A rapid generation system for the motion envelope of an automotive chassis suspension spring

Abstract

The present invention relates to the technical field of hardware equipment, and specifically to a system for quickly generating the motion envelope of an automotive chassis suspension spring. First, a multi-modal sensor array is used to collect multi-modal spring data in real time. Secondly, the deviation of the multi-modal spring data is identified; the multi-modal spring data is corrected according to the error; the multi-modal spring data is subjected to multi-dimensional data fusion to generate fusion data. Then, edge computing is used to process the fusion data to obtain time-frequency features; based on the time-frequency data, a motion envelope parallel optimization method is used to generate the spring motion envelope. Next, the data within the spring motion envelope window is smoothed to remove abnormal fluctuation values. Finally, the working states of each unit are continuously monitored, and the coordination degree of each unit is evaluated in real time; the corrected control parameters are transmitted back to each unit through a feedback loop for adjustment. The present invention can reduce computing resources and improve the efficiency and accuracy of spring motion envelope generation.

Description

Technical Field

[0001] The present invention relates to the technical field of hardware equipment, and particularly to a system for quickly generating the motion envelope of an automotive chassis suspension spring. Background Art

[0002] Automotive chassis suspension springs are usually located in the vehicle's suspension system, connecting the vehicle body and the wheels. They are mainly arranged between the suspension arms, shock absorbers, and between the wheels and the body, aiming to buffer the impacts and vibrations during vehicle driving and absorb the external forces of road unevenness. The springs ensure the smooth driving and comfort of the vehicle by supporting the vehicle body weight and maintaining the relative positions of the wheels and the body. Generating the motion envelope of the automotive chassis suspension spring can help optimize the suspension performance, ensure a reasonable deformation range of the spring under different working conditions, avoid overloading, and improve the comfort and safety of the vehicle.

[0003] Regarding the generation of the spring envelope, traditional methods usually rely on geometric modeling and analysis based on mathematical models. First, by modeling the deformation of the spring under different working conditions, combining the geometric parameters, load, and elastic characteristics of the spring, finite element analysis or traditional analytical methods are used to calculate the motion trajectory of the spring during compression or stretching. Traditional methods usually use simplified mechanical models and describe the dynamic changes and motion envelope of the spring through a series of equations. These calculation processes require detailed input of the physical parameters of the spring and obtain the results through theoretical derivation or numerical simulation. However, the traditional methods have cumbersome calculation processes and limited accuracy, and it is difficult to accurately simulate the dynamic changes of the spring under complex environmental conditions. They ignore the real-time monitoring and dynamic adjustment of environmental factors and lack real-time performance and adaptability.

[0004] The prior art has introduced more digital tools and intelligent calculation methods to improve the accuracy of generating the motion envelope of the chassis suspension spring. Some advanced solutions use finite element simulation software for dynamic simulation and perform precise analysis in combination with the geometric shape and loading conditions of the spring. These methods can not only simulate the deformation of the spring but also comprehensively analyze the spring considering the influence of multiple physical fields. In addition, with the development of sensor technology, modern automotive suspension systems can collect real-time working state data of the suspension through pressure sensors, acceleration sensors, etc. Combining these real-time data, the accuracy of the envelope is improved through model calibration and correction. The prior art also uses machine learning and optimization algorithms to automatically optimize the spring design parameters based on a large amount of data and improve the performance. Modern optimization technologies such as particle swarm optimization algorithms and genetic algorithms are also applied to the design and adjustment of the suspension system, further improving the accuracy of the envelope. However, the prior art faces the problem of low calculation efficiency. Especially under complex working conditions, the calculation time of finite element analysis and multi-physical field simulation is relatively long. In addition, the accuracy of the envelope is still limited, and it is difficult to fully consider the influence of real-time environmental changes on the spring performance.

[0005] To this end, a rapid generation system for the motion envelope of an automotive chassis suspension spring is proposed. Summary of the Invention

[0006] The object of the present invention is to provide a rapid generation system for the motion envelope of an automotive chassis suspension spring, which is used to improve the efficiency and accuracy of the rapid generation of the spring motion envelope. First, a multi-modal sensor array is used to collect multi-modal spring data in real time. Secondly, the deviation of the multi-modal spring data is identified; the multi-modal spring data is corrected according to the error; the multi-modal spring data is subjected to multi-dimensional data fusion to generate fusion data. Then, edge computing is used to process the fusion data to obtain time-frequency features; based on the time-frequency data, a motion envelope parallel optimization method is used to generate the spring motion envelope. Next, the data within the spring motion envelope window is smoothed to remove abnormal fluctuation values. Finally, the working states of each unit are continuously monitored, and the coordination degree of each unit is evaluated in real time; the corrected control parameters are transferred back to each unit through a feedback loop for adjustment.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A rapid generation system for the motion envelope of an automotive chassis suspension spring, comprising:

[0009] A sensor data acquisition unit, which uses a multi-modal sensor array and is arranged at multiple positions of the suspension to collect multi-modal spring data in real time;

[0010] An error correction and sensor data fusion unit, which identifies the deviation of the multi-modal spring data; corrects the multi-modal spring data according to the error; and performs multi-dimensional data fusion on the multi-modal spring data through a data fusion algorithm to generate fusion data;

[0011] A motion envelope generation unit, which uses edge computing to process the fusion data to obtain time-frequency features; based on the time-frequency data, uses a motion envelope parallel optimization method to generate the spring motion envelope; the motion envelope parallel optimization method includes steps: data distribution, initialization, local optimization, information exchange, global convergence, and envelope generation;

[0012] A motion envelope optimization unit, which smooths the data within the spring motion envelope window to remove abnormal fluctuation values;

[0013] A feedback and monitoring unit, which continuously monitors the working states of each unit, and evaluates the coordination degree of each unit in real time; based on the real-time evaluation, transfers the corrected control parameters back to each unit through a feedback loop for adjustment.

[0014] Specifically, the multimodal sensor array includes pressure, acceleration, and fiber optic sensors; the multimodal sensor array is arranged at two endpoints of the spring, around the suspension arm, and at the part where the suspension is connected to the vehicle body; the sensor data includes the deformation amount, stress, and vibration information of the spring.

[0015] Further, identify the deviation of the multimodal spring data; correct the multimodal spring data according to the error, and the specific steps are as follows:

[0016] Include the principal component analysis method and the autoencoder; use the principal component analysis method to extract the key features of the multimodal spring data, input the key features into the autoencoder, and capture the non-linear error;

[0017] Calculate the correction weight based on the error and perform data correction.

[0018] Further, through the data fusion algorithm, perform multi-dimensional data fusion on the multimodal spring data to generate fusion data, and the specific steps are as follows:

[0019] Perform data preprocessing on the multimodal spring data, and perform data alignment and spatio-temporal synchronization;

[0020] Calculate the fusion weight according to the signal-to-noise ratio of the multimodal spring data, and fuse the multimodal spring data based on the fusion weight to generate fusion data.

[0021] Further, the processing flow of the edge computing specifically includes:

[0022] Extract time-domain features from the fusion data, including amplitude, periodicity, and maximum displacement;

[0023] Convert the time-domain features into the frequency domain and extract frequency-domain features, including frequency components, as well as the corresponding peak frequencies and harmonic information;

[0024] Extract the periodicity information, amplitude characteristics, and transient characteristics of the spring movement from the time-domain features; extract the frequency response characteristics and vibration modes of the spring movement from the frequency-domain features; use these features as time-frequency features and use them as the input of the motion envelope parallel optimization method.

[0025] Further, the specific steps to generate the spring motion envelope using the motion envelope parallel optimization method are as follows:

[0026] Data allocation: Allocate the time-frequency features to generate multiple different time-frequency data sets; allocate the time-frequency data sets to multiple computing nodes;

[0027] Initialization: On each of the computing nodes, a part of the particle swarm is initialized; each particle represents a potential motion envelope solution and has a position, a velocity, and a fitness value.

[0028] Local optimization: Each of the computing nodes performs particle swarm optimization on the corresponding time-frequency data set respectively, calculates the positions and velocities of the particles, and updates the fitness values of the particles; each of the computing nodes independently optimizes the corresponding time-frequency data set to obtain a local optimal position.

[0029] Furthermore, the specific steps of generating the spring motion envelope by using the motion envelope parallel optimization method further include:

[0030] Information exchange: During each cycle of particle swarm optimization on each of the computing nodes, each of the computing nodes exchanges the local optimal positions of the particles with each other, and selects the optimal solution from all the local optimal positions as the global best position.

[0031] Global convergence: Update the particle swarms of each of the computing nodes according to the global best position and the local optimal positions, and gradually converge until the set number of iterations is reached.

[0032] Envelope generation: In the final iteration, the particle swarms of each of the computing nodes all find the optimal solutions; summarize all the optimal solutions to generate the final spring motion envelope.

[0033] Furthermore, use a Gaussian kernel function to perform weighted smoothing on the data within the spring motion envelope window to remove abnormal fluctuation values, including high-frequency noise.

[0034] Furthermore, continuously monitor the working states of each unit, and evaluate the coordination degree of each unit in real time. The specific steps are as follows:

[0035] Monitor the computing loads of each unit, the normal transfer of data streams, the correctness of task execution, and time delays, and give an early warning if an abnormality occurs.

[0036] Evaluate the coordination degree of each unit, including real-time response, data accuracy, and consistency of processing results, and give an early warning if an abnormality occurs.

[0037] Furthermore, in the feedback and monitoring unit, the feedback loop is a PID control.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] 1. The data fusion algorithm of the present invention integrates spring data from multiple sensors into unified fusion data, which can eliminate possible local errors, noises, and biases of each sensor, thereby improving the accuracy and reliability of the data. The multi-dimensional data fusion takes into account the information of different sensors, can more comprehensively and accurately reflect the actual state of the suspension spring, provides clearer and more comprehensive basic data for subsequent analysis, reduces errors caused by a single data source, and effectively improves the overall stability and reliability of the system.

[0040] 2. The present invention uses edge computing to push data processing and analysis tasks to the data source side, reducing latency and improving real-time performance. By extracting time-domain features and frequency-domain features, the key dynamic features of spring movement can be accurately captured. These time-frequency features provide precise inputs for subsequent spring movement envelope generation, helping to improve the accuracy and efficiency of envelope generation in the system.

[0041] 3. By distributing time-frequency feature data to multiple computing nodes for parallel processing, the present invention can quickly explore different solution spaces, avoiding the computational bottleneck of traditional single-node optimization methods. The local optimization and global information exchange mechanism ensure the gradual convergence of the global optimal solution, can accurately capture the dynamic characteristics of the spring and generate accurate movement envelopes. In addition, considering multiple possible solutions simultaneously can more effectively handle the non-linear envelope generation problem and reduce the computational time. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a system structure diagram of a system for quickly generating the movement envelope of an automotive chassis suspension spring according to the present invention;

[0043] Figure 2 It is a schematic diagram of the processing flow of edge computing according to the present invention;

[0044] Figure 3 It is a method flow chart of the parallel optimization method for the movement envelope according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0046] In order to reduce computing resources and improve the efficiency and accuracy of spring movement envelope generation, the present invention provides a system for quickly generating the movement envelope of an automotive chassis suspension spring. To illustrate the function of the present invention, the effectiveness of the present invention will be described from the following embodiments.

[0047] Example 1

[0048] In order to improve the R & D efficiency, shorten the product launch cycle, and ensure that the movement performance of the springs in the suspension system meets the design requirements under different driving conditions, a certain new energy vehicle R & D department uses a rapid generation system for the movement envelope of automotive chassis suspension springs. As Figure 1 shown, it shows the system structure diagram of a rapid generation system for the movement envelope of automotive chassis suspension springs.

[0049] Referring to Figure 1 , this system includes a sensor data acquisition unit that uses a multi-modal sensor array arranged at multiple positions on the suspension to collect multi-modal spring data in real time.

[0050] Specifically, the multi-modal sensor array includes pressure, acceleration, and fiber optic sensors; the multi-modal sensor array is arranged at the two endpoints of the spring, around the suspension arm, and at the part where the suspension is connected to the vehicle body; the sensor data includes the deformation amount, stress, and vibration information of the spring.

[0051] In this embodiment, the pressure sensor is used to detect the external force and pressure changes applied to the spring and monitor the stress state of the spring. Pressure sensors are installed at the endpoints of the spring and around the suspension arm to measure the external pressure on the spring in real time. Based on the pressure changes, the deformation amount and corresponding stress of the spring can be deduced.

[0052] The acceleration sensor measures the acceleration of the spring under the stressed state and detects the spring vibration. The acceleration sensors are placed at both ends of the spring and on the suspension arm to monitor the movement state of the suspension system under different working conditions. The acceleration sensor can accurately capture the vibration characteristics of the spring, including the frequency, amplitude, and acceleration fluctuations of the spring, and further analyze the dynamic characteristics of the resonant frequency and vibration mode of the spring.

[0053] The fiber optic sensor is used to measure the deformation amount and stress of the spring with high precision, and detects the deformation situation by the change of the wavelength of the reflected light. When the spring is stressed, the deformation of the optical fiber will cause a change in the reflected wavelength, so that the deformation amount of the spring can be accurately calculated. As shown in Table 1, it shows the relevant data obtained by the pressure, acceleration, and fiber optic sensors.

[0054] Table 1 Relevant data obtained by pressure, acceleration, and fiber optic sensors

[0055]

[0056] By combining the use of pressure, acceleration, and fiber optic sensors, the deformation, stress, and vibration information of the spring can be comprehensively obtained. The combination of these sensors provides a multi-dimensional monitoring solution that can accurately capture the dynamic response of the spring under different working conditions, helping to improve the accuracy of spring motion envelope generation.

[0057] Furthermore, referring to Figure 1 , the system also includes an error correction and sensor data fusion unit that identifies the deviations in the multi-modal spring data; corrects the multi-modal spring data according to the errors; and performs multi-dimensional data fusion on the multi-modal spring data through a data fusion algorithm to generate fusion data. The specific steps of this unit are as follows:

[0058] It includes the principal component analysis method and the autoencoder; uses the principal component analysis method to extract the key features of the multi-modal spring data:

[0059] Calculate the covariance matrix of the multi-modal spring data, perform eigenvalue decomposition on the covariance matrix to obtain the eigenvectors and eigenvalues. In this embodiment, the eigenvalues represent the variation trends between different physical quantities such as force, deformation, and acceleration. Then, select the eigenvectors corresponding to the first largest eigenvalues to reduce the dimension of the data and reconstruct the data to obtain the key features. In this embodiment, the principal component analysis method helps to remove unnecessary noise and improve the calculation efficiency.

[0060] Furthermore, input the key features into the autoencoder to capture the non-linear errors. Among them, the autoencoder includes an encoder and a decoder. The encoder compresses the input, that is, the key features , into a low-dimensional space using a non-linear transformation , where represents the non-linear activation function, represents the low-dimensional feature representation, represents the parameters of the encoder. The decoder restores the low-dimensional feature representation to the original data space: , where represents the reconstructed data, represents the parameters of the decoder. Then, based on the key features and the reconstructed data , calculate the error , and the formula is , where the subscript represents the th sensor.

[0061] Furthermore, based on the error , calculate the correction weight , and the calculation formula is , where represents a control factor used to adjust the influence of the error on the correction weight. represents the th error of the sensor. represents the natural constant. This formula ensures that data with larger errors will receive more weight, thus helping to better correct the data and improve the overall accuracy. Then, based on the correction weight data correction is performed to obtain the corrected multi-modal spring data .

[0062] Furthermore, the multi-modal spring data is preprocessed, including denoising, missing value filling, and normalization. Since different types of sensors have different sampling rates and timestamps, time alignment is required, and the methods include but are not limited to interpolation and resampling. Further, since different sensors are arranged at different spatial positions, spatial coordinate transformation is required to ensure that data from different sensors of the same physical quantity can be compared and fused in the same spatial framework. The methods of spatial alignment include but are not limited to coordinate transformation and sensor calibration methods, which will not be elaborated here.

[0063] Furthermore, the signal-to-noise ratio determines the reliability of each sensor data. Therefore, the fusion weight is calculated based on the signal-to-noise ratio of the multi-modal spring data:

[0064]

[0065] where represents the signal-to-noise ratio of the th sensor data, represents the variance of the th sensor data, represents the expected value of the th sensor data.

[0066] Calculate the fusion weight based on the signal-to-noise ratio :

[0067] ;

[0068] where represents the total number of sensors.

[0069] Fuse the multi-modal spring data based on the fusion weight to generate the fusion data , and the calculation formula is:

[0070] ;

[0071] Through data preprocessing, alignment, and spatio-temporal synchronization, the consistency and accuracy of multi-modal sensor data are ensured, effectively reducing errors caused by sensor delay or data mismatch. By calculating the fusion weights based on the signal-to-noise ratio, weighted fusion can be performed according to the quality of each sensor's data, improving the accuracy and robustness of data fusion, and thus generating more reliable fusion data. This not only optimizes the computational efficiency but also improves the accuracy rate in the subsequent spring motion envelope generation process.

[0072] Furthermore, referring to Figure 1 , the system further includes a motion envelope generation unit that uses edge computing to process the fusion data to obtain time-frequency features; based on the time-frequency data, a spring motion envelope is generated using a parallel optimization method for the motion envelope; the parallel optimization method for the motion envelope includes the steps of: data distribution, initialization, local optimization, information exchange, global convergence, and envelope generation. As Figure 2 shown, it shows a schematic diagram of the processing flow of edge computing.

[0073] Specifically, the processing flow of the edge computing specifically includes:

[0074] Distribute the computing tasks to edge devices near the suspension sensors; the fusion data is a time-series signal, and time-domain features are extracted from the fusion data, including amplitude, periodicity, and maximum displacement. The amplitude is the difference between the maximum value and the minimum value in the fusion data; the maximum displacement is the maximum offset value of the fusion data.

[0075] Furthermore, use the Fourier transform formula to transform the time-domain features into the frequency domain and extract frequency-domain features, including frequency components, as well as the corresponding peak frequencies and harmonic information. The frequency components are represented by discrete values in the frequency domain; the peak frequency is the position with the largest amplitude in the spectrum; the harmonic information includes harmonic frequencies, which are integer multiples of the fundamental frequency.

[0076] Furthermore, extract the periodic information, amplitude characteristics, and transient characteristics of the spring motion from the time-domain features. The periodic information is obtained by calculating the periodic changes in the time-domain features; the amplitude characteristics are obtained by calculating the mean or peak value of the time-domain features; the transient characteristics are characterized by extracting the instantaneous frequency or instantaneous amplitude of the time-domain features. Extract the frequency response characteristics and vibration modes of the spring motion from the frequency-domain features. The frequency response characteristics are the amplitude or phase changes in each frequency band in the frequency-domain features; the vibration modes are extracted by extracting different vibration modes from different harmonic components in the frequency features.

[0077] Take these features as time-frequency features and use them as the input of the parallel optimization method for the motion envelope.

[0078] Edge computing extracts time-domain and frequency-domain features to comprehensively analyze the periodicity, amplitude, transient, etc. of spring motion, improving the accurate recognition and analysis ability of the motion state, providing effective input for the motion envelope parallel optimization method, and optimizing the system response accuracy. Additionally, edge computing can process the data collected by sensors in real time, avoiding the transmission of large-scale data to remote servers, thereby reducing communication latency and the load on central computing.

[0079] Furthermore, based on the time-frequency data, a motion envelope parallel optimization method is used to generate the spring motion envelope. As Figure 3 shown, the method flow chart of the motion envelope parallel optimization method is presented. The specific steps are as follows:

[0080] Data allocation: Allocate the time-frequency features to generate multiple different time-frequency data sets; allocate the time-frequency data sets to multiple computing nodes. Specifically, assume there are computing nodes, and each computing node is allocated a part of the time-frequency data set, denoted as , where , and the total time-frequency data set is denoted as , that is, . By dividing the data into multiple subsets and performing independent calculations on each subset, the computing bottleneck is reduced, and the overall performance is improved. Moreover, the computing tasks are reasonably allocated, the computing process is optimized, and the time consumption of intermediate links is reduced.

[0081] Initialization: On each of the computing nodes, initialize a part of the particle swarm; each particle represents a potential motion envelope solution and has a position, velocity, and fitness value. Specifically, for each particle , where , is the number of particles on the th computing node. The state of particle consists of the following three parts: position : , represents the dimension index of the particle position; velocity : ; fitness value . Among them, the initial position and velocity are usually generated from a random distribution:

[0082] ;

[0083] where, and represent the boundary values of the position respectively; and represent the boundary values of the velocity respectively, and represent random numbers.

[0084] Provide diverse initial solutions for the optimization algorithm to ensure the breadth of the exploration space and avoid local optima.

[0085] Local optimization: Each of the computing nodes performs particle swarm optimization on the corresponding time-frequency data set to calculate the position and velocity of the particles and update the fitness value of the particles. The velocity of the particle is updated by the following formula:

[0086] ;

[0087] where represents the inertia weight; and represent the learning factors; and represent random numbers; represents the local optimal position of particle ; represents the global optimal position.

[0088] The position of the particle is updated by the following formula:

[0089] ;

[0090] The fitness of the particle is updated according to the obtained time-frequency data set to evaluate the quality of the particle position.

[0091] Each of the computing nodes independently optimizes the corresponding time-frequency data set, where each particle updates its local optimal position according to the fitness function , that is .

[0092] Through local optimization search, the quality of the solution can be improved, rough solutions can be avoided, and the approximation of the solution to the optimal state can be ensured.

[0093] Information exchange: During each cycle of particle swarm optimization of each of the computing nodes, each of the computing nodes exchanges the local optimal positions of the particles with each other , selects the optimal solution from all the local optimal positions as the global best position , that is , where

[0094] Sharing information between computing nodes helps avoid local optimality and improve global search capabilities. By exchanging solution information, the algorithm can better adjust the search direction and improve global exploration capabilities.

[0095] Global convergence: updating the particle swarm of each computing node according to the global optimal position and the local optimal position, and gradually converging until the set number of iterations is reached.

[0096] Envelope generation: In the final iteration, the particle swarm of each computing node finds its optimal solution , A set of parameters is a mathematical representation of the spring motion envelope, where All the optimal solutions are summarized to generate the final spring motion envelope , expressed as .

[0097] Envelope generation provides an accurate description of the overall behavior of the system by merging local solutions to form a continuous solution space.

[0098] Further, refer to Figure 1 The system also includes a motion envelope optimization unit, which smoothes the data in the spring motion envelope window and removes abnormal fluctuation values. Specifically, a moving average algorithm is selected to slide in the data window, gradually calculate and replace the value of each data point, and use the weighted value of the neighboring data to smooth the current data point. Further, a low-pass filter is applied to the spring motion envelope data to remove high-frequency noise, and the low-pass filter includes but is not limited to a Butterworth filter and a Kalman filter. The filtered data is corrected to further smooth the data and ensure the stability of the spring motion envelope.

[0099] Further, refer to Figure 1 The system also includes a feedback and monitoring unit, which continuously monitors the working status of each unit and evaluates the coordination degree of each unit in real time; based on the real-time evaluation, the corrected control parameters are transmitted back to each unit through a feedback loop for adjustment.

[0100] Specifically, monitor the computing load of each unit, the normal transfer of data streams, the correctness of task execution, and time delay. In case of anomalies, issue a warning. Use performance monitoring tools to monitor the computing load, such as Prometheus, Grafana, or New Relic, to collect metrics such as CPU usage, memory consumption, and disk I / O of each unit in real time. These data will be continuously monitored in the background, and tasks will be assigned through load balancing algorithms such as weighted round-robin and least connections to avoid overloading a certain unit. Use network latency monitoring tools to monitor the normal transfer of data streams, such as Ping, Traceroute, and Zabbix, to monitor the transmission process of data packets and ensure the smooth transmission of data between units. If packet loss or latency is detected, the system will trigger an alarm. Use a logging system such as the ELK Stack or a monitoring service such as Datadog to record the logs of task execution in real time, including input and output data, and error exceptions. Conduct unit tests through an automated test framework to ensure that each unit performs its intended function. Use timestamps and real-time performance analysis tools, such as the latency monitoring function of Apache Kafka, to monitor the transit time of data in each processing unit. If a certain link exceeds the set response time threshold, the system will trigger an alarm.

[0101] Furthermore, use time series analysis methods to analyze the response time of each unit to ensure that each unit responds to requests on time. By setting threshold alarms, when a certain unit has a too long response time, the system will issue an alarm. And adopt data verification techniques to ensure the accuracy of the data output by each unit by comparing it with a known standard data set. Combine data consistency algorithms to ensure that no errors occur during data processing by each module. Through data integrity checks and differential comparison algorithms, check whether the processing results between different modules are consistent. In this embodiment, the methods used in the feedback and monitoring unit are all existing technologies and will not be elaborated here.

[0102] Monitor the workload, data stream transmission, task execution correctness, and time delay of each unit in real time to ensure the efficient operation of the system. Evaluate the collaborative work effect between units to avoid dysfunction of each unit.

[0103] Evaluate based on real-time monitoring information, and transfer the corrected control parameters back to the respective units through a feedback loop. Specifically, calculate the deviation between the current state and the ideal target by real-time monitoring of system performance. For example, if the computational load of a certain unit is too high, the system will calculate the gap between it and the preset standard load. Based on proportional, integral, and derivative control algorithms, calculate the adjustment value for each unit. The PID algorithm generates an adjustment output based on the current error, the accumulation of historical errors, and the rate of change of the error. According to the adjustment value output by the PID control, update the configuration parameters of each unit in real time. For example, if the response time of a certain unit is slow, the PID controller can increase the computing resources of the unit or adjust its task priority. Through automation tuning tools such as the adaptive scheduling methods of Ansible, Chef, or Kubernetes, based on the feedback of PID control, dynamically adjust the resource allocation, task scheduling, etc. of the system. All adjustments will be secondarily optimized based on the feedback results.

[0104] Adjust the operating parameters of each unit through PID control to ensure that the system operates efficiently and stably within the target range. Through a closed-loop control mechanism, realize the dynamic optimization of the system, so that the system always maintains high efficiency and stability under changing workloads, thereby generating a spring motion envelope with high accuracy.

[0105] The system for quickly generating the motion envelope of the automotive chassis suspension spring significantly improves the efficiency and accuracy of generating the spring motion envelope through multi-modal sensor data acquisition, error correction and data fusion, edge computing processing, and parallel optimization methods. The system can provide real-time feedback and optimization of the motion envelope to ensure that the motion performance of the suspension meets the design requirements in different driving environments, shorten the R & D cycle, and improve the R & D efficiency and product performance of new energy vehicles.

[0106] Embodiment 2

[0107] A certain B automotive after-sales department used a system for quickly generating the motion envelope of the automotive chassis suspension spring to analyze the performance of the automotive suspension system and identify potential faults or abnormalities during spring motion.

[0108] The system includes a sensor data acquisition unit that uses a multi-modal sensor array arranged at multiple positions on the suspension to collect multi-modal spring data in real time.

[0109] Specifically, the multi-modal sensor array includes pressure, acceleration, and fiber optic sensors; the multi-modal sensor array is arranged at both ends of the spring, around the suspension arm, and at the part where the suspension is connected to the vehicle body; the sensor data includes the deformation amount, stress, and vibration information of the spring.

[0110] Furthermore, the system also includes an error correction and sensor data fusion unit, which identifies the deviation of the multimodal spring data; corrects the multimodal spring data according to the error; and generates fusion data by performing multi-dimensional data fusion on the multimodal spring data through a data fusion algorithm. The specific steps of this unit are as follows:

[0111] It includes the principal component analysis method and the autoencoder; uses the principal component analysis method to extract the key features of the multimodal spring data:

[0112] Calculate the covariance matrix of the multimodal spring data, perform eigenvalue decomposition on the covariance matrix to obtain eigenvectors and eigenvalues. In this embodiment, the eigenvalues represent the change trends between different physical quantities, such as force, deformation, and acceleration. Then, select the eigenvectors corresponding to the first several largest eigenvalues to reduce the dimension of the data and reconstruct the data to obtain the key features.

[0113] Furthermore, input the key features into the autoencoder to capture the non-linear error. Among them, the autoencoder includes an encoder and a decoder. The encoder compresses the input, that is, the key features , into a low-dimensional space, using a non-linear transformation , where represents the non-linear activation function, represents the low-dimensional feature representation, represents the parameters of the encoder. The decoder restores the low-dimensional feature representation to the original data space: , where represents the reconstructed data, represents the parameters of the decoder. Then, calculate the error based on the key features and the reconstructed data , and the formula is , where the subscript represents the th sensor.

[0114] Furthermore, calculate the correction weight based on the error , and the calculation formula is , where represents the control factor, which is used to adjust the influence of the error on the correction weight, represents the error of the th sensor, represents the natural constant. Then, perform data correction according to the correction weight to obtain the corrected multimodal spring data .

[0115] Further, preprocess the multimodal spring data, including denoising, missing value imputation, and normalization. Since different types of sensors have different sampling rates and timestamps, time alignment is required, including but not limited to interpolation and resampling methods to synchronize the data of all sensors to a unified time axis. Further, since different sensors are arranged at different spatial positions, spatial coordinate transformation is required to ensure that the data of different sensors for the same physical quantity can be compared and fused in the same spatial framework. The methods of spatial alignment include but are not limited to coordinate transformation and sensor calibration methods.

[0116] Further, the signal-to-noise ratio determines the reliability of each sensor data. Therefore, calculate the fusion weights according to the signal-to-noise ratio of the multimodal spring data:

[0117] ;

[0118] where represents the signal-to-noise ratio of the -th sensor data, represents the variance of the -th sensor data, represents the expected value of the -th sensor data.

[0119] Calculate the fusion weights based on the signal-to-noise ratio :

[0120] ;

[0121] where represents the total number of sensors.

[0122] Fuse the multimodal spring data based on the fusion weights to generate fused data , and the calculation formula is:

[0123] ;

[0124] Further, the system further includes a motion envelope generation unit, which processes the fused data using edge computing to obtain time-frequency features; based on the time-frequency data, uses a motion envelope parallel optimization method to generate a spring motion envelope; the motion envelope parallel optimization method includes the steps: data distribution, initialization, local optimization, information exchange, global convergence, and envelope generation. As Figure 2 shown, it shows a schematic diagram of the processing flow of edge computing.

[0125] Specifically, the processing flow of the edge computing specifically includes:

[0126] Distribute the computing tasks to the edge devices near the suspension sensors; the fused data is a time-series signal, and extract the time-domain features from the fused data, including amplitude, periodicity, and maximum displacement. The amplitude is the difference between the maximum value and the minimum value in the fused data; the maximum displacement is the maximum offset value of the fused data.

[0127] Further, use the Fourier transform formula to transform the time-domain features into the frequency domain, and extract the frequency-domain features, including frequency components, as well as the corresponding peak frequency and harmonic information. The frequency components are represented by discrete values in the frequency domain; the peak frequency is the position with the largest amplitude in the spectrum; the harmonic information includes harmonic frequencies, which are integer multiples of the fundamental frequency.

[0128] Further, extract the periodicity information, amplitude characteristics, and transient characteristics of the spring movement from the time-domain features. The periodicity information is obtained by calculating the period change in the time-domain features; the amplitude characteristics are obtained by calculating the mean or peak value of the time-domain features; the transient characteristics are characterized by extracting the instantaneous frequency or instantaneous amplitude of the time-domain features. Extract the frequency response characteristics and vibration modes of the spring movement from the frequency-domain features. The frequency response characteristics are the amplitude or phase changes in each frequency band in the frequency-domain features; the vibration modes are extracted by extracting different vibration modes from different harmonic components in the frequency features.

[0129] Take these features as time-frequency features and use them as the input of the parallel optimization method for the motion envelope.

[0130] Further, based on the time-frequency data, use the parallel optimization method for the motion envelope to generate the spring motion envelope. As Figure 3 shown, the method flow chart of the parallel optimization method for the motion envelope is presented, and the specific steps are as follows:

[0131] Data distribution: Distribute the time-frequency features to generate multiple different time-frequency data sets; distribute the time-frequency data sets to multiple computing nodes. Specifically, assume there are computing nodes, and each computing node is assigned a part of the time-frequency data set, denoted as , where , and the total time-frequency data set is denoted as , that is, .

[0132] Initialization: On each of the computing nodes, initialize a part of the particle swarm; each particle represents a potential motion envelope solution and has a position, velocity, and fitness value. Specifically, for each particle , where , is the number of particles on the th computing node. The state of particle consists of the following three parts: position : , represents the dimensional index of the particle position; velocity : ; fitness value . Among them, the initial position and velocity are usually generated from a random distribution:

[0133] ;

[0134] Among them, and respectively represent the boundary values of the position; and respectively represent the boundary values of the velocity, and represent random numbers.

[0135] Local optimization: Each of the computing nodes performs particle swarm optimization on the corresponding time-frequency data set , calculates the position and velocity of the particles, and updates the fitness value of the particles. The velocity of the particle is updated by the following formula:

[0136] ;

[0137] Among them, represents the inertia weight; and represent the learning factors; and represent random numbers; represents the local optimal position of the particle ; represents the global optimal position.

[0138] The position of the particle is updated by the following formula:

[0139] ;

[0140] The fitness of the particle is updated according to the obtained time-frequency data set to evaluate the quality of the particle position.

[0141] Each of the computing nodes independently optimizes the corresponding time-frequency data set, where each particle updates its local optimal position according to the fitness function , that is .

[0142] Information exchange: During each period of particle swarm optimization of each computing node, each computing node exchanges the local optimal positions of the particles with each other , select the optimal solution from all the local optimal positions as the global optimal position ,Right now ,in, Represents the local optimal solution of each computing node. At the end of each iteration, each computing node sends its local optimal solution to other nodes so that the search of each particle can be adjusted according to the global optimal solution.

[0143] Global convergence: updating the particle swarm of each computing node according to the global optimal position and the local optimal position, and gradually converging until the set number of iterations is reached.

[0144] Envelope generation: In the final iteration, the particle swarm of each computing node finds its optimal solution , A set of parameters is a mathematical representation of the spring motion envelope, where All the optimal solutions are summarized to generate the final spring motion envelope , expressed as .

[0145] Furthermore, the system also includes a motion envelope optimization unit, which smoothes the data in the spring motion envelope window to remove abnormal fluctuation values. Specifically, a moving average algorithm is selected to slide in the data window, gradually calculate and replace the value of each data point, and use the weighted value of the adjacent data to smooth the current data point. Furthermore, a low-pass filter is applied to the spring motion envelope data to remove high-frequency noise, and the low-pass filter includes but is not limited to a Butterworth filter and a Kalman filter. The filtered data is corrected to further smooth the data and ensure the stability of the spring motion envelope.

[0146] Furthermore, the system also includes a feedback and monitoring unit, which continuously monitors the working status of each unit and evaluates the degree of coordination of each unit in real time; based on the real-time evaluation, the corrected control parameters are transmitted back to each unit through a feedback loop for adjustment.

[0147] Specifically, the computing load, normal transmission of data flow, correctness of task execution and time delay of each unit are monitored, and an alarm is issued if an abnormality occurs. Use performance monitoring tools to monitor the computing load, such as Prometheus, Grafana or New Relic to collect CPU usage, memory consumption, disk I / O and other indicators of each unit in real time. These data will be continuously monitored in the background, and tasks will be allocated through load balancing algorithms such as weighted polling and minimum number of connections to avoid overloading of a certain unit. Use network delay monitoring tools to monitor the normal transmission of data flow, such as Ping, Traceroute, and Zabbix to monitor the transmission process of data packets to ensure smooth transmission of data between units. If packet loss or delay is detected, the system will trigger an alarm. Use a logging system, such as ELK Stack, or a monitoring service, such as Datadog, to record task execution logs in real time, including input and output data, abnormal errors, etc. Perform unit testing through an automated testing framework to ensure that each unit performs the intended function. Use timestamps and real-time performance analysis tools, such as Apache Kafka's delay monitoring function, to monitor the flow time of data in each processing unit. If a link exceeds the set response time threshold, the system will trigger an alarm.

[0148] Furthermore, the response time of each unit is analyzed by time series analysis method to ensure that each unit responds to the request on time. By setting the threshold alarm, the system will sound an alarm when the response time of a unit is too long. And adopt data verification technology to ensure that the data output by each unit is accurate by comparing with the known standard data set. Combined with the data consistency algorithm to ensure that each module does not produce errors during data processing. Through data integrity check and differential comparison algorithm, check whether the processing results between different modules are consistent. In this embodiment, the methods used in the feedback and monitoring units are all prior art, and no further elaboration is made here.

[0149] Evaluate based on the information from real-time monitoring, and transmit the corrected control parameters back to the respective units through a feedback loop. Specifically, calculate the deviation between the current state and the ideal target by real-time monitoring the system performance. For example, if the computing load of a certain unit is too high, the system will calculate the gap between it and the preset standard load. Based on the proportional, integral, and derivative control algorithms, calculate the adjustment value for each unit. The PID algorithm generates an adjustment output according to the current error, the cumulative historical error, and the error change rate. According to the adjustment value output by the PID control, update the configuration parameters of each unit in real time. For example, if the response time of a certain unit is slow, the PID controller can increase the computing resources of the unit or adjust its task priority. Through automated tuning tools, such as the adaptive scheduling methods of Ansible, Chef, or Kubernetes, based on the feedback of PID control, dynamically adjust the resource allocation, task scheduling, etc. of the system. All adjustments will be secondarily optimized based on the feedback results.

[0150] This system effectively improves the generation efficiency and accuracy of the spring motion envelope by real-time collecting multi-modal sensor data and performing error correction and data fusion. In the application of the after-sales department, it can quickly identify potential faults or abnormalities in the suspension system. Through edge computing and parallel optimization methods, the system reduces the computing time and improves the accuracy of fault diagnosis. The optimized motion envelope data helps to accurately analyze the spring performance, timely detect problems, improve the safety and reliability of the vehicle, and reduce the maintenance cost.

[0151] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fast motion envelope generation system for automobile chassis suspension springs, characterized in that: include: A sensor data acquisition unit uses a multi-modal sensor array arranged at multiple locations of the suspension to collect multi-modal spring data in real time; The error correction and sensor data fusion unit identifies the deviation of the multi-modal spring data; corrects the multi-modal spring data according to the error; and performs multi-dimensional data fusion on the multi-modal spring data through a data fusion algorithm to generate fused data; A motion envelope generation unit processes the fused data using edge computing to obtain time-frequency features; Based on the time-frequency characteristics, a spring motion envelope is generated using a motion envelope parallel optimization method; the motion envelope parallel optimization method includes the steps of data allocation, initialization, local optimization, information exchange, global convergence and envelope generation; A motion envelope optimization unit is used to smooth the data in the spring motion envelope window and remove abnormal fluctuation values; A feedback and monitoring unit, which continuously monitors the working status of each unit and evaluates the coordination degree of each unit in real time; Based on the real-time evaluation, the modified control parameters are transmitted back to the various units through the feedback loop for adjustment.

2. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: The multimodal sensor array includes pressure, acceleration and optical fiber sensors; the multimodal sensor array is arranged at the two end points of the spring, around the suspension arm and the part where the suspension is connected to the vehicle body; the sensor data includes the deformation, stress and vibration information of the spring.

3. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: Identify the deviation of the multi-modal spring data; and correct the multi-modal spring data according to the error, the specific steps are: It includes a principal component analysis method and an autoencoder; using the principal component analysis method to extract key features of the multimodal spring data, inputting the key features into the autoencoder to capture nonlinear errors; A correction weight is calculated based on the error to perform data correction.

4. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: Through the data fusion algorithm, the multi-modal spring data is subjected to multi-dimensional data fusion to generate fused data. The specific steps are as follows: Preprocessing the multi-modal spring data, and performing data alignment and spatiotemporal synchronization; A fusion weight is calculated according to a signal-to-noise ratio of the multi-modal spring data, and the multi-modal spring data is fused based on the fusion weight to generate fused data.

5. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: The processing flow of edge computing specifically includes: Extracting time domain features from the fused data, including amplitude, periodicity and maximum displacement; Convert the time domain features into the frequency domain, extract the frequency domain features, including frequency components, and corresponding peak frequencies and harmonic information; The periodic information, amplitude characteristics and transient characteristics of the spring motion are extracted from the time domain characteristics; the frequency response characteristics and vibration modes of the spring motion are extracted from the frequency domain characteristics; these characteristics are used as time-frequency characteristics and as inputs of the motion envelope parallel optimization method.

6. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: The specific steps of using the motion envelope parallel optimization method to generate a spring motion envelope are as follows: Data allocation: allocating the time-frequency features to generate multiple different time-frequency data sets; allocating the time-frequency data sets to multiple computing nodes; Initialization: Initialize a part of the particle swarm on each of the computing nodes; each particle represents a potential motion envelope solution, having a position, a velocity, and a fitness value; Local optimization: Each computing node performs particle swarm optimization on the corresponding time-frequency data set, calculates the position and velocity of the particles, and updates the fitness value of the particles; each computing node independently optimizes the corresponding time-frequency data set to obtain a local optimal position.

7. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 6, characterized in that: The specific steps of using the motion envelope parallel optimization method to generate a spring motion envelope also include: Information exchange: During each round of particle swarm optimization of each computing node, each computing node exchanges the local optimal position of particles with each other, and selects the optimal solution from all the local optimal positions as the global optimal position; Global convergence: updating the particle swarm of each computing node according to the global optimal position and the local optimal position, and gradually converging until the set number of iterations is reached; Envelope generation: In the final iteration, the particle swarm of each computing node finds the optimal solution; all the optimal solutions are summarized to generate the final spring motion envelope.

8. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: The data within the spring motion envelope window is weighted and smoothed using a Gaussian kernel function to remove abnormal fluctuation values, including high-frequency noise.

9. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: Continuously monitor the working status of each unit and evaluate the coordination degree of each unit in real time. The specific steps are: Monitor the computing load of each unit, the normal transmission of data flow, the correctness of task execution and time delay, and issue an early warning if an abnormality occurs; Assess the degree of coordination among various units, including the real-time response, accuracy of data and consistency of processing results, and issue early warnings when abnormalities occur.

10. The system for quickly generating motion envelope of a vehicle chassis suspension spring according to claim 1, characterized in that: In the feedback and monitoring unit, the feedback loop is PID control.

Images