Sensor-based road surface quality detection system

Abstract

The application relates to the technical field of road surface monitoring, in particular to a road surface quality detection system based on a sensor, which comprises a disturbance extraction module, a track analysis module, a consistency identification module, a mutation reduction module and a crack confirmation module.In the application, through time sequence correlation analysis of vertical structure vibration acceleration, GPS direction vector and transverse angular velocity signal during vehicle driving, disturbance behavior and track deviation trend change during vehicle contact with the road surface can be identified, real disturbance response is further screened from the change consistency, a space transfer model of crack mutation is constructed in combination with the joint state structure of disturbance intensity and deviation amplitude, joint abnormal intervals of structure damage and track deviation in space-time are identified, and accurate determination and space calibration of the crack area are realized through multidimensional index distribution density and time continuity discrimination criteria.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pavement monitoring, in particular to a sensor-based pavement quality detection system. BACKGROUND

[0002] The technical field of pavement monitoring relates to technical methods for information collection and identification of road surface conditions, mainly including crack detection, pothole identification, flatness state perception, structural response measurement, etc.

[0003] Among them, the sensor-based pavement quality detection system refers to a technical method for real-time or periodic collection of physical states of the road surface by using ground or vehicle-mounted sensing devices to identify physical damage such as road surface flatness, cracks, potholes, and settlement.

[0004] The prior art only collects the physical state of the vehicle when driving on the road surface by using ground or vehicle-mounted sensing devices, and mainly relies on single physical parameters such as vertical acceleration or strain to identify road surface cracks, settlement and other damage during the detection process. Due to the instability of the vehicle operating state and environmental noise in actual operation, there are often many non-structural disturbances in the vibration signal, which can easily lead to misjudgment. In the case of frequent road conditions or sparse crack distribution, crack information is often masked by dynamic fluctuations and cannot be effectively identified. For example, slight structural mutations formed under non-uniform load are easily confused with vehicle self-vibration, causing uncertainty and insufficient coverage accuracy of the detection results, and thus affecting the accurate presentation of road disease distribution and the development of subsequent maintenance strategies. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art and to provide a sensor-based pavement quality detection system.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: the sensor-based pavement quality detection system comprises:

[0007] A disturbance extraction module collects vertical direction structural vibration acceleration of a vehicle during driving in a specified period of a target road section, judges the dynamic disturbance behavior in the vertical structural response of the road surface under the action of the vehicle dynamic load, and marks the disturbance jump point sequence;

[0008] A trajectory analysis module obtains the GPS direction vector and the lateral angular velocity signal of the adjacent position points of the vehicle driving path in the same period, analyzes the deflection stability of the vehicle driving trajectory, and outputs the trajectory offset feature sequence;

[0009] A consistency identification module compares the change direction between each jump point in the disturbance jump point sequence and the trajectory offset in the trajectory offset feature sequence, and screens the change consistent index set;

[0010] A mutation reduction module obtains the disturbance jump feature segment and the trajectory offset feature segment corresponding to each index in the change consistent index set, identifies the joint mutation position of the two segments, and records as a crack response reduction sequence;

[0011] A crack confirmation module judges whether the crack response reduction sequence meets the set aggregation standard in terms of time continuity and number density, selects a crack area of the target road section, and obtains a road surface quality detection result.

[0012] As a further scheme of the present application, the disturbance jump point sequence includes a disturbance amplitude change position, a direction reversal marker, and a time index number, the trajectory offset feature sequence includes a path offset angle change amount, a yaw angle integral value, and a trajectory change index, the change consistent index set is specifically a position number, a continuous change alignment marker, and a synchronous change index group, the crack response reduction sequence includes a joint change position point, a structure and trajectory mutation corresponding segment, and a sampling trend change result, and the road surface quality detection result includes a crack area start and end position and a crack aggregation determination section.

[0013] As a further scheme of the present application, the disturbance extraction module includes:

[0014] An acceleration collection sub-module collects a structure vibration acceleration sequence recorded by a vertically arranged acceleration sensor during vehicle driving in a specified period of a target road section and arranges the structure vibration acceleration sequence into continuous time sequence data to generate an acceleration sequence data set;

[0015] A structure response judgment sub-module sets a fixed time window for segmented processing according to the acceleration sequence data set, extracts an acceleration sequence in each segment, calculates a standard deviation of each acceleration sequence, judges whether a change direction of a plurality of continuous segment standard deviation values appears a reversal trend and does not return to a first segment standard deviation reference value, and obtains a disturbance direction change judgment result;

[0016] A disturbance jump marker sub-module obtains and marks a time index position in the disturbance direction change judgment result that meets a continuous reversal of a direction and does not return to a fluctuation level, and generates a disturbance jump point sequence.

[0017] As a further scheme of the present application, the trajectory analysis module includes:

[0018] A direction vector acquisition sub-module obtains GPS coordinate data of continuous position points of a vehicle driving path in a same period, calculates a direction included angle between coordinate sequences of adjacent position points to construct a path direction vector sequence, extracts a continuous offset angle change value of a target vector sequence in a time index dimension, and generates a path offset angle sequence;

[0019] An angular velocity integration sub-module is configured to record a lateral angular velocity signal output by a vehicle inertial measurement device, calculate an integral of the lateral angular velocity with respect to time in a time index order, obtain a lateral angle change value in a continuous time period, and construct a lateral angle sequence.

[0020] A trajectory offset construction sub-module is configured to generate a trajectory offset feature sequence by calling a difference between the path offset angle sequence and the lateral angle sequence at the same time index.

[0021] As a further scheme of the present application, the consistency identification module comprises:

[0022] A position extraction sub-module is configured to extract a trajectory offset feature in the trajectory offset feature sequence at the same index position based on each jump point index number recorded in the disturbance jump point sequence, establish a change correspondence relationship between the two types of data at the unified index, and generate an index corresponding feature comparison group.

[0023] A direction comparison sub-module is configured to determine whether the disturbance change direction and the trajectory offset change direction at each index position in the index corresponding feature comparison group continuously present the same change direction at the current index point, identify the index positions that simultaneously have the direction consistency feature, and obtain a direction consistent index sequence.

[0024] An index screening sub-module is configured to screen an index set that satisfies index continuity and a direction consistent number of no less than three groups according to a distribution interval between adjacent index numbers in the direction consistent index sequence, eliminate isolated index points and data segments with a change interval exceeding a range, and obtain a change consistent index set.

[0025] As a further scheme of the present application, the mutation regularization module comprises:

[0026] A feature segment acquisition sub-module is configured to acquire each index number in the change consistent index set, extract a disturbance amplitude change value of the target index number at the time points before and after the disturbance jump point sequence to form a disturbance jump feature segment, simultaneously extract a continuous offset value of the target index number in the trajectory offset feature sequence to form a trajectory offset feature segment, and generate a double-feature change segment group.

[0027] A joint state construction sub-module is configured to call the disturbance amplitude change value and the trajectory offset change value corresponding to each segment in the double-feature change segment group, set the disturbance intensity and the offset amplitude as joint input parameters, and construct a joint state change structure.

[0028] A mutation position identification sub-module is configured to sample a state trajectory according to the continuous index positions of the joint state change structure, calculate a state transition probability of the disturbance change state and the offset change state in a joint space by using a Markov chain Monte Carlo algorithm, and screen a crack response regularization sequence.

[0029] As a further scheme of the present application, the crack confirmation module comprises:

[0030] An index distribution analysis submodule acquires all index numbers in the crack response sequence, identifies time intervals and index distances between consecutive adjacent indexes, counts the concentration of continuously distributed indexes, and generates an index distribution density value group;

[0031] An aggregation judgment submodule calculates the index quantity and time coverage interval of each paragraph in the continuous index paragraph according to the index distribution density value group, sets a mutation aggregation quantity reference value and a time continuity threshold, judges whether there is an index paragraph that simultaneously satisfies density aggregation and time continuity, and obtains a crack aggregation position judgment result;

[0032] A crack region calibration submodule acquires the index paragraph position in the crack aggregation position judgment result, calibrates the corresponding crack region in the target road section, and obtains a road surface quality detection result.

[0033] Compared with the prior art, the present application has the following advantages and positive effects:

[0034] In the present application, through time sequence correlation analysis of the vertical structure vibration acceleration, the GPS direction vector and the lateral angular velocity signal during the vehicle driving process, the disturbance behavior and the trajectory deviation trend change during the vehicle and road surface contact process can be identified. After correlating and comparing the structure disturbance and the trajectory variation, the real disturbance response is further screened from the consistency of the changes. The spatial transfer model of crack mutation is constructed by combining the joint state structure of disturbance intensity and deviation amplitude, the joint abnormal interval of structure damage and trajectory deviation in time and space is identified, the joint features of disturbance and deviation are normalized and aggregated by the multidimensional index distribution density and time continuity judgment standard, the accurate determination and spatial calibration of the crack region are realized, compared with the crack information obtained by only single-dimensional collection based on physical sensors, not only the crack identification accuracy and anti-interference ability are improved, but also the perception ability of slight or intermittent cracks is enhanced, the misjudgment and missed detection risk caused by the influence of interference fluctuation is overcome, and the system identification efficiency of complex road surface quality problems is improved. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The system flowchart of the present application;

[0036] Figure 2 The flowchart of the disturbance extraction module of the present application;

[0037] Figure 3 The flowchart of the trajectory analysis module of the present application;

[0038] Figure 4 The flowchart of the consistency identification module of the present application;

[0039] Figure 5 The flow chart of the mutation reduction module of the present application;

[0040] Figure 6 The flow chart of the crack confirmation module of the present application. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0042] Please refer to Figure 1 The sensor-based road surface quality detection system comprises:

[0043] The disturbance extraction module collects the vertical direction structural vibration acceleration of the vehicle during the driving process in the specified period of the target road section, judges the dynamic disturbance behavior in the vertical structure response of the road under the action of the vehicle dynamic load, and marks the disturbance jump point sequence;

[0044] The trajectory analysis module obtains the GPS direction vector and the lateral angular velocity signal of the adjacent position points of the vehicle driving path in the same period, analyzes the deflection stability of the vehicle driving trajectory, and outputs the trajectory deviation feature sequence;

[0045] The consistency identification module compares the change direction between each jump point of the disturbance jump point sequence and the trajectory deviation in the trajectory deviation feature sequence, and screens the change consistent index set;

[0046] The mutation reduction module obtains the disturbance jump feature segment and the trajectory deviation feature segment corresponding to each index in the change consistent index set, identifies the joint mutation position of the two segments, and records it as the crack response reduction sequence;

[0047] The crack confirmation module judges whether the crack response reduction sequence meets the set aggregation standard in terms of time continuity and number density, selects the crack area of the target road section, and obtains the road surface quality detection result;

[0048] The disturbance jump point sequence includes the disturbance amplitude change position, the direction reversal mark, and the time index number, the trajectory deviation feature sequence includes the path deviation angle change amount, the yaw angle integral value, and the trajectory change index, the change consistent index set specifically includes the position number of the consistent direction of the structural disturbance and the trajectory deviation, the continuous change alignment mark, and the synchronous change index group, and the crack response reduction sequence includes the joint change position point, the corresponding segment of the structural and trajectory mutation, and the sampling trend change result. The road surface quality detection result includes the start and end positions of the crack area and the crack aggregation determination section.

[0049] Please refer to Figure 2 The disturbance extraction module comprises:

[0050] an acceleration acquisition sub-module, which collects structural vibration acceleration sequences recorded by acceleration sensors arranged in the vertical direction during the driving of the vehicle in the specified period of the target road section and arranges the continuous time series data to generate an acceleration sequence data set;

[0051] When collecting the data recorded by the acceleration sensors arranged in the vertical direction during the driving of the vehicle in the specified period of the target road section, the spatial correspondence between the sensor arrangement points and the road structure units must be determined by the geographic information system for the structural arrangement of the target road section, and then the collection period is determined according to the road traffic schedule, the sampling frequency and channel sensitivity are set by using the data collection terminal, the original analog signal from the vertical acceleration sensor is converted into a digital signal, the sampling arrangement is performed through the analog-digital conversion module, and the time stamp calibration is used to ensure that all acceleration data sequences can be aligned to form a complete time series. The missing section, sensor jitter or system error code in the data must be detected and removed in real time. For example, when collecting the vertical vibration data of the bridge structure excited by heavy vehicles passing through the bridge in the period from 23:00 at night to 1:00 in the morning on a viaduct in a city, the sampling frequency is set to 500 Hz per second, the static section data without vehicle passing through is removed through 5-minute window translation verification, and finally the continuous effective section is combined into the unit acceleration sequence of the vehicle driving process according to the time range of the vehicle entering and leaving the road section.

[0052] a structural response judgment sub-module, which sets a fixed time window for segmented processing according to the acceleration sequence data set, extracts the acceleration sequence in each segment, calculates the standard deviation of each segment, and judges whether the change direction of the standard deviation values of multiple continuous segments appears a reverse trend and does not return to the standard deviation benchmark value of the first segment to obtain a disturbance direction change judgment result;

[0053] According to the acceleration time series data set, the data is processed in a set time window, and each segment has a fixed length, such as 10 seconds. Before segmentation, the data points are divided according to the sampling frequency, and then the continuous segments are combined in a sliding window manner. Then the acceleration sequence is extracted for standard deviation calculation. The standard deviation can be obtained by traversing all acceleration values in each segment, first calculating the mean value, then calculating the square deviation mean value, and then taking the square root. For example, the mean value of the data points in a segment is μ, and the mean value of (x-μ)² for each data point x is calculated and taken the square root, that is, the standard deviation σ can be obtained. The standard deviation reflects the structural vibration fluctuation level of the segment. After constructing the sequence of standard deviations of multiple adjacent segments in time order, it is judged whether the change direction exists inversion. The increase and decrease trend of adjacent three segments can be used for judgment. If the previous segment increases and the next segment decreases, it is an inversion trend. If the inversion occurs twice and does not return to the initial standard deviation baseline value, it is marked as a disturbance trend abnormal segment. The baseline value is set as the mean value of the standard deviation of the first five minutes of sampling, which is used as the normal operation reference line. For example, the standard deviation of the normal segment of the bridge is 0.02g. If the segment value changes to 0.03g→0.01g→0.025g, it is not returned after inversion, and it is determined that the structural response direction disturbance occurs in the segment.

[0054] The disturbance jump marker submodule obtains and marks the time index position that meets the condition of continuous inversion of direction and non-return of fluctuation level in the disturbance direction change judgment result, and generates a disturbance jump point sequence.

[0055] According to the direction inversion judgment result, the time index position that meets the condition is extracted, that is, the start time of each data is taken as a reference, and the start time point of the segment that meets the condition of direction inversion trend and non-return of standard deviation fluctuation is recorded. First, whether the inversion is formed is judged according to the standard deviation trend of adjacent three segments, and then whether the inversion appears continuously at least twice is judged. The difference between the standard deviation value of the current segment and the initial standard deviation baseline value is greater than the set tolerance interval. The tolerance can be set to ±5% of the baseline value range. For example, if the baseline is 0.02g, the tolerance is 0.001g. If the standard deviation of the current segment is 0.03g, it is out of the tolerance range, that is, it is not returned. The start time index of the segment is recorded as a jump point. For example, when a vehicle passes through a bridge segment and causes a sudden change in structural response, the standard deviation sequence is 0.02g→0.03g→0.01g→0.028g. It is judged that continuous inversion occurs and the last segment is higher than the baseline. The time index of the segment corresponding to 0.028g is recorded. Finally, the multiple time index points that meet the condition are summarized as a disturbance jump point sequence.

[0056] Please refer to Figure 3 , the trajectory analysis module comprises:

[0057] The direction vector acquisition submodule acquires GPS coordinate data of continuous location points along the vehicle's driving path within the same time period, calculates the direction angle between the coordinate sequences of adjacent location points to construct a path direction vector sequence, extracts the continuous offset angle change value of the target vector sequence in the time index dimension, and generates a path offset angle sequence.

[0058] GPS coordinate data is continuously collected during the vehicle's journey, resulting in a series of location points along the vehicle's path. After arranging these points in chronological order, spatial relationships between any two adjacent points need to be calculated. By calculating the directional angle between them, a path direction vector sequence is constructed segment by segment. This calculation process first converts the coordinate differences into directional components on a two-dimensional plane. The difference between east and north directions represents the tilt of the vehicle's driving direction. Then, an inverse trigonometric function is used to determine the angle between two adjacent vectors. This angle is the directional offset angle between the vehicle's path segment and the previous segment. This process requires combining Earth coordinate transformation methods to convert the original latitude and longitude into planar projected coordinates to avoid errors caused by geographical curvature. The coordinates are then indexed and arranged using timestamps to form a time series. Subsequently, the changing trends of the directional angles of all vectors within a continuous time period are extracted to obtain the minute offset of the path direction at each moment. The offset angle sequence is calculated by the difference between the offset angle and the previous vector direction angle. In bridge ramps, curved road sections, or when vehicles merge, this offset angle will show a stable increase or abrupt change, forming a typical change pattern. The path offset angle sequence can be recorded accordingly.

[0059] The angular velocity integral submodule synchronously records the lateral angular velocity signal output by the vehicle's inertial measurement equipment, calculates the integral of the lateral angular velocity with respect to time in time index order, obtains the yaw angle change value within a continuous time period, and constructs a yaw angle sequence.

[0060] During vehicle operation, the lateral angular velocity value output by the inertial measurement unit is synchronously read. This value represents the vehicle's rotational speed around its vertical axis. By reading the lateral angular velocity data at each time point, integration is performed on a continuous time axis to accumulate the vehicle's yaw angle change. This process can be achieved by first unifying the sampling time interval to a fixed value, then multiplying the average speed between every two sampling points by the time interval to obtain the small rotation angle within that time period, and then summing them segment by segment to obtain the cumulative yaw angle. The change of this yaw angle over time forms a yaw angle sequence, which is used to describe the rotational change trend of the vehicle in different time periods. For example, when the vehicle is turning right, the lateral angular velocity is positive in several consecutive time slices, causing the yaw angle to gradually increase. When traveling in a straight line, the angular velocity approaches zero, and the yaw angle remains stable. During the integration process, outliers need to be corrected or removed, and filtering methods are used to smooth the angular velocity data to avoid error accumulation that causes the yaw angle to deviate from the actual path change. Finally, a yaw angle value sequence at each time point can be obtained.

[0061] The trajectory deviation construction submodule calls the difference between the path deviation angle sequence and the yaw angle sequence at the same time index to generate a trajectory deviation feature sequence;

[0062] The path deviation angle sequence and the yaw angle sequence are extracted, and a point-by-point difference operation is performed under the same time index condition to obtain the angle difference between the path direction and the vehicle body orientation at each moment. The difference represents the trajectory deviation feature and can reflect whether the actual running path of the vehicle is consistent with its inertial orientation. The time stamps of the two sequences need to be unified, and the corresponding values at the same time point are subtracted. If the path deviation angle is greater than the yaw angle, it means that the path turning is faster than the vehicle turning action, otherwise the vehicle body deflection is greater than the path change, causing the deviation of the actual trajectory. For example, during the process of the vehicle entering a curve, if the GPS path display deflection is small and the yaw angle increases, it means that the vehicle turns ahead or with a larger amplitude, forming an inside deviation. Finally, the difference values at all time points are recorded to form the trajectory deviation feature sequence.

[0063] Please refer to Figure 4 The consistency identification module includes:

[0064] The position extraction submodule extracts the trajectory deviation feature at the same index position in the trajectory deviation feature sequence based on the index number of each jump point recorded in the disturbance jump point sequence, establishes the change correspondence relationship between the two types of data under the unified index, and generates an index corresponding feature comparison group.

[0065] According to the index number in the disturbance jump point sequence, the corresponding trajectory deviation feature value is extracted in the trajectory deviation feature sequence at the same index position. The index alignment processing of the two types of sequences needs to be completed according to a unified time reference, so that each disturbance point has a clear matching position in the trajectory deviation sequence. The extraction process directly locates the corresponding trajectory deviation value in the trajectory deviation sequence by reading the time label or number in the disturbance jump point record as a key value. Then each pair of data forms a feature combination item, including the disturbance change direction label and the trajectory deviation angle difference at the corresponding moment, and the index information of the pair of data is saved to form an index corresponding feature comparison group. For example, during the driving process of a vehicle, a disturbance jump occurs within a certain number of seconds. The deviation angle data corresponding to the number of seconds is extracted and paired with the disturbance direction. Each row of index data in the table is visualized to form a corresponding change set of trajectory disturbance and deviation.

[0066] The direction comparison submodule judges whether the disturbance change direction and the trajectory deviation change direction at each index position in the index corresponding feature comparison group continuously present the same change direction in the vicinity of the current index point, identifies the index positions that simultaneously have the direction consistency feature, and obtains a direction consistent index sequence.

[0067] The index corresponding feature comparison group is analyzed, and the direction consistency is determined according to the change trend of the disturbance direction at each index point and the change trend of the trajectory deviation direction. The positive and negative values of the disturbance direction change amount at the continuous index points are calculated, and the same sign determination is performed with the corresponding change amount in the trajectory deviation sequence. If the continuous two or more index points show the same direction change, it is considered that the directions are consistent. In the determination process, the number threshold of adjacent indexes can be set, for example, the continuous three points show positive change, that is, the disturbance standard deviation continuously increases, and the trajectory deviation angle also continuously increases, so the three points can be determined as a direction consistent segment. The processing needs to traverse all index items of the comparison group and record the starting index position that meets the condition. In the process, the change amplitude also needs to be preliminarily determined to exclude the false direction consistency caused by the small change of the nearly static state. For example, when the vehicle changes lanes, the trajectory deviation angle in the disturbance jump point segment continuously increases, and the disturbance direction standard deviation is also in the increasing state, which can be considered as the same direction change, so the direction consistent index sequence is identified as the output result.

[0068] The index screening submodule screens an index set that meets the conditions of index continuity and direction consistency of not less than three groups according to the distribution interval between adjacent index numbers in the direction consistent index sequence, eliminates isolated index points and data segments with change intervals exceeding the range, and obtains a change consistent index set.

[0069] The data in the direction consistent index sequence is analyzed in structure. The continuity of index distribution is identified mainly by calculating the time interval between adjacent index numbers. When traversing the entire index sequence, if the interval between the two adjacent index numbers is greater than the preset distance threshold, it is considered to be discontinuous, and the position is segmented as a boundary. Then, it is determined whether the number of data pairs with continuous direction consistency in each segment reaches the set lower limit, such as at least three groups or higher standards. The continuous index set that meets the condition is retained, and isolated points are eliminated. A single direction consistent index does not have continuity and is not representative. In the elimination process, it is also necessary to check whether there is a significant reversal of change trend in the index segment. If there is, it is marked as an invalid segment and is not retained. Finally, the index set with continuous change trend and consistent direction is output, which can be applied to further behavior feature modeling or trajectory deviation identification processing procedures, such as the continuous direction consistent feature appearing in the bridge section entering process, which is marked as a characteristic stable section.

[0070] Please refer to Figure 5 The mutation normalization module includes:

[0071] The feature fragment acquisition submodule acquires each index number in the change consistent index set, extracts disturbance amplitude change values of the target index number at adjacent time points before and after the disturbance jump point sequence to form a disturbance jump feature fragment, and simultaneously extracts continuous offset values of the target index number in the trajectory offset feature sequence to form a trajectory offset feature fragment, thereby generating a double-feature change fragment group;

[0072] After obtaining the change consistent index set, each index number is read one by one, a certain number of adjacent time point data before and after the index number in the disturbance jump point sequence is searched, the disturbance amplitude change value of each time point in the time range is extracted, and a disturbance jump feature fragment is formed. The extraction process needs to be performed through setting a fixed-length time window, for example, taking the target index as the center, taking the same number of disturbance amplitude values before and after, and arranging the disturbance amplitude values in time sequence to form a continuous disturbance change sequence. In the same way, the continuous offset values corresponding to the target index in the trajectory offset feature sequence are simultaneously extracted, that is, the offset feature data is read in the same time window range to obtain the trajectory offset feature fragment. This operation needs to ensure that the two groups of feature fragments are completely aligned in time, so as to facilitate subsequent joint analysis. For example, when a vehicle appears disturbance jump and continuous offset trend in the bridge section, a plurality of time points before and after the change consistent index point are expanded to obtain the complete disturbance change curve and the offset trajectory change curve constituting the phenomenon process. The two curves respectively form a double-feature change fragment group.

[0073] The joint state construction submodule calls the disturbance amplitude change value and the trajectory offset change value corresponding to each fragment in the double-feature change fragment group, sets the disturbance intensity and the offset amplitude as joint input parameters, and constructs a joint state change structure.

[0074] The feature change fragment group is taken as input data, the disturbance amplitude change value and the trajectory offset change value of each fragment are read in sequence, a disturbance-offset feature pair is formed at each time point, the feature pair is defined as the basic dimension of the state input, the disturbance intensity level and the trajectory offset amplitude level are set as the joint state parameters, the disturbance change value is divided into amplitude intervals in the formed feature fragment, for example, it can be divided into low amplitude, medium amplitude and high amplitude, which correspond to the absolute value size range of the disturbance, and the offset value is divided into similar levels, each time point feature pair is mapped to a state node in the joint state space, and then a state change path is constructed. The path is the joint evolution trajectory of the disturbance and the offset in the time fragment, and the continuity of the path represents the feature sequence of the state transition. For example, when a vehicle drives through a structure uneven area, the disturbance change is changed from low amplitude to high amplitude, and the offset amplitude is also significantly increased. The continuous state nodes formed in this process can be represented as a state transition path.

[0075] The mutation location identification submodule samples the state trajectory based on the continuous index position of the joint state change structure, calculates the state transition probability of the perturbation change state and the offset change state in the joint space using the Markov chain Monte Carlo algorithm, and filters the crack response normalization sequence.

[0076] Using joint state change segments as input, this submodule performs joint analysis on the changes in the amplitude of disturbances and the changes in trajectory offsets generated by vehicles within the continuous time periods corresponding to the consistent change index set, in order to discover the time locations where structural disturbances and path offsets change significantly simultaneously. This submodule first constructs multiple disturbance-offset joint state trajectories composed of consecutive time points by sliding sampling along the joint state path. Then, based on the frequency of transitions between states, it constructs a joint state transition probability matrix and calculates and evaluates the energy level of state changes. Furthermore, it extracts high-energy change paths using a Markov chain Monte Carlo algorithm, selecting time index points with abrupt change characteristics from these paths.

[0077] First, within the time series covered by the consistent change index set, the corresponding perturbation jump feature segments and trajectory offset feature segments are invoked. At each index point, a joint state trajectory containing multiple time nodes is constructed by sliding forward and backward at fixed time steps. The state at each time node consists of the current perturbation amplitude change value and trajectory offset change value, forming a joint state sequence. Next, the jump relationships between any two consecutive states in this state sequence are counted, and the number of times the state changes from one combination to another is recorded, forming a joint state transition matrix. The calculation formula is as follows:

[0078] ;

[0079] in, This indicates that the vehicle transitions from the joint state in the time series. Jump to The probability, This represents the total number of times the transition occurs across all state trajectories. Indicates from state The total number of all possible transitions at the start. This matrix serves as an index for other states in the state space. It reflects the evolution trend of the linkage between disturbances and offsets during vehicle operation.

[0080] To further identify abrupt state changes, the rate of perturbation and the rate of shift are calculated at each state point. These are then normalized and weighted summed to form a potential energy value, which is used to assess the degree of anomalousness of the state. The potential energy function is expressed as follows:

[0081] ;

[0082] in, Joint anomaly potential representing current state, Rate of change of perturbation jump magnitude at adjacent time point (intensity change per unit time), Rate of change of trajectory deviation angle per unit time, and Max and min values of perturbation rate and deviation rate statistically obtained in all segments, used to complete normalization operation. Parameters and Control the contribution size of perturbation change and deviation change to potential value respectively. When the detection task is more focused on structural response, can set , if more attention is paid to vehicle trajectory deviation behavior, can set .

[0083] After constructing the state transition matrix and the potential value corresponding to each state, execute Markov chain Monte Carlo sampling, and judge whether to jump to a new candidate state through the following Metropolis-Hastings acceptance function:

[0084] ;

[0085] Where, is the current state, is the next candidate state, represent the potential values of the current and candidate states respectively, if the new state potential is lower, that is, the change is not significant, then directly accept the jump, if the new state potential is higher, then accept in the form of probability, to ensure that the algorithm can jump out of the stable segment and identify the mutation trend.

[0086] After completing a large number of sampling, collect all the time points of potential sudden rise, aggregate and analyze the distribution characteristics of these points, combine adjacent time points into segments, filter out discrete abnormal values that only appear at individual time, and only keep time sections with significant perturbation and deviation changes, finally form the crack response sequence, provide effective input index section for vehicle trajectory behavior identification and structure perturbation analysis. This method can be adapted by adjusting the weight and window parameters under different road conditions, structure types and vehicle categories, and has wide application flexibility.

[0087] State transition probability calculation formula:

[0088] ;

[0089] Suppose the joint state sequence (perturbation level + deviation level) has been extracted from a certain continuous 10-second time segment of the vehicle, where the simplified state set is as follows, : medium perturbation + medium deviation, : high perturbation + large deviation, : Low disturbance + small offset. In order to divide the disturbance amplitude change value and the trajectory offset change value into high, medium and low levels, interval demarcation needs to be carried out based on the statistical characteristics of the full data. Specifically, first, the disturbance change rate (such as the acceleration change per unit time) and the offset angle change rate (such as the offset angle change per unit time) collected by the vehicle in multiple time segments are calculated respectively, and the maximum value, minimum value and median value in all segments are extracted, and the distribution curve of the change rate is constructed. On this basis, the data can be divided into three intervals by using the equidistant method or the segmented strategy based on the percentile, for example, the disturbance change rate is divided into low (close to the minimum value to the 33rd percentile), medium (33rd to 66th percentile), and high (66th percentile to the maximum value), and the trajectory offset is also divided in the same way. When the disturbance or offset rate at a certain time point is calculated during the actual detection process, it can be classified according to the interval in which the value is located. This division method is universal and adjustable, suitable for different roads, vehicles or sensor precision scenes, and only needs to be re-estimated once at the beginning of each batch of tasks to obtain new division standards and ensure the consistency and accuracy of the level classification.

[0090] The number of transition relationships observed in the state path is as follows: , , That is, the total number of jumps from is: ;

[0091] The state transition probability from to is: ;

[0092] It is shown that the probability of a vehicle changing from a medium disturbance-medium offset state to a high disturbance-large offset state in the current time period is 60%.

[0093] Joint abnormal potential function (normalized):

[0094] ;

[0095] Let the disturbance change rate at the current time point be , and the trajectory offset angle rate be , the minimum and maximum values of the disturbance rate in all data segments are: , , and the minimum and maximum values of the offset rate are: , , and the structure-oriented detection weight setting is: , .

[0096] The normalized disturbance rate is: ;

[0097] The normalized shift rate is: ;

[0098] The final joint potential value is calculated as: ;

[0099] The potential value of this state is 0.6541, indicating that it is in a high mutation risk state.

[0100] Metropolis-Hastings acceptance function:

[0101] ,

[0102] Let the current state be The potential energy is , the perturbation of the next candidate state is further improved, and the potential energy is calculated as .

[0103] The energy difference is calculated as: ;

[0104] The acceptance probability is calculated as: ;

[0105] It means that although the potential energy is increased, this jump has a probability of 87.9% to be sampled and accepted.

[0106] First, by counting the number of jumps of the joint state in the path sequence, the transition probability between each state is calculated, and the frequency ratio of the state jumping from one perturbation-offset combination to another combination is obtained, which is used to describe the evolution trend of the structural perturbation and trajectory offset of the vehicle in a certain driving stage. Subsequently, the perturbation change rate and trajectory offset change rate of each state point are calculated, and these two values are normalized with the maximum and minimum values in the whole data range, respectively. Then, according to the set weight coefficient, the two values are combined to obtain the joint potential value of the state, which is used to measure the abnormality and mutation risk. Finally, by comparing the potential difference between the current state and the candidate state, the jump probability is calculated using the acceptance function. If the energy of the new state is lower, it is always accepted. If the energy is higher, it is accepted with a certain probability, so as to construct a state sample chain with global exploration ability.

[0107] Please refer to Figure 6 , the crack confirmation module comprises:

[0108] The index distribution analysis submodule obtains all index numbers in the crack response normalized sequence, identifies the time interval and index distance between consecutive adjacent indexes, and calculates the concentration of continuous distribution indexes to generate an index distribution density value group.

[0109] The time distribution characteristics of all mutation response indexes in the normalized sequence are analyzed to identify whether there is a high-frequency and short-period mutation aggregation phenomenon in the detection period. The module first obtains all index numbers in the normalized sequence, sorts them according to the actual sampling time stamps corresponding to each index, and then divides the mutation points with adjacent time intervals and continuous index numbers into several index segments, each of which represents a continuous mutation segment that may have an aggregation behavior. For each paragraph, the system calculates the concentration degree of the mutation points in the paragraph for subsequent aggregation judgment.

[0110] To accurately evaluate the aggregation characteristics of the response points on the time axis, the index distribution analysis submodule not only considers the number and duration of responses, but also introduces a regulation mechanism for the uniformity of time distribution. The density calculation uses a weighted expression to comprehensively reflect the number, time span and time distribution dispersion of the mutation points in the paragraph, and the formula is as follows:

[0111]

[0112] The weighted distribution density of the segment, with a unit of “per second”; is the number of crack response indexes contained in the segment, is the actual sampling time difference between the first and last response index points in the segment, with a unit of seconds; is the standard deviation of the time stamps corresponding to all indexes in the segment, also with a unit of seconds, used to measure the dispersion degree of the mutation points on the time axis; is a preset standard reference value, usually set to the time standard deviation under ideal uniform distribution according to the system sampling interval, for normalization; is an adjustment coefficient used to control the weighting degree of the standard deviation on the density, the greater the value, the more the system values the concentration of the mutation points in time distribution, usually set between 0 and 1.

[0113] For example, in a certain vehicle driving segment, if 6 crack response indexes in a segment are concentrated within 4 seconds, and the time distribution standard deviation is significantly lower than the system reference value, the density value of the segment will be significantly amplified, indicating that the mutation behavior in this area is significantly concentrated. On the contrary, if the response points are dispersed, even if the number is the same, the density value will be adjusted lower, effectively suppressing the interference of long-period low-frequency responses on the mutation recognition accuracy.

[0114] All continuous paragraphs are calculated according to the above formula to form a set of index distribution density values marked by time segments, providing quantifiable input for the subsequent aggregation judgment submodule to support the identification of whether the crack response shows a significant concentration trend in a certain area. This process ensures accurate measurement of the time distribution structure of the mutation points and enhances the ability to extract local abnormal behavior. ​​

[0115] Suppose that during a pavement monitoring process, the vehicle is driving on a bridge section, and the system detects the following 6 crack response mutation points, with index numbers as follows: The corresponding time stamps (in seconds) are as follows: the 1st point: 12.00, the 2nd point: 12.25, the 3rd point: 12.48, the 4th point: 13.15, the 5th point: 13.36, and the 6th point: 13.55.

[0116] The system sets this section as a continuous mutation section (index distance less than or equal to 2, time interval less than 2 seconds), and needs to calculate the distribution density thereof.

[0117] Mutation point number This section contains 6 response points, so: ,

[0118] Time span The time difference between the first and last index of this section is: seconds;

[0119] First, calculate the squared mean deviation of each point time from the average time:

[0120] The time mean is: ;

[0121] Sum of each squared deviation: ;

[0122] Calculate the standard deviation: seconds;

[0123] The system sets that under ideal conditions, the 6 points are evenly distributed within 1.5 seconds, and the standard deviation is about: seconds, and the time concentration adjustment factor is: .

[0124] Apply the formula: ;

[0125] Substitute the numerical values: points / second;

[0126] The weighted distribution density of this crack response section is about 5.36 points / second.

[0127] The aggregation judgment submodule calculates the index number and time coverage interval of each section in the continuous index section according to the index distribution density value group, sets the mutation aggregation number reference value and time continuity threshold, judges whether there is an index section that simultaneously satisfies density aggregation and time continuity, and obtains the crack aggregation position judgment result.

[0128] The index distribution density value group is received, and each index paragraph corresponding to the density is analyzed to evaluate whether it exhibits obvious mutation aggregation characteristics in quantity and time. Specifically, the system first reads the response point quantity of each index paragraph and its corresponding time coverage interval, i.e., the time difference value corresponding to the earliest and latest mutation points in the paragraph, and then compares these data with the preset mutation aggregation quantity reference value and time continuity threshold to determine whether the paragraph meets both the "quantity reaches the mutation standard" and "time period is sufficiently continuous" conditions. For example, if a paragraph contains more than a set reference value, such as 6 or more, and the time span of these response points is less than the system-set upper limit of continuity, such as 2 seconds, the paragraph is considered to be a highly continuous response paragraph with high mutation intensity. All index paragraphs that meet the dual conditions are marked as mutation aggregation paragraphs, and the system adds them to the crack aggregation position judgment result set to provide target index information for subsequent area calibration. This process ensures that isolated high-density response points are excluded from the identification results, while capturing real possible structural abnormal distribution intervals.

[0129] The crack area calibration sub-module obtains the index paragraph positions in the crack aggregation position judgment result set, calibrates the corresponding crack area in the target road segment, and obtains the road surface quality detection result.

[0130] After obtaining the crack aggregation positions identified by the aggregation judgment sub-module, the mutation index paragraphs are mapped back to the original road segment structure layout one by one to calibrate their actual position ranges in the road monitoring area. This process is based on the previously established index-coordinate relationship or time-path mapping rules, and is positioned according to the GPS coordinates or road segment numbers associated with the start and end indices, thereby realizing the physical space attribution judgment of high-risk structure areas. For example, a crack aggregation paragraph composed of indices 1503 to 1509 corresponds to the 2nd to 4th meter range on the east side of Bridge A segment through the space-time correspondence rule conversion, and the system generates corresponding crack area annotation information based on this, which is output as the final road surface quality detection result. This result can be used for subsequent tasks such as bridge maintenance, road repair scheduling, or structure safety level evaluation, providing direct support for actual road asset management and operation.

[0131] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms. Any person skilled in the art can modify or change the above disclosed technical content to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments within the technical solution content of the present application still falls within the protection scope of the present application.

Claims

1. A sensor-based road surface quality detection system, characterized in that, The system includes: The disturbance extraction module collects the vertical structural vibration acceleration during vehicle travel in a specified time period on the target road section, determines the dynamic disturbance behavior in the vertical structural response of the road surface under the dynamic load of the vehicle, and marks the disturbance jump point sequence. The trajectory analysis module acquires the GPS direction vector and lateral angular velocity signals of adjacent points along the vehicle's travel path within the same time period, analyzes the deflection stability of the vehicle's travel trajectory, and outputs a trajectory offset feature sequence. The consistency identification module compares the change direction between each jump point in the disturbance jump point sequence and the trajectory offset in the trajectory offset feature sequence, and filters the change consistency index set. The mutation reduction module obtains the perturbation jump feature segment and trajectory offset feature segment corresponding to each index in the change consistency index set, identifies the joint mutation position of the two segments, and records it as the crack response reduction sequence. The crack confirmation module determines whether the crack response normalization sequence reaches the set threshold value for the number of abrupt clusters and the threshold value for time continuity and quantity density, selects the crack area of ​​the target road section, and obtains the road surface quality detection result. The disturbance jump point sequence includes the location of disturbance amplitude change, direction reversal marker, and time index number. The trajectory offset feature sequence includes the path offset angle change, yaw angle integral value, and trajectory change index. The change consistency index set specifically includes the location number where the structural disturbance and trajectory offset direction are consistent, the continuous change alignment identifier, and the synchronous change index group. The crack response normalization sequence includes joint change location points, corresponding segments of structural and trajectory abrupt changes, and sampling trend change results. The pavement quality detection results include the start and end positions of the crack area and the crack aggregation judgment section.

2. The sensor-based road surface quality detection system according to claim 1, characterized in that, The disturbance extraction module includes: The acceleration acquisition submodule collects the structural vibration acceleration sequence recorded by vertically deployed acceleration sensors during vehicle travel in a specified time period on the target road segment, organizes it into continuous time-series data, and generates an acceleration sequence dataset. The structural response judgment submodule processes the data in segments within a set time window based on the acceleration sequence dataset. Before segmentation, the data points are divided according to the sampling frequency. Then, the continuous segments are spliced ​​and combined in a sliding window manner. The acceleration sequence is extracted segment by segment for standard deviation calculation. After the standard deviations of multiple adjacent segments are arranged in chronological order to form a sequence, it is judged whether the direction of change is reversed. The increase or decrease trend of three adjacent segments is used for judgment. That is, if the previous segment increases and the next segment decreases, it is a reversal trend. If there are two consecutive reversals and the standard deviation of the initial segment does not return to the baseline value, it is marked as an abnormal segment of disturbance trend. The baseline value is set as the average standard deviation of the first five minutes of sampling, which is used as the normal operation reference line to obtain the judgment result of the change in disturbance direction. The perturbation jump marker submodule extracts the start time of each data segment based on the direction reversal judgment result. It records the start time of the segment where the direction reversal trend is established and the standard deviation fluctuation has not returned to the benchmark value. This includes judging whether a reversal has formed by the standard deviation trend of three adjacent segments, judging whether the reversal has occurred at least twice consecutively, and the difference between the current segment's standard deviation value and the initial standard deviation benchmark value is greater than the set tolerance range. If it exceeds the tolerance range, it is considered non-regression. The start time index of this segment is recorded as the jump point, and the results are summarized into a perturbation jump point sequence.

3. The sensor-based road surface quality detection system according to claim 2, characterized in that, The trajectory analysis module includes: The direction vector acquisition submodule acquires GPS coordinate data of continuous location points along the vehicle's driving path within the same time period, calculates the direction angle between the coordinate sequences of adjacent location points to construct a path direction vector sequence, extracts the continuous offset angle change value of the target vector sequence in the time index dimension, and generates a path offset angle sequence. The angular velocity integration submodule synchronously reads the lateral angular velocity value output by the inertial measurement device during vehicle operation. By reading the lateral angular velocity data at each time point, it performs integration calculations on a continuous time axis to accumulate the yaw angle change of the vehicle. During the integration process, outliers need to be corrected or removed, and filtering methods are used to smooth the angular velocity data to construct a yaw angle sequence. The trajectory offset construction submodule calls the difference between the path offset angle sequence and the yaw angle sequence at the same time index to generate a trajectory offset feature sequence.

4. The sensor-based road surface quality detection system according to claim 3, characterized in that, The consistency identification module includes: The location extraction submodule extracts trajectory offset features at the same index position in the trajectory offset feature sequence based on the index number of each jump point recorded in the perturbation jump point sequence, establishes the correspondence between the changes of the two types of data under the same index, and generates an index-corresponding feature control group. The direction comparison submodule determines whether the perturbation change direction and trajectory offset change direction of each index position in the index-corresponding feature control group are continuously changing in the same direction at the current index point. It calculates the positive and negative values ​​of the perturbation direction change at consecutive index points and determines the same sign with the corresponding change in the trajectory offset sequence. If two or more consecutive index points show the same direction change, they are considered to be consistent in direction. The module identifies index positions that simultaneously have the characteristic of consistent direction and obtains the consistent direction index sequence. The index filtering submodule filters the index set that satisfies the condition of continuous indexes and consistent orientation with no less than three groups based on the distribution spacing between adjacent index numbers in the direction-consistent index sequence, removes isolated index points and data segments with change intervals exceeding the range, and obtains the change-consistent index set.

5. The sensor-based road surface quality detection system according to claim 4, characterized in that, The mutation reduction module includes: The feature fragment acquisition submodule acquires each index number in the change consistent index set, extracts the change value of the disturbance amplitude at the time points before and after the disturbance jump point sequence of the target index number to form a disturbance jump feature fragment, and simultaneously extracts the continuous offset value of the target index number in the trajectory offset feature sequence to form a trajectory offset feature fragment, generating a dual feature change fragment group. The joint state construction submodule calls the disturbance amplitude change value and trajectory offset change value corresponding to each segment in the dual feature change segment group to form a disturbance-offset feature pair at each time point. The feature pair is defined as the basic dimension of the state input. The disturbance intensity level and trajectory offset amplitude level are set as joint state parameters. In the formed feature segment, the disturbance amplitude change value is first divided into amplitude ranges. The trajectory offset is also divided in the same way. The feature pair at each time point is mapped to a state node in the joint state space to construct the joint state change structure. The mutation location identification submodule samples the state trajectory based on the continuous index positions of the joint state change structure, constructs multiple disturbance-offset joint state trajectories composed of continuous time points, and constructs a joint state transition probability matrix based on the jump frequency between states. The calculation formula is as follows: ； in, This indicates that the vehicle transitions from a joint state in the time series. Jump to The probability, This represents the total number of times the transition occurs in all state trajectories. Indicates from state The total number of all possible transitions at the start. For indexes of other states in the state space; To identify abrupt state changes, the rate of change of state disturbance and the rate of change of state offset are calculated at each state point. After normalization, these two values ​​are weighted and summed to form the potential energy value. The potential energy function is expressed as follows: ； in, The joint anomalous potential energy represents the current state. This indicates the rate of change of the disturbance jump amplitude at adjacent time points. This represents the rate of change of the trajectory offset angle per unit time. and These represent the maximum and minimum values ​​of the perturbation rate and offset rate obtained statistically across all segments, respectively. and The contributions of disturbance and offset changes to the potential energy value are controlled separately. After constructing the state transition matrix and the potential energy value corresponding to each state, Markov chain Monte Carlo sampling is performed, and the Metropolis-Hastings acceptance function is used to determine whether to jump to a new candidate state: ； in, This is the current state. The next candidate state, These represent the potential energy values ​​of the current state and the candidate state, respectively. Based on the degree of joint anomalous potential energy and mutation risk, the crack response normalization sequence is screened by comparing the potential energy difference between the current state and the candidate state and using the acceptance function to calculate the state transition probability.

6. The sensor-based road surface quality detection system according to claim 5, characterized in that, The crack confirmation module includes: The index distribution analysis submodule obtains all index numbers in the crack response normalization sequence, sorts them according to the actual sampling timestamp corresponding to each index, divides the mutation points with adjacent time intervals and consecutive index numbers into multiple index segments, each segment representing a continuous mutation fragment with clustering behavior, and calculates the concentration of the continuously distributed indexes using the following formula: ； in, This represents the weighted distribution density of consecutive mutant fragments. It is the number of crack response indices for consecutive mutation fragments. It is the sampling time difference between the first and last response index points within a continuous mutation segment. It is the standard deviation of the timestamps corresponding to all indices of consecutively mutated fragments. It is a preset standard reference value. It is the adjustment coefficient; Integrate all weighted distribution density values ​​to generate an indexed distribution density value group; The clustering determination submodule calculates the number of indexes and time coverage intervals for each segment in a continuous index segment based on the index distribution density value group, sets a baseline value for the number of abrupt clusters and a time continuity threshold, determines whether there are index segments that simultaneously satisfy density clustering and time continuity, and obtains the crack clustering location determination result. The crack area calibration submodule obtains the index segment position in the crack cluster location determination result, calibrates the corresponding crack area in the target road segment, and obtains the road surface quality detection result.

Images