METHOD FOR AUTOMATIC AUTONOMOUS CONTROL OF A PACKING MACHINE
A method for automatic autonomous control of a packing machine (C) having a position-measuring device (WMS, GPS, 32) for precise detection of the position of the track-building machine in a track, and signal detection by actuators of working assemblies (23, bv, 18, 26) of the packing machine (C). Track ballast data are detected by sensors (23, bv, 18, 26) during the packing and the current track ballast parameters are detected therefrom and stored for a subsequent work pass and analysed by a device for machine learning (17, ML). An analysis of the track ballast state data (EF7, S9, A3) is created on the basis of machine learning methods (ML, 17) and the track ballast parameters are analysed in view of a drop in compression forces that occurs in the longitudinal track direction and work instructions (EF7, S9, A3) for an optimal work approach are ascertained therefrom and stored. In a subsequent work pass, depending on the current position in the track and on the associated work instruction data, the packing machine carries out the work instructions automatically and autonomously.
The invention relates to a method for the automatic autonomous control of a track-building machine having a position-measuring device and precise synchronization to the track, position detection of the working assemblies of the packing machine, with the aid of which the control computer of a packing machine is given positionally accurate work instructions for each sleeper area to be packed and the packing machine carries this out fully automatically and autonomously depending on the current position in the track and the associated work instruction data.
DESCRIPTION OF THE PRIOR ARTMost of the tracks for railroads are designed as ballasted track superstructure. The sleepers lie in the ballast. The wheel forces exerted by the trains running over them cause the ballast to round off, partially break off and wear away. This results in irregular settlements in the ballast and shifts in the lateral and vertical position geometry of the track. Due to the settlements of the ballast bed, errors occur in the longitudinal height, superelevation (in curves), twisting, track and directional position. These errors in turn increase the forces acting on the ballast, which in turn have a destructive effect on the ballast.
If certain comfort limit values or safety limit values of these geometric variables set by the railroad directorates are exceeded, then maintenance work is planned and carried out in a timely manner. Currently, track-building machines are mostly used to correct and rectify these geometric track defects. To control the process, there are measuring systems for recording the current track position for the parameters straightening, lifting, twisting and cross slope. To ensure that the track can be released for operation again after such maintenance work, the permanent way machines are equipped with so-called acceptance measuring systems or acceptance recorder systems. For the quality of the track position after improvement by permanent way machines or other methods, the railroad administrations have defined so-called acceptance tolerances. Acceptance tolerances represent the minimum requirements of the quality of the geometric improvements produced.
In addition to the acceptance tolerances, there are also the safety-relevant release tolerances. These represent the limits that must be complied with so that the processed track can be safely reopened for train traffic. Compliance with these tolerances is verified by the acceptance recorder systems. On track maintenance machines (track packing machines), there is the track car driver who is responsible for controlling the machine with regard to the target geometry and with regard to recording the track position left behind after the maintenance work by the recorder system.
Currently, track maintenance is planned on the basis of the track geometry, which is recorded via the position of the rails. Track measurement vehicles drive over the tracks at regular intervals and record their geometric position. The track position is usually divided into sections of about 200 m in length and the standard deviation of the elevation, direction, superelevation and twist is recorded. In addition to these statistical values, singular individual errors are also measured. If the statistical values exceed certain comfort tolerances then maintenance work is planned and carried out. If the individual faults exceed certain critical values, immediate action is taken to rectify them, otherwise slow running points or track closures will have to be imposed because of the danger to train traffic.
The maintenance machine is given the target track geometry, previously recorded and measured track defects and the area to be maintained. No further specifications are made.
A second operator, i.e. the tamper, is provided for the packing operation. Regardless of the type of defect in the ballast, he usually performs a standard packing. The methodology whether multiple packing, lifting over the positions etc. is left to him.
How exactly the tamper packs the track, e.g. which packing pressure he uses, which packing time, whether he packs the same sleeper several times, whether he works on a spot with a little more lift, whether he chooses a slightly larger opening width, etc. is left to the tamper's assessment, experience and motivation.
Before the use of permanent way machines, the actual track position is measured with various known measuring systems and compared with the target track position. The differences in height and direction are transferred to the packing machines as track correction values with the track set geometry.
There are packing machines specialized in packing of switches (divisible packing units—so called splithead units, additional lifting devices for the branching line, pivotable compacting picks etc.) and packing machines which are preferably built for line packing. Packing machines are known in cyclic but also in continuous working advance. In addition, there are single-sleeper and multi-sleeper packing machines. Multiple sleeper packing machines pack several sleepers at once in one working cycle. However, they can also be used in such a way that only one sleeper is packed.
At present, construction sites are planned on the basis of the track geometry position measured with track measurement cars but irrespective of what caused the track defects, e.g. the ballast may be destroyed, rounded and crushed in the area of a rail joint, similar defects occur on very hard substrates and are referred to as “white spots”. In these places, the driving dynamics cause ballast to be pulverized and these spots are indicated by escaping mineral dust. The ballast can be very damaged if it lies for a long time. A large amount of fines and organic material or soil pressed up from the subsoil may have filled the interstices of the ballast grains. It is known from practice that the track position of such ballast structures cannot be durably corrected with track packing machines. It is also known from practice that individual faults occur randomly distributed in the track. About 40% of these local faults can be permanently corrected. 60% of these faults develop again within a short time. Tracks with a good ballast condition are packed on average about every four years. Individual defects that indicate destruction of the ballast require maintenance measures approximately every 1-3 months. During each packing operation, the ballast is damaged by the packing tools due to the high compression forces. The use of packing machines generates costs for maintenance work and is a hindrance to operation. As long as work is being carried out, no trains can run, and if there is a busy neighboring track, slow speed sections are set up for it. It is also known that a highly contaminated ballast bed requires high compression forces. There is practically no room for movement between the ballast bed grains because they are filled with fine material. This increases the compression forces that have to be applied to move the ballast. At the same time, such contaminated track beds exhibit reduced durability of the corrected track geometry because the frictional forces and interlocking between the ballast bed grains are low. Track geometry defects are usually recorded by independent measurement methods prior to a packing operation, stored, and transferred to the packing machine computers in electronic form. Track geometry faults typically have 10-25 m wavelengths with amplitudes of 10-40 mm. Long wavelength faults in the 25-70 m range also occur and have higher fault amplitudes.
Packing units fix the position of a track during a maintenance measure. This is done by packing tools, so-called packing picks, which plunge into the ballast next to the sleepers and compact the ballast under the sleeper by means of a linear closing movement superimposed by a compression vibration. As standard, the linear closing movement is superimposed by a hydraulic cylinder and the vibration amplitude mechanically generated by an eccentric shaft. Newer fully hydraulic packing drives generate the linear closing motion and the vibration simultaneously.
From WO2019091681 A1 a track-building machine is known, which collects network data and transmits them to a system control center. The track-building machine has a sensor system and collects raw data. From this, it is to be planned when and where operations of the track-building machine are to be performed. During the navigation process, raw data is collected to update the network data, i.e. data such as rebuilds or faults and the like, and not specific ballast parameters recorded during packing. The course of the compression forces cannot be obtained from the collection of network data. A fully hydraulic drive of a packing unit is disclosed in AT 513 973 A, for example. To regulate and control this drive, the adjusting movement is recorded by means of integrated displacement sensors. The packing pressure is measured by pressure sensors. As described in AT 520 117 A, parameters such as compression work, ballast hardness, ballast bed contamination, compression force, compression times and ballast stiffness, etc. can be measured and derived. It is known from AT 515 801 A how optimum packing times can be specified depending on measurements. The opening width of the packing tools can also be freely and continuously adjusted via these fully hydraulic packing drives.
The packing operator is currently responsible for selecting the correct setting of the packing unit, such as packing pressure, packing time, lowering speed of the packing unit, opening width, packing depth, elevation of the track or multiple packing, etc. Further planning of the work, such as the packing work itself, but also preparatory work such as ballast replacement in the area of local disturbances, local drainage improvement, etc., does not take place. This increases the track maintenance costs and reduces the durability of the achieved track position.
Points in the track with high bedding hardness form high points, which change little in elevation due to train traffic. The more different the stiffness variations in the track bed, the greater the force interaction between wheel and rail, the higher the load on the track and the faster the track geometry deteriorates. Singular short faults in the track have the tendency to expand longitudinally under the high dynamic forces acting in the track, to increase in the height of the track fault and to produce consequential faults caused by the excited track vehicles.
Sensors are known that can determine the position of the sleepers in the track when a packing machine passes over them. With the aid of such devices, the machine can be positioned correctly for packing fully automatically. Machines that operate fully automatically are thus known from practice.
The storage of target track geometry data in databases of the infrastructure managers and the possibility of downloading them or returning results is also possible in some cases. Machine learning systems are state of the art. Machine learning is a generic term for computer-aided generation of knowledge from experience. For this purpose, algorithms build a statistical model based on training data. Patterns and regularities are recognized in the learning data. This enables the system to assess even unknown data. With the help of GPS systems installed on packing machines, an exact assignment of the sleepers and the recorded measurement parameters to the track kilometer can be made via the GPS coordinates.
Known are virtual GPS correction data services that send correction data to suitable GPS receivers. This means that only one moving GPS-supported measuring vehicle moving on the track is required. RTK-GPS has the advantage that it can determine the absolute location very precisely (about 5 mm in position and 10-15 mm in height) using RTK correction data. The more satellites and satellite systems are received simultaneously by a GPS receiver, the more accurate the results. Modern satellite receivers simultaneously receive and utilize the GPS, GLONASS, GALILEO, BeiDou, QZSS, IRNSS and SBAS satellite systems. They can send data to the correction service and receive correction data on a second channel. The accuracy in the range of 5-15 mm is too inaccurate for calculating correction values for the track packing machine for uplift or direction, but it is sufficient to define absolute reference points of the track geometry. Sleepers in the track and other places can be located just as precisely and provided with GPS coordinates. These places or sleepers can be precisely and unambiguously located with permanent way machines equipped with an RTK-GPS system.
OBJECT OF THE INVENTIONThe invention is based on the object of providing a method for the automatic autonomous control of a track-building machine which avoids the above-mentioned disadvantages. The method is to supply the track-building machine not only generally with nominal geometry data and track position correction data, but also with exact locally unambiguously assigned work instructions, so that these are autonomously packed with high quality adapted also to the properties and requirements of the ballast bed and thus avoid the error susceptibility by humans. At the same time, the packing machine should record the ballast bed parameters during the work, analyze them with the computer and, at the end of the work, transfer them preferably to an infrastructure operator in preparation for the next pass.
The invention solves the given object with the features of claim 1. Advantageous further developments of the invention are shown in the subclaims. In particular, the invention solves the object in that during packing the ballast bed data is recorded via sensors and from this the current ballast bed parameters are recorded and stored for a subsequent work pass and analyzed with a machine learning device, wherein an analysis of the ballast bed state data is made on the basis of machine learning methods and the ballast bed parameters are analyzed with regard to a drop in the compression forces occurring in the longitudinal direction of the track and work instructions for an optimum working method are ascertained therefrom and stored, wherein, in a subsequent work pass, depending on the current position in the track and the associated work instruction data, the packing machine carries out said work instructions fully automatically and autonomously.
For automatic and autonomous control of the operation of a packing machine and its packing units and lifting-straightening units, the following is provided: A packing machine control computer is given positionally accurate (via GPS coordinates, for example) work instructions for each sleeper area to be packed (this may include: Multiple packing, larger opening width of the packing tools, packing pressure, overlifting, specification of the maximum compression force, packing time, automatic packing time depending on the compression, etc. or specification of the work sequence in switches—at which points, for example, the splithead units are to be split in the case of a switch packing machine and the outer part is to be swung outwards, etc.). These working parameters were recorded in a previous work pass, a complete or partial packing of a track, and stored for a following work pass.
The packing machine is positioned precisely at the sleeper areas to be packed via automatic sleeper detection or GPS coordinates. At the position reached, the packing machine can then carry out this work fully automatically and autonomously, depending on the specified work instructions, generating new work instructions for a next pass if necessary and then moving to the next sleeper area via an automatic travel system, where the sequence is repeated accordingly until the entire intended work area has been worked through.
The predetermined work instructions do not have to be performed fully automatically, but can be displayed to an operator for each sleeper area, with the operator setting and performing the predetermined work modes.
According to the invention, the ballast bed data and work data are recorded during packing with the aid of the fully hydraulic packing drive and its sensors, and the current ballast bed parameters (such as ballast bed hardness, compression force, packing time, penetration time of the packing units, deceleration acceleration of the packing units during the penetration process, current GPS position or track km, current lifting and leveling value, current lifting force and leveling force, etc.) are calculated, stored and analyzed using a machine learning device with machine learning techniques.
According to the invention, a ballast bed state record is generated during the work and displayed to the packing operator or pre-car operator for information, and a ballast state report is generated from the measurement data after the work, both of which are sent to the infrastructure manager as a basis for work preparation for the upcoming packing pass.
If the work instructions have not been handed over by the infrastructure manager or the responsible work planner, then the packing is provided with appropriate instructions for the optimum working method from the analysis of the ballast bed data that runs along with the packing work.
According to the invention, the measurement data of the packing work are analyzed by a rule-based expert system (AI system or other machine learning program) with regard to a sudden drop in the compression forces in the longitudinal direction (individual error) or statistical parameters such as standard deviation, mean value, correlation with the track position level error, etc., and instructions for the optimum mode of operation are determined and specified from this.
Useful and valuable information can be derived from the data obtained with the hydraulic packing drive and its sensors. Algorithms from the field of artificial intelligence (AI) are used for analysis. AI systems are able to find correlations and patterns in differently structured data sets that the human interpreter can hardly or not at all grasp. The AI system is used to make a prediction regarding the occurrence of track deterioration and track faults, and from this to indicate maintenance suggestions that will increase the durability of the track. Other machine learning (ML) techniques (rule-based learning) are also suitable for this purpose.
A rule-based expert system (XPS) can support the operator by providing concrete suggestions. XPS have a great advantage in areas where profound expertise is available for the interpretation of the algorithmic models and data situation.
The following is an example of what such a work instruction could look like. This could be computer-generated by the work scheduler. The list would include all the sleepers to be packed.
The various work instructions can be coordinated and standardized in consultation between the infrastructure manager and the machine operator. The work instructions could mean the following:
EF7—Single fault in the track. Working parameters: 3-fold packing; enlarged opening width (7 5cm); packing time 3×0.65 sec
S9—Highly polluted ballast area. Working parameters: Single packing; packing pressure 180 bar; packing time 1.2 sec.
A3—General default. Working parameters: Limitation of maximum compression force to 35 kN; packing with automatic optimized packing time selection; packing depth 485 mm.
By means of the general default (working parameters to be applied if there is no specific requirement) and the work instructions for special places and their nomenclature, it is easy for the control computer of modern packing machines to interpret and execute the corresponding defined instructions.
In the drawing, the subject matter of the invention is shown by way of example, wherein:
An example of a ballast bed analysis report is shown below.
This is preceded by a statistical evaluation that provides general statements about the processed section. The analysis with machine learning system ML provides statements about the durability of the track position and the ballast bed hardness. If there are any faults 9,10, they are indicated with their type, exact location, length and characteristic values. The transmission of these data to the infrastructure manager or a responsible work scheduler forms the basis for the specification of the work instructions for the next pass. The analysis also gives an estimate of the track deterioration rate which is essential for the timing of the next pass. This data is also easily converted into a machine-readable form and transmitted.
Ballast Bed Report:
Statistical Evaluation
Durability of the track position
The ballast bed has defects. There is a low durability of the track layer.
The estimate results in a track deterioration rate of 1.6 mm/year.
Ballast bed hardness
The mean value of the ballast bed hardness was 254 Nm.
The ballast bed is in borderline highly contaminated condition. Track bed cleaning is recommended. A critical fault (with crushed/rounded ballast was found in the tamped area).
Replacement of the ballast in area 76.580 over 11 sleepers is recommended.
Fault 1
Claims
1. A method for automatic autonomous control of a packing machine having a position-measuring device detecting a position of a track-building machine in a track, with signal detection by actuators of working assemblies of the packing machine, said method comprising:
- acquiring ballast bed data via sensors during packing; and
- acquiring current ballast bed parameters from the ballast bed data;
- storing the current ballast bed parameters for a subsequent work pass; and
- analyzing the current ballast bed parameters with a machine learning device so as to create an analysis of ballast bed state data based on machine learning methods, wherein the current ballast bed parameters are analyzed with regard to a drop in compression forces occurring in a longitudinal direction of the track; and
- determining and storing work instruction data defining work instructions of an optimum mode of operation from the current ballast bed parameters; and
- wherein, in a subsequent work pass, based on a current position in the track and associated work instruction data, the packing machine carries out the work instructions of said work instruction data fully automatically and autonomously.
2. The method according to claim 1, wherein the method further comprises supplying a control computer of the packing machine with positionally accurate work instructions for each sleeper region to be packed, and
- wherein the packing machine moves to a respective longitudinal position in the track fully automatically based on the detected corresponding longitudinal position in the track and the specified work instruction data and autonomously performs work in the sleeper area to be packed based on the work instruction data,
- whereupon the next sleeper area to be processed is accessed via an automatic travel system in accordance with the work instruction data and the cycle of processing the work instruction data and moving to the next sleeper area to be processed is repeated until an intended work area has been processed.
3. The method according to claim 1, wherein the method further comprises displaying predetermined work instructions for each sleeper area to an operator, and the operator sets and executes the predetermined work instructions.
4. The method according to claim 2, wherein the method further comprises generating a ballast bed state record during the work and generating a ballast bed state report with the result of the analysis of the ballast bed data by the machine learning device and transmitting both the ballast bed state record and the ballast bed state report to an infrastructure manager.
5. The method according to claim 1, wherein the analysis of the ballast bed data by the machine learning device, which runs during the packing work, provides an operator with indications for an optimum working procedure.
6. The method according to claim 3, wherein the method further comprises generating a ballast bed state record during the work and generating a ballast bed state report with the result of the analysis of the ballast bed data by the machine learning device and transmitting both the ballast bed state record and the ballast bed state report to an infrastructure manager.
7. The method according to claim 6, wherein the analysis of the ballast bed data by the machine learning device, which runs during the packing work, provides an operator with indications for an optimum working procedure.
8. The method according to claim 2, wherein the analysis of the ballast bed data by the machine learning device, which runs during the packing work, provides an operator with indications for an optimum working procedure.
9. The method according to claim 3, wherein the analysis of the ballast bed data by the machine learning device, which runs during the packing work, provides an operator with indications for an optimum working procedure.
10. The method according to claim 4, wherein the analysis of the ballast bed data by the machine learning device, which runs during the packing work, provides an operator with indications for an optimum working procedure.
Type: Application
Filed: Jun 4, 2021
Publication Date: Jul 20, 2023
Inventor: Bernhard Lichtberger (Wien)
Application Number: 18/008,433
