Clustering road surfaces from sensor data from vehicle bumpers

Abstract

The invention relates to a system for classifying unevenness in a road surface profile, comprising a deflection detection unit (1) for determining a temporal course of a deflection of a spring-damper system of a vehicle during its travel, and comprising a computing unit (3) which is designed to assign sections in the temporal course of the deflection to at least one specific one of a finite number of predefined classes of unevenness by means of a pattern recognition-based comparison of the temporal course of the deflection with explicit or implicit pre-stored information about the finite number of classes of unevenness.

Description

The invention relates to a system for classifying asperities in a road surface profile, and to a method for generating a machine learning model for classifying asperities in a road surface profile.Vehicles such as passenger cars or trucks typically have a spring-damper system, such that dynamic excitations due to unevennesses in the road are transmitted as attenuated as possible to the vehicle body above it. The deflection of this spring-damper system on the body is typically dependent on physical parameters, in particular a profile of the road surface, longitudinal accelerations and transverse accelerations, the mass of the vehicle portion above the spring-damper system, and the dynamic properties of the spring-damper system itself, in particular a spring stiffness and a damper constant (or their nonlinear equivalents). If these external influences and the parameters of the spring-damper system are known, this deflection can also be calculated by using analytically detachable relationships.By reverse application of this relationship, it is possible to draw a very precise conclusion about the geometric profile of the road surface from a measured deflection, the detection of longitudinal accelerations and transverse accelerations and again on the basis of the parameters of the spring-damper system.In the prior art, it is at least known to detect unevenness in the geometric roadway profile lying ahead by means of optical sensors in order to be able to adjust a chassis of a motor vehicle in a predictive manner.DE 10 2018 003 129 A1 relates to a method for adjusting a chassis of a motor vehicle in which roadway information, which characterizes an orientation and / or a local unevenness of a roadway lying in front of the motor vehicle, is received by means of a control device of the motor vehicle and the chassis is adjusted as a function of the roadway information, wherein the roadway information is received by a vehicle-external computing device by means of the control device.If, as explained at the beginning, the deflection of the spring-damper system is intended to be measured and the geometric profile of the currently traveled road surface is deduced therefrom, this can advantageously be carried out in a central processing unit. The advantage of using a central processing unit compared to a local processing unit arranged in the vehicle is that the algorithm used for this purpose can be changed very easily and centrally, i.e. at once for all vehicles used. A change in the algorithm in a control device of a plurality of vehicles, on the other hand, entails higher outlay and greater time delays.A prerequisite for the execution of such an algorithm in a central processing unit is, however, naturally the knowledge of the values of all the above-described physical parameters in order to be able to determine the geometric profile of the road surface. If no measurement or reliable estimation is present only in one of these values, or if a data transmission to the central processing unit is not implemented, the algorithm cannot be successfully solved on the central processing unit. The geometric profiles of the traveled roads cannot be determined with sufficient accuracy using the above-mentioned analytical solution.There is therefore a need to be able to estimate elements of a geometric profile of a road surface. It is the object of the invention to identify geometric anomalies such as unevennesses in a road surface, and to use only a deflection profile of a spring-damper system for this purpose.The invention results from the features of the independent claims. Advantageous refinements and refinements are the subject matter of the dependent claims.A first aspect of the invention relates to a system for classifying unevennesses in a road surface profile, having a deflection detection unit for determining a temporal profile of a deflection of a spring-damper system of a vehicle during its travel, and having a computing unit which is designed to assign sections in the temporal profile of the deflection to at least one specific information item from a finite number of predefined classes of unevennesses by a comparison, based on pattern recognition, of the temporal profile of the deflection with explicit or implicit information items stored beforehand about the finite number of classes of unevennesses.The pattern recognition makes it possible to recognize specific amplitude characteristics in the time characteristic of the deflection of a spring-damper system of the vehicle because of its characteristic pattern and to assign them to a specific class of unevennesses. Such predefined classes can be, for example: striking hole, road damage, shaft cover, transition in the road surface, end or beginning of an asphalted or concreted roadway, outward curvature from the roadway upward;For this purpose, the pattern recognition analyzes the amplitude curve over time. The analysis can be explicit or implicit. On the one hand, for example, a Fourier transformation can allow the frequency analysis normalized via a speed, and on the other hand, implicit methods can be used, such as, for example, the use of an artificial neural network. The term implicit is that the parameters in a predefined artificial neural network are generally no longer intuitively comprehensible to a human user and are completely abstracted from the physical principles.In each case, probabilities are obtained with which a respective class is present per unevenness. A corresponding unevenness is therefore classified into such a class with the highest probability of its presence. In this case, the class "no unevenness" can also be used in a particular embodiment, which corresponds to a discarding of the amplitude profile under consideration over time.The deflection of the spring-damper system describes, in particular, a relative change in distance between a reference point fixed to the vehicle body and a reference point on the rim side. A spring travel can thus be determined. Since a damper has a speed-dependent resistance force (this corresponds to the movement speed of the deflection of the spring-damper system), the computing unit is advantageously also informed of a speed of the vehicle along the surface of the earth, so that the pattern recognition can take place depending on the vehicle speed.Advantageously, the pattern recognition-based classification of unevennesses into specific classes can take place solely with the aid of an amplitude profile of a deflection of a spring-damper system of a vehicle. Further data, in particular optical data, are not necessary for carrying out pattern recognition.The data obtained according to the invention can be made available, for example, to authorities or road infrastructure operators. Customers from the so-called "data as a service" environment can thus obtain a more accurate statement as to what type is a respective unevenness of the road surface. The customer can thus plan repair measures better, for example.According to an advantageous embodiment, the system further comprises an identification unit which is designed to transmit information about the vehicle type and / or information about a spring damper system installed on the vehicle to the computing unit, wherein the computing unit is a central, stationary computing unit.The computing unit can have a database in which an assignment of parameters of a respective spring-damper system of a vehicle type is stored in order to be able to read out the parameters of the spring-damper system with the aid of information about the vehicle type. Alternatively, parameters of the spring-damper system can also be transmitted directly from the vehicle to the computing unit.The computing unit is preferably a stationary computing unit which carries out the task of classification for a plurality of vehicles. This backend implementation has the advantage that the algorithms on the computing unit can easily be changed at any time, which could not happen promptly by changing the software of the control units in the vehicles.According to a further advantageous embodiment, the system further comprises a speed detection unit for determining a respectively current speed of the vehicle, wherein the computing unit is designed to take into account the respective speed during the pattern recognition.According to a further advantageous embodiment, the system further comprises a locating unit for determining a respectively current position of the vehicle, wherein the computing unit is designed to assign a respective unevenness assigned to a specific class to a respective geoposition on the basis of a position of the vehicle.The highly accurate geoposition associated with a respective classified unevenness may make it easier for a customer of the information about the classified unevenness to find and repair the respective unevenness, for example.According to a further advantageous embodiment, the deflection detection unit is designed to detect a respective temporal profile of a deflection of a left and a right spring-damper system of a vehicle during its travel.By viewing the left and right sides of the vehicle in order to determine a respective separate amplitude profile of the deflection, more accurate classifications of the unevennesses can be carried out, since in particular information about the width thereof is available.According to a further advantageous embodiment, the computing unit is designed to execute a machine learning model for pattern recognition.According to a further advantageous embodiment, the machine learning model comprises an artificial neural network.The following can be used here: similarity learning, unuputrified learning, self-organized maps, convolutional neural networks, autoencoder, conformal prediction, probability theory.According to a further advantageous embodiment, the computing unit is designed to preprocess the respective temporal profile of the deflection in a transformation in order to obtain a latent space, and to transfer the latent space by means of classification clustering into a probability distribution of the assignment to a respective specific class from the finite number of classes.A further aspect of the invention relates to a method for generating a machine learning model for classifying unevennesses in a road surface profile, comprising the steps of:executing a test trip using a test vehicle and thereby: optically detecting a roadway surface and detecting a time curve of a deflection of a spring-damper system of the vehicle;automatically identifying individual unevennesses of the road surface as belonging to a respectively specific one of a plurality of predefined classes of unevennesses by image recognition;training the machine learning model with the time curve of the deflection as an input variable, and with the identified unevennesses, with assignment to the time curve of the deflection as an output variable;The test vehicle has a special sensor system compared to the vehicles in connection with the system described above and below, in order to be able to obtain corresponding optical data about the road surface. The test lighter is preferably also equipped with an accurate locating unit in order to be able to enable highly accurate documentation of the data acquisition during the test trips. In contrast, usual vehicles which are used within the scope of the system described above and below do not require a highly accurate optical sensor system for detecting the road surface, nor a computer-implemented image recognition for classifying the unevennesses.With the asperities identified by image recognition, highly reliable rankings are available to classes that form a ground truth for training the machine learning model. Preferably, a speed of the test vehicle is also logged in order to be able to assign the corresponding amplitudes of the deflections of the spring-damper system to a current speed and to be able to take them into account in the training process of the machine learning model.This advantageously enables automatic data acquisition, which, due to its efficiency, allows a plurality of vehicle types, a plurality of speeds and other parameters to be taken into account in the training of the machine learning model.According to a further advantageous embodiment, after the detection of the time profile of the deflection of the spring-damper system of the vehicle, a statistical method is applied in a computer-implemented manner in order to recognize predictive regions in the time profile of the deflection, wherein input variables for training the machine learning model are formed only with the predictive regions of the time profile of the deflection and only with unevennesses identified as belonging to the predictive regions.Advantages and preferred refinements of the proposed method result from an analogous and analogous transfer of the statements made above in connection with the proposed system.Further advantages, features and details are evident from the following description, in which--possibly with reference to the drawing--at least one exemplary embodiment is described in detail. Identical, similar and / or functionally identical parts are provided with the same reference numerals.The following are shown: FIG. 1 : A method for generating a machine learning model for classifying unevennesses in a road surface profile according to an exemplary embodiment of the invention. FIG. 2 : shows a system for classifying unevennesses in a road surface profile according to an exemplary embodiment of the invention. FIG. 3 : Exemplary classes of unevennesses in the road surface profile.FIG. 1 illustrates a method of generating a machine learning model for classifying asperities in a road surface profile. First, a test trip is carried out S 1 using a special test vehicle with a camera directed at the front road section, which can be a 3D camera. This serves to detect the road surface. A time profile of a deflection of a spring-damper system of the vehicle is likewise recorded in a temporally associated manner. By means of image recognition, furthermore, individual unevennesses of the road surface are identified S 2 in an automated manner as belonging to a respectively specific one of a multiplicity of predefined classes of unevennesses. Exemplary surface profiles of roadway with various predefined classes are described in FIG. 3. This provides sufficient information to carry out the training S 3 of the machine learning model with the temporal profile of the deflection as an input variable, and with the identified unevennesses, associated with the profile, as an output variable. The obtained machine learning model may be applied in a system as described in FIG. 2. In this case, a multiplicity of vehicles can use temporal profiles of deflections of a respective spring-damper system of a vehicle during their ongoing operation in order to have the classification carried out at a computing unit 3.FIG. 2 shows such a system for classifying unevennesses in a road surface profile. A deflection detection unit 1 of a vehicle determines a temporal profile of a deflection of a spring-damper system of the vehicle during its travel by sensor. By wireless data transmission, the time profile is transmitted to a stationary central processing unit 3. The arithmetic unit 3 serves to assign sections in the time profile of the deflection to at least one specific information item from a finite number of predefined classes of unevennesses by a comparison, based on pattern recognition, of the time profile of the deflection with explicit or implicit information stored beforehand about the finite number of classes of unevennesses. In other words, the central processing unit 3 executes an algorithm for each of the vehicles with such a deflection detection unit 1 in order to detect specific classes of unevennesses by means of pattern recognition. Camera data or other sensor data is not necessary for classification. Here, a machine learning model which was generated by the method as described under FIG. 1 is preferably applied.FIG. 3 shows possible unevennesses in the curves (A) to (D) which belong to the finite number of predefined classes. In partial image (A), a striking hole is shown. In partial image (B), on the other hand, a ground wave, in partial image (C) a roadway damage, and in partial image (D) a transition of the roadway surface, since, for example, a newly asphalted road section begins and the latter has a changed height to the preceding road section.Although the invention has been illustrated and explained in more detail by preferred exemplary embodiments, the invention is not restricted by the disclosed examples and other variations can be derived therefrom by the person skilled in the art without departing from the scope of protection of the invention. It is therefore clear that a large number of possible variations exist. It is also clear that embodiments mentioned by way of example represent only examples which are not to be understood in any way as limiting, for example, the scope of protection, the possible applications or the configuration of the invention. Rather, the preceding description and the description of the figures enable the person skilled in the art to implement the exemplary embodiments in concrete terms, wherein the person skilled in the art, knowing the disclosed inventive concept, can make various changes, for example with regard to the function or the arrangement of individual elements mentioned in an exemplary embodiment, without departing from the scope of protection defined by the claims and their legal equivalents, such as further explanations in the description.List of reference characters1 Displacement Detection Unit 3 Calculation Unit S 1 Executing S 2 Identifying S 3 TrainingReferences included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedDE 10 2018 003 129 A1

[0005]

Claims

System for classifying unevennesses in a road surface profile, having a deflection detection unit (1) for determining a temporal profile of a deflection of a spring-damper system of a vehicle during its travel, and having a computing unit (3) which is designed to assign sections in the temporal profile of the deflection to at least one specific information from a finite number of predefined classes of unevennesses by a comparison, based on pattern recognition, of the temporal profile of the deflection with explicit or implicit information stored beforehand about the finite number of classes of unevennesses.The system according to claim 1, further comprising an identification unit configured to transmit information about the vehicle type and / or information about a spring damper system installed on the vehicle to the computing unit (3), wherein the computing unit (3) is a central stationary computing unit (3).System according to Claim 2, furthermore having a speed detection unit for determining a respectively current speed of the vehicle, wherein the arithmetic unit (3) is designed to take into account the respective speed during the pattern recognition.System according to one of the preceding claims, further comprising a locating unit for determining a respectively current position of the vehicle, wherein the computing unit (3) is designed to assign a respective unevenness assigned to a specific class to a respective geoposition on the basis of a position of the vehicle.The system according to any one of the preceding claims, wherein the deflection detection unit (1) is configured to detect a respective time profile of a deflection of a left and a right spring-damper system of a vehicle during its travel.System according to one of the preceding claims, wherein the arithmetic unit (3) is designed to execute a machine learning model for pattern recognition.The system of claim 6, wherein the machine learning model comprises an artificial neural network.The system according to claim 7, wherein the computing unit (3) is configured to preprocess the respective temporal course of the deflection in a transformation in order to obtain a latent space, and to transfer the latent space by means of classification clustering into a probability distribution of the assignment to a respective specific class from the finite number of classes.Method for generating a machine learning model for classifying unevennesses in a road surface profile, having the steps: - carrying out (S1) a test journey using a test vehicle and in the process: optically detecting a road surface and detecting a temporal profile of a deflection of a spring-damper system of the vehicle; - automatically identifying (S2) individual unevennesses of the road surface as belonging to a respectively specific one of a multiplicity of predefined classes of unevennesses by image recognition; - training (S3) the machine learning model with the temporal profile of the deflection as input variable and with the identified unevennesses, with assignment to the profile as output variable;Method according to Claim 9, wherein after the detection of the time profile of the deflection of the spring-damper system of the vehicle, a statistical method is applied in a computer-implemented manner in order to identify predictive regions in the time profile of the deflection, input variables for training the machine learning model being formed only with the predictive regions of the time profile of the deflection and only with unevennesses identified as belonging to the predictive regions.

Images