method. The formula is as follows: , This is the influence coefficient matrix. This is the measured amplitude vector; For a general equilibrium scenario with M correction planes, a multi-plane equilibrium equation is established. QR decomposition yields ,in Let M be the amplitude vector. For unbalanced quantity vectors, For matrix The false reversal; Design amplitude loss function ,in This is the measured amplitude vector. Given the theoretical amplitude vector, an update rule for the influence coefficient matrix is ​​established based on the gradient descent principle: , For learning rate, These are the updated influence coefficient matrix and the unupdated influence coefficient matrix, respectively. Set the convergence criteria for the iteration. When the number of iterations reaches the preset upper limit and the change in the loss function of two consecutive iterations is less than the set threshold, the iteration process is terminated and the final optimized influence coefficient matrix is ​​output.

8. The method for processing and optimizing wheel assembly dynamic balancing test data based on the Industrial Internet of Things as described in claim 1, characterized in that, When a sudden amplitude change exceeding a preset threshold is detected, the testing process is paused and a collaborative analysis mechanism with the cloud is triggered. Based on the standard deviation analysis results, the dynamic balance tolerance range is dynamically optimized, including: During the wheel hub rotation test, the vibration amplitude change is monitored in real time. When the current amplitude value exceeds the preset multiple threshold of the historical cycle average amplitude, it is determined to be an abnormal state. The edge computing gateway immediately suspends the current test process and uploads the abnormal event timestamp and complete vibration signal record to the cloud analysis platform. After receiving abnormal data, the cloud server reconstructs the mass distribution map of the wheel hub based on historical mass distribution data using the Kriging spatial interpolation algorithm; Based on the range of standard deviation of the quality distribution, a tiered tolerance adjustment strategy is implemented: when the standard deviation exceeds the first threshold, the tolerance is adjusted using an amplification factor; when the standard deviation is in the middle threshold range, the baseline tolerance remains unchanged; when the standard deviation is below the second threshold, the tolerance is tightened using a reduction factor.

Images