Gap measurement method for commercial vehicle steering system

Abstract

The invention discloses a clearance measurement method for a commercial vehicle steering system, and relates to the technical field of vehicle testing, and the method comprises the steps: constructing a physical transmission environment, fixing a vehicle frame and a steering gear, and lifting a front axle to enable wheels to leave the ground; a laser radar and a 4D millimeter wave radar are arranged on the observation side, and a spatial transformation relation is determined; synchronously acquiring rotation angle, point cloud and speed data in the process of driving the steering wheel to rotate; space transformation alignment data is utilized, boundary frames of components such as a steering drop arm and a longitudinal pull rod are identified through a laser radar, and the transverse movement speed of the components is obtained by fusing millimeter wave radar speed information; and finally, calculating an angle gap between the first transmission section and the second transmission section according to the response delay time of the motion starting moment of each component and the action moment of the steering wheel in combination with the steering wheel rotation angular velocity. According to the method, the geometric positioning and speed monitoring fusion technology is utilized, and accurate segmentation quantification of the internal clearance of the steering system under the non-contact condition is achieved.

Description

Technical Field

[0001] This application relates to the field of vehicle testing technology, and in particular to a method for measuring clearance in commercial vehicle steering systems. Background Technology

[0002] Commercial vehicle steering systems typically consist of a multi-stage mechanical transmission structure, from the steering wheel and steering gear to the steering arms, tie rods, and wheels. Existing clearance detection methods mostly assess the overall system's play by measuring the free-turn angle of the steering wheel while the vehicle's tires are in contact with the ground. However, the heavy axle loads of commercial vehicles result in significant static friction between the tires and the ground. This external resistance can easily mask minute mechanical clearances within the transmission chain, leading to distorted measurement results. Furthermore, this method of measuring the entire system cannot break down the total clearance and pinpoint it to specific transmission stages, making it difficult to accurately distinguish whether the clearance originates from loose gear meshing within the steering gear or from wear on the tie rod ball joints. In actual repair and inspection, it often relies on manual disassembly or experience-based judgment, lacking the technical means to quantify the independent clearances of each transmission stage.

[0003] To improve detection efficiency, non-contact sensors are increasingly being used in chassis testing. However, existing technologies have limitations in capturing the extremely minute motion responses of mechanical components. For example, if only point cloud data from lidar is used for monitoring, determining whether a component has started moving typically relies on calculating the position difference between consecutive frames. Due to the discrete nature of point cloud data, this difference calculation method is sensitive to measurement noise and struggles to accurately and precisely define the instantaneous turning point from rest to movement within the extremely short time corresponding to minute gaps. This leads to deviations in determining the start of movement, thus affecting the accuracy of gap calculation.

[0004] Furthermore, existing automated testing solutions have shortcomings in terms of versatility and geometric calibration. To identify specific linkage components or obtain accurate motion data, current methods often require testers to pre-apply dedicated visual markers to oil-covered chassis components or install contact displacement sensors, increasing preparation time and complexity. Simultaneously, when using radar-type sensors with speed measurement capabilities for non-contact observation, the radar beam direction usually forms an angle with the actual lateral movement direction of the linkage, meaning the directly read radial velocity data cannot accurately reflect the component's actual speed. Without an effective automatic identification and geometric projection correction mechanism, measurement results often exhibit uncontrollable geometric errors when facing test scenarios with different vehicle models and installation angles. Summary of the Invention

[0005] The purpose of this invention is to provide a clearance measurement method for commercial vehicle steering systems, which solves a technical problem in existing commercial vehicle steering system clearance measurement: it is difficult to segment and quantify the internal clearance of multi-stage transmission chains under non-contact conditions, and a single sensor is difficult to simultaneously ensure component identification accuracy and motion response capture sensitivity.

[0006] This invention provides the following solution:

[0007] A method for measuring clearance in commercial vehicle steering systems is provided, the method comprising the following steps:

[0008] S1. Construct a physical transmission environment for a commercial vehicle steering system that includes a first transmission segment and a second transmission segment. Fix the connection position between the vehicle frame and the steering gear and constrain the position of the steering wheel. Raise the front axle so that the wheels are out of contact with the ground and establish a mechanical transmission link from steering wheel input to wheel response.

[0009] S2. Arrange a corner acquisition device at the steering wheel, and arrange a lidar and a 4D millimeter-wave radar on the observation side of the mechanical transmission link, and establish the spatial transformation relationship between the lidar coordinate system and the 4D millimeter-wave radar coordinate system.

[0010] S3. During the process of driving the steering wheel to rotate, simultaneously record the steering wheel's rotation angle and angular velocity data sequence, timestamp data sequence, LiDAR point cloud data, and 4D millimeter-wave radar data.

[0011] S4. Align the point cloud data of the lidar with the data of the 4D millimeter-wave radar using the spatial transformation relationship, identify the bounding boxes of each key component in the mechanical transmission link, and fuse the speed information in the 4D millimeter-wave radar data into the bounding boxes to obtain the average movement speed of each key component in the lateral movement direction.

[0012] S5. Determine the starting time of movement of each key component based on the average movement speed, calculate the response lag time between the starting time of the angle change and the starting time of movement of each key component in the steering wheel angle and angular velocity data sequence, and multiply the response lag time by the angular velocity in the steering wheel angle and angular velocity data sequence to obtain the angular clearance of each stage of the mechanical transmission link.

[0013] Preferably, in step S1, when constructing the physical transmission environment of the commercial vehicle steering system, a rigid frame is selected as the reference platform. The steering gear is fixed to the rigid frame with bolts to limit the displacement of the housing. The mounting bracket of the steering wheel is fixed to constrain the spatial position of the steering column. The front axle of the commercial vehicle is raised as a whole using a lifting device. The first transmission segment is configured to include at least a steering wheel, steering column, drive shaft, steering gear, and steering drop arm; the second transmission segment is configured to include at least a steering drop arm, tie rod, steering knuckle, and wheel. This configuration eliminates the obstruction of ground friction to the measurement of minute gaps, providing a physical basis for segmented observation.

[0014] Preferably, in step S2, when arranging the sensors, a position on one side of the vehicle body is selected as the observation side and a lidar is installed there. The X-axis of the lidar coordinate system is adjusted to point vertically to the side of the vehicle body and the Z-axis is adjusted to point vertically upward, so that the lidar scanning range covers the outer side of the steering arm, the tie rod, and the wheel to obtain the geometric contour. At the same time, the 4D millimeter-wave radar is installed above the lidar, and the geometric centers of the two are aligned on the vertical line to reduce the observation parallax.

[0015] Preferably, in step S4, the processing of multi-source data employs a spatial alignment and feature fusion strategy. First, the 4D millimeter-wave radar data is input into the coordinate space of the lidar data, and coordinate mapping is completed using spatial transformation relationships. Next, using the lidar data as a geometric reference, a three-dimensional region of interest is set, and background noise points are removed. Then, a KD-Tree spatial index structure is constructed, and a region growing clustering algorithm based on Euclidean distance is used to traverse the data and search for neighboring points that meet the distance threshold to form point cloud clusters. Finally, effective point cloud clusters representing potential key components are selected according to a preset point cloud quantity threshold, and their axis-aligned bounding boxes are calculated to generate bounding boxes. This process utilizes lidar data to determine the spatial location of components.

[0016] Preferably, in step S4, in order to distinguish each component, semantic classification is performed on the bounding box: spatial standardization is performed on the effective point cloud cluster, and geometric center translation and scale normalization are performed; the standardized data is input into the pre-trained PointNet classification network model, spatial alignment is performed using the input transformation network, high-dimensional point-by-point features are extracted through a multilayer perceptron, and global feature vectors are aggregated through a max pooling layer; finally, the probability distribution is output using the Softmax function, and corresponding semantic labels are assigned to the steering arm, tie rod and wheel according to the confidence threshold.

[0017] Preferably, in step S4, the lateral motion velocity is obtained using the Doppler velocity measurement characteristics of 4D millimeter-wave radar. After coordinate alignment, all 4D millimeter-wave radar points falling within the bounding boxes of each key component are retrieved to construct a point set. For each point in the point set, velocity vector projection is performed to recover the velocity, and the cosine of the angle between the radial velocity vector and the lateral motion direction is calculated. The projected velocity is obtained through multiplication. The projected velocities of all valid points in the point set are summed and the arithmetic mean is taken to obtain the average lateral motion velocity of each key component. This method corrects for single-point measurement bias and inconsistencies between the radar radial velocity and the actual motion direction.

[0018] Preferably, in step S5, the motion start time is determined by time series analysis: extracting data from the stationary phase, calculating statistical characteristics, and setting a stationary determination threshold; constructing a window that slides along the time axis for each key component, calculating the average amplitude level of the average motion speed within the window to generate a motion intensity operator; when the motion intensity operators for multiple consecutive time steps all exceed the stationary determination threshold, the component is determined to enter the motion state, and the first time point in the sequence that meets the condition is recorded.

[0019] Preferably, in step S5, the angular clearance is calculated using a segmented calculation logic based on time-domain response lag. For the first transmission segment, the start time of steering wheel angle change and the start time of steering arm movement are determined, and the time difference between the two is calculated as the response lag time of the first transmission segment. This lag time is then multiplied by the average angular velocity of the steering wheel rotation to obtain the internal clearance between the steering wheel and the steering arm. For the second transmission segment, the start time of the tie rod movement is determined, and the time difference between this start time and the start time of the steering arm movement is calculated as the inter-stage response lag time. This inter-stage lag time is also multiplied by the average angular velocity of the steering wheel rotation to obtain the internal clearance between the steering arm and the tie rod. This method utilizes the time difference of the motion response of each stage component to measure the clearance distribution within the steering system.

[0020] Preferably, the data fusion process in step S4 also includes a time dimension alignment operation: a unified time reference is established based on the ROS platform, with the lidar frame time as the main alignment reference, 4D millimeter-wave radar data in the cache queue is retrieved, the frame with the smallest time difference and meeting the synchronization threshold is selected for binding, redundant data is removed, and the data during spatial fusion comes from the same physical time.

[0021] This invention utilizes the aforementioned technical solution to locate component positions using lidar and combines this with 4D millimeter-wave radar speed measurement to capture minute movements, constructing a fusion processing flow that integrates geometric positioning and speed monitoring. By converting mechanical clearances into motion response time differences between input and output ends, as well as between stage output ends, it achieves the measurement and segmented quantification of clearances at each stage of the transmission chain in a commercial vehicle steering system under non-contact conditions.

[0022] The above solution achieves the following beneficial technical effects:

[0023] This invention constructs a physical test environment with an elevated front axle and converts mechanical clearances into time-dimensional data, eliminating the inhibitory effect of ground tire resistance on the minute movements of the steering system. It uses the time difference between steering wheel input and the response of each stage of transmission components to characterize the clearance size, realizing the segmented measurement and quantification of the internal clearances of multiple stages of the transmission chain, such as from the steering wheel to the steering arm and from the steering arm to the tie rod, in a non-disassembled state of the whole vehicle. This solves the problem that traditional static measurement methods are difficult to separate the independent clearances of each stage of the transmission segment.

[0024] This invention fuses spatiotemporal data from lidar and 4D millimeter-wave radar. LiDar utilizes the high spatial resolution of lidar to accurately define the three-dimensional boundaries of components such as steering arms and tie rods, while the high-sensitivity Doppler velocity measurement characteristics of 4D millimeter-wave radar directly acquire the motion state. The combination of geometric positioning and velocity monitoring avoids noise interference caused by relying solely on point cloud position difference calculations for velocity. It can sensitively capture the instantaneous changes of components from stationary to moving under non-contact conditions, improving the accuracy of determining the start of motion.

[0025] This invention introduces the PointNet deep learning network for component semantic classification and velocity vector projection recovery algorithm, which realizes automatic identification of key components of steering systems in different vehicle models and corrects the geometric deviation between radar radial velocity and the actual lateral movement direction of the component. This process does not require manual pasting of markers or installation of contact sensors on the component, and can automatically compensate for the loss of velocity components caused by the observation angle, thereby improving the automation and versatility of detection while ensuring measurement accuracy. Attached Figure Description

[0026] Figure 1 This is a flowchart of a clearance measurement method for commercial vehicle steering systems provided by one or more embodiments of the present invention.

[0027] Figure 2 This is a schematic diagram of a steering system test bench and radar layout provided in a specific embodiment of the present invention.

[0028] Figure 3 This is a schematic diagram of obtaining high radar point objects in the environment through radar clustering processing according to a specific embodiment of the present invention.

[0029] Figure 4 This is a schematic diagram of the steering clearance calculation timing principle based on motion response timing difference provided in a specific embodiment of the present invention. Detailed Implementation

[0030] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] See attached document Figure 1 This invention provides a method for measuring clearance in commercial vehicle steering systems. Based on the ROS platform, this method involves constructing a test bench equipped with non-contact sensors to perform segmented observations of the commercial vehicle steering system and calculating the angular clearance in the steering drivetrain using multi-source data fusion technology. The method includes the following steps:

[0032] S1. Construct a test bench for the commercial vehicle steering system. The physical environment is built according to the structural logic of the steering wheel, steering column, drive shaft, steering gear, and steering fork as the first segment, and the steering fork, tie rod, steering knuckle, and wheel as the second segment. Specifically, the connection points between the chassis and the steering gear are fixed, and the steering wheel position is fixed to constrain the steering column; the front axle is raised so that the wheels are no longer in contact with the ground; the steering column, fork, tie rod, and steering fork (knuckle arm) are installed sequentially to construct a complete mechanical transmission system, enabling the steering wheel to be turned mechanically to deflect the wheels.

[0033] S2. Deploy a multi-source sensor system and perform joint calibration. Install signal lines at the steering wheel and connect them to the controller and data processing terminal, setting the sampling frequency to 50Hz to output the steering wheel angle and corresponding timestamp in real time. Deploy a LiDAR and a 4D millimeter-wave radar at the observation point of the commercial vehicle's steering system. The LiDAR is positioned on the left side of the vehicle with its X-axis perpendicular to the left side of the vehicle to acquire 3D point information of environmental objects in a 360° angle. The 4D millimeter-wave radar is centered and positioned directly above the LiDAR. The LiDAR sampling frequency is set to 10Hz, and the 4D millimeter-wave radar sampling frequency is set to 13Hz. Using the Lidar_align open-source package, extrinsic parameter calibration is performed by using an inertial measurement unit as an intermediate reference or directly calculating the geometric relationship between the two radar data, thus obtaining the extrinsic parameter transformation matrix between the LiDAR coordinate system and the 4D millimeter-wave radar coordinate system.

[0034] S3. Acquire multi-dimensional motion data during the steering process. The test bench is enclosed to prevent external interference. Simultaneously, the radar system and steering wheel angle controller collect data. As the steering wheel is slowly turned under manual force, the data processing terminal synchronously records the steering wheel angle data sequence and timestamp data sequence, as well as frame-by-frame point cloud data from the lidar and frame-by-frame velocity data from the 4D millimeter-wave radar. The turning operation continues until wheel rotation is observed and then stops.

[0035] S4. Spatiotemporal fusion and feature extraction of multi-source sensor data. On the ROS platform, an extrinsic parameter transformation matrix is ​​input to align the timestamps of LiDAR data frames, 4D millimeter-wave radar data frames, and steering wheel angle data, removing redundant data frames without a corresponding relationship. The K-Nearest Neighbors (k-NN) algorithm is used to cluster the LiDAR point cloud, searching and defining a threshold for the number of point clouds to generate bounding boxes surrounding each component. The PointNet point cloud deep learning network is applied, based on a pre-trained target set and classification set, to extract features and perform semantic classification on the clustered point cloud bounding boxes, outputting category labels for the steering arm, tie rod, or wheel. Simultaneously, the radar point velocity information acquired by the 4D millimeter-wave radar is extracted, mapping radar points with velocity values ​​to the corresponding component bounding boxes. The average velocity component in the Y direction (lateral movement direction) of each component bounding box is obtained through vector calculation.

[0036] S5. Calculate the clearance of each stage of the transmission chain in the commercial vehicle steering system. Based on the speed changes of components detected by 4D millimeter-wave radar, determine the moment when each component begins to generate effective movement. Align the steering wheel angle data with the movement times of the steering arm, tie rod, and wheels. When a speed jump is detected in a component (such as the steering arm), record the corresponding steering wheel angle at that moment. Calculate the difference between this angle and the angle corresponding to the movement time (or starting time) of the previous stage component; this difference is the angular clearance of that transmission chain segment. Alternatively, when the steering wheel speed is approximately uniform, calculate the time difference between the component's movement time and its starting time, and combine this with the steering wheel speed to estimate the corresponding angular clearance.

[0037] See attached document Figure 1 and attached Figure 2 In order to implement the gap measurement method of the present invention, in this embodiment, the precise construction of the physical test bench and the spatial deployment of sensors provide a data basis for subsequent gap calculation.

[0038] The specific implementation process for setting up a test bench for the commercial vehicle steering system in step S1 is as follows:

[0039] The test bench was constructed following the transmission logic of a commercial vehicle steering system: the first segment is from the steering wheel to the steering arm, and the second segment is from the steering arm to the wheel. First, a rigid chassis was selected as the reference platform, and the steering gear was rigidly fixed to a designated position on the chassis with bolts, ensuring that the steering gear housing remained stationary relative to the chassis under load. Next, the steering wheel mounting bracket was fixed to constrain the spatial position of the steering column, simulating the steering column posture in a real vehicle's cab. To eliminate the masking effect of ground friction on the measurement of minute gaps, a lifting device was used to raise the entire front axle of the commercial vehicle, completely detaching the left and right front wheels from the ground and placing them in a suspended state. Based on this, components were assembled sequentially according to the mechanical torque transmission path: the steering wheel was connected to the steering column, the steering column was connected to the drive shaft via a universal joint fork, and the input end of the drive shaft was connected to the steering gear; the output shaft of the steering gear was connected to the steering arm, the steering arm was connected to one end of the tie rod via a ball joint, and the other end of the tie rod was connected to the steering knuckle arm, ultimately driving the wheel mounted on the wheel hub. Through this mechanical connection, a complete physical transmission chain from steering wheel input to wheel response was established.

[0040] Regarding step S2, the deployment and joint calibration process of the multi-source sensor system is as follows: For signal acquisition, a steering angle sensor is installed at the steering wheel shaft and connected to the real-time controller and data processing terminal via a signal line. The controller's sampling frequency is configured to 50Hz to record the steering wheel angle. and its corresponding timestamp sequence.

[0041] In terms of environmental perception, based on the observation requirements of this embodiment, a non-contact sensor array is deployed at specific locations on the test bench. A lidar is installed on the left side of the vehicle body, its mounting orientation adjusted so that the X-axis of the lidar coordinate system points vertically to the left side of the vehicle body, and the Z-axis points vertically upwards. This layout utilizes the lidar's 360° scanning capability to directly cover the outer surfaces of the steering arm, tie rod, and wheels in a test bench environment without vehicle body skin obstruction, acquiring three-dimensional point cloud data for each component. A 4D millimeter-wave radar is physically installed directly above the lidar, its position adjusted so that the geometric center of the 4D millimeter-wave radar is aligned vertically with the geometric center of the lidar to minimize spatial parallax. The 4D millimeter-wave radar also faces the steering transmission assembly on the left side of the vehicle body, used to capture the radial velocity of the components using the Doppler effect. To meet the time density requirements of data fusion, the sampling frequency of the lidar is set to 10Hz, and the sampling frequency of the 4D millimeter-wave radar is set to 13Hz.

[0042] During the sensor calibration preparation phase, if an inertial measurement unit (IMU)-based calibration scheme is adopted, the IMU is rigidly fixed to the radar mounting bracket, keeping it relatively stationary with the lidar and 4D millimeter-wave radar. Using the `Lidar_align` package, the rigid body transformation relationship between the lidar coordinate system and the 4D millimeter-wave radar coordinate system is calculated by collecting the trajectory differences between the IMU data and the radar data, or by directly using the observation data of the two radars on the same static scene. This yields the extrinsic parameter transformation matrix, completing the spatial joint calibration of the multi-source sensor system. All sensor data transmission lines converge to the same data processing terminal to ensure data acquisition synchronization.

[0043] See attached document Figure 1 and attached Figure 2 After completing the setup of the test bench and the physical connection of the sensors, step S2 is performed for joint calibration and time synchronization to solve the problem of unification of multi-source data in spatial and temporal dimensions.

[0044] For joint spatial calibration, since the lidar and 4D millimeter-wave radar are installed separately, it is necessary to establish a rigid body transformation relationship between the two. The lidar coordinate system is set as follows: Its origin is located at the optical center of the lidar; the 4D millimeter-wave radar coordinate system is set as follows: Its origin is located at the geometric center of the 4D millimeter-wave radar. Using the Lidar_align calibration method, with the inertial measurement unit (IMU) as the motion reference or intermediate reference, the test bench or sensor support is driven to perform six-degree-of-freedom motion. The LiDAR point cloud trajectory, the 4D millimeter-wave radar velocity field trajectory, and the inertial data of the IMU are collected during this motion. The error between the trajectories of each sensor is minimized using a nonlinear optimization algorithm, and the extrinsic parameter transformation matrix from the 4D millimeter-wave radar coordinate system to the LiDAR coordinate system is calculated. .

[0045] Extrinsic transformation matrix The coordinate transformation is used to unify the spatial coordinates of target points detected by 4D millimeter-wave radar to the lidar coordinate system. The calculation is based on the following formula:

[0046] ;

[0047] in, The target point is represented in the lidar coordinate system. The three-dimensional coordinate vector below is represented as ; Representing the same target point in the 4D millimeter-wave radar coordinate system The three-dimensional coordinate vector below is represented as ; The result of the calibration Homogeneous transformation matrix; represent A rotation matrix describes the angular deviation between the axes of two coordinate systems. represent Translation vector, describing the spatial offset between the two radar centers; represent Zero vector; 1 represents the homogeneous coordinate scaling factor.

[0048] For time synchronization, given the differences in sampling frequencies among the various sensors (steering wheel angle sensor 50Hz, 4D millimeter-wave radar 13Hz, LiDAR 10Hz), a unified time reference is established based on the ROS platform. The data processing terminal uses the LiDAR data frame time as the primary alignment reference because LiDAR provides the main geometric contour information and has the lowest frequency. In data stream processing, when the data processing terminal receives a frame with a timestamp of... When retrieving lidar point cloud data, the timestamp in the cache queue is... 4D millimeter-wave radar data frames. Calculation of time difference. Select The smallest millimeter-wave radar frame that is lower than the preset synchronization threshold is bound to the current lidar frame. Other 13Hz millimeter-wave radar data frames that are not selected are discarded as asynchronous redundant data. Meanwhile, for the 50Hz steering wheel angle data, nearest neighbor interpolation or linear interpolation is used to obtain the time. Corresponding steering wheel angle Step S2 outputs a fused data packet containing unified timestamps, aligned point cloud data, velocity data, and corner data, providing spatiotemporally consistent data input for subsequent gap calculations.

[0049] See attached document Figure 1 and attached Figure 3 In step S4, in order to extract the geometric features of key components such as steering arm, tie rod and wheel from the massive messy data obtained by lidar, point cloud preprocessing and clustering operations are first required.

[0050] The data processing terminal first reads the LiDAR data frame processed in steps S2 and S3. Considering the presence of non-target echoes from the ground, support columns, and walls around the test bench, a pass-through filter is set in the ROS system to define a three-dimensional region of interest. This three-dimensional region of interest is based on the LiDAR coordinate system. The designation is to retain only the point cloud data whose X, Y, and Z axis coordinates are located within the effective transmission space of the test bench, i.e., to remove background noise points outside the coordinate range.

[0051] For the retained point cloud after filtering the 3D region of interest, a region growing clustering algorithm based on Euclidean distance is adopted. To improve the search efficiency of massive point clouds, a KD-Tree spatial index structure is constructed using the retained point cloud data.

[0052] The clustering process is as follows: Create an empty list of clusters. And a set of visited point markers. Traverse the point cloud data; if a certain point... If it is not visited, it will be added to the current cluster as the initial seed point. And search for neighborhoods that meet the distance threshold using KD-Tree All points. Among them, two points and Spatial Euclidean distance between The calculation formula is:

[0053] ;

[0054] in, Point With point The spatial Euclidean distance between them; , , Point Three-dimensional coordinates in the lidar coordinate system; , , Point Three-dimensional coordinates in the lidar coordinate system.

[0055] If the calculation result satisfies Then the point Join current And the queue to be searched. The region growing clustering algorithm continuously takes new points from the queue to be searched and repeats the above neighborhood search process until the queue to be searched is empty, which marks the completion of clustering an independent connected region (i.e., a potential component).

[0056] Subsequently, for each generated cluster Perform quantity filtering. Set a point cloud quantity threshold. and If the number of points within a cluster satisfy If the number of points is too low, the cluster is considered a valid commercial vehicle steering system component (such as a steering arm or wheel) and is retained; otherwise, if the number of points is too high, it is considered an air noise or discrete interference, and if the number of points is too high, it is considered an unfiltered background wall and is rejected.

[0057] For each valid point cloud cluster after filtering, its axis-aligned bounding box is calculated. This involves extracting the maximum and minimum values ​​of all points in the cluster along the X, Y, and Z dimensions to generate a rectangular bounding box that encloses the component. The rectangular bounding box not only locks the spatial position of the component, but its geometric center coordinates and dimensional parameters also serve as the basic input for subsequent deep learning classification networks and velocity vector fusion.

[0058] See attached document Figure 1 and attached Figure 3 Following the point cloud clustering process in step S4, in order to accurately distinguish key components such as steering arms, tie rods, and wheels from the candidate bounding boxes generated by geometric clustering, this embodiment uses the PointNet deep learning network for semantic classification.

[0059] The data processing terminal first performs spatial normalization on each candidate point cloud cluster. Since the original point cloud data is based on the LiDAR global coordinate system, the spatial positions of various components differ significantly; directly inputting this data into the network would affect the stability of feature extraction. The processing unit calculates the geometric center coordinates of the current point cloud cluster. and all points within the cluster The coordinates are translated to In a local coordinate system with the origin as the origin, scale normalization is performed simultaneously to scale the point cloud coordinate values ​​to the unit sphere, eliminating the influence of spatial location and size differences on the classification results.

[0060] The standardized point cloud data is input into a pre-trained PointNet classification network model. This model has already undergone offline parameter training using a labeled dataset of commercial vehicle steering system components. During inference, the network first spatially aligns the input point cloud using an input transformation network, then uses a multilayer perceptron to upscale the features of each point, extracting high-dimensional point-by-point features. Next, a global feature vector, insensitive to the order of input points, is aggregated from the high-dimensional point-by-point features using a symmetric function, max pooling. This global feature vector represents the overall three-dimensional shape of the point cloud cluster. After mapping through a fully connected layer, the global feature vector outputs a vector of length [length missing]. score vector ,in Corresponding to the total number of target categories to be classified, in this embodiment These correspond to the steering arm, tie rod, and wheel, respectively.

[0061] To obtain the confidence scores for each component category, the Softmax function is used to convert the network output score vector. Convert to a probability distribution. Predicted probabilities for each category The calculation formula is as follows:

[0062] ;

[0063] in, Indicates that the input point cloud belongs to the first... The predicted probabilities of each category range from [0, 1]. The base of the natural logarithm; Represents the network output score vector The Middle The value of the nth element, i.e., the nth... The original log odds of each category; Represents the network output score vector The Middle The value of each element; Corresponding to the total number of target categories to be classified, in this embodiment These correspond to the steering arm, tie rod, and wheel, respectively. This represents the sum of the exponential operations on all elements in the score vector, used as the normalized denominator.

[0064] The data processing terminal calculates the predicted probability. Perform classification decision-making. Set confidence thresholds. Select the category index corresponding to the maximum value in the predicted probability vector. .like Then, assign the category label to the current point cloud cluster and its corresponding bounding box, and add it to the list.

[0065] See attached document Figure 1 and attached Figure 3 During step S4, the data processing terminal has obtained a lidar point cloud bounding box with semantic tags for the steering arm, tie rod, or wheel, and their three-dimensional spatial positions. In order to obtain the dynamic response data of the above-mentioned components during the steering clearance test, especially the movement speed of the components along the Y-axis (i.e., the lateral movement direction of the test bench), this embodiment fuses the speed information captured by the 4D millimeter-wave radar into the above-mentioned geometric object.

[0066] The data processing terminal first maps the current frame point cloud data acquired by the 4D millimeter-wave radar to the lidar coordinate system based on the extrinsic parameter calibration matrix obtained in step S2, in order to unify the spatial reference. The processing unit traverses all bounding boxes with semantic labels. It retrieves all millimeter-wave radar points that fall within the bounding box with semantic labels. If a certain radar point... Spatial coordinates located in the bounding box Within the geometric range, it is marked as the effective velocity observation point of the component and added to the point set of the component. .

[0067] Considering that the physical measurement characteristics of 4D millimeter-wave radar can only provide the radial velocity of the target relative to the radar optical center, while the clearance quantization of commercial vehicle steering systems depends on the linear velocity of components along a specific mechanical transmission direction (limited to the Y-axis direction in this embodiment), the data processing terminal performs point set... For each radar point in the dataset, velocity vector projection is performed to reconstruct the data. For the point set... Included One effective radar point, the average Y-axis velocity of the component at the current moment. The calculation formula is as follows:

[0068] ;

[0069] in, This represents the calculated average velocity of the component in the Y-axis direction; This indicates the total number of valid millimeter-wave radar points that fall within the boundary frame of the component. The coefficient representing the arithmetic mean of the sum of velocities at all valid points; This indicates an operation that sums the results within the parentheses; the index... Traverse from 1 to ; The first direct measurement output of the 4D millimeter-wave radar represents the... Radial velocity values ​​of each radar point; Indicates the first The cosine of the angle between the radial velocity vector of each radar point and the Y-axis direction of motion is used as the geometric factor for velocity projection reconstruction.

[0070] To accurately solve this formula, The calculation is based on the geometric position of the radar point in the radar coordinate system, and its expansion formula is as follows:

[0071] ;

[0072] in, Indicates the first The Y-axis coordinate components of each radar point in the millimeter-wave radar's own coordinate system; Indicates the first The X-axis coordinate components of each radar point in the millimeter-wave radar's own coordinate system; Indicates the first The z-axis coordinate components of a radar point in the millimeter-wave radar's own coordinate system; , , These represent the square operations for each coordinate component; Indicates the first The straight-line Euclidean distance from each radar point to the radar optical center is the modulus of the denominator.

[0073] Through the above calculation steps, the data processing terminal transforms discrete radar data containing only radial information into vector velocities characterizing the actual motion state of the components, and averages the values ​​of all observation points within the same component to eliminate measurement noise. Finally, the data processing terminal outputs fused state data containing timestamps, semantic categories, 3D bounding boxes, and Y-axis motion velocity, which serves as the key input parameter for the subsequent step S5 to calculate the steering clearance.

[0074] See attached document Figure 1 and attached Figure 4 In order to accurately extract the first-level response time and the second-level response time in step S5 and eliminate the interference of random noise in the radar signal on the determination of the motion start time, this embodiment uses an average amplitude detection algorithm based on a sliding time window to perform time domain analysis on the velocity data output in step S4.

[0075] During the data processing initialization phase, the processing unit first extracts the radar velocity data from the stationary phase in step S3 when the steering wheel angle is zero, and calculates the average value of this segment of radar velocity data. and standard deviation Based on statistical principles, the processing unit sets a threshold for determining stillness. In this embodiment, it is set This ensures the confidence level distinguishes between background noise and valid signals. Subsequently, for the time-series velocity data of components such as the steering arm and tie rod, the processing unit constructs a data structure with a length of [length missing]. The time window is scrolled forward frame by frame along the timeline.

[0076] For any time The processing unit calculates the average level of the component velocity amplitude within the current sliding time window, denoted as the motion intensity operator. The formula for calculating the motion intensity operator is as follows:

[0077] ;

[0078] in, Indicates at time The calculated average velocity amplitude within the current time window is used to quantify the motion intensity of the components; This indicates the number of data frames contained in the sliding time window, which is set to 5 frames in this embodiment; This represents the coefficient used to calculate the arithmetic mean of the accumulated velocity amplitudes within the window. This indicates an accumulation and summation operation on the data within the sliding window, index. Traverse from 0 to |...| indicates the absolute value operation, which converts the positive and negative velocities into amplitude values ​​to prevent sign cancellation; This represents the calculated average velocity of the component in the Y-axis direction; This indicates the time interval between adjacent radar data frames.

[0079] The processing unit will calculate the motion intensity operator in real time. Compared with the preset static judgment threshold A comparison is performed. To further prevent misjudgment due to sporadic noise pulse triggering, the processing unit introduces a continuity determination mechanism: only when continuous... Three time steps (set to 3 in this embodiment) All values ​​are strictly greater than Only when the condition is met does the processing unit determine that the component has entered a motion state. At this point, the processing unit traces back to the first time point in this continuous sequence that meets the condition and marks it as the start time of the component's motion response.

[0080] Based on this logic, for the observation data of the steering boom, if the judgment condition is met, the processing unit records that moment as... For the observation data of the longitudinal tie rod, if the judgment condition is met, the processing unit records that moment as... By combining statistical threshold setting and sliding window smoothing, the processing unit ensures the accuracy of the time parameters substituted into the gap calculation formula in step S5. Finally, it combines the starting time of the steering wheel angle recorded in step S3. The processing unit completes the precise calculation of the transmission clearance at each stage.

[0081] See attached document Figure 1 and attached Figure 4 In step S5, the data processing terminal uses the key motion time points of each component determined in the aforementioned steps, combined with the motion parameters of the input end, to perform segmented quantitative calculation of the mechanical clearance inside the commercial vehicle steering system. This calculation process is based on the time-domain response lag principle of the mechanical transmission chain, that is, when the input end is in continuous motion, the delay in the response time of the output end directly corresponds to the physical clearance existing inside the transmission segment.

[0082] In step S3, to ensure the stability of data acquisition, the testers constructed a closed test environment to eliminate external interference. The steering wheel angle sensor, LiDAR, and 4D millimeter-wave radar were synchronously triggered via an integrated controller. After the test began, the controller drove the steering wheel to rotate at a preset angular velocity in one direction at a constant speed, or the operator could manually rotate it slowly. During this period, the data processing terminal synchronously recorded the steering wheel angle sequence and system timestamps. Simultaneously, based on the component velocity vector fusion results output in step S4, the data processing terminal continuously monitored the average movement speed of the steering arm, tie rod, and wheels in the Y-axis direction. .

[0083] For the first stage of the drivetrain, namely the mechanical link between the steering wheel and the steering arm, the data processing terminal calculates the cumulative steering clearance angle of this segment. The calculation formula is as follows:

[0084] ;

[0085] in, This indicates the internal steering clearance angle value of the first-stage transmission section (from the steering wheel to the steering arm), in degrees. This represents the average angular velocity of the steering wheel rotation during the test. This value is obtained by taking the derivative of the steering wheel angle data recorded in step S3 with respect to time and averaging it. The unit is degrees per second. This indicates the initial response time of the steering boom motion determined by the sliding time window algorithm in step S5. This indicates the moment when the steering wheel angle data obtained in step S5 begins to change; This indicates the response lag time of the first-stage transmission section, which is the time difference between when the input end has generated angular displacement and when the steering arm has not yet generated effective linear motion.

[0086] For the second-stage drivetrain, namely the mechanical link between the steering drop arm and the tie rod, the data processing terminal calculates the steering clearance angle within this segment. The calculation formula is as follows:

[0087] ;

[0088] in, This indicates the internal steering clearance angle value of the second-stage drive section (from the steering drop arm to the tie rod), in degrees; This represents the average angular velocity of the steering wheel rotation during the test. This value is obtained by taking the derivative of the steering wheel angle data recorded in step S3 with respect to time and averaging it. The unit is degrees per second. This indicates the initial response time of the longitudinal tie rod motion determined by the sliding time window algorithm in step S5; This indicates the initial response time of the steering boom motion determined by the sliding time window algorithm in step S5. It represents the interstage response lag time of the second-stage transmission section relative to the previous stage transmission, and characterizes the mechanical free travel during the movement of the steering arm driving the longitudinal tie rod.

[0089] The data processing terminal, through the aforementioned calculation steps, maps the time-domain lag data into angle-domain gap data, and outputs the internal steering gap angle values ​​of the first-stage transmission section (steering wheel to steering arm). Internal steering clearance angle value of the second-stage drive section (steering arm to tie rod) This result enables independent decoupling and quantitative evaluation of the clearances between the steering wheel and the drop arm, and between the drop arm and the tie rod, ultimately generating a clearance distribution report for the commercial vehicle steering system.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for measuring clearance in a commercial vehicle steering system, characterized in that, Includes the following steps: S1. Construct a physical transmission environment for a commercial vehicle steering system that includes a first transmission segment and a second transmission segment. Fix the connection position between the vehicle frame and the steering gear and constrain the position of the steering wheel. Raise the front axle so that the wheels are out of contact with the ground and establish a mechanical transmission link from steering wheel input to wheel response. S2. Arrange a corner acquisition device at the steering wheel, and arrange a lidar and a 4D millimeter-wave radar on the observation side of the mechanical transmission link, and establish the spatial transformation relationship between the lidar coordinate system and the 4D millimeter-wave radar coordinate system. S3. During the process of driving the steering wheel to rotate, simultaneously record the steering wheel's rotation angle and angular velocity data sequence, timestamp data sequence, LiDAR point cloud data, and 4D millimeter-wave radar data. S4. Align the point cloud data of the lidar with the data of the 4D millimeter-wave radar using the spatial transformation relationship, identify the bounding boxes of each key component in the mechanical transmission link, and fuse the speed information in the 4D millimeter-wave radar data into the bounding boxes to obtain the average movement speed of each key component in the lateral movement direction. S5. Determine the starting time of movement of each key component based on the average movement speed, calculate the response lag time between the starting time of the angle change and the starting time of movement of each key component in the steering wheel angle and angular velocity data sequence, and multiply the response lag time by the angular velocity in the steering wheel angle and angular velocity data sequence to obtain the angular clearance of each stage of the mechanical transmission link.

2. The clearance measurement method for commercial vehicle steering systems according to claim 1, characterized in that, In step S1, the steps of constructing the physical transmission environment of the commercial vehicle steering system, which includes the first transmission segment and the second transmission segment, include: A rigid frame is selected as the reference platform. The steering gear is fixed to the rigid frame with bolts. The mounting bracket for fixing the steering wheel is fixed, thereby constraining the spatial position of the steering column. The lifting device is used to raise the front axle of the commercial vehicle as a whole. The first transmission section includes at least a steering wheel, a steering column, a drive shaft, a steering gear, and a steering arm; The second transmission section includes at least a steering arm, a tie rod, a steering knuckle, and a wheel.

3. The clearance measurement method for commercial vehicle steering systems according to claim 2, characterized in that, In step S2, the step of deploying lidar and 4D millimeter-wave radar on the observation side of the mechanical transmission link includes: Select one side of the vehicle body as the observation side of the mechanical transmission link, install the lidar on one side of the vehicle body, adjust the X-axis of the lidar coordinate system to point vertically to the side of the vehicle body and the Z-axis to point vertically upward, so that the lidar scanning range covers the steering arm, the tie rod and the outer side of the wheel. The 4D millimeter-wave radar is mounted above the lidar, and the geometric center of the 4D millimeter-wave radar is aligned with the geometric center of the lidar on a vertical line.

4. The clearance measurement method for commercial vehicle steering systems according to claim 2, characterized in that, In step S4, the step of identifying the bounding boxes of each key component in the mechanical transmission link includes: The 4D millimeter-wave radar data is input into the coordinate space of the point cloud data of the lidar, and the spatial coordinates of the 4D millimeter-wave radar data are mapped to the coordinate system of the point cloud data of the lidar using the spatial transformation relationship, thereby completing the data alignment. Using the point cloud data of the lidar as a geometric reference, a three-dimensional region of interest is set for the spatial range where the mechanical transmission link is located, and background noise points are removed. A KD-Tree spatial index structure is constructed, and a region growing clustering algorithm based on Euclidean distance is adopted to traverse the point cloud data of the lidar and search for points in the neighborhood that meet the Euclidean distance threshold. Points that meet the conditions are added to the current cluster. Valid point cloud clusters representing potential key components are selected based on a preset point cloud quantity threshold, and the axis-aligned bounding boxes of the valid point cloud clusters are calculated and the bounding boxes are generated.

5. The clearance measurement method for commercial vehicle steering systems according to claim 4, characterized in that, In step S4, identifying the bounding boxes of each key component in the mechanical transmission link further includes semantic classification of the bounding boxes. Specifically, semantic classification of the bounding boxes includes: Spatial standardization is performed on the effective point cloud cluster, the coordinates of the geometric center are calculated, and the coordinates of all points in the effective point cloud cluster are translated to the local coordinate system, and scale normalization is performed. The standardized point cloud data of the LiDAR is input into a pre-trained PointNet classification network model. Spatial alignment is performed through an input transformation network. High-dimensional point-by-point features are extracted using a multilayer perceptron. Global feature vectors are aggregated through a max pooling layer. The score vector output by the PointNet classification network model is converted into a probability distribution using a Softmax function. The semantic labels of the steering arm, the tie rod, and the wheel in the mechanical transmission link are determined based on the confidence threshold.

6. The clearance measurement method for commercial vehicle steering systems according to claim 1, characterized in that, In step S4, the step of obtaining the average movement speed of each key component in the lateral movement direction includes: Based on the aligned coordinate system, retrieve all 4D millimeter-wave radar points that fall within the bounding box of each key component, and construct the point set of each key component. For each of the 4D millimeter-wave radar points in the point set, velocity vector projection restoration is performed. The cosine value of the angle between the radial velocity vector of the 4D millimeter-wave radar point and the lateral movement direction is calculated. The radial velocity value is multiplied by the cosine value to obtain the projected velocity. The projected velocities of all valid 4D millimeter-wave radar points in the point set are summed and divided by the total number of valid 4D millimeter-wave radar points to obtain the average movement velocity of each key component.

7. The clearance measurement method for commercial vehicle steering systems according to claim 1, characterized in that, In step S5, the step of determining the motion start time of each key component based on the average motion speed includes: The average value and standard deviation of the 4D millimeter-wave radar data during the stationary phase are calculated, and a stationary determination threshold is set. For each key component, a sliding time window is constructed along the time axis, and the average amplitude level of the average motion speed of each key component within the current sliding time window is calculated to obtain the motion intensity operator; The motion intensity operator calculated in real time is compared with the static determination threshold. When the motion intensity operator for multiple consecutive time steps is greater than the static determination threshold, it is determined that each key component has entered a motion state, and the first time point in the continuous sequence that meets the condition is recorded as the motion start time.

8. The clearance measurement method for commercial vehicle steering systems according to claim 2, characterized in that, In step S5, the step of obtaining the angular clearance of each stage of the mechanical transmission link includes: Determine the moment when the steering angle begins to change in the steering wheel's angle and angular velocity data sequence, and the moment when the steering arm, as one of the key components, begins to move; The time difference between the starting moment of the movement of the steering arm and the moment when the steering angle begins to change in the steering wheel angle and angular velocity data sequence is calculated and used as the response lag time of the first transmission segment of the commercial vehicle steering system. Multiply the response lag time of the first transmission segment by the average angular velocity of the steering wheel's angle and angular velocity data sequence to obtain the internal steering clearance angle value of the first transmission segment between the steering wheel and the steering arm.

9. The clearance measurement method for commercial vehicle steering systems according to claim 8, characterized in that, In step S5, obtaining the angular clearance of each stage of the mechanical transmission link further includes: Determine the start time of the movement of the longitudinal tie rod, which is one of the key components; Calculate the time difference between the starting time of the movement of the longitudinal tie rod and the starting time of the movement of the steering drop arm, and use it as the interstage response lag time of the second transmission segment of the commercial vehicle steering system relative to the first transmission segment; Multiplying the interstage response lag time by the average angular velocity yields the internal steering clearance angle value of the second transmission segment between the steering drop arm and the longitudinal tie rod.

10. The clearance measurement method for commercial vehicle steering systems according to claim 1, characterized in that, In step S4, aligning the point cloud data of the lidar and the 4D millimeter-wave radar data using the spatial transformation relationship also includes alignment in the time dimension. This time dimension alignment specifically includes: A unified time reference is established based on the ROS platform, and the frame time of the point cloud data of the lidar is used as the main alignment reference. When a frame of point cloud data from the lidar is received, the 4D millimeter-wave radar data in the cache queue is retrieved, the acquisition time difference between the two is calculated, the millimeter-wave radar frame with the smallest time difference and less than the preset synchronization threshold is selected and bound to the current lidar frame, and redundant data that is not selected is removed.

Images