CN112539816B - Dynamic weighing correction method based on deep neural network in digital twin environment - Google Patents
Dynamic weighing correction method based on deep neural network in digital twin environment Download PDFInfo
- Publication number
- CN112539816B CN112539816B CN202011397361.6A CN202011397361A CN112539816B CN 112539816 B CN112539816 B CN 112539816B CN 202011397361 A CN202011397361 A CN 202011397361A CN 112539816 B CN112539816 B CN 112539816B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- weighing
- dynamic
- data
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01G—WEIGHING
- G01G19/00—Weighing apparatus or methods adapted for special purposes not provided for in the preceding groups
- G01G19/02—Weighing apparatus or methods adapted for special purposes not provided for in the preceding groups for weighing wheeled or rolling bodies, e.g. vehicles
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a dynamic weighing correction method based on a deep neural network in a digital twin environment, which is characterized by comprising the following steps of: when the vehicle passes through the weighing area, the quartz weighing sensor converts the pressure signal into an electric signal to be output, the wheel axle identifier identifies the number of wheel axles of the vehicle, and the camera device records the running process of the vehicle; the road-vehicle-weighing digital twin system with the multi-scale numerical simulation technology is used for carrying out refined simulation and data twin, various parameters which can be measured by a quartz weighing sensor when a vehicle passes through a weighing area are simulated, a part of data is twinned, and the sample capacity is enriched; and then inputting the collected data and twin data into a trained dynamic weighing correction system based on a deep neural network for correction, finally outputting the dynamic vehicle weight, and then comparing the dynamic vehicle weight with the latest 'over-limit and overload determination standard of road freight vehicles' to judge whether the vehicle is overweight.
Description
Technical Field
The invention belongs to the technical field of intelligent traffic application, and particularly relates to a dynamic weighing correction method for a road freight vehicle in a digital twin environment and an identification scheme for a driver weight-loss fee-evading behavior.
Background
At present, the mature weighing technologies in the market are mainly divided into two types, one type is static weighing, and the other type is dynamic weighing. The static weighing generally adopts a method of weighing by using a weighing platform, and the vehicle can be weighed by decelerating and stopping, so that the method has the advantages of low speed and efficiency, congestion and traffic accidents are easily caused, and the traffic and transportation efficiency is influenced. However, the dynamic weighing has the advantages that the vehicle does not need to be stopped in the whole weighing process, namely, the whole vehicle weighing is completed in the normal running process, and the weighing efficiency is greatly improved. The most main problems of domestic dynamic weighing systems, particularly medium-high speed dynamic weighing systems, are that the precision is not high, but the imported system is expensive in cost, is not unified with domestic judgment standards, and cannot be widely applied.
The dynamic weighing is that a weighing sensor is laid on a road surface, the dynamic pressure of a vehicle is measured when the vehicle passes through the weighing sensor, then the weight of a wheel on one side is calculated, the axle weight of the vehicle is further solved, and the weight of the whole vehicle is the sum of the axle weights of all axles. The biggest problem of the whole dynamic weighing system comes from three aspects: firstly, weighing sensor's problem because in the actual engineering application, weighing sensor can face various harsh environment, so require weighing sensor to the strong adaptability of environment, just so can guarantee sensitivity and the high accuracy of weighing, consequently the more weighing sensor who commonly uses in the existing market generally is piezoelectric quartz weighing sensor. The second problem is data acquisition, and in real life, many truck drivers adopt various skills to pass through weighing channels for avoiding expenses, such as winding and walking an S-shaped route, which has a serious influence on accurate collection of dynamic data. In addition, the time consumption for collecting sample data is long, and the low efficiency is also a big factor for limiting the development of dynamic weighing. Thirdly, dynamic data processing problem, because dynamic weighing is superior to ordinary weighing mode, so this technique has received much attention, but the precision of weighing is the biggest factor that restricts its development, because the interference that factors such as environment, vehicle and driver action were brought is contained in the data of weighing. Although many technologies are used for analyzing dynamic weighing data at present, the data level is simply processed, and hidden characteristics of the data are not deeply mined, so that the problem of low weighing precision is caused. How to correct errors of dynamic weighing by deeply mining hidden features in dynamic weighing data, how to acquire a large amount of sample data in a relatively short time and how to identify weight reduction and fee evasion behaviors of a truck driver are important problems which need to be solved when the method is widely applied.
Disclosure of Invention
The invention aims to provide a dynamic weighing and correcting method based on a deep neural network in a digital twin environment, which is characterized in that a high-speed dynamic weighing and correcting system is established by constructing a digital-physical twin system of a dynamic weighing system, and the problem that the traditional dynamic weighing system is low in measurement precision is optimized. And the real-time property, the accuracy and the data processing capability of the system measurement data are improved by developing an intelligent algorithm with self-adaptability or learning capability.
On one hand, a quartz weighing sensor is laid in a weighing area to carry out real-time measurement, and data are acquired; erecting a camera in the weighing area, and collecting image information; on the other hand, a road-vehicle-weighing digital twin system with a multi-scale numerical simulation technology is used for carrying out fine simulation and data twin, various parameters which can be measured by a sensor when a vehicle passes through a weighing area are simulated, and data are twinned, so that a sample database is expanded. And then inputting the acquired data and the twin data into a high-speed dynamic weighing correction system based on a deep neural network for correction, and finally outputting a result.
The technical scheme of the invention is as follows: a dynamic weighing correction method based on a deep neural network in a digital twin environment is characterized by comprising the following steps: when a vehicle passes through a preset weighing area, the quartz weighing sensor senses that the vehicle passes through, then pressure signals applied to the quartz weighing sensor by the vehicle are converted into corresponding electric signals, a wheel axle identifier in the weighing area identifies the number of wheel axles of the detected vehicle, and a camera device shoots pictures of the passing vehicle and records the running process of the vehicle; the road-vehicle-weighing digital twin system with the multi-scale numerical simulation technology is used for carrying out fine simulation and data twin, various parameters which can be measured by a sensor when a vehicle passes through a weighing area are simulated, partial data are twin, and the sample capacity is enriched; and then inputting the collected data and twin data into a trained dynamic weighing correction system based on a deep neural network for correction, finally outputting the dynamic vehicle weight, and then comparing the dynamic vehicle weight with the over-limit and overload determination standard of the freight vehicle specified by the relevant department to judge whether the vehicle is overweight.
Further, the specific method is as follows:
(1) and (4) building a detection device of the dynamic weighing correction system so as to construct a digital twin system.
Firstly, reasonably planning the distribution positions of quartz weighing sensors: the quartz weighing sensors are distributed on two sides of the weighing area in a staggered manner, the wheel axle identifier is positioned in the middle of the second sensor and the fourth sensor, and ground induction coils are respectively arranged in front of and behind the weighing area and used for inducing vehicle signals;
when the vehicle passes through the weighing area, the quartz weighing sensor acquires data; in order to identify the abnormal driving behavior of the driver, a camera device is arranged in the weighing area and is used for shooting the route of the driver passing through the weighing area as an image signal; the quartz weighing sensors are reasonably arranged, so that weighing data can be accurately collected, and a digital twin system can be simulated conveniently; the influence of the roadbed and the road surface on the sensors can be reduced by the staggered distribution of the sensors, and the simulation data is more accurate; the ground induction coil can avoid the interference of other vehicles, and is favorable for establishing a man-vehicle-road closed-loop driving model.
(2) Building road-vehicle-weighing digital twin system
The specific method comprises the following steps:
1) establishing a vehicle subsystem: firstly, building a tire model according to geometric description, vehicle load application and simulation of contact with the ground, and performing dynamic simulation; then, a man-vehicle-road closed-loop driving model suitable for normal driving and abnormal driving behavior research is established according to steering wheel corners, horizontal/longitudinal acceleration and horizontal/longitudinal displacement, multi-body dynamics analysis is carried out on the closed-loop driving model under the normal driving and abnormal driving conditions, and a data set A is finally obtained;
2) establishing a road subsystem: analyzing the dynamic response characteristics of the roadbed and the pavement according to the three-dimensional representation of the ground, then analyzing the dynamic constitutive of the roadbed and the pavement through experiments, and calculating by combining the vehicle load, the driving behavior and the fatigue resistance of the pavement to finally obtain a data set B;
3) establishing a measurement subsystem: performing dynamic simulation on the sensor and the roadbed pavement according to the temperature, the humidity and the strain rate of the ground, and then performing dynamic simulation on the sensor through analysis on the prestress applied by the sensor and the performance of the piezoelectric material to obtain a data set C;
4) establishing a road-vehicle-balance digital twin system through refined simulation of a vehicle subsystem, a road subsystem and a measurement subsystem, completing refined simulation according to an established tire model, a man-vehicle-road closed-loop driving model, a pressure signal acquired by a sensor and a data set A, B, C, twining partial data, and enriching sample capacity;
(3) and (4) building a dynamic weighing correction system based on a deep neural network, and calculating the dynamic vehicle weight of the vehicle.
The specific method comprises the following steps:
1) screening the collected data: when the vehicle passes through the whole dynamic weighing and correcting system, the pressure signal of the single-side tire is a curve chart with a waveform rule, when the pressure signal reaches the maximum, a relatively stable period exists, and the data of the stable period is taken as an effective data segment;
2) extracting network input characteristic values, and processing by using a normalization method to eliminate the influence caused by different input units of singular values and the characteristic values;
for each vehicle passing through the dynamic weighing correction system, taking the pressure of each tire passing through each quartz weighing sensor as an input characteristic value of a network, taking the speed of each axle of the vehicle passing through a weighing area as an input characteristic value of the network, and taking an image signal shot by a camera device as one characteristic value of the network;
the pressure characteristic value of each tire adopts the average value of the effective data section in the stable period, and the speed of each shaft passing through the weighing area is calculated by using the time of the middle position in the stable period;
3) and training a deep neural network correction system by taking the collected tire pressure characteristic value, the collected axle speed characteristic value, the collected image signal and the collected twin data as input and taking the dynamic vehicle weight as output. The method comprises the steps of recognizing whether a vehicle is normally driven or abnormally driven when the vehicle passes through a weighing area according to image signals, matching a normal correction model for the normal driving and matching an abnormal driving model for the abnormal driving, then respectively training, and putting the vehicle into use after the training is finished;
4) calculating dynamic vehicle weight of vehicle based on trained deep neural network correction system
When the vehicle passes through the weighing area, the sensor inputs the acquired data into the trained neural network, the neural network processes the data and outputs the processed data to the weighing instrument, a microprocessor is arranged in the weighing instrument and can calculate the axle load of the vehicle to be measured according to a corresponding mathematical model and dynamic weighing data processing software, the dynamic vehicle weight is finally output by accumulating the axle loads, the output dynamic vehicle weight is compared with the allowable upper load limit in the 'road freight vehicle overrun overload determination standard', and whether the vehicle to be measured has overrun problem or overload condition is judged.
Further, the neural network is trained according to the following method: the image signal is input as a characteristic value for identifying a running state of the vehicle passing through the weighing area, and the normal driving matches the normal correction model and the abnormal driving matches the abnormal correction model.
Processing the extracted network input characteristic value and twin data by a normalization method, and eliminating the influence caused by different input units of singular values and characteristic values; respectively inputting the processed characteristic values acquired under the normal driving state and the characteristic values acquired under the abnormal driving state into a normal correction model and an abnormal correction model, then adjusting the weight attenuation coefficient of the neural network correction model, comparing the output result of the correction model with the static vehicle weight, continuing to train the correction model if the error rate does not reach the preset accuracy rate, and ending the neural network training and putting into use until the error rates of the dynamic vehicle weight and the static vehicle weight output by the models reach the preset standard.
Further, the speed of each shaft as it passes through the weighing area is calculated as follows:
when the vehicle passes through the weighing area, when a first tire a passes through a first quartz weighing sensor, taking the average value of the stable stages when the first tire a is completely weighed as a characteristic value a1, when the tire a passes through a third quartz weighing sensor, taking the average value of the stable stages when the tire is completely weighed as a characteristic value a3, and so on, obtaining b2, b4, c1, c3, d2 and d4 in sequence;
the axle speed is calculated as follows: assuming that the time t is the middle of the plateau when the first tire a passes the first quartz load cell1The time of the middle position of the stationary phase when the first tire a passes through the third quartz weighing sensor is t3And S represents a weighing areaLength of the first tire a, velocity v of the first tire aaThe calculation method of (2) is as follows:
the velocity v of the second tire b can be obtained in the same wayb，
The speed v of the first wheel axle1Can be calculated from the mean of the speeds of the tires a, b, i.e.
The speed v of the third tire c and the fourth tire d can be calculated in the same wayc、vdAnd the speed v of the second wheel axle2And so on.
The working principle of the invention is as follows: the quartz weighing sensor senses that the vehicle passes by, then pressure signals applied to the quartz weighing sensor by the vehicle are converted into corresponding electric signals to be output, the wheel axle identifier identifies the number of wheel axles of the vehicle, and the camera device records the running process of the vehicle. Various parameters which can be measured by a sensor when a vehicle passes through a weighing area are simulated by a multi-scale fine simulation technology, partial data are generated, and the sample capacity is enriched. And then, carrying out characteristic value extraction, normalization processing and correction of a weighing correction system on the collected data and the twin partial data, and finally inputting the data into a weighing instrument to obtain the dynamic vehicle weight of the vehicle.
Compared with the existing dynamic weighing method on the market, the advantages of the invention can be summarized as the following points:
1. the neural network has a multilayer structure, so that the neural network can find the nonlinear relation among a large amount of data, and the dynamic weighing correction system of the deep neural network can accurately identify and extract required data in complex data information and measure the dynamic vehicle weight of the vehicle through the processing of the data acquisition and processing unit. The method can improve the measurement precision, reduce the measurement time and has the advantages of convenience and accuracy.
2. In order to identify the behavior of weight loss and fee evasion of a truck driver, the invention also inputs the image signal of the running state of the vehicle into the network, namely the weighing process characteristic of a single vehicle is input into the network, so that the network can deeply analyze the influence of the running state on the weighing precision when the vehicle is weighed, can effectively identify the violation behavior of weight loss and fee evasion of the truck driver, provides a basis for law enforcement of related departments, and has certain social significance.
3. The multi-scale numerical simulation technology can perform dynamic simulation on tires, pavements, sensors and vehicles, and then perform data twinning by a digital twinning system, so that the sample capacity of a sample database is enriched, and the influences of long time consumption of data acquisition and insufficient sample data on weighing precision can be effectively compensated.
Drawings
FIG. 1 is a hardware installation profile of a vehicle dynamic weighing system.
FIG. 2 is a schematic diagram of dynamic data acquisition for a two-axle truck.
Fig. 3 is a model structure diagram of the deep neural network dynamic weighing correction system.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be described in detail below. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the examples of the present invention without any inventive step, are within the scope of the present invention.
The invention relates to a dynamic weighing correction method based on a deep neural network in a digital twin environment, which comprises the steps of processing collected data and twin data, then correcting by a deep neural network dynamic weighing correction system, finally outputting dynamic vehicle weight, and then comparing with vehicle overrun overload determination standards specified by relevant departments to judge whether the vehicle is overweight.
As shown in fig. 1, the hardware device of the dynamic weighing correction system provided by the present invention mainly includes: quartz weighing sensor, wheel axle recognizer, ground induction coil, charge amplifier, data acquisition and processing unit, weighing instrument, etc. Wherein, other hardware equipment such as charge amplifier, data acquisition processing unit, weighing instrument all install in outdoor rack.
The specific implementation mode of the invention is as follows:
(1) firstly, reasonably arranging a quartz sensor and a camera device, and building a detection device of a dynamic weighing correction system so as to build a road-vehicle-balance digital twin system;
because the traditional quartz weighing sensor has the defects of uneven pressure sensitivity, sensitivity to environmental change, rough processing, low assembly precision and the like, the high-speed dynamic weighing system optimizes the used quartz sensor, firstly adopts the installation position of the piezoelectric quartz plate to avoid uneven pressure sensitivity, secondly selects the aluminum type material with lower influence on environmental change, and finally adopts the finish machining pressure plane, the quartz group assembly plane and the high-precision assembly to improve the defects of the traditional quartz weighing sensor.
The sensor arrangement is shown in fig. 1, and assuming that the weighing area S is 5 meters, the four quartz weighing sensors used are distributed in a staggered manner in the weighing area. The wheel axle identifier is positioned in the middle of the second sensor and the fourth sensor. Place ground induction coil respectively around presetting weighing area for the response vehicle signal, when the vehicle approached the weighing area, leading ground induction coil sensed vehicle signal, opened weighing signal transmission channel, carried out data acquisition. When the vehicle drives away from the weighing area, the rear ground induction coil closes the weighing signal transmission channel, one-time information acquisition is completed, and then the ground induction coil processes and transmits acquired data. The arrangement mode of the double ground induction coils can ensure that the transmission of a single signal belongs to the same vehicle, the identification precision of the wheel axle identifier can be effectively improved, the number of axles of the vehicle can be identified more accurately, and the influence on the measurement precision caused by the fact that the distance between two trucks is too close to the data confusion is avoided.
In order to recognize the abnormal behavior of the driver, a camera device is arranged in the weighing area and is used for shooting the route of the driver passing through the weighing area as an image signal.
(2) The establishment of a complete road-vehicle-balance digital twin system is an important step for establishing a dynamic weighing correction system, and is related to whether the system can accurately measure the weight of a dynamic vehicle.
The specific method comprises the following steps:
1) vehicle subsystems are established. Firstly, building a tire model according to geometric description, vehicle load application, simulation of contact with the ground and the like, and performing dynamic simulation; and then establishing a man-vehicle-road closed-loop driving model suitable for normal driving and abnormal driving behavior research according to the steering wheel corner, the horizontal/longitudinal acceleration and the horizontal/longitudinal displacement, and performing multi-body dynamics analysis on the closed-loop driving model under the normal driving and abnormal driving conditions to finally obtain a data set A.
2) And establishing a road subsystem. And (3) carrying out processing such as CT scanning, geometric description, model reconstruction and the like on the sample ground, analyzing the three-dimensional representation of the ground, further analyzing the dynamic response characteristics such as the loading rate, the temperature, the humidity and the like of the roadbed-road surface, analyzing the dynamic structure of the roadbed-road surface through a microscopic experiment, carrying out high-performance calculation by combining the vehicle load, the driving behavior and the fatigue resistance of the road surface, and finally obtaining a data set B.
3) A measurement subsystem is established. And performing dynamic simulation on the sensor-pavement interface according to the temperature, the humidity and the strain rate of the ground, and performing dynamic simulation on the sensor through analysis on prestress applied by the sensor, the performance of a piezoelectric material and a piezoelectric mode, so as to realize fine simulation analysis on a series of parameters related to the sensor, the ground and the like, and obtain a data set C.
4) A road-vehicle-balance digital twin system is established through fine simulation of a vehicle subsystem, a road subsystem and a measurement subsystem, fine simulation is completed according to the established model, pressure signals collected by sensors and a data set A, B, C, and partial data are twinned.
(3) And (4) building a dynamic weighing correction system based on a deep neural network, and calculating the dynamic vehicle weight of the vehicle.
1) And screening the collected data. In the vehicle weighing raw data collected by the dynamic weighing system, a large amount of noise may exist in the weighing signal due to different actual environments and vehicle states. Although the real data may be affected by some noise and the waveforms of the sampled data may be different, the waveforms of the sampled data are regular under relatively accurate measurement conditions. As shown in fig. 2, for a two-axle truck as an example, when the vehicle runs through the weighing system, the pressure signal of the tire on one side is a waveform curve, and it can be seen in the graph that there is a relatively stable period when the pressure signal reaches the maximum, and the data of the stable period can be used as a valid data segment for weighing.
2) And extracting the network input characteristic value. To obtain the input characteristic value of the network, some relevant information in the dynamic pressure signal needs to be extracted, and for each vehicle to be measured which passes through the weighing system, the pressure of the tire passing through each sensor, the speed of the vehicle when passing through and the image of the weighing area where the vehicle passes through can be used as the input characteristic value of the network.
The pressure characteristic of each tire is taken as the mean of the valid data segments for the plateau and the time at the middle position of the plateau is used to calculate the speed at which each axle passes through the weighing area.
Considering that the maximum number of axles of the freight vehicles is 6, the length of the freight vehicles can reach about 18 meters, and the states of different vehicles passing through the weighing area are different, the speed of each axle of the vehicles passing through the weighing area is selected as the input characteristic of the network, so that the precision of dynamic weighing is improved.
When the network input characteristic value is extracted, in order to identify whether the driver has abnormal driving behaviors such as cheating, weight reduction and fee evasion, an image signal shot by the camera device is also used as one characteristic value of the system.
3) The method comprises the steps of taking collected tire pressure characteristic values, wheel axle speed characteristic values, image signals and twin data as input, taking dynamic vehicle weight as output, training a deep neural network, taking the image signals as characteristic value input, and being used for identifying the running state of a vehicle passing through a weighing area, matching a regular correction model in regular driving and matching an abnormal correction model in abnormal driving. Respectively inputting the processed characteristic values acquired under the normal driving state and the characteristic values acquired under the abnormal driving state into a normal correction model and an abnormal correction model, then adjusting the weight attenuation coefficient of the neural network correction model, comparing the output result of the correction model with the static vehicle weight, continuing to train the correction model if the error rate does not reach the preset accuracy rate, and ending the training of the neural network correction model until the error rates of the dynamic vehicle weight and the static vehicle weight output by the models reach the preset standard, and putting the models into use.
Because the neural network has a multilayer structure and is a highly complex nonlinear dynamic learning system, the neural network can process a large amount of complex data, and a nonlinear relation between the complex data is obtained by a nonlinear fitting correction method.
As shown in fig. 3, on one hand, the collected images and data are screened, and then the extracted network characteristic values are processed by using a normalization method, so as to eliminate the influence caused by the difference between the singular value and the characteristic value input unit. On the other hand, twin data is subjected to the same processing. And then, inputting the processed data obtained by the two methods into a dynamic weighing correction system of the deep neural network for carrying out error analysis and weight updating when the training network reversely propagates.
Because the specifications of freight vehicles are different, the number of characteristic values finally input into a network by each type of vehicle is different, the freight vehicles have 2-6 axles according to the currently used standard for determining the overrun and overload of the road freight vehicles, the characteristic values of 2-axle trucks are 11, the characteristic values of 6-axle trucks are 31 according to the extraction method of the characteristic values, and other types of vehicles have the same principle. In order to unify the number of the characteristic values input by the network, the number of the characteristic values input by the network is uniformly set to be 31, different characteristic values are input according to different axle numbers of the vehicle, and the rest are replaced by 0.
Taking a two-axle truck as an example, four tires of the truck are respectively numbered as a, b, c and d, and an axle is numbered as (i) and (ii). When the truck passes through the weighing area, when the tire a passes through the first quartz weighing sensor, the pressure signal is as shown in fig. 2, the average value of the stable stage when the tire a is completely weighed is taken as a characteristic value a1, when the tire a passes through the third quartz weighing sensor, the average value of the stable stage when the tire a is completely weighed is taken as a characteristic value a3, and the like, so that b2, b4, c1, c3, d2 and d4 are obtained in sequence.
The axle speed is calculated as follows: let t be the time at which the tire a passes the middle position of the stationary phase when passing the first quartz load cell1And the time of the middle position of the stationary phase when the tire a passes through the third quartz weighing sensor is t3And S represents the length of the weighing area. The velocity v of the tire aaThe calculation method of (2) is as follows:
the velocity v of the tire b can be obtained in the same wayb，
The speed v of the axle (r)1Can be calculated from the mean of the speeds of the tires a, b, i.e.
The speed v of the tires c and d can be calculated by the same methodc、vdAnd speed v of the wheel axle-2。
In order to recognize the driving behavior of the driver, a camera device is installed in the weighing area, and the driving process of the vehicle can be recorded as an image signal.
The method comprises the steps of taking 11 collected characteristic values a1, a3, b2, b4, c1, c3, d2, d4, image signals and twin data as input, taking dynamic vehicle weight as output, training a deep neural network, taking the image signals as characteristic value input, identifying the running state of a vehicle passing through a weighing area, matching normal driving with a normal correction model, and matching abnormal driving with an abnormal correction model. Respectively inputting the processed characteristic values acquired under the normal driving state and the characteristic values acquired under the abnormal driving state into a normal correction model and an abnormal correction model, then adjusting the weight attenuation coefficient of the neural network correction model, comparing the output result of the correction model with the static vehicle weight, continuing to train the correction model if the error rate does not reach the preset accuracy rate, and ending the training of the neural network correction model until the error rates of the dynamic vehicle weight and the static vehicle weight output by the models reach the preset standard.
4) Calculating dynamic vehicle weight of vehicle based on trained deep neural network correction system
The trained neural network is put into use, when the vehicle passes through a weighing area, data collected by the sensor is processed by the trained network and then output to the weighing instrument, the microprocessor is arranged in the weighing instrument and can calculate the axle load of the vehicle to be measured according to a corresponding mathematical model and dynamic weighing data processing software, and finally the dynamic vehicle weight is output by accumulating the axle loads.
And comparing the output dynamic vehicle weight with the upper limit of the load allowed in the latest standard for determining the overload and overload of the road freight vehicle, determining whether the tested vehicle has an overload problem or overload condition, and checking and enforcing the law according to the overload problem or overload condition.
The described embodiments of the invention are only some, but not all embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the examples of the present invention without any inventive step, are within the scope of the present invention.
Claims (2)
1. The dynamic weighing correction method based on the deep neural network in the digital twin environment is characterized by comprising the following steps of: when a vehicle passes through the weighing area, the quartz weighing sensor senses that the vehicle passes, then pressure signals applied to the quartz weighing sensor by the vehicle are converted into corresponding electric signals, the wheel axle identifier in the weighing area identifies the number of wheel axles of the detected vehicle, meanwhile, the camera device shoots pictures of the passing vehicle, and the running process of the vehicle is recorded; the road-vehicle-weighing digital twin system with the multi-scale numerical simulation technology is used for carrying out fine simulation and data twin, various parameters which can be measured by a quartz weighing sensor when a vehicle passes through a weighing area are simulated, a part of data is twinned, and the sample capacity is enlarged; then inputting the collected data and twin data into a trained dynamic weighing correction system based on a deep neural network for correction, finally outputting dynamic vehicle weight, and then comparing the dynamic vehicle weight with an overrun overload determination standard specified by a relevant department to judge whether the detected vehicle is overweight; the specific method comprises the following steps:
(1) building a detection device of a dynamic weighing and correcting system so as to build a road-vehicle-balance digital twin system;
firstly, reasonably planning the distribution positions of quartz weighing sensors: the quartz weighing sensors are distributed on two sides of the weighing area in a staggered manner, the wheel axle identifier is positioned in the middle of the second quartz weighing sensor and the fourth quartz weighing sensor, and ground induction coils are respectively arranged in front of and behind the weighing area so as to induce vehicle signals; when the vehicle passes through the weighing area, the quartz weighing sensor collects required data; in order to identify the abnormal behavior of the driver, a camera device is arranged in the weighing area and is used for shooting the route of the driver passing through the weighing area as an image signal;
(2) building road-vehicle-weighing digital twin system
The specific method comprises the following steps:
1) establishing a vehicle subsystem: firstly, building a tire model according to geometric description, vehicle load and simulation of contact with the ground, and performing dynamic simulation; then, a man-vehicle-road closed-loop driving model suitable for normal driving and abnormal driving behavior research is established according to the steering angle, the horizontal/longitudinal acceleration and the horizontal/longitudinal displacement of a vehicle steering wheel, multi-body dynamics analysis is carried out on the closed-loop driving model under the normal driving and abnormal driving conditions, and a data set A is finally obtained;
2) establishing a road subsystem: analyzing the dynamic response characteristics of the roadbed and the road surface according to the three-dimensional representation of the ground, then analyzing the dynamic constitutive of the roadbed and the road surface through experiments, and calculating by combining the vehicle load, the driving behavior and the fatigue resistance of the road surface to finally obtain a data set B;
3) establishing a measurement subsystem: performing dynamic simulation on a quartz weighing sensor and a roadbed pavement according to the temperature, the humidity and the strain rate of the ground, and performing dynamic simulation on prestress applied to the quartz weighing sensor and the performance of a piezoelectric material of the quartz weighing sensor to obtain a data set C;
4) establishing a road-vehicle-weighing digital twin system through refined simulation of a vehicle subsystem, a road subsystem and a measurement subsystem, completing refined simulation according to an established tire model, a man-vehicle-road closed-loop driving model and a pressure signal and a data set A, B, C acquired by a quartz weighing sensor, twining partial data to serve as partial training data samples, and enriching the sample capacity;
(3) building a dynamic weighing correction system based on a deep neural network, and calculating the dynamic vehicle weight of the vehicle;
the specific method comprises the following steps:
1) screening the collected data: when a vehicle passes through a weighing area, a pressure signal of a single-side tire is a curve graph with a waveform rule, when the pressure signal reaches the maximum, a relatively stable period exists, and data of the stable period is used as an effective data segment;
2) extracting network input characteristic values, and processing by using a normalization method to eliminate the influence caused by different input units of singular values and the characteristic values;
for each vehicle passing through the dynamic weighing correction system, the pressure of each tire passing through each quartz weighing sensor is used as an input characteristic value of a network, the speed of each axle of the vehicle passing through a weighing area is used as an input characteristic value of the network, and an image signal shot by a camera device is also used as an input characteristic value of the network; the pressure characteristic value of each tire adopts the average value of the effective data section in the stable period, and the speed of each shaft passing through the weighing area is calculated by using the time of the middle position in the stable period;
3) taking the collected tire pressure characteristic value, the collected wheel axle speed characteristic value, the collected image signal and the collected twin partial data as input, taking the dynamic vehicle weight as output, training a deep neural network, and finishing the training and putting the neural network into use after the error reaches a preset standard;
4) calculating dynamic vehicle weight of vehicle based on trained deep neural network correction system
When the vehicle passes through the weighing area, the quartz weighing sensor inputs the acquired data into the trained neural network, the neural network processes the data and outputs the processed data to the weighing instrument, a microprocessor is arranged in the weighing instrument and calculates the axle load of the vehicle to be tested according to a corresponding mathematical model and dynamic weighing data processing software, the dynamic vehicle weight is finally output by accumulating the axle loads, the output dynamic vehicle weight is compared with the allowable upper load limit in the over-limit and overload determination standard of the road freight vehicle, and whether the vehicle to be tested has the over-limit problem or the overload condition is judged.
2. The dynamic weighing correction method based on the deep neural network in the digital twin environment as claimed in claim 1, wherein the deep neural network is trained according to the following method: the image signal is used as a characteristic value input and used for identifying the running state of the vehicle passing through the weighing area, and the normal driving is matched with the normal correction model, and the abnormal driving is matched with the abnormal correction model; processing the extracted network input characteristic value and twin data by a normalization method, and eliminating the influence caused by different input units of singular values and characteristic values; respectively inputting the processed characteristic values acquired under the normal driving state and the characteristic values acquired under the abnormal driving state into a normal correction model and an abnormal correction model, then adjusting the weight attenuation coefficient of the neural network correction model, comparing the output result of the correction model with the static vehicle weight, continuing to train the correction model if the error rate does not reach the preset accuracy rate, and ending the neural network training and putting into use until the error rates of the dynamic vehicle weight and the static vehicle weight output by the models reach the preset standard.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011397361.6A CN112539816B (en) | 2020-12-03 | 2020-12-03 | Dynamic weighing correction method based on deep neural network in digital twin environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011397361.6A CN112539816B (en) | 2020-12-03 | 2020-12-03 | Dynamic weighing correction method based on deep neural network in digital twin environment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112539816A CN112539816A (en) | 2021-03-23 |
| CN112539816B true CN112539816B (en) | 2022-03-01 |
Family
ID=75015578
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011397361.6A Active CN112539816B (en) | 2020-12-03 | 2020-12-03 | Dynamic weighing correction method based on deep neural network in digital twin environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112539816B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113240821B (en) * | 2021-04-12 | 2022-08-26 | 西安科技大学 | Dynamic weighing truck non-stop charging system and method based on multi-information fusion |
| CN113507494A (en) * | 2021-05-26 | 2021-10-15 | 杭州金毅科技有限公司 | Method and system for source treatment and super-collection |
| CN113720429A (en) * | 2021-10-13 | 2021-11-30 | 武汉市路安电子科技集团有限公司 | Vehicle separation method based on axle data in dynamic truck scale |
| CN115127652A (en) * | 2022-02-17 | 2022-09-30 | 武汉理工大学 | High-precision narrow-strip vehicle dynamic weighing system based on intelligent algorithm and weighing method thereof |
| CN114543964A (en) * | 2022-02-21 | 2022-05-27 | 平安国际智慧城市科技股份有限公司 | Method and device for detecting material weighing cheating and computer equipment |
| CN115294370B (en) * | 2022-10-08 | 2023-05-09 | 南通琦欣供应链管理有限公司 | Warehouse-in and warehouse-out detection device and detection method based on twin model |
| CN115859837B (en) * | 2023-02-23 | 2023-05-16 | 山东大学 | Digital twin modeling-based fan blade dynamic impact detection method and system |
| CN116242465B (en) * | 2023-05-12 | 2023-07-18 | 深圳亿维锐创科技股份有限公司 | Dynamic vehicle weighing method and system |
| CN117542208A (en) * | 2023-11-22 | 2024-02-09 | 广东泓胜科技股份有限公司 | Dynamic speed measuring system and method for automobile |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101237755B1 (en) * | 2011-12-01 | 2013-02-28 | 주식회사 에스아이엠티 | Management system for calibrating weigher of axle using of wire/wireless network |
| CN103852147A (en) * | 2012-11-30 | 2014-06-11 | 北京万集科技股份有限公司 | Integrated dynamic weighing system for rectifying illegal driving and method |
| CN108061595A (en) * | 2017-12-14 | 2018-05-22 | 四川奇石缘科技股份有限公司 | A kind of traffic police administers the non-at-scene law enforcement detecting system of outline measuring overload and method |
| CN109000769A (en) * | 2018-08-16 | 2018-12-14 | 成都深蓝安捷交通设备有限公司 | One level quartz crystal formula dynamic highway vehicle automatic weighing instrument |
| CN109668610A (en) * | 2019-01-11 | 2019-04-23 | 东南大学 | The system of vehicle dynamically weighting method and its use based on neural net regression |
| CN109918972A (en) * | 2017-12-13 | 2019-06-21 | 北京万集科技股份有限公司 | A kind of driving weight intelligent control method and system |
| CN111144039A (en) * | 2019-12-04 | 2020-05-12 | 东南大学 | Train dynamic weighing system and weighing method based on deep learning |
| CN111735523A (en) * | 2020-08-27 | 2020-10-02 | 湖南大学 | Vehicle weight detection method and device based on video identification and storage medium |
| CN111882882A (en) * | 2020-07-31 | 2020-11-03 | 浙江东鼎电子股份有限公司 | Method for detecting cross-lane driving behavior of automobile in dynamic flat-plate scale weighing area |
-
2020
- 2020-12-03 CN CN202011397361.6A patent/CN112539816B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101237755B1 (en) * | 2011-12-01 | 2013-02-28 | 주식회사 에스아이엠티 | Management system for calibrating weigher of axle using of wire/wireless network |
| CN103852147A (en) * | 2012-11-30 | 2014-06-11 | 北京万集科技股份有限公司 | Integrated dynamic weighing system for rectifying illegal driving and method |
| CN109918972A (en) * | 2017-12-13 | 2019-06-21 | 北京万集科技股份有限公司 | A kind of driving weight intelligent control method and system |
| CN108061595A (en) * | 2017-12-14 | 2018-05-22 | 四川奇石缘科技股份有限公司 | A kind of traffic police administers the non-at-scene law enforcement detecting system of outline measuring overload and method |
| CN109000769A (en) * | 2018-08-16 | 2018-12-14 | 成都深蓝安捷交通设备有限公司 | One level quartz crystal formula dynamic highway vehicle automatic weighing instrument |
| CN109668610A (en) * | 2019-01-11 | 2019-04-23 | 东南大学 | The system of vehicle dynamically weighting method and its use based on neural net regression |
| CN111144039A (en) * | 2019-12-04 | 2020-05-12 | 东南大学 | Train dynamic weighing system and weighing method based on deep learning |
| CN111882882A (en) * | 2020-07-31 | 2020-11-03 | 浙江东鼎电子股份有限公司 | Method for detecting cross-lane driving behavior of automobile in dynamic flat-plate scale weighing area |
| CN111735523A (en) * | 2020-08-27 | 2020-10-02 | 湖南大学 | Vehicle weight detection method and device based on video identification and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 动态车辆称重系统的算法研究;张惠芳等;《研究与开发》;20170731;第36卷(第7期);52-61 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112539816A (en) | 2021-03-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112539816B (en) | Dynamic weighing correction method based on deep neural network in digital twin environment | |
| CN106198058B (en) | Real-time vertical wheel impact force measurement method based on tire pressure monitoring | |
| CN112201038B (en) | Road network risk assessment method based on risk of bad driving behavior of single vehicle | |
| CN104164829B (en) | Detection method of road-surface evenness and intelligent information of road surface real-time monitoring system based on mobile terminal | |
| CN106441530B (en) | A kind of bridge dynamic weighing method and dynamic weighing system based on long gauge length optical fibre grating sensing technique | |
| Liu et al. | On-line estimation of road profile in semi-active suspension based on unsprung mass acceleration | |
| CN100533073C (en) | Dynamic weighting system and method for vehicle | |
| CN109668610A (en) | The system of vehicle dynamically weighting method and its use based on neural net regression | |
| CN104792937A (en) | Bridge head bump detection evaluation method based on vehicle-mounted gravitational acceleration sensor | |
| CN111833604B (en) | Vehicle load state identification method and device based on driving behavior feature extraction | |
| CN111058360B (en) | Road surface flatness detection method based on driving vibration data | |
| CN112528208B (en) | Weighing-free AI intelligent recognition truck overload estimation method, device and system | |
| CN107957259A (en) | Wheelmark cross direction profiles measuring system and measuring method | |
| CN108871788A (en) | A kind of automatic transmission shift attribute test rack and its method of calibration and shift quality evaluation method | |
| CN103471865A (en) | Train suspension system failure isolation method based on LDA method | |
| CN108229832A (en) | Pure electric bus selection method based on road operation test and Fuzzy Hierarchy Method | |
| CN103344395B (en) | A kind of confirmation method of bridge strengthening target bearing capacity and device | |
| CN104864949A (en) | Vehicle dynamic weighing method and device thereof | |
| CN110867075A (en) | Method for evaluating influence of road speed meter on reaction behavior of driver under rainy condition | |
| CN104165676A (en) | Dynamic vehicle high-accuracy weighing method achieved in axle dynamic monitoring mode and axle set weighing mode | |
| Brzozowski et al. | A weigh-in-motion system with automatic data reliability estimation | |
| CN115544763A (en) | Road surface flatness prediction method, system and medium | |
| Ding et al. | Non-contact vehicle overload identification method based on body vibration theory | |
| CN109146261B (en) | Three-dimensional garage parking distribution method based on 3-parameter Weibull distribution model | |
| WO2019232737A1 (en) | Iteration-based quasi-static bridge influence line identification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230410 Address after: 710000, 20th floor, Block 4-A, Xixian Financial Lane, Fengdong New City Energy Jinmao District, Xixian New District, Xi'an City, Shaanxi Province Patentee after: Shaanxi Wisdom Luheng Electronic Technology Co.,Ltd. Address before: 710054 No. 58, middle section, Yanta Road, Shaanxi, Xi'an Patentee before: XI'AN University OF SCIENCE AND TECHNOLOGY |