WO2004101300A1

WO2004101300A1 - Method for detection and monitoring of the movements of vehicles

Info

Publication number: WO2004101300A1
Application number: PCT/DE2004/000976
Authority: WO
Inventors: Jakob Schillinger; Thorsten Gollewski; Dietmar Kohn
Original assignee: Conti Temic Microelectronic Gmbh
Priority date: 2003-05-08
Filing date: 2004-05-10
Publication date: 2004-11-25
Also published as: DE10320809A1

Abstract

The invention relates to a method for the detection and monitoring of the movements of a vehicle, whereby several sensors are mounted on the vehicle and chassis for recording measured values of accelerations, speeds, distances, forces or noises in the chassis and the measured values, as recorded by the sensors, are processed by means of a pattern recognition method, in particular, a method known from speech processing.

Description

METHOD FOR DETECTING AND MONITORING MOTION IN VEHICLES

The invention relates to a method for detecting and monitoring movement in vehicles, according to the preamble of patent claim 1.

So far, the chassis condition has been measured by various different sensors, which record the various parameters in a wide variety of ways. One of the measured variables is determined while driving, the other measured variables are determined when the vehicle is at a standstill. The tire pressure of a tire is detected by a pressure sensor that measures the tire pressure while driving. There are also measurement procedures for test benches that check the tire condition and / or the dampers when the vehicle is stationary. Simple monitoring and control algorithms for damper adjustment, such as the sky hook damper, are also known. The

The condition of the road and thus the bumps in the road can be measured using ultrasound. Various methods are also known to record the weather conditions.

The methods known in connection with chassis diagnostics and controller optimization are based on modeling the chassis using differential equations and embedding them in condition observers. These methods require considerable computational and metrological effort to assess the condition of the entire vehicle, so that these methods only for reduced Practical requirements have been incorporated. The following references and patent specifications are listed by way of example: DE3608420C2, DE4218087A1, DE4019501A1, DE3883212T2, DE19909157A1, DE4213221A1, DE4235809C1, DE19543137A1, DE4440413A1, DE4431794A1, DE44354A1.

In the publication: "Dynamics of Motor Vehicles" by Manfred Mitschke, the third edition in Volume B describes and analyzes vehicle vibrations that a four-wheeled vehicle generates, which drives over an uneven road surface and thereby represents a vibration system. It shows what influence the various parameters have on the vehicle vibration system. In this book, the parameters (e.g. tire pressure, wheel suspension, damper bumps, engine excitation, engine speed) that influence the vibration system are presented. Vibration equations are set up depending on these parameters. The solution or solutions to these vibration equations are accelerations, relative distances and forces, which are then used to assess the behavior of the chassis and thus to determine comfort and driving safety.

The disadvantage of these known methods for determining the chassis condition is the high costs which are caused by the direct sensing of the measurement parameters. Another disadvantage arises from the sometimes very time-consuming and costly calculation of complicated vibration systems and the large amounts of data that have to be processed within a very short time. Another unfavorable circumstance is that multiple use of the sensor signals is not possible.

DE 4218087 A1 discloses a method for regulating the damping of the chassis of a motor vehicle and / or for diagnosing the chassis. Here, process variables that are related to a vertical movement of the vehicle are fed to a mathematical model that determines the relationship between the vertical acceleration of a wheel or a structure of the vehicle Vehicle on the one hand and other process variables and parameters on the other. The parameters of the model are estimated using a parameter estimation method known per se and used to generate a manipulated variable for setting the damping and / or as diagnostic data.

Another disadvantage here is the complex calculation. First, several equations are set up that describe the physical relationships between the measured values and the parameters. These equations contain several unknown parameters. Since the number of equations is not sufficient to calculate each unknown parameter exactly, a parameter estimation method is used. The elaborately estimated parameters are then used in turn to calculate the physical process variables that are required for the control or diagnosis of the chassis.

EP 0554131 A1 discloses methods for sensing the tire condition by means of a sensor attached to the tire. The sensors record the speed of rotation of each tire. Another sensor detects the speed or acceleration of the vehicle. By comparing these values with each other, conclusions can be drawn about the tire pressure.

In addition to the disadvantages mentioned above, the disadvantage here is that a large number of sensors are required to detect a state.

The object of the invention is to monitor the chassis condition, in particular the tire pressure, the wheel and tire condition, the damper condition and the damper settings, the condition of the roadway or rail, and the body movement quickly, reliably and inexpensively. Another object of the invention is to optimize the setting parameters of the damping system in the field of road, rail and aircraft. The object is achieved by the features in claim 1. Here, sensors for detecting accelerations, speeds, distances and / or forces are attached to the chassis and the body. The sensor signals are evaluated using a parameter estimation method. The parameters obtained from this are evaluated with the aid of methods for pattern recognition, the fuzzi techniques and / or neural networks.This means that the causes are not recorded, which cause movements on the chassis and the movements of the vehicle are calculated using vibration equations, whereby the The result obtained is used to assess the chassis condition. The chassis state is formed from the individual, different measured values. These vectors, which not only reflect individual measured values themselves, but also the relationship of several measured values to one another, can then be compared with stored vectors or vector fields to which a vehicle state is assigned.

The advantage of the invention is that, with a small amount of data compared to previous methods, very reliable statements can be made about the chassis status, since the different constellations of the measured values relative to one another also have meaningfulness about the vehicle status. The data processing for previous methods is also very low, since no complicated calculations are required, but only result vectors are compared with one another. The evaluation takes place very quickly, so that in the event of malfunctions, the system can be intervened at an early stage to prevent major damage, for example. Cost-effective multiple use of the sensors used is also an advantage, which increases safety, the early detection of dangerous conditions (eg aging or loss of pressure in the tires) and improves driving comfort. This also forms a way of recognizing the road condition as an input signal for intelligent braking algorithms. The presented method is now based on a completely different approach: The sensor signals are processed without direct reference to mathematical or physical models using the speech recognition method and the comparison of the quantized states is used to determine the condition of the road surface and the overall chassis as well as individual components closed.

In a known manner, sensors are attached to suitable points of the vehicle, which directly or indirectly record the required measured variables. Sensors attached in the area of the wheel hub, the subframe or the damper dome primarily detect the excitation which, generated by the road surface, acts on the body via the tires and spring / damper strut. Acceleration sensors are preferably used here, but sensors for speed, structure-borne noise, pressure, force, stroke, travel or angle can also be used. The number and spatial distribution depends on the requirements. The minimum equipment is one sensor per vehicle, the maximum equipment is one sensor per measurement variable per wheel. In addition, at least one sensor can be attached to the chassis to detect the excitation that acts on the body via the chassis. With the help of motion sensors, such as acceleration (preferably), speed, pressure, travel, stroke, angle, position, yaw rate, pitching motion or rolling motion, a measure of the load or the adjustment of the dampers can be obtained from the movement of the chassis , With the help of acoustic sensors, such as a microphone or structure-borne sound sensor, vibrations of the chassis, which result from brakes, mechanical bearings, tire tread, stone chips or splash water, are recorded. Critical brake conditions (wear, rubbing, squeaking) can be determined and evaluated triggered by the brake application. An acoustic sensor, attached to or near the wheel arch, can use voice analysis to identify critical road surface conditions, such as roll split or water film. Certain measured variables can be recorded in an amplified or weakened manner via the spatial arrangement of the sensor or sensors. These measuring principles are known and described in the specialist literature. In conventional practice, one or more sensors are assigned to each measurement task. The algorithms, based on digital filters, statistical methods and / or RMS calculations, are tailored to the measurement task.

The new approach assumes multiple use of the sensors using speech recognition methods. This results in a significant cost, space, u. Weight advantage. The technical effort for speech recognition remains within reasonable limits.

The application of the methods of speech recognition for the assessment of the chassis and the road is based on the idea that time and. Set quasi-steady state sections at the sensor outputs. The sequence of quasi-steady-state driving conditions results in sections of steady-state signal signatures on the sensors, comparable to the sequence of sounds within a word, each sound corresponding to such a quasi-stationary driving state.

The driving noise differs depending on whether you drive with summer or winter tires on cobblestone or paved roads. These four different sounds can be assigned to four different class names (= sounds). The number of different class names has to be determined application-specifically. According to the patent claim, the output signals of the sensors are processed with the methods of speech recognition as they are applied to a microphone signal in a known manner.

A general description of speech recognition can be found in the literature at:

Schukat-Talamazzini: Automatic speech recognition, Vieweg Verlag 1995 Quartieri, Thomas: Speech Signal Processing, Prentice Hall 2001.

Zeil, A .: Simulation of Neural Networks, Addison Verlag 1994.

The transfer of these procedures into the proposed application of the evaluation of the chassis and the condition of the road takes place in these three layers:

Layers 1 and 2 are sensor-specific, layer 3 represents the overall context with regard to the vehicle.

The sensors are preferably used optionally as sensors for the measured variables: acceleration, structure-borne noise, speed, force, pressure, stroke, path, angle, distance, pitching movement or rolling movement.

The methods of linear prediction analysis, cepstrum analysis, short time fourier analysis, filter bank analysis, sinusoidal analysis or frequency domain pitch estimation are preferably used for the feature extraction.

The estimation vectors for defined class names are preferably mapped using methods of neural networks, fuzzy logic, assignment tables, linear discriminant analysis or hidden Markov models.

The measurement values are preferably sampled both equidistant in time and equidistant in distance.

Monitoring and detection using a classifier is preferably carried out on the basis of the methods of neural networks, fuzzy logic, assignment tables, linear discriminant analysis or hidden Markov models, which carries out the overall evaluation from the comparison of the class names for each measurement channel. An orthogonal estimation method is preferably used for the feature generation.

The estimation vectors are preferably calculated using AR-Lattice (autoregressive), ARMA-Lattice (autoregressive moving average) or joint lattice / ladder methods.

The sensor signals, the estimation vectors, the class names and / or the overall evaluation via bus coupling are preferably passed on to other diagnostic, reporting or control systems or generated in at least one of these systems and a defined action is started there.

A value for

- the condition of the road and / or

- the tire condition and / or

- the damper condition and / or

- the brake condition and / or

- determines the load.

Preferably, driving over the edge strip is also recognized.

The invention can be used for road vehicles, rail vehicles and aircraft.

The functionality of individual sensors can also preferably be inferred from the comparison of the estimation vectors.

In addition, the results are preferably stored and the aging condition is evaluated from the time comparison. The invention is to be illustrated using exemplary embodiments and the figures.

Figure 1 shows the model for the detection and monitoring of movements on vehicles. In this exemplary embodiment, it is a vehicle with four wheels, four dampers, a chassis and five sensors, which generally measure accelerations. However, speed, distance, length, force or noise sensors can also be used. Essentially two transmission systems are shown in the upper part of the figure. One transmission system is the unit of road and wheel. The other transmission system is represented by dampers and chassis. Both transmission systems are interconnected. Four chassis sensors - one on each subframe - are arranged between the transmission systems roadway / wheel and damper / chassis. The fifth sensor, the chassis sensor, is located on the chassis. The transmission systems change a signal WR present at the input of the transmission systems to change the white noise. In the first step of the method, measurement signals are recorded at suitable points, which were selected on the basis of empirical values.

The second stage of this model illustration shows five inputs for the sensor signals. Here, the measurement results of the four chassis sensors and the chassis sensor are fed into the pre-filter.

After the pre-filter, the values together with the white noise are fed to one or more estimation methods.

Each sensor delivers series of measurements. Measurement series are a number of measurement results that are recorded during a time period. The components of vectors are calculated from the measurement series of a sensor using an orthogonal estimation method. There are three estimation vectors for each chassis sensor and two estimation vectors for the chassis sensor. It is fundamental to the exemplary embodiment that the dynamic behavior of the mechanical elements can be modeled with sufficient accuracy using a sectionally linear transmission system. Orthogonal estimation methods are used in speech recognition.

These properties can also be used to describe a driving state of a vehicle. Depending on the structure of the transmission system, the AR (autoregressive, ARMA (autoregressive moving average) or joint ladder method can be selected. These estimation methods form at least two vectors from a measurement series of a sensor, the number of elements of which depends on the order of the system, which The system sensor is a series of measurements consisting of the time-dependent measured values: rτiι (tι),

generate m ₃ (t3), ... m _n (t _n ). If no changes occur during the journey, the following applies in the steady state: mι (tι) = m ₂ (t ₂ ) = m ₃ (t ₃ ) = ... = nrin (tn). Otherwise, the values may differ more or less depending on the event. Two vectors K _c and K _{d are} calculated from this number of measured values, these vectors each having μ elements with:

The number μ of elements depends on the change in the measured values over time.

The values of k _c ι ... k _{c //} and k _d i — k ^ depend on the one hand on the individual measured values and on the relation of the measured values to one another. For example:

With this method, new series of measurements with the current measured values are continuously formed and new vectors are calculated from them.

In the exemplary embodiment, two vectors are then formed from the measured values of the chassis sensor using the AR estimation method, as described above, and from the measured value series:

of the chassis sensors F1, F2, F3, F4 and the chassis sensor, three estimation vectors K _f , K _b , K _{g are} formed per chassis sensor according to the joint ladder estimation method

In total, 14 vectors are obtained in the application example:

Fl - Fl _f K? .. KΓ. F2, K _b , K. κ * iF2 κ ^' -F _f 3, κ r _b F3, κ _e , κ - F _f 4, κ - _b F4 K - Fi K c' d ^•

On the one hand, these vectors are compared with one another and on the other hand with stored values, element by element, and then they are classified. For the classification, it is necessary that values, ranges of values, and possibly relations of the elements of different vectors with one another, are already stored in a memory for the most varied driving situations and conditions, which are reflected in vectors and there in particular in the elements.

After the classification, the results are fed to a map. The map in turn influences the prefilter.

The estimation vectors are converted into physical quantities. Functions result from the conversion. A function H (z) = B (z) / A (z) results from the joint ladder estimation of a chassis sensor.

After the conversion, the AR estimate of the chassis sensor results in a function H (z) = 1 / E (z). The physical values then result, which are then fed to control loops, diagnostic systems or driver information systems in the third stage of the model in order to influence an undesired driving state, to recognize errors at an early stage or at least to report errors.

Figure 2 describes the course of the estimation. In the system illustration, the white noise is fed into the function 1 / A (z). The chassis sensor detects what this function or the transmission system does with the white noise. On the one hand, this series of measured values reaches a pre-filter, and on the other hand it becomes another Transmission system or function B (z) fed. The chassis sensor detects what the second transmission system does with the white noise. These series of measured values go into two different pre-filters, if necessary. In this example, three calculation strings are formed, which result from two different measurement series. Two of the calculation strings are linked to each other and influence each other by using the value or vector from one strand as the basis for calculating the value or vector in the other strand. In one and the other independent calculation strand, two processes are started after the pre-filter , A first value arrives at point T and then in a first circle. Between T and the first circle, this branches to a second circle. The first value also goes directly to the second circle and branches back to the first circle

A diagnostic table is shown in FIG. If you want to detect a fault on the wheel at the front right on the chassis sensor F1, you have to, for example,

of the corresponding vector K ^F1 with the corresponding elements k ² k _f ^F lk, which were recorded by the other chassis sensors F2, F3, F4. Is now

it can be concluded from this that there is a different condition on the wheel, for example insufficient air pressure.

Fig. 4 outlines a modified evaluation.

5 illustrates the two-stage basic principle of parameter estimation and pattern recognition, as it is ultimately used in speech processing, here applied to the analysis of the vehicle and road condition.

The transfer of a pattern recognition process, as described in the

Speech recognition has so far been used in the proposed application of the assessment of the chassis and the Road surface quality takes place in these three layers and is visualized in FIG. 6.

Layer 1 is used to generate features. The measurement signals are sampled at time and / or path equidistant and processed separately. For the generation of features, the known methods are optionally used on these samples without any claim to completeness: Linear Prediction Analysis (LPC), Cepstrum Analysis, Short Time Fourier Analysis, Filter Bank Analysis, Sinusoidal Analysis, Frequency-Domain Pitch Estimation, combinations thereof or further developments. Methods that lead to orthogonal estimation vectors are advantageous.

The LPC method estimates the coefficients of the equivalent transfer function. The order of the transfer function is determined by the underlying physical model. Ladder / lattice processes are preferably used here. There are different approaches for modeling the transfer function using LPC:

The AR (autoregressive) lattice method is used when a pulse train or white noise has to be used as the input signal of the transmission system to be modeled. Because of the generality of the approach, this method can be applied to any sensor output. The estimation vector describes the coefficients of the approximated AR model.

ARMA (autoregressive moving average) - Lattice methods require two measurement signals that can be interpreted as the input and output of a transfer function. The estimate vector describes the coefficients of the ARMA transfer function between these two measurement channels.

Joint ladder / lattice methods can be used to advantage if the signal on channel 1 using AR lattice estimation and the Transmission path between channel 1 and channel 2 can be simulated using an MA model. For example, the signal from the chassis sensor can be connected to channel 1 and the signal from the chassis sensor to channel 2. In this case, the AR estimate vector describes the entire transmission path between road and tire, and the MA estimate vector describes the spring / damper transmission path.

Filter Bank Analysis determines the RMS values of the output signals of a series of bandpass filters with specific passband ranges that are adapted to the application. Cepstrum analysis and the different Fourier analysis methods determine the weighted spectral components in the recorded sensor signal.

The sample values as well as the estimation vectors can be forwarded to higher-level control loops if required.

The arithmetic operations usually result in an infinite number of different estimation vectors. Therefore, in the further embodiment, the infinite number of solutions are mapped to a finite number of class names in the second layer using the classification methods known from speech recognition. In speech recognition, this corresponds to the assignment to sounds. The best-known classification methods are not exhaustive: Neural networks, fuzzy logic, linear discriminant analysis, assignment tables or hidden Markov models. The number of class names ("sounds") is determined by the desired classification depth. This in turn is determined by the downstream diagnosis and control systems. This means that both the estimation vectors and the quantized class names, to which controller parameters are assigned in a table, are available The classification takes place as part of a learning process, whereby both simulation methods and driver testing are used.

In the further embodiment, these class names of the individual sensors are transferred to the third layer. This takes place in this layer mutual comparison of class names. Here, too, the classification methods of speech recognition are used (without claiming to be complete): Neural networks, fuzzy logic, linear discriminant analysis, assignment tables or hidden Markov models.

For example, if all sensors attached to the chassis report the same status (= same class name, "= loud"), this information about the road surface and tires can be passed on to the higher-level control system as a current estimate. The class name of a wheel sensor signal differs significantly from the class name of the other wheel sensor signals can be concluded that there is a defect on the single wheel (imbalance, carcass tear, insufficient tire pressure, different tire profile, etc.). The fact that the class names are derived from the path-equidistant scanning is helpful in distinguishing between the condition of the road surface and the tire For which tire properties are independent of the wheel speed, this makes it possible, for example, to clearly distinguish between tire imbalance and a wavy road surface, in which the class names change depending on the vehicle speed, with the former there being essentially no influence Another criterion relates to the speed at which the class names change: sizes dependent on the road surface or tire pressure can change very briefly. Wear-dependent variables, such as tire tread depth, or damping coefficient usually change very slowly. Trend tracking of class names or estimation vectors is required here.

If the class names change in pairs (left vehicle half: VL and HL versus right vehicle half: VR and HR), it can be concluded that the vehicle is changing lanes or driving along the verge, whereby the different class names can be used to determine which side of the vehicle is leaving the lane Has.

In conjunction with a soft-set damper, the vehicle's load can be deduced from the vehicle's own movement, detected by a suitable chassis sensor. In order to avoid evaluation errors, it is advantageous to take into account the excitation frequency of the undercarriage.

Acoustic sensors, attached to the chassis or chassis, are particularly suitable for signal analysis using speech recognition. With this, the profile properties of the tires, which influence the braking process, can be detected, bearing damage due to the characteristic spectral components can be detected and damage to the brake can be recognized.

Claims

Patentansprüehe

1. A method for detecting and monitoring the movement of a vehicle on which a plurality of sensors for detecting measured values of accelerations, speeds, distances, forces or noises are attached to the chassis and the body, characterized in that the measured values detected by the sensors are carried out using a pattern recognition method to be edited.

2. The method according to claim 1, characterized in that the measured values detected by the sensors are processed using a pattern recognition method known from speech processing.

3. The method according to claim 1 or 2, characterized in that the measured values detected by the sensors are processed using an estimation method, wherein - several measured values are acquired over time from each sensor and then, - numerical values are calculated from the measured values of one sensor which form the components of different estimation vectors and - the estimation vectors are compared and classified with other vectors, which are stored in a storage unit and which are each assigned a defined driving state.

4. The method according to any one of the preceding claims, characterized in that the components of an estimation vector are calculated using an orthogonal estimation method, in particular the ladder or lattice method.

5. The method according to any one of the preceding claims, characterized in that the components of an estimation vector are estimated using autoregressive, moving average or joint ladder methods.

6. The method according to any one of the preceding claims, characterized in that the measured values for calculating the

Components of an estimation vector are repeated at defined time intervals.

7. The method according to any one of the preceding claims, characterized in that the measured values for calculating the components of an estimation vector after covering defined

Distances are repeated.

8. Structure for a method according to one of the preceding claims, characterized in that at least one sensor is arranged on each subframe and on the chassis, the measurements of which are used to calculate the components of the vector.