Edge-computing-based contactless vehicle weighing system and weighing method

Abstract

The application discloses a non-contact vehicle weighing system and method based on edge computing. The weighing system comprises a detection layer, which comprises a pair of parallel sliding rails, a detection area formed between the pair of parallel sliding rails, and at least four non-contact detection units. At least two non-contact detection units are installed on each parallel sliding rail. Each non-contact detection unit corresponds to the detection of one tire. Each non-contact detection unit comprises a sliding trolley capable of moving along the parallel sliding rail to follow the corresponding tire and a parameter sensor carried on the sliding trolley. The parameter sensor comprises a radar tire pressure sensor and a vertical deflection sensing assembly carried on the sliding trolley. An edge computing layer calculates the weight of the vehicle according to the tire parameters detected by the non-contact detection unit. The application provides a non-contact vehicle weighing system and method based on edge computing, which can realize the rapid measurement of the weight of the vehicle without direct contact with the vehicle.

Description

Technical Field

[0001] This invention relates to the field of vehicle weight measurement technology, specifically to a contactless vehicle weighing system and method based on edge computing. Background Technology

[0002] Vehicle weighing is a core component of traffic management, cargo measurement, and overload prevention. Current mainstream weighing methods rely on contact-type equipment such as weighbridges and axle load meters, which present the following significant drawbacks:

[0003] 1. Poor adaptability and limited by vehicle type: Traditional weighbridges require vehicles to be parked on the weighing platform, and the platform size is fixed (e.g., 3m×18m), which cannot adapt to various vehicle types with wheelbases of 2.5m-4.5m (e.g., light trucks and heavy semi-trailers); axle load meters require vehicles to run over each axle, which can easily lead to missed measurements for vehicles with longer wheelbases, and the compatibility coverage is less than 60%.

[0004] 2. Contact measurement is prone to wear and tear and has high maintenance costs: Weighbridges rely on strain gauges on the weighing platform to sense pressure. After long-term use, the strain gauges are prone to failure due to vehicle crushing and dust corrosion. The maintenance cycle is short (1-2 months / time), and the cost of a single maintenance exceeds 2,000 yuan. The axle load meter sensor is exposed to the outside. The failure rate increases by 30% when it gets wet in rainy weather, and the annual maintenance cost accounts for 25% of the total equipment cost. Summary of the Invention

[0005] The purpose of this invention is to provide a contactless vehicle weighing system and method based on edge computing, which can achieve rapid measurement of vehicle weight without direct contact with the vehicle.

[0006] This invention discloses a contactless vehicle weighing system based on edge computing, comprising: a detection layer, wherein the detection layer includes:

[0007] A pair of slide rails are arranged in parallel and spaced apart, so that a detection area for vehicles to pass through is formed between the pair of slide rails;

[0008] Two detection groups are respectively set on a pair of slide rails. Each detection group includes at least two non-contact detection units. Each non-contact detection unit includes a radar tire pressure sensor and a vertical deflection detection component, as well as a sliding trolley slidably mounted on the corresponding slide rail. The radar tire pressure sensor and the vertical deflection detection component are both mounted on the sliding trolley, so that when the vehicle passes through the detection area, the sliding trolley slides along with one of the vehicle's tires and detects the tire pressure and vertical deflection of the tire respectively.

[0009] An edge computing layer, connected to each of the detection units, is used to acquire the tire pressure and vertical deflection of all tires, and calculate the total weight M of the vehicle according to the following formula:

[0010] Fi = K base *(P real -P0) + K def *δ (1)

[0011] (2)

[0012] Among them, F i For a single wheel load, P real The measured actual tire pressure of the following tire, P0 is the tire pressure when unloaded, which is a preset value and depends on the tire model, K base K is the reference value for tire stiffness. def The deformation correction coefficient is obtained through a lightweight CNN-LSTM fusion model, where δ is the vertical deflection and g is the gravitational acceleration.

[0013] A further improvement of the present invention is that the vertical deflection detection component includes: a camera mounted on the sliding trolley; and a vertical deflection calculation unit connected to the camera, wherein the vertical deflection calculation unit obtains the current profile of the tire based on the tire photograph captured by the camera, and compares the current profile of the tire with the initial profile of the tire to obtain the vertical deflection of the tire.

[0014] A further improvement of the present invention is that the contactless detection unit further includes a tire temperature sensor mounted on the sliding trolley and used to detect the tire temperature of the following tire when the vehicle passes through the detection area; the edge computing layer further includes a temperature correction unit that corrects the calculation results of the edge computing layer based on the tire temperature detected by the temperature sensor.

[0015] A further improvement of the present invention is that it also includes a linkage control module, which includes: a wheel speed sensor mounted on the sliding trolley for detecting the movement speed of the following tire; a drive motor mounted on the sliding trolley for driving the sliding trolley to move and adjusting the movement speed of the sliding trolley; and a control unit connected to the wheel speed sensor and the drive motor for controlling the drive motor according to the movement speed of the tire to adjust the movement speed of the sliding trolley, thereby ensuring that the sliding trolley can closely follow the corresponding tire.

[0016] A further improvement of the invention is that it also includes an application output layer for storing and outputting the vehicle weight calculated by the edge computing layer.

[0017] A further improvement of the present invention is that the application output layer includes an alarm module for determining whether the total weight of the vehicle passing through the detection unit exceeds a maximum threshold and for issuing an alarm when the maximum threshold is exceeded.

[0018] A further improvement of the present invention is that the alarm module includes an identification unit and an alarm unit connected to the identification unit; the identification unit pre-stores maximum load threshold information corresponding to different vehicle models, which is used to identify the model of the vehicle passing through the detection area and output the corresponding maximum load according to the identified model; the alarm unit is used to compare the maximum load with the vehicle weight, and issue an alarm when the vehicle weight is greater than the maximum load.

[0019] A further improvement of the weighing system of the present invention is that the non-contact detection unit further includes a vibration sensor mounted on the sliding trolley for detecting ground vibration; the edge computing layer further includes a vibration correction unit for correcting the calculation results of the edge computing layer based on the ground vibration detected by the vibration sensor.

[0020] A further improvement of the present invention is that it includes the following steps:

[0021] S1. After the vehicle to be weighed enters the detection area and travels within the detection area, select the same number of non-contact detection units as the vehicle tires, and slide the corresponding sliding trolleys to follow each tire. Use these non-contact detection units to detect the tire pressure P and vertical deflection δ of the corresponding tires respectively.

[0022] S2, the edge computing layer receives the tire pressure P of each tire. real Combined with the vertical deflection δ, the total vehicle weight M is obtained using the following formula:

[0023] F i = K base *(P real -P0) + K def *δ,

[0024]

[0025] Among them, F i K represents the single-wheel load of the tire. base K is the reference value for tire stiffness. def P0 is the deformation correction coefficient obtained by the lightweight CNN-LSTM fusion model, P0 is the tire pressure when the tire is unloaded, which is obtained according to the tire model, and g is the acceleration due to gravity.

[0026] Where n is the total number of tires (e.g., n=6 for a 6-wheeled truck), M is in kg, and g is the acceleration due to gravity.

[0027] A further improvement of the present invention is that the contactless detection unit further includes a tire temperature sensor mounted on the sliding trolley; during step S1, the tire temperature of the following tire is detected by the tire temperature sensor, and the result is fed back to the temperature correction unit of the edge computing layer; before step S2:

[0028] The temperature correction unit corrects the tire pressure detected by the tire pressure sensor based on the tire temperature, using the following correction formula:

[0029] P corr = P real -α* (T - 25)

[0030] Where 25℃ is the reference temperature, and α=0.3kPa / ℃ is the temperature correction factor;

[0031] After obtaining the corrected tire pressure P corr Then, in step S2, the single-wheel load F of the tire is obtained. i At that time, the deformation correction coefficient K was obtained through a lightweight CNN-LSTM fusion model. def During the process, tire temperature T is added as a parameter to obtain a more accurate K. def ;

[0032] And the formula for calculating Fi is adjusted as follows:

[0033] F i =K base *(P corr -P0) + K def *δ.

[0034] This invention discloses a contactless vehicle weighing system and method based on edge computing. It achieves contactless detection of tire parameters through a contactless detection unit, preventing wear caused by contact measurement. At the same time, it accurately obtains the vehicle weight through an edge computing layer. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0036] Figure 2 This is a schematic diagram of the weighing process of the present invention;

[0037] Figure 3 This is a schematic diagram illustrating the specific parameters of the present invention;

[0038] Figure 4 Diagram of the CNN-LSTM fusion model architecture; Detailed Implementation

[0039] like Figures 1-4As shown, the present invention discloses a contactless vehicle weighing system based on edge computing, comprising:

[0040] The detection layer includes:

[0041] A pair of slide rails are arranged in parallel and spaced apart, so that a detection area for vehicles to pass through is formed between the pair of slide rails;

[0042] Two detection groups are respectively set on a pair of slide rails. Each detection group includes at least two non-contact detection units. Each non-contact detection unit includes a radar tire pressure sensor and a vertical deflection detection component, as well as a sliding trolley slidably mounted on the corresponding slide rail. The radar tire pressure sensor and the vertical deflection detection component are both mounted on the sliding trolley, so that when the vehicle passes through the detection area, the sliding trolley slides along with one of the vehicle's tires and detects the tire pressure and vertical deflection of the tire respectively.

[0043] An edge computing layer, connected to each of the detection units, is used to acquire the tire pressure and vertical deflection of all tires, and calculate the total weight M of the vehicle according to the following formula:

[0044] F i = K base *(P real -P0) + K def *δ (1)

[0045] (2)

[0046] Among them, F i For a single wheel load, P real The measured actual tire pressure of the following tire, P0 is the tire pressure when unloaded, which is a preset value and depends on the tire model, K base K is the reference value for tire stiffness. def The deformation correction coefficient is obtained through a lightweight CNN-LSTM fusion model, where δ is the vertical deflection and g is the gravitational acceleration.

[0047] like Figure 1 As shown, the vertical deflection sensing component includes: a camera mounted on the sliding trolley; and a vertical deflection calculation unit connected to the camera. The vertical deflection calculation unit obtains the current profile of the tire based on the tire photograph captured by the camera, and compares the current profile of the tire with the initial profile of the tire to obtain the vertical deflection of the tire.

[0048] In this embodiment, the radar tire pressure sensor is a millimeter-wave radar tire pressure sensor (model: TPMS-Radar-02, detection distance 0.3-2m, accuracy ±2kPa, sampling rate 10Hz, protection level IP68).

[0049] In this embodiment, a 2K vision camera is selected, with a frame rate of 30fps, a field of view of 120°, and a lens with a dustproof coating.

[0050] like Figure 1 As shown, the contactless detection unit further includes a tire temperature sensor mounted on the sliding trolley and used to detect the tire temperature of the following tire when the vehicle passes through the detection area; the edge computing layer further includes a temperature correction unit that corrects the calculation results of the edge computing layer based on the tire temperature detected by the temperature sensor.

[0051] Preferably, in this embodiment, the millimeter-wave radar tire pressure sensor (model: TPMS-Radar-02) can detect both tire pressure and temperature, so the temperature sensor is also a millimeter-wave radar tire pressure sensor.

[0052] In this embodiment, the edge computing layer hardware uses an industrial-grade edge box (4 TOPS computing power, 8GB memory, 64GB storage, supports 4G / 5G networks, and operates at a temperature of -30℃ to 70℃). The software includes a pre-loaded lightweight CNN-LSTM fusion model, an improved Hertz contact formula, an environmental error correction algorithm, and a data processing latency of ≤100ms.

[0053] like Figure 1 As shown, it also includes a linkage control module, which includes: a wheel speed sensor mounted on the sliding trolley for detecting the movement speed of the following tire; a drive motor mounted on the sliding trolley for driving the sliding trolley and adjusting the movement speed of the sliding trolley; and a control unit connected to the wheel speed sensor and the drive motor for controlling the drive motor according to the movement speed of the tire to adjust the movement speed of the sliding trolley, thereby ensuring that the sliding trolley can closely follow the corresponding tire.

[0054] like Figure 1 As shown, it also includes an application output layer for storing and outputting the vehicle weight calculated by the edge computing layer.

[0055] like Figure 1As shown, the application output layer includes an alarm unit, and the alarm module includes an identification unit and an alarm unit connected to the identification unit. The identification unit has pre-stored maximum load threshold information corresponding to different vehicle models, which is used to identify the model of the vehicle passing through the detection area and output the corresponding maximum load according to the identified model. The alarm unit is used to compare the maximum load with the total weight of the vehicle and issue an alarm when the vehicle weight is greater than the maximum load.

[0056] In this embodiment, as Figure 1 As shown, the alarm unit is an audible and visual alarm device. When the calculated total weight of the vehicle is more than 10% higher than the maximum load, the audible and visual alarm device emits a red light and an audible alarm. When the total weight of the vehicle is 4% to 10% higher than the maximum load, it emits a yellow light and an audible alarm.

[0057] like Figure 1 As shown, the application output layer also includes a storage unit that stores the weighed vehicle, its tire parameters, and the vehicle's weight.

[0058] Preferably, in this embodiment, as Figure 1 As shown, the application output layer also includes a touch screen, which is connected to the edge computing layer to display the calculation results of the edge computing layer and the various parameters detected by the detection layer.

[0059] like Figure 1 As shown, the non-contact detection unit also includes a vibration sensor mounted on the sliding trolley for detecting ground vibration; the edge computing layer also includes a vibration correction unit for correcting the calculation results of the edge computing layer based on the ground vibration detected by the vibration sensor.

[0060] Preferably, in this embodiment, as Figure 1 As shown, it also includes an infrared beam sensor installed on a pair of parallel slide rails and located at the starting point of the detection area. The infrared beam sensor can detect vehicles entering the detection area. When a vehicle is detected, it matches the detected vehicle / tire with the vehicle / tire model in the database to obtain the vehicle / tire model information. Then, it transmits the vehicle entry signal information to the detection layer and the edge computing layer, enabling the detection layer and the edge computing layer to enter the working state. When no vehicle enters, the detection layer and the edge computing layer are in standby state to reduce unnecessary energy waste.

[0061] like Figure 2 As shown, a weighing method for a contactless vehicle weighing system based on edge computing as described in claim 1 is characterized by comprising the following steps:

[0062] S1. After the vehicle to be weighed enters the detection area and travels within the detection area, select the same number of non-contact detection units as the vehicle tires, and slide the corresponding sliding trolleys to follow each tire. Use these non-contact detection units to detect the tire pressure P and vertical deflection δ of the corresponding tires respectively.

[0063] S2, the edge computing layer receives the tire pressure P of each tire. real Combined with the vertical deflection δ, the total vehicle weight M is obtained using the following formula:

[0064] F i = K base *(P real -P0) + K def *δ,

[0065]

[0066] Among them, F i K represents the single-wheel load of the tire. base K is the reference value for tire stiffness. def P0 is the deformation correction coefficient obtained by the lightweight CNN-LSTM fusion model, P0 is the tire pressure when the tire is unloaded, which is obtained according to the tire model, and g is the acceleration due to gravity.

[0067] Where n is the total number of tires (e.g., n=6 for a 6-wheeled truck), M is in kg, and g is the acceleration due to gravity.

[0068] like Figure 2 As shown, the contactless detection unit also includes a tire temperature sensor mounted on the sliding trolley; during step S1, the tire temperature sensor detects the temperature of the following tire and feeds the result back to the temperature correction unit of the edge computing layer; before step S2:

[0069] The temperature correction unit corrects the tire pressure detected by the tire pressure sensor based on the tire temperature, using the following correction formula:

[0070] P corr = P real -α* (T - 25)

[0071] Where 25℃ is the reference temperature, and α=0.3kPa / ℃ is the temperature correction factor;

[0072] After obtaining the corrected tire pressure P corr Then, in step S2, the single-wheel load F of the tire is obtained. i At that time, the deformation correction coefficient K was obtained through a lightweight CNN-LSTM fusion model. defDuring the process, tire temperature T is added as a parameter to obtain a more accurate K. def ;

[0073] And the formula for calculating Fi is adjusted as follows:

[0074] F i =K base *(P corr -P0) + K def *δ.

[0075] Preferably, in this embodiment, before performing step S1, a tire positioning baseline is marked along the ground in the detection area, and calibration is performed using a standard weight cart (error ±0.1%, load 10t, 20t, 30t): the weight cart passes through the detection area at a speed of 3km / h, and tire pressure (P) is collected under different loads. cal Vertical deflection (δ) cal The data was used to fit and obtain the tire stiffness reference value K. base The formula is as follows: K base =F cal / (ΔP cal +δ cal )

[0076] Among them, F cal The load on a single wheel of a standard weight cart is measured in N.

[0077] ΔP cal =P cal -P0 (P0 is the tire pressure when unloaded, in kPa).

[0078] δ ca l represents the vertical deflection of the wheel under calibrated conditions (in mm).

[0079] Preset environmental correction parameters: Temperature correction coefficient α = 0.3 kPa / ℃ (compensates for 0.3 kPa tire pressure per ℃), Road vibration correction threshold f th =3Hz (Data smoothing is initiated when the vibration frequency exceeds 3Hz);

[0080] K base P0, α, f th The parameters are stored in the edge computing terminal database.

[0081] Preferably, after the vehicle enters the inspection area, the camera photographs the sidewall of the tire, and extracts the tire model (e.g., "11R22.5") and rated tire pressure (P) using OCR character recognition technology. rated Automatically matches the tire's basic parameters (section width b, aspect ratio ε) in the database.

[0082] Preferably, when the linkage control module is working, the wheel speed sensor collects the vehicle speed v (e.g., 3km / h) in real time, and drives the motor to match the sliding speed of the non-contact detection unit with v, ensuring that the millimeter-wave radar and vision camera are always aligned with the center area of ​​the tire sidewall (deviation ≤5mm).

[0083] Preferably, in this embodiment, the actual tire pressure P of the tire is collected in step S1. real During the vehicle's movement within the detection area, the radar tire pressure sensor repeatedly collected the actual tire pressure P. real And obtain multiple sets of P at different time points. real Values, and take multiple sets of P. real The average value is used to obtain the average tire pressure P. avg After obtaining the average tire pressure P avg Then, in calculating the single wheel load F i Then, the calculation formula is modified as follows:

[0084] F i = K base *(P avg -P0) + K def *δ,

[0085] Calculate the corrected tire pressure P corr When, modify the calculation formula to

[0086] P corr = P avg -α* (T - 25)

[0087] Multiple sets of actual tire pressure P were collected. real At that time, the millimeter-wave radar collected the real-time tire pressure P at a frequency of 10Hz. real The temperature T is continuously sampled for 0.5 seconds, and samples exceeding the normal range (P) are discarded. rated After identifying outliers (±200 kPa), the average value P is then calculated. avg ;

[0088] Preferably, the vertical deflection sensing component specifically includes the following steps: the camera synchronously captures the tire profile, noise is removed by Gaussian filtering, the tire region is segmented by a region growing algorithm, and the tire vertical deflection δ—that is, the tire radius R under load—is calculated. load The difference between the tire radius R0 and the unloaded tire radius is calculated using the following formula:

[0089] δ = R0 - R load

[0090] R0 is obtained from the tire model parameters (e.g., R_0=584mm for 11R22.5 tire), R load Calculated by contour fitting (accuracy ±0.1mm);

[0091] The contactless detection unit completes the P of all tires avg The system collects T and δ data, and automatically resets to the initial position after the data collection is completed.

[0092] Preferably, the corrected tire pressure P is obtained after passing through the temperature correction unit. corr Then, further correction is performed by the vibration correction unit. The vibration correction specifically includes the following steps: the vibration sensor detects the road surface vibration under the corresponding tire within the detection area; if the vibration sensor detects a road surface vibration frequency f > f th =3Hz, using the P of the adjacent tires of this tire corr The mean δ is interpolated and corrected using the following formula (taking the left front tire abnormality as an example):

[0093] P corr左前轮修正 =(P corr右前轮 +P corr左中轮 ) / 2

[0094] δ 修正 =(δ 右前轮 +δ 左中轮 ) / 2

[0095] Single wheel load fitting: The corrected P corr δ is input to a lightweight CNN-LSTM fusion model, and K is output as a deformation correction coefficient. def (Model input features: P) corr -P0, δ, T, output K def (Inference time ≤ 0.2 seconds); combined with the improved Hertz contact formula, the single-wheel load F is calculated. i :

[0096] F i = K base *(P corr -P0) + K def *δ

[0097] Among them, F i Let K be the real-time load (N) of the i-th tire. base K is the reference value for tire stiffness. def Deformation correction coefficients output for the edge model;

[0098] Vehicle total weight calculation: Sum the loads on all tires and divide by the gravitational acceleration g = 9.81 m / s² to obtain the vehicle total weight M.

[0099]

[0100] Where n is the total number of tires (e.g., n=6 for a 6-wheeled truck), and M is in kg; the entire calculation process takes ≤0.5 seconds.

[0101] The edge computing layer will store the vehicle's total weight M and the parameters of each tire (P). corr The calculation process data is transmitted to the application output layer. Specifically, in this embodiment, the above calculation results are simultaneously displayed on the on-site touch screen and transmitted to the management platform via 4G / 5G.

[0102] In this embodiment, if M exceeds the vehicle's maximum load (retrieved from the vehicle model database) by 10%, the application will trigger an audible and visual alarm (red light + buzzer) and simultaneously send a text message to the administrator (content: "License plate XX, current weight XXkg, overloaded by XX%, please check").

[0103] All data (including raw sensor data, calculation results, and early warning records) are generated into encrypted hash values ​​and stored in the local storage module to support subsequent traceability queries.

[0104] like Figure 4 As shown, in this embodiment, the edge computing layer incorporates the following algorithm model:

[0105] 1. Lightweight CNN-LSTM fusion model (for output K) def )

[0106] Model architecture: Input layer (3-dimensional features: P) corr -P0, δ, T) → CNN branch (2 convolutional layers + 1 pooling layer, extracting spatial features) → LSTM branch (1 LSTM layer, extracting temporal features) → Fusion layer (feature concatenation) → Output layer (1-dimensional K) def );

[0107] Model optimization: The parameters are compressed to less than 30MB using model quantization technology to adapt to the computing power of edge terminals; the model is trained using 1000 sets of on-site measured data (covering different vehicle types and loads), and the model error rate is ≤±1%.

[0108] Inference efficiency: single-round inference time ≤ 0.2 seconds, supports parallel computing for 6 tires.

[0109] 2. Improved Hertz contact formula (for calculation)

[0110] The traditional Hertz formula only considers the linear relationship between tire pressure and load. This invention introduces a tire deformation correction term K. def ×δ, solves the error problem caused by tire deformation under heavy loads—as the load increases, the tire's vertical deflection δ increases, K def Dynamic adjustment (K under heavy load) def Increase the weight of deformation contribution to the load to make the formula adaptable to a wide load range of 10t-50t.

[0111] 3. Data smoothing algorithm (for vibration correction)

[0112] A sliding window filtering algorithm is used to filter P under vibration disturbance. real The δ data is smoothed with a window size of 5 (the average of 5 consecutive sampling points), as shown in the following formula:

[0113]

[0114] Among them, X j For the original sampled data (P) real Or δ), X smooth The smoothed data can reduce the error caused by vibration by 40%.

[0115] The weighing process of the present invention will be explained more intuitively below with reference to a specific embodiment:

[0116] • Equipment Deployment: In this embodiment, the equipment is deployed at the freight station. A pair of parallel sliding rails (6m long) are laid on both sides of the freight station entrance passage (4m wide). Each sliding rail is equipped with 3 sliding detection units (millimeter-wave radar TPMS-Radar-02, 2K vision camera); the edge computing terminal (4TOPS computing power) is installed in a protective box next to the passage.

[0117] • Parameter initialization:

[0118] Calibration was performed using 10t, 20t, and 30t standard weight carts to obtain K. base =120N / (kPa•mm), a weight cart (20t, 6 wheels, tire model 11R22.5) passes through the detection area at a speed of 3km / h, collecting P. cal =950kPa, T=25℃, δ cal =5mm; Calculate K base =(20000×9.81 / 6) / (950-800+5)=120N / (kPa•mm), stored in the edge terminal;

[0119] ◦ Preset temperature correction factor α = 0.3 kPa / ℃, vibration threshold f th =3Hz;

[0120] ◦ Enter common model parameters into the tire database: 11R22.5 tire P0=800kPa, R0=584mm; 12R24.5 tire P0=820kPa, R0=615mm;

[0121] • Management platform configuration: Set the overload threshold (110% of the vehicle's rated load capacity) and bind the administrator's mobile phone number (for receiving warning SMS messages). Specific Implementation

[0122] Vehicle identification and positioning

[0123] At 09:00 on July 10, 2024, a 6-wheeled truck with license plate number “A12345” (tire model 11R22.5) entered the detection area, triggering the process of the infrared beam sensor; the camera identified the tire model and retrieved P0=800kPa and R0=584mm; the wheel speed sensor detected the vehicle speed v=3km / h, and the sliding rail drove the detection unit to slide synchronously.

[0124] Millimeter-wave radar acquires P data from the left front tire. real (A total of 5 data points were collected during the process, with values ​​of 940, 942, 938, 941, and 939 kPa respectively), and the mean P was taken. avg =940kPa, T=30℃; the vertical deflection sensing component calculated δ=4.8mm (R load =R0-δ=579.2mm); at the same time, five other tires were sampled to obtain all Pavg, T, and δ.

[0125] Data correction

[0126] Calculate P for the left front tire corr =940 - 0.3 × (30 - 25) = 938.5 kPa; the vibration sensor detects f = 2.5 Hz < 3 Hz, so no vibration correction is needed; other tires are corrected using the same logic.

[0127] Edge computing and weight calculation

[0128] Edge terminal input: left front tire P corr -P0=138.5kPa, δ=4.8mm, T=30℃ to the CNN-LSTM model, output K def =80N / mm; Calculate the single wheel load F i =120×138.5+80×4.8=16620+384=17004N; Total load ΣF of 6 tires i =17004×6=102024N; Total weight M=102024 / 9.81≈10400kg (10.4t).

[0129] Results Output and Early Warning

[0130] The touchscreen displays "License plate A12345, total weight 10.4t, rated load 10t, overload 4%"; triggering a yellow audible and visual alarm (not exceeding the 10% threshold), and simultaneously uploading the data to the management platform and storing it in the local storage module (hash value: 0x7a2b...c3d4).

[0131] Implementation effect verification:

[0132] • Accuracy verification: Calibrated using a traditional weighbridge (error ±0.3%), the error rate between the 10.4t calculated by this method and the weighbridge value of 10.3t is 0.96%, which meets the ±3% accuracy requirement;

[0133] • Efficiency verification: An average of 220 trucks are inspected per day, with an average time of 8 seconds per inspection, and no lane congestion. This represents a 633% improvement in efficiency compared to weighbridges (30 trucks per day).

[0134] • Environmental adaptability verification: In rainy weather (rainfall of 20mm / h) and high temperature (40℃) environments, the accuracy attenuation was 0.3% and 0.2% respectively, with no failures;

[0135] • Cost verification: The equipment purchase cost is 80,000 yuan (traditional weighbridge costs 150,000 yuan), the average annual maintenance cost is 2,000 yuan (weighbridge costs 20,000 yuan), the labor cost is reduced by 60,000 yuan / year, and the overall annual cost is reduced by 98,000 yuan.

[0136] This application has the following beneficial effects:

[0137] 1. Significantly improved vehicle compatibility: By setting up a pair of sliding rails to form a sliding rail weighing system, and through the linkage control module to achieve dynamic tire tracking, it is compatible with various vehicle types such as trucks and buses with wheelbases of 2.5m-4.5m. The weighing coverage rate is increased from 60% of the traditional weighbridge to 100%, without the need to replace the weighing platform or adjust the device.

[0138] 2. Non-contact measurement and low maintenance cost: All sensors do not come into contact with the vehicle, eliminating wear and water ingress risks and extending the maintenance cycle to 6 months (reducing maintenance frequency by 83% compared to weighbridges); the cost of a single maintenance is only 200 yuan (90% lower than the 2,000 yuan cost of weighbridges), resulting in an average annual maintenance cost saving of 18,000 yuan.

[0139] 3. Superior weighing efficiency and accuracy: The weighing time for a single weighing is ≤10 seconds, which is 30 times more efficient than the traditional weighbridge (5 minutes / vehicle). The average number of vehicles inspected per day is ≥200, avoiding channel congestion. Combining tire pressure and deformation dual parameters with edge fitting, the error rate is stable within ±3%, which is better than the weighing solutions based on pure vision (±5%) and pure tire pressure (±8%).

[0140] 4. Strong anti-interference capability and good environmental adaptability: Both millimeter-wave radar and vision camera are IP68 protected and are not affected by dust or rain; temperature correction and vibration smoothing algorithms eliminate environmental interference, and the accuracy attenuation is ≤0.5% in harsh environments such as -30℃~70℃ and PM2.5=500μg / m³.

[0141] 5. High real-time performance and support for unattended operation: The edge computing terminal processes data locally with a latency of ≤100ms, without relying on the cloud; automatic audible and visual alarm for overload and SMS push notification enable unattended weighing at freight stations and construction sites, reducing labor costs (an average annual saving of 60,000 yuan per site).

[0142] 6. Data traceability and high security: All raw data and calculation results are stored in encrypted form and generate a unique hash value. They can be traced and queried through the management platform to prevent data tampering and meet the needs of measurement and auditing.

[0143] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A contactless vehicle weighing system based on edge computing, characterized in that, include: The detection layer includes: A pair of slide rails are arranged in parallel and spaced apart, so that a detection area for vehicles to pass through is formed between the pair of slide rails; Two detection groups are respectively set on a pair of slide rails. Each detection group includes at least two non-contact detection units. Each non-contact detection unit includes a radar tire pressure sensor and a vertical deflection detection component, as well as a sliding trolley slidably mounted on the corresponding slide rail. The radar tire pressure sensor and the vertical deflection detection component are both mounted on the sliding trolley, so that when the vehicle passes through the detection area, the sliding trolley slides along with one of the vehicle's tires and detects the tire pressure and vertical deflection of the tire respectively. An edge computing layer, connected to each of the detection units, is used to acquire the tire pressure and vertical deflection of all tires, and calculate the total weight M of the vehicle according to the following formula: F i = K base *(P real -P0) + K def *δ (1) （2） Among them, F i For a single wheel load, P real The measured actual tire pressure of the following tire, P0 is the tire pressure when unloaded, which is a preset value and depends on the tire model, K base K is the reference value for tire stiffness. def The deformation correction coefficient is obtained through a lightweight CNN-LSTM fusion model, where δ is the vertical deflection and g is the gravitational acceleration.

2. The contactless vehicle weighing system based on edge computing as described in claim 1, characterized in that, The vertical deflection detection component includes: a camera mounted on the sliding trolley; and a vertical deflection calculation unit connected to the camera. The vertical deflection calculation unit obtains the current profile of the tire based on the tire photograph captured by the camera, and compares the current profile of the tire with the initial profile of the tire to obtain the vertical deflection of the tire.

3. The contactless vehicle weighing system based on edge computing as described in claim 1, characterized in that, The contactless detection unit also includes a tire temperature sensor mounted on the sliding trolley and used to detect the tire temperature of the following tires when the vehicle passes through the detection area; the edge computing layer also includes a temperature correction unit that corrects the calculation results of the edge computing layer based on the tire temperature detected by the temperature sensor.

4. The contactless vehicle weighing system based on edge computing as described in claim 1, characterized in that, It also includes a linkage control module, which includes: a wheel speed sensor mounted on the sliding trolley for detecting the movement speed of the following tire; a drive motor mounted on the sliding trolley for driving the sliding trolley and adjusting the movement speed of the sliding trolley; and a control unit connected to the wheel speed sensor and the drive motor for controlling the drive motor according to the movement speed of the tire to adjust the movement speed of the sliding trolley, thereby ensuring that the sliding trolley can closely follow the corresponding tire.

5. The contactless vehicle weighing system based on edge computing as described in claim 1, characterized in that, It also includes an application output layer for storing and outputting the vehicle weight calculated by the edge computing layer.

6. The contactless vehicle weighing system based on edge computing as described in claim 5, characterized in that, The application output layer includes an alarm module for determining whether the total weight of vehicles passing through the detection unit exceeds a maximum threshold and for issuing an alarm when the maximum threshold is exceeded.

7. The contactless vehicle weighing system based on edge computing as described in claim 6, characterized in that, The alarm module includes an identification unit and an alarm unit connected to the identification unit; the identification unit has pre-stored maximum load threshold information corresponding to different vehicle models, which is used to identify the model of the vehicle passing through the detection area and output the corresponding maximum load according to the identified model; the alarm unit is used to compare the maximum load with the vehicle weight, and issue an alarm when the vehicle weight is greater than the maximum load.

8. The contactless vehicle weighing system based on edge computing as described in claim 1, characterized in that, The contactless detection unit also includes a vibration sensor mounted on the sliding trolley for detecting ground vibrations; the edge computing layer also includes a vibration correction unit for correcting the calculation results of the edge computing layer based on the ground vibrations detected by the vibration sensor.

9. A weighing method for a contactless vehicle weighing system based on edge computing as described in claim 1, characterized in that, Includes the following steps: S1. After the vehicle to be weighed enters the detection area and travels within the detection area, select the same number of non-contact detection units as the vehicle tires, and slide the corresponding sliding trolleys to follow each tire. Use these non-contact detection units to detect the tire pressure P and vertical deflection δ of the corresponding tires respectively. S2, the edge computing layer receives the tire pressure P of each tire. real Combined with the vertical deflection δ, the total vehicle weight M is obtained using the following formula: F i = K base *(P real -P0) + K def *δ, Among them, F i K represents the single-wheel load of the tire. base K is the reference value for tire stiffness. def The deformation correction coefficients are obtained through the lightweight CNN-LSTM fusion model, P0 is the tire pressure under no-load conditions, which is obtained according to the tire model, and g is the acceleration due to gravity. Where n is the total number of tires (e.g., n=6 for a 6-wheeled truck), M is in kg, and g is the acceleration due to gravity.

10. The weighing method of the edge computing-based contactless vehicle weighing system as described in claim 8, characterized in that, The contactless detection unit also includes a tire temperature sensor mounted on the sliding trolley. During step S1, the tire temperature sensor detects the temperature of the following tire and feeds the result back to the temperature correction unit of the edge computing layer. Before step S2, the temperature correction unit corrects the tire pressure detected by the tire pressure sensor based on the tire temperature, using the following correction formula: P corr = P real -α* (T - 25) Where 25℃ is the reference temperature, and α=0.3kPa / ℃ is the temperature correction factor; After obtaining the corrected tire pressure P corr Then, in step S2, the single-wheel load F of the tire is obtained. i At that time, the deformation correction coefficient K was obtained through a lightweight CNN-LSTM fusion model. def During the process, tire temperature T is added as a parameter to obtain a more accurate K. def ; And the formula for calculating Fi is adjusted as follows: F i =K base *(P corr -P0) + K def *δ.

Images