A preprocessing method for flatness calibration measurement based on neural networks

Abstract

This invention relates to a preprocessing method for road surface smoothness calibration and measurement based on neural networks, comprising the following steps: Step 1, installing an acceleration sensor on a vehicle; Step 2, constructing a standardized calibration section for a vibratory road surface smoothness test vehicle; Step 3, recording the points at different speeds where the front wheel just touches the raised calibration block, the rear wheel touches the raised calibration block, and the rear wheel moves away to a smooth index point; Step 4, calculating the index value corresponding to the highest amplitude point at each speed using Python; Step 5, calculating the average value of the three nodes to the highest point at each speed based on the data obtained in Step 4; Step 6, obtaining the corrected distance of the front wheel from the highest amplitude point at different speeds based on Step 5; Step 7, running the test on the previously calibrated section again, and obtaining the xyz axis data and GPS data, combined with the three linear target relationship functions previously predicted, yields three points.

Description

Technical Field

[0001] This invention relates to the field of road engineering technology, and in particular to a preprocessing method for roughness calibration and measurement based on neural networks. Background Technology

[0002] Road surface smoothness is an important parameter in road inspection, reflecting the bumpiness of the road surface. Vibration sensors installed on vehicles are used to calculate road surface smoothness, collecting vibration data along the Z-axis. This vibration data can be used to predict road surface smoothness, primarily because it shows a correlation with the International Roughness Index (IRI) of road surfaces, and the equipment cost is significantly reduced.

[0003] Currently, the calibration method for flatness measurement based on vibration sensors involves manually selecting a range of vibration data for calibration calculation. This method relies on manually observing changes in the peak values ​​of the vibration data to extract the data, which results in large errors, susceptibility to subjective human factors, slow speed, and low efficiency, thus limiting the application of vibration sensors in flatness measurement.

[0004] To ensure the accuracy of the calculated International Roughness Index (IRI) value, the preprocessing of the roughness calibration is crucial. To achieve a better fit between the root mean square error (RMS) of the vehicle's vertical acceleration statistics and the IRI, vibration data from low to high levels are needed to test the fit. For optimal results, a trapezoidal convex calibration block can be selected. The data should be obtained at the first point where the front wheel just touches the block, the second point where the rear wheel just touches the block, and the third point where the rear wheel leaves the block and reaches a smooth surface. Then, a neural network method is used to optimize and calculate a linear relationship between the RMS of the vehicle's vertical acceleration statistics and the IRI.

[0005] Currently, there isn't a well-established system or method for preprocessing flatness measurement data that can easily and accurately preprocess the data. Finding these three points by observing vehicles driving over trapezoidal protrusion calibration blocks at different speeds still relies on manual recording and visual comparison. Therefore, manual calibration is difficult, cumbersome, inefficient, heavily subjective, and inaccurate. It also relies heavily on manual labor, resulting in high measurement costs and hindering large-scale, high-frequency calibration. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a preprocessing method for roughness calibration measurement based on neural networks. This method uses a neural network algorithm to automatically identify the first point where the front wheels just touch the trapezoidal raised calibration block, the second point where the rear wheels just touch the block, and the crucial third point where the rear wheels leave the block and the surface becomes smooth. Based on these three automatically identified points at different speeds, optimization and calculation are performed using platforms such as Matlab to obtain a more accurate International Roughness Index (IRI) value.

[0007] The above-mentioned objective of this invention is achieved through the following technical solutions:

[0008] A preprocessing method for flatness calibration measurement based on neural networks includes the following steps:

[0009] Step 1: Install an acceleration sensor on the vehicle;

[0010] Step 2: Construct a standardized calibration road section for the vibratory road surface smoothness test vehicle;

[0011] Step 3: Drive the test vehicle to be calibrated at a constant speed of 10, 20, 30, 40, 50, and 60 km / h (each speed can be ±30%), with one wheel driving over the raised calibration block. Calculate the vertical acceleration obtained using a language tool to obtain the z-axis data and GPS data of each speed passing through the raised calibration block during this time. Merge and connect the z-axis data and GPS data according to time and visualize them. Based on the visualized waveform, record the time at which the front wheel just touches the raised calibration block, the rear wheel touches the raised calibration block, and the rear wheel leaves the flat index point at different speeds.

[0012] Step 4: Repeat step 3 5 times to obtain acceleration data (x, y, z axes) and GPS data at speeds of 10, 20, 30, 40, 50, and 60 km / h (each speed ±30%). Use a language tool to extract the z-axis data, speed, azimuth angle, etc., and visualize the curves using Python. Record the 6 points where the front wheel touches the raised marker at the 6 different speeds. Calculate the 6 index points corresponding to the highest amplitude at each speed using Python.

[0013] Calculate the distance from the point where the front wheel touches the raised marker at each speed to the point of maximum amplitude. Then record the six points where the rear wheel touches the raised marker at six different speeds, and calculate the distance from the point where the rear wheel touches the raised marker to the point of maximum amplitude at each speed. Simultaneously, record the six points where the rear wheel leaves the raised marker and reaches a stable position at six different speeds, and calculate the distance from the point where the rear wheel leaves the raised marker and reaches a stable position to the point of maximum amplitude at each speed.

[0014] Step 5: Calculate the average value of the three nodes to the highest point at each speed based on the data obtained in Step 4;

[0015] Step 6: Based on the values ​​obtained in Step 5, V1 (speed) - SA1 (distance), V2 (speed) - SA2 (distance), V3 (speed) - SA3 (distance), V4 (speed) - SA4 (distance), V5 (speed) - SA5 (distance), V6 (speed) - SA6 (distance), that is, the corrected distance between the front wheel and the highest point of the amplitude at different speeds;

[0016] By using a neural network to perform regression prediction on V-SA, we can obtain the index points where the front wheel touches the raised marker at different speeds. By predicting in sequence, we can obtain the index points where the rear wheel touches the raised marker at different speeds. By predicting in sequence, we can obtain the index points where the rear wheel leaves the raised marker at different speeds.

[0017] Step 7: Run the route again on the previously calibrated section (at a speed of 0~80 km / h). The obtained xyz axis data and GPS data, combined with the three linear target relationship functions predicted earlier, can give us three points.

[0018] Furthermore, in step 1,

[0019] a. The accelerometer is capable of measuring up to 3 axes;

[0020] b. The resolution of the accelerometer is ±1g;

[0021] c. The range of the accelerometer: ±10g;

[0022] d. The output frequency of the accelerometer: greater than 200Hz;

[0023] e. The installation method of the acceleration sensor: adhesive bonding;

[0024] f. The installation location of the acceleration sensor: installed above the rear wheel of the vehicle;

[0025] g. The number of acceleration sensors installed: one above each of the two left wheels and one above each of the two right wheels;

[0026] h. The Z-axis of the accelerometer is oriented upwards.

[0027] Furthermore, in step 2, the method for constructing a standardized calibration section for a vibratory road surface roughness testing vehicle includes the following steps:

[0028] Find an asphalt road surface longer than 1km (a smooth and flat surface without damage, potholes, speed bumps, etc.) and construct a standardized calibration section using raised calibration blocks with a height of 5cm, a top edge length of 5cm, and a bottom edge length of 40cm (the size of the raised blocks is not fixed and can be selected by yourself).

[0029] Furthermore, in step 6,

[0030] The neural network prediction yields three linear objective relation functions, with the following conclusions:

[0031] SA = w1 * (speed) + b1;

[0032] SB = w2 * (speed) + b2;

[0033] SC = w3 * speed + b3;

[0034] Where speed is the velocity, and SA represents the distance (index distance) from the point where the front wheel touches the raised marker block to the point of maximum amplitude.

[0035] SB represents the distance (index distance) from the rear wheel touching the raised marker block to the highest point of the amplitude;

[0036] SC represents the distance from the rear wheel leaving the raised marker block to the point of smoothness from the highest point of amplitude (index distance);

[0037] w — weighting coefficient;

[0038] b — constant.

[0039] In summary, the present invention has at least one of the following beneficial technical effects:

[0040] The method described in this invention utilizes Python and neural networks to quickly and accurately preprocess road surface data, thereby saving time and costs for subsequent road surface calibration. During routine inspections, only one vehicle calibration and preprocessing step are required, followed by optimized calculations to obtain the relationship between the vehicle's root mean square error (RMS) of vertical acceleration and the International Roughness Index (IRI), thus yielding the road's IRI value.

[0041] In terms of execution efficiency, compared to existing manual recording and visualization comparison methods, the method described in this patent can quickly and accurately complete the preprocessing of smoothness data during routine inspections, obtaining the road's iri value more rapidly. Furthermore, the method described in this patent does not require personnel to have algorithmic expertise or coding skills; simply running a Python script can preprocess the smoothness data and then perform optimization calculations to obtain the calibration file.

[0042] In terms of economic benefits, the method mentioned in this patent mainly relies on neural network algorithms and lightweight accelerator sensing equipment, which greatly increases the frequency of inspections. At the same time, by investing in equipment on the basis of existing daily inspections, the IRI value of roads can be measured, thus eliminating the related costs of special testing. Attached Figure Description

[0043] Figure 1 This is a schematic diagram illustrating an asphalt pavement for the present invention.

[0044] Figure 2 This is a schematic diagram illustrating the structure of the raised calibration block for the present invention.

[0045] Figure 3 This is a schematic diagram illustrating the test vehicle's single-sided wheel running over the raised calibration block at different speeds, as shown in this invention.

[0046] Figure 4 The present invention demonstrates the waveforms of the test vehicle at different speeds.

[0047] Figure 5 This invention presents waveform diagrams showing the index values ​​corresponding to the highest amplitude points of the test vehicle at various speeds.

[0048] Figure 6 The waveform diagram shows the corrected distance between the point where the front wheel of the test vehicle touches the raised marker block and the point of highest amplitude, as presented in this invention.

[0049] Figure 7 This is a schematic diagram illustrating how the front wheels of the test vehicle touched the index point of the raised marker block at different speeds, as shown in this invention.

[0050] Figure 8 This is a schematic diagram illustrating how the rear wheels of the test vehicle touched the index point of the raised marker block at different speeds, as shown in this invention.

[0051] Figure 9 This is a schematic diagram illustrating the index point where the rear wheel of the test vehicle leaves the raised marker block at different speeds, as shown in this invention.

[0052] Figure 10 The waveforms of three linear objective relation functions are shown for this invention. Detailed Implementation

[0053] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0054] Example 1:

[0055] This invention discloses a preprocessing method for flatness calibration and measurement based on neural networks, comprising the following steps:

[0056] Step 1, (Basic Structure 1) Install an acceleration sensor on the vehicle, wherein:

[0057] a. The accelerometer can measure 3 axes;

[0058] b. Accelerometer resolution: ±1g;

[0059] c. Accelerometer range: ±10g;

[0060] d. Accelerometer output frequency: greater than 200Hz;

[0061] e. Installation method of the accelerometer sensor: adhesive mounting;

[0062] f. Installation location of the acceleration sensor: Installed above the rear wheel of the vehicle;

[0063] g. Number of accelerometers installed: one above each of the two left wheels and one above each of the two right wheels;

[0064] h. The Z-axis of the accelerometer is oriented upwards;

[0065] Step 2, (Basic Structure 2) Construct a standardized calibration section for the vibratory road surface smoothness test vehicle;

[0066] Find an asphalt road surface longer than 1km (a smooth, flat surface without damage, potholes, speed bumps, etc.). Figure 1 ) will be as Figure 2 A standardized calibration road section is constructed using raised calibration blocks with a height of 5cm, a top bottom edge length of 5cm, and a bottom bottom edge length of 40cm (the size of the raised blocks is not fixed and can be selected by the user).

[0067] Step 3, (Basic Structure 3): Drive the test vehicle to be calibrated at a constant speed of 10, 20, 30, 40, 50, and 60 km / h (each speed can be ±30%), with one wheel driving over the raised calibration block (see...). Figure 3 The acquired vertical acceleration is used to calculate the z-axis data and GPS values ​​of each velocity passing through the raised calibration block during this time using language tools. The z-axis data and GPS data are then fused and concatenated according to time and visualized. Based on the visualized waveform, the following times are recorded at different speeds: the front wheel just touching the raised calibration block, the rear wheel touching the raised calibration block, and the rear wheel leaving the flat index point (see...). Figure 4 ).

[0068] Step 4: Repeat step 3 5 times to obtain acceleration data (x, y, z axes) and GPS data at speeds of 10, 20, 30, 40, 50, and 60 km / h (each speed ±30%). Use a language tool to extract the z-axis data, speed, azimuth angle, etc., and visualize the curves using Python. Record the 6 points where the front wheel touches the raised marker at the 6 different speeds. Calculate the 6 index points corresponding to the highest amplitude at each speed using Python.

[0069] Calculate the distance from the point where the front wheel touches the raised marker at each speed to the point of maximum amplitude. Then record the six points where the rear wheel touches the raised marker at six different speeds, and calculate the distance from the point where the rear wheel touches the raised marker to the point of maximum amplitude at each speed.

[0070] Simultaneously, the rear wheel was recorded at six points from leaving the raised marker block to reaching a stable position at six different speeds, and the distance from the point of leaving the raised marker block to the point of stability at each speed was calculated to the value of the highest amplitude.

[0071] Step 5: Calculate the average value of the three nodes to the highest point at each speed based on the data obtained in Step 4;

[0072] Step 6: Based on the values ​​obtained in Step 5, V1 (speed) - SA1 (distance), V2 (speed) - SA2 (distance), V3 (speed) - SA3 (distance), V4 (speed) - SA4 (distance), V5 (speed) - SA5 (distance), V6 (speed) - SA6 (distance), that is, the corrected distance between the front wheel and the highest point of the amplitude at different speeds;

[0073] By performing regression prediction on V-SA using a neural network, the index point where the front wheel touches the raised marker block at different speeds can be obtained. Figure 7 By predicting in sequence, the index points where the rear wheel touches the raised marker block at different speeds can be obtained. Figure 8 By predicting in sequence, we can obtain the index points where the rear wheel leaves the raised marker block at different speeds. Figure 9 ).

[0074] The neural network prediction yields three linear objective relation functions, with the following conclusions:

[0075] SA = w1 * (speed) + b1;

[0076] SB = w2 * (speed) + b2;

[0077] SC = w3 * speed + b3;

[0078] Note: speed is the speed, and SA represents the distance (index distance) from the front wheel touching the raised mark block to the highest point of the amplitude.

[0079] SB represents the distance (index distance) from the rear wheel touching the raised marker block to the highest point of the amplitude;

[0080] SC represents the distance from the rear wheel leaving the raised marker block to the point of smoothness from the highest point of amplitude (index distance);

[0081] w — weighting coefficient;

[0082] b — constant;

[0083] Step 7: Run the route again on the previously calibrated section (at speeds of 0-80 km / h). Combine the obtained x, y, and z axis data and GPS data with the three linear target relationship functions predicted earlier to obtain three points. Figure 10 From left to right, the first dotted line is the first point where the front wheel just touches the raised indicator block, the second dotted line is the second point where the rear wheel touches the raised indicator block, and the third dotted line is the third point where the rear wheel leaves the raised indicator block and reaches a stable position.

[0084] Example 2:

[0085] The method and implementation methods proposed in this invention are as follows: Figure 3 As shown, acceleration sensors and other equipment are installed on the vehicle. The selection and installation method of the acceleration sensors are described in Basic Structure 1.

[0086] The implementation process can be summarized as follows:

[0087] 1. Find a smooth, straight asphalt road surface with a length greater than 1 km;

[0088] 2. A test vehicle to be calibrated, with acceleration sensors installed on its rear wheels;

[0089] 3. Place a raised calibration block on the road surface. Drive the test vehicle at a constant speed of 10, 20, 30, 40, 50, and 60 km / h (each speed can be ±30%), with one wheel driving over the raised calibration block (see...). Figure 3 (Length, width, and height can be defined by the user). The acquired vertical acceleration is used to calculate the z-axis data and GPS values ​​of each velocity passing through the raised calibration block during this time period using a language tool. The z-axis and GPS data are then fused and concatenated according to time and visualized. Based on the visualized waveform, the following times are recorded at different speeds: the front wheel just touching the raised calibration block, the rear wheel touching the raised calibration block, and the rear wheel leaving the flat index point (see...). Figure 4 );

[0090] 4. Repeat step 3 5 times to obtain acceleration data (x, y, z axes) and GPS data at speeds of 10, 20, 30, 40, 50, and 60 km / h (each speed ±30%). Use a programming language to extract the z-axis data, speed, azimuth angle, etc., and visualize the curves using Python. Record the points (indexA11, indexA12, indexA13, indexA14, indexA15, indexA16) where the front wheel touches the raised marker at different speeds (see...). Figure 4 The first point), calculated using Python, corresponds to the index values ​​of the highest amplitude points at each speed: M11, M12, M13, M14, M15, M16 (see...). Figure 5 );

[0091] Calculate the distance from the point where the front wheel touches the raised marker block to the point of maximum amplitude at each speed: A11=|indexA11-M11|, A12=|indexA12-M12|, A13=|indexA13-M13|, A14=|indexA14-M14|, A15=|indexA15-M15|, A16=|indexA16-M16|;

[0092] Then record the points where the rear wheel touches the raised markers at different speeds: indexB11, indexB12, indexB13, indexB14, indexB15, and indexB16 (see...). Figure 4 (The second point), calculate the distance from the point where the rear wheel touches the raised marker block to the point of maximum amplitude at each speed; B11=|indexB11-M11|, B12=|indexB12-M12|, B13=|indexB13-M13|; B14=|indexB14-M14|, B15=|indexB15-M15|, B16=|indexB16-M16|;

[0093] Simultaneously record the points at which the rear wheel leaves the raised marker block and reaches a stable position at different speeds: indexC11, indexC12, indexC13, indexC14, indexC15, and indexC16 (see...). Figure 4 (The third point), calculate the distance from the raised marker block to the stable point and the highest amplitude at each speed: C11=|indexC11-M11|, C12=|indexC12-M12|, C13=|indexC13-M13|; C14=|indexC14-M14|, C15=|indexC15-M15|, C16=|indexC16-M16|;

[0094] Note: An1n2, Bn1n2, and Cn1n2 represent nodes respectively, where A is the front wheel touching the raised block, B is the rear wheel touching the raised block, and C is the rear wheel leaving the raised block. n1 is the number of repetitions, and n2*10 is the speed at the time of the experiment. For example, A33 represents the distance between the point where the front wheel touches the raised block at a speed of 30 in the third repetition and the point of highest amplitude, and so on.

[0095] 5. Based on the data obtained in step 4, calculate the average value of the three nodes to the highest point at each speed: SA1 = (A11+A21+A31+A41+A51) / 5, SA2 = (A12+A22+A32+A42+A52) / 5,...

[0096] Following this pattern, SA1, SA2, SA3, SA4, SA5, SA6 will be calculated respectively; SB1 = (B11 + B21 + B31 + B41 + B51) / 5... and so on, SB1, SB2, SB3, SB4, SB5, SB6 will be calculated respectively. SC1 = (C11 + C21 + C31 + C41 + C51) / 5,... and so on, SC1, SC2, SC3, SC4, SC5, SC6 will be calculated respectively.

[0097] Note: SA1: Corrected distance from the point where the front wheel touches the raised marker block to the highest point of amplitude when calibrating the test vehicle at a speed of 10km / h (±30%); SA6: Corrected distance from the point where the front wheel touches the raised marker block to the highest point of amplitude when calibrating the test vehicle at a speed of 60km / h (±30%) (see...) Figure 6 SBn represents the corrected distance from the highest point of amplitude when the rear wheel touches the raised marker at a speed of n*10. SCn represents the corrected distance from the highest point of amplitude when the rear wheel leaves the raised marker at a speed of n*10. The distances mentioned in the text are indexed distances. For example, if the accelerometer is 250Hz, it will output 250 times per second, and the indexed distance would be 1-250, and so on.

[0098] 6. Based on the distances obtained in step 5 (V1(speed)-SA1(distance), V2(speed)-SA2(distance), V3(speed)-SA3(distance), V4(speed)-SA4(distance), V5(speed)-SA5(distance), V6(speed)-SA6(distance), which represent the corrected distances between the front wheel and the point of highest amplitude at different speeds), the index point (index) of the front wheel touching the raised marker block at different speeds can be obtained by performing regression prediction on V-SA using a neural network. Figure 7By predicting in sequence, the index points where the rear wheel touches the raised marker block at different speeds can be obtained. Figure 8 By predicting in sequence, we can obtain the index points where the rear wheel leaves the raised marker block at different speeds. Figure 9 ).

[0099] The neural network prediction yields three linear objective relation functions, with the following conclusions:

[0100] SA = w1 * (speed) + b1;

[0101] SB = w2 * (speed) + b2;

[0102] SC = w3 * speed + b3;

[0103] Note: speed is the velocity, and SA represents the distance (index distance) from the front wheel touching the raised marker block to the highest point of the vibration.

[0104] SB represents the distance (index distance) from the rear wheel touching the raised marker block to the highest point of the vibration.

[0105] SC represents the distance from the rear wheel leaving the raised marker block to the point of smoothness from the highest point of amplitude (index distance).

[0106] w — weighting coefficient

[0107] b — constant.

[0108] The implementation principle of this invention is as follows: Currently, the preprocessing of flatness calibration data mainly relies on manual labor. This involves manually recording and visually comparing the first point where a single front wheel of a car just touches the trapezoidal raised calibration block at different speeds, the second point where a single rear wheel just touches the trapezoidal raised calibration block, and the third point where a single rear wheel leaves the trapezoidal raised calibration block and reaches a smooth surface. Then, based on the obtained preprocessed data, the root mean square error (RMS) of the vehicle's vertical acceleration statistical index and the International Roughness Index (IRI) are fitted. However, this method suffers from problems such as low search efficiency, slow speed, high time cost, low accuracy, and inability to perform calibration over a large area.

[0109] Therefore, this invention uses a neural network algorithm to automatically identify the first point where the front wheels of the car just touch the trapezoidal raised calibration block, the second point where the rear wheels just touch the trapezoidal raised calibration block, and the third point where the rear wheels leave the trapezoidal raised calibration block and the surface becomes smooth. Based on these three automatically identified points at different speeds, optimization and calculation are performed using platforms such as Matlab to obtain a more accurate International Roughness Index (IRI) value.

[0110] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A pre-processing method for flatness calibration measurement based on neural network, characterized in that: The method comprises the following steps: Step 1, installing an acceleration sensor on a vehicle; Step 2, constructing a standardized calibration section for a vibratory pavement flatness test vehicle; Step 3, driving the test vehicle at a constant speed of 10, 20, 30, 40, 50, and 60 km / h over the raised calibration block on one side, obtaining the vertical acceleration, and using a language tool to calculate the z-axis data and GPS data of the test vehicle at each speed, fusing the z-axis data and GPS data according to time, and visualizing the fused data, and recording the index points at which the front wheel just touches the raised calibration block, the rear wheel touches the raised calibration block, and the rear wheel leaves the raised calibration block at a constant speed, respectively; Step 4, repeating step 3 five times to obtain the acceleration xyz-axis data and GPS data of the test vehicle at speeds of 10, 20, 30, 40, 50, and 60 km / h, using a language tool to extract the z-axis data, speed, and azimuth angle data, and using python to visualize the curve and record the six points at which the front wheel touches the raised calibration block at the six different speeds, and using python to calculate the six index value points corresponding to the highest amplitude points at each speed; respectively calculating the value of the point at which the front wheel touches the raised calibration block at each speed from the highest amplitude point, then recording the six points at which the rear wheel touches the raised calibration block at the six different speeds, respectively calculating the value of the point at which the rear wheel touches the raised calibration block at each speed from the highest amplitude point, and recording the six points at which the rear wheel leaves the raised calibration block at the six different speeds, respectively calculating the value of the point at which the rear wheel leaves the raised calibration block at each speed from the highest amplitude point; Step 5, calculating the average value of the three nodes to the highest point at each speed according to the data obtained in step 4; Step 6, obtaining the corrected distance of the front wheel from the highest amplitude point at different speeds according to V1-SA1, V2-SA2, V3-SA3, V4-SA4, V5-SA5, V6-SA6 obtained in step 5, wherein V1, V2, V3, V4, V5, and V6 represent speed, and SA1, SA2, SA3, SA4, SA5, and SA6 represent distance; According to the neural network, the index index points at which the front wheel touches the raised calibration block at different speeds are predicted, and the index index points at which the rear wheel touches the raised calibration block at different speeds are predicted in sequence, and the index index points at which the rear wheel leaves the raised calibration block at different speeds are predicted in sequence; In the step 6, Three linear target relationship functions are obtained through the prediction of the neural network, and the conclusions are as follows: SA = w1 * speed + b1; SB = w2 * speed + b2; SC = w3 * speed + b3; wherein speed is speed, SA represents the distance of the front wheel touching the raised calibration block from the highest amplitude point, SB represents the distance of the rear wheel touching the raised calibration block from the highest amplitude point, and SC represents the distance of the rear wheel leaving the raised calibration block to the highest amplitude point. ​ ​ w1, w2, w3 - weight coefficients; b1, b2, b3 - constants; Step 7, run again on the previous calibration section, 0~80 km / h speed of the xyz axis data and gps data, combined with the three linear target relationship function obtained by the previous prediction to obtain three points.

2. The pre-processing method for flatness calibration measurement based on neural network according to claim 1, characterized in that: In the step 1, a. the acceleration sensor can measure the number of axes is 3 axes; b. the resolution accuracy of the acceleration sensor: ±1g; c. the range of the acceleration sensor: ±10g; d. the output frequency of the acceleration sensor: greater than 200Hz; e. the installation method of the acceleration sensor: adhesive installation; f. the installation position of the acceleration sensor: installed above the rear wheel of the car; g. the installation number of the acceleration sensor: one on the left wheel and one on the right wheel; h. the Z axis of the acceleration sensor is upward.

3. The pre-processing method for flatness calibration measurement based on neural network according to claim 1, characterized in that: In the step 2, the method for constructing the standardized calibration section of the vibration type road surface flatness test vehicle includes the following steps: Find a length of more than 1km of asphalt pavement, wherein the asphalt pavement is smooth and flat without damage, pit, speed bump, and construct a convex calibration block with a height of 5cm, an upper bottom length of 5cm and a lower bottom length of 40cm to construct a standardized calibration section.

Images