JP7443271B2 - working machine - Google Patents

Description

TECHNICAL FIELD The present invention relates to working machines.

In information-based construction and construction automation using work machines, it is extremely important to grasp the ever-changing topographical information of the site. In addition, improving the accuracy of topographical information is essential as it affects construction accuracy at the work site.

Conventional techniques related to calculation of terrain information in such working machines include those described in Patent Documents 1 and 2, for example. Patent Document 1 discloses an object detection section that is attached to a working machine, detects an object, and outputs information about the object, and uses the information of the object detected by the object detection section to detect the three-dimensional image of the object. A construction management system is disclosed that includes a shape detection unit that outputs shape information representing an original shape, and a communication unit that transmits information regarding construction including at least one of the shape information and information obtained from the shape information. has been done.

Further, Patent Document 2 discloses a selection unit that selects a target image pair to be processed from among image pairs captured by an imaging device provided in a working machine, and a selection unit that selects a target image pair to be processed, and a selection unit that selects a target image pair to be processed, and A construction management device is disclosed that includes a stereo measurement unit that generates three-dimensional data representing the topography around the working machine from the selected pair of target images using camera parameters related to the imaging device.

In addition, as a conventional technique related to speeding up arithmetic processing, for example, Patent Document 3 describes the delay time of a radio frequency signal output from a transmitting device of an active system (a system that is normally used) based on a predetermined signal. a detection unit that detects a delay time of a radio frequency signal output from a transmission device of a protection system (a system used as a protection system); Adjusting the signal delay time in the working transmitting device or the signal delay time in the backup transmitting device so that the difference between the delay time and the delay time of the radio frequency signal output from the transmitting device in the active system is reduced. A delay time adjustment device is disclosed that includes an adjustment section that performs the following steps.

JP 2017-71915 Publication JP2019-49121A JP2017-22658A

In the conventional technology described in Patent Document 3, a plurality of arithmetic units perform arithmetic processing in parallel on a task-by-task basis. However, when there is a large imbalance in the amount of calculation for each task, such as the calculation of topographical information performed by a work machine at a work site, the conventional method of performing calculations in parallel for each task cannot achieve high-speed calculation processing. It was not possible to fully implement the

The present invention has been made in view of the above, and an object of the present invention is to provide a working machine that can perform calculations based on a large amount of information at higher speed.

The present application includes a plurality of means for solving the above problems, and one example is a work machine that performs work at a work site and measures topographical information of the work site , which has a single master calculation device. , a plurality of slave arithmetic devices, and a storage device, and based on an information group consisting of a plurality of pieces of information stored in the storage device, the one master arithmetic device and the plurality of slave arithmetic devices a control system that performs arithmetic processing of topographical information on the work site; In response to a request for execution of arithmetic processing based on the information group, the information group is divided into a plurality of small information groups based on a predetermined division rule, and execution of the arithmetic processing based on the small information group is performed on the plurality of small information groups. Each of them is set in any one of the plurality of slave processing devices, and the plurality of slave processing devices respectively execute arithmetic processing based on the small information group set in the master processing device, and perform the calculation together with the result of the arithmetic processing. Information on the calculation load required to execute the process is returned to the master calculation device, and the master calculation device integrates the calculation results returned from the plurality of slave calculation devices, and uses the integrated calculation results as the information group. The information group is returned as the calculation result of the information group to the slave calculation device that has sent the execution request for the calculation processing based on When changing the division rule so that the calculation load is uniform, the average calculation load, which is the calculation load per slave calculation device, is calculated, and the small information group is assigned to each of the plurality of slave calculation devices. When the calculation processing based on the small information group is assigned, the current total calculation load and the total calculation load before the calculation processing based on the small information group are compared to see which is closer to the average calculation load, and based on the comparison result, Therefore, the above-mentioned division rule shall be changed .

According to the present invention, calculations based on a large amount of information can be performed faster.

1 is a diagram schematically showing the appearance of a hydraulic excavator that is an example of a working machine. FIG. 1 is a diagram schematically showing the overall configuration of a control system according to a first embodiment. 3 is a flowchart illustrating processing contents of an object detection application of a slave CPU. 7 is a flowchart showing the processing contents of normal vector calculation processing in the object detection application of the slave CPU. FIG. 7 is a diagram illustrating an example of a processing request for distribution processing that is transmitted in the request transmission processing. FIG. 7 is a diagram illustrating an example of a processing instruction request for arithmetic processing that is transmitted in processing instruction request transmission processing. FIG. 7 is a diagram illustrating an example of a processing command response transmitted in a processing command response transmission process. It is a flowchart which shows the processing content of the processing distribution part of the topography management part of master CPU. 12 is a flowchart showing the processing content of processing distribution processing in the processing distribution section of the topography management section of the master CPU. 12 is a flowchart showing the processing contents of the normal vector calculation execution unit of the terrain processing unit of the slave CPU. It is a flowchart which shows the processing content of the processing result collection part of the topography management part of master CPU. 12 is a flowchart illustrating the processing content of processing distribution processing in the processing distribution section of the topography management section of the master CPU according to the second embodiment. 12 is a flowchart showing processing contents of a processing result collection unit of a topography management unit of a master CPU according to a second embodiment. FIG. 3 is a diagram schematically showing the overall configuration of a control system according to a third embodiment. 12 is a flowchart showing the processing contents of the processing distribution section of the terrain management section of the master CPU according to the third embodiment. 3 is a flowchart showing the processing contents of database partial synchronization processing.

Embodiments of the present invention will be described below with reference to the drawings. Note that in this embodiment, a hydraulic excavator equipped with a front work machine will be described as an example of a work machine, but the present invention is not limited to this, and the present invention can be applied to other work machines that operate at a work site. It is also possible.

<First embodiment>
A first embodiment of the present invention will be described in detail with reference to FIGS. 1 to 8.

FIG. 1 is a diagram schematically showing the appearance of a hydraulic excavator, which is an example of a working machine according to the present embodiment.

In FIG. 1, a hydraulic excavator 101 includes an articulated front working machine 110 configured by connecting a plurality of driven members (boom 111, arm 112, bucket 113) that rotate in the vertical direction, and a vehicle body 130. The upper rotating body 131 and the lower traveling body 132 are provided to be rotatable with respect to the lower traveling body 132. Further, the base end of the boom 111 of the front work equipment 110 is supported vertically rotatably at the front part of the upper revolving body 131, and one end of the arm 112 is connected to a different end from the base end of the boom 111 (the tip ), and a bucket 113 is supported at the other end of the arm 112 so as to be rotatable in the vertical direction. The boom 111, the arm 112, the bucket 113, the upper revolving body 131, and the lower traveling body 132 include a boom cylinder 121, an arm cylinder 122, a bucket cylinder 123, a swing motor 124, and left and right travel motors 125, 126 (hydraulic actuators). In FIG. 1, only one on the right side is shown, and the other one is driven by only the reference numerals shown in parentheses).

A driver's cab 151 in which an operator rides is arranged at the front of the upper part of the revolving upper structure 131. The operator's cab 151 has a plurality of operating levers (not shown) that output operating signals for operating the hydraulic actuators 121 to 126, functions as a display device for notifying the operator of information, and displays information for the operator. A monitor (not shown), etc., which functions as an input device that allows input of information, is arranged. For example, a touch panel is formed on the surface of the monitor screen, and the function of this touch panel allows input from an operator to be received.

Although not shown, the operation of the boom cylinder 121, arm cylinder 122, bucket cylinder 123, swing motor 124, and left and right travel motors 125, 126 is controlled by a hydraulic pump driven by a prime mover such as an engine from each hydraulic actuator 121 to 126. This is done by controlling the direction and flow rate of hydraulic oil supplied to 126 with a control valve. The control valve is controlled by a drive signal (pilot pressure) output from the pilot pump via the pilot valve. The operation of each hydraulic actuator 121 to 126 is controlled by controlling the pilot valve with the control system based on the operation signal from the operation lever.

At the rear of the upper part of the upper revolving body 131, two GNSS (Global Navigation Satellite System) receivers 171 and 172 are arranged side by side. Hereinafter, when distinguishing between these GNSS receivers 171 and 172, the one placed on the left side will be referred to as the left GNSS receiver 171, and the one placed on the right side will be referred to as the right GNSS receiver 172. The GNSS receivers 171 and 172 receive positioning signals output from GNSS satellites flying far above, calculate the position of the hydraulic excavator 101 in the earth coordinate system based on the received positioning signals, and provide the position information as position information. It has a function to output. Furthermore, since the relative positions of the GNSS receivers 171 and 172 with respect to the upper rotating body 131 are fixed, the orientation of the upper rotating body 131 can be calculated from the deviation of the position information measured by the two GNSS receivers 171 and 172. I can do it.

Here, the GNSS receivers 171 and 172 constitute a position measuring device that measures the position of the hydraulic excavator 101, which is a working machine, at a work site and outputs the measurement result as position information.

Inertial measurement units (IMUs) 181 to 184 are arranged on the boom 111, the arm 112, the bucket 113, and the upper revolving body 131, respectively. Hereinafter, when it is necessary to distinguish these inertial measuring devices 181 to 184, they will be referred to as a boom inertial measuring device 181, an arm inertial measuring device 182, a bucket inertial measuring device 183, and an upper revolving body inertial measuring device 184, respectively. .

The inertial measurement devices 181 to 184 measure angular velocity and acceleration. Considering the case where the driven members 111 to 113 on which the inertial measurement devices 181 to 184 are arranged and the upper rotating body 131 are stationary, the direction of the gravitational acceleration in the IMU coordinate system set to the inertial measurement devices 181 to 184 ( In other words, based on the vertical downward direction) and the mounting state of the inertial measuring devices 181 to 184 (that is, the relative positional relationship between the inertial measuring devices 181 to 184 and the driven members 111 to 113 and the upper rotating body 131). , the orientations (ground angles) of the driven members 111 to 113 and the upper rotating body 131 can be detected as attitude information. Then, the relative angles of the driven members 111 to 113 and the upper rotating body 131 can be calculated from the detection results of the inertial measurement devices 181 to 184. Further, the rotation angle of the upper rotating body 131 can be detected based on the angular velocity detected by the inertial measurement device 184, and the inclination angle of the upper rotating body 131 can be detected based on the direction of the gravitational acceleration. That is, based on the detection results of the inertial measurement devices 181 to 184, the posture of the working machine in the vehicle body coordinate system fixedly set with respect to the upper revolving body 131 (for example, the position of each part constituting the hydraulic excavator 101, the position of the bucket, 113 toe position, etc.).

A terrain measuring device 100 that measures the terrain around the hydraulic excavator 101 and outputs the measurement results as terrain data is disposed in the upper revolving body 131, for example, above the driver's cab 151. The terrain measuring device 100 is, for example, an external sensor such as a stereo camera, a laser scanner, or a millimeter wave radar. The terrain data output from the terrain measuring device 100 is, for example, point group data indicating the surface position of the terrain or object in a vehicle body coordinate system that is fixedly set with respect to the upper revolving body 131. For example, the terrain measuring device 100 automatically measures the terrain around the hydraulic excavator 101 periodically and outputs it as terrain data. Note that the mounting position of the terrain measuring device 100 is not limited to the above, and it may be placed in a position and posture that allows the terrain around the hydraulic excavator 101 to be sufficiently measured. Note that position information (coordinates) in the earth coordinate system, the IMU coordinate system, and the vehicle body coordinate system can be mutually converted between the coordinate systems.

FIG. 2 is a diagram schematically showing the overall configuration of the control system.

In FIG. 2, the control system includes one master CPU (master processing unit) 10, a plurality of slave CPUs (slave processing units) 20, 30, and a database server (storage device) 40. The operation of the hydraulic excavator 101 is controlled using the calculation results by the master CPU 10 and the plurality of slave CPUs 20 and 30 based on the stored information. Note that in this embodiment, only two of the plurality of slave CPUs are illustrated and explained.

The master CPU 10 (not shown) includes a CPU (Central Processing Unit) as a processing device, as well as a storage device (for example, ROM) that stores applications executed by the processing device and setting information necessary for executing the applications. , semiconductor memory such as RAM), which is equivalent to a computer. Similarly, each of the plurality of slave CPUs 20 and 30 is an application equivalent to a computer, which has a storage device in which, in addition to a CPU serving as a processing device, an application executed by the processing device and setting information necessary for executing the application are stored. It is.

The database server 40 is composed of, for example, a hard disk drive or a large-capacity flash memory, and stores information ( (described later) etc. are stored.

The master CPU 10, slave CPUs 20, 30, and database server 40 of the control system are connected by an interface (not shown), and can exchange signals with each other at high speed. It is assumed that communication between each of the CPUs 10 to 30 uses, for example, the TCP/IP protocol. The control system is also connected to a monitor in the driver's cab 151 and the drive system of the hydraulic actuators through an interface (not shown), and displays information on the monitor based on calculation results and controls the operation of the hydraulic actuators 121 to 126. control.

Here, the configuration and arithmetic processing of the control system will be explained in more detail. In this embodiment, a case where the application 23 is executed on the slave CPU 20 will be described as an example. The application 23 is, for example, an object detection application, which calculates the normal vector of each point in the point cloud information stored in the database server 40 and detects a set of points whose normal vectors have continuous directions as an object. It is something to do.

The master CPU 10 is a functional unit that is in charge of managing processing related to the topography of the work site, such as calculation processing related to the illustrated object detection application, and is a topography management unit 11 that is realized by an application executed on the master CPU 10. and one or more other applications 15. The terrain management unit 11 also includes a processing distribution unit 12, an information storage unit 13, and a processing result collection realized by an application executed on the master CPU 10, as functional units in charge of processing in the terrain management unit 11. It has a section 14.

The slave CPU 20 is a functional unit that performs processing related to the topography of the work site, such as calculation processing related to the illustrated object detection application, and includes a topography processing unit 21 realized by an application executed on the slave CPU 20, and a plurality of It is composed of other applications 23 and 24. Additionally, the terrain processing unit 21 serves as a functional unit in charge of executing calculation processing (for example, normal vector calculation processing) in the terrain processing unit 21, and executes normal vector calculation realized by an application executed on the slave CPU 20. It has a section 22. In this embodiment, a case will be exemplified in which the application 23 of the slave CPU 20 is an object detection application.

Similarly, the slave CPU 30 is a functional unit that performs processing related to the topography of the work site, such as calculation processing related to the illustrated object detection application, and includes a topography processing unit 31 realized by an application executed on the slave CPU 30. , and a plurality of other applications 33 and 34. Additionally, the terrain processing unit 31 serves as a functional unit in charge of executing calculation processing (for example, normal vector calculation processing) in the terrain processing unit 31, and executes normal vector calculation realized by an application executed on the slave CPU 30. It has a section 32.

In the control system of this embodiment configured as described above, the application 23 (object detection application) of the slave CPU 20 has a plurality of arithmetic processes (normal vector arithmetic processes to be described later) that should be processed in parallel. When the application 23 (object detection application) is started on the slave CPU 20, the application 23 first distributes processing to the processing distribution unit 12 of the terrain management unit 11 of the master CPU 10 in order to process multiple calculation processes in parallel. Send a processing request request. When the processing distribution unit 12 receives a processing request for distribution processing from the slave CPU 20, it distributes the calculation processing to each slave CPU 20, 30 based on a predetermined distribution processing procedure, and the terrain processing unit 21 of the slave CPU 20, 30 , 31 to send a processing instruction request for arithmetic processing. In addition, in the distribution process, the processing distribution unit 12 acquires the amount of processing calculation (computation load) stored in the database server 40, and performs the distribution process based on this amount of processing calculation. When the terrain processing units 21 and 31 of the slave CPUs 20 and 30 each receive a processing instruction request for arithmetic processing, they execute the arithmetic processing based on the information of the processing instruction request, and calculate the arithmetic result and the computational load required for the arithmetic processing. and is transmitted to the processing result collection unit 14 of the topography management unit 11 of the master CPU 10. The processing result collection unit 14 receives and combines the calculation results from all the slave CPUs 20 and 30 that have sent the processing instruction request, and sends the processing request request as the calculation result of the plurality of calculation processes to be processed in parallel. Send to application 23 (object detection application). Furthermore, the processing result collection unit 14 updates the amount of processing calculation (computation load) stored in the database server 40 based on the amount of processing calculation from all the slave CPUs 20 and 30 that have transmitted the processing command request.

FIG. 3 is a flowchart showing the processing contents of the object detection application of the slave CPU. An object detection application detects the presence or absence of an object and its position by extracting a set of points indicating an object from point cloud information.

As shown in FIG. 3, the application 23 (object detection application) of the slave CPU 20 first acquires the measurement results (point cloud information) measured by the terrain measuring device 100 and stored in the database server 40, and acquires the point cloud information. The normal vector of each point constituting is calculated (step S100). In this embodiment, as will be described in detail later, the calculation of the normal vector for the point cloud information in step S100 is distributed and performed in parallel by the plurality of slave CPUs 20 and 30, thereby speeding up the calculation process. It has been realized.

Next, the object detection application searches for the nearest other point for each point forming the point cloud information (step S110), and finds a set of neighboring points in which the normal vector direction of each point in the point cloud is continuous. (hereinafter referred to as a cluster) is extracted (step S120). Hereinafter, such extraction processing will be referred to as clustering.

Next, the object detection application determines whether the number of clusters extracted in step S120 is 2 or more (step S130), and if the determination result is NO, that is, if the number of clusters is less than 2. In this case, it is determined that there is an error, an error value is returned as the calculation result (step S132), and the process ends.

Further, if the determination result in step S130 is YES, that is, if the number of clusters is 2 or more, the calculated cluster is returned as the detected object (step S131), and the process ends.

FIG. 4 is a flowchart showing the processing contents of normal vector calculation processing in the object detection application of the slave CPU.

As shown in FIG. 4, the application 23 (object detection application) of the slave CPU 20 first performs normal vector calculation processing (step S100 in FIG. 2) of the normal vectors of each point constituting the point cloud information. A processing request for distribution processing to be executed in parallel is generated and transmitted to the processing distribution unit 12 of the topography management unit 11 of the master CPU 10 (step S200).

Next, the object detection application waits until it receives the normal vector calculation result from the processing result collection unit 14 of the topography management unit 11 of the master CPU 10 (step S210), and then receives the normal vector calculation result from the master CPU 10. When received (step S220), the received calculation result is used to proceed to the subsequent processing of the object detection application (step S110 in FIG. 2).

FIG. 5 is a diagram illustrating an example of a processing request request for distribution processing transmitted in the request transmission processing. The processing request is transmitted, for example, by transmitting a data file written in JSON (JavaScript Object Notation) format, which is an object language.

As shown in FIG. 5, in the request transmission process (step S200 in FIG. 4), a processing request is generated to be transmitted to the processing distribution unit 12 of the topography management unit 11. The processing request includes a request source CPU name, a request source IP address, a request source port number, a request ID, request processing details, and target terrain area information.

The request source CPU name (requester), the request source IP address (IP-addr), and the request source port number (Port num) are used to convey information to the topography management unit 11 about the destination of the calculation result. be. The terrain management section 11 stores this information in the information storage section 13.

The request processing content (target_region) is used to convey information about the content that the topography management unit 11 wants to process. In this embodiment, normal vector calculation is performed.

The target terrain area information (process_type) is used to convey information on the range of terrain desired to be processed to the terrain management unit 11. In this embodiment, the object detection application indicates the area of the work site where the object should be detected. The area in which the hydraulic excavator 101, which is a working machine, works is divided in advance into a plurality of small areas, and can be uniquely specified, for example, by coordinates given as the site coordinate system. In this embodiment, this small area is referred to as a grid. For example, if the grid area indicated by the target terrain area information is a rectangle with four coordinates (0, 0) (0, 10) (20, 0) (20, 10) as vertices, "0: 20,0:10”.

The request ID is used to link the transmitted processing request and response (computation result) on a one-to-one basis. For example, this is necessary when sending multiple requests with the same application.

FIG. 8 is a flowchart showing the processing contents of the processing distribution section of the topography management section of the master CPU.

As shown in FIG. 8, when the processing distribution unit 12 of the terrain management unit 11 of the master CPU 10 receives a processing request request from the application 23 of the slave CPU 20, it first sends the terrain processing load (processing calculation amount) from the database server 40. is acquired (step S300). The processing load is given to each terrain grid. The processing load is displayed as an integer value for each grid.

Next, based on the obtained processing load, an area to be allocated to each slave CPU 20, 30 is determined (step S310). In this processing distribution process, area allocation is performed in grid units.

Next, a terrain filter is acquired from the database server 40 (step S320). The terrain filter is information indicating which slave CPU of the plurality of slave CPUs 20 and 30 each grid is assigned to. For example, you can assign an integer value to each grid as a terrain filter, and create a rule that the first slave CPU will be in charge of the calculation processing for a grid whose number is 1, and the second slave CPU will be responsible for the calculation processing of that grid if the number is 2. Decide in advance. In this embodiment, a currently usable terrain filter is acquired from among a plurality of terrain filters stored in advance in the database server 40. The database server 40 stores a predetermined number of terrain filters in order to use the plurality of terrain filters simultaneously.

Next, a terrain filter indicating the area allocated to each grid is created, and information in the database server 40 is updated (step S330).

Subsequently, a processing command request for arithmetic processing is transmitted to each of the slave CPUs 20 and 30 in charge of arithmetic processing for each grid (step S340), and the processing is ended.

FIG. 6 is a diagram illustrating an example of a processing instruction request for arithmetic processing that is transmitted in the processing instruction request transmission process. The processing instruction request is transmitted, for example, by transmitting a data file written in the JSON (JavaScript Object Notation) format, which is an object language.

As shown in FIG. 6, in the request transmission process (step S340 in FIG. 8), a processing command request to be transmitted to each slave CPU 20, 30 is generated. The processing command request includes a request ID, a request sub ID, target terrain area information, a terrain filter name, and request processing details.

The request ID (request ID), target terrain region information (target_region), and request processing content (process_type) are the same as the information in the processing request request (see FIG. 5).

The request sub ID (request_sub ID) is assigned to each slave CPU 20, 30 that shares processing, and is used to confirm that there has been a response (reception of calculation results) from all slave CPUs 20, 30. It is. That is, the number of all processing requests can also be determined from the request sub-ID.

The terrain filter name (mask) is a name for identifying the terrain filter stored in the database server 40, and is used by the slave CPUs 20, 30 to acquire terrain information from the database server 40.

FIG. 9 is a flowchart showing the processing contents of the processing distribution process in the processing distribution section of the topography management section of the master CPU.

As shown in FIG. 9, as a process distribution process (step S310 in FIG. 8), the processing distribution unit 12 of the terrain management unit 11 of the master CPU 10 first adds up the processing calculation amount of all grids, and By dividing by the number of , the average amount of processing calculations processed by one slave CPU is calculated (step S400).

Subsequently, initialization is performed by setting the grid processed by each slave CPU 20, 30 to 0 (zero) (step S410), and setting the variable i representing each slave CPU to 0 (zero) (step S420).

Next, it is determined whether (i+1)<N (N: total number of slave CPUs), that is, whether there is a slave CPU that has not performed grid allocation processing (step S430), and the determination result is If YES, it is determined whether there is an unallocated grid (step S440). If the determination result in step S440 is NO, that is, if there is no unallocated grid to be allocated to the i-th slave CPU, the process ends, and the subsequent process of the process distribution unit 12 (see FIG. The process moves to step S320).

Further, if the determination result in step S440 is YES, that is, if there is an unallocated grid, the unallocated grid is allocated to the i-th slave CPU (step S450).

Next, it is determined whether the total amount of processing calculation of the grid assigned to the i-th slave CPU is larger than the average amount of processing calculation calculated in step S400 (step S460), and if the determination result is NO, Then, the process returns to step S440.

Further, if the determination result in step S460 is YES, that is, if the total processing complexity of the grid assigned to the i-th slave CPU is larger than the average processing complexity calculated in step S400, the Then, it is determined whether the current total processing amount of the i-th slave CPU is closer to the average processing amount than the total processing amount of calculation before adding the latest grid (step S470). That is, in step S470, a comparison is made to determine which of the current total processing amount of the i-th slave CPU and the total processing amount of calculation before adding the latest grid is closer to the average processing amount.

If the determination result in step S470 is NO, and if the total processing amount before adding the latest grid is closer to the average processing amount, the latest grid is returned to unassigned (step S480), and step Returning to the process of S430, the process moves to the allocation process for the (i+1)th slave CPU.

Further, if the determination result in step S470 is YES, that is, if the current total processing amount is closer to the average processing amount, the process returns to step S430 and allocation to the (i+1)th slave CPU is performed. Move to processing.

Furthermore, if the determination result in step S430 is NO, that is, if the allocation processing to all slave CPUs has been completed, it is then determined whether or not there is an unallocated grid (step S431).

If the determination result in step S431 is YES, that is, if there is an unallocated grid, the unallocated grid is allocated to a slave CPU whose total processing amount is lower than the average processing amount (step S432). , the process returns to step S431. That is, the processes of steps S431 and S432 are repeated until there are no unallocated grids. If there is an unallocated grid, as in the case where the determination result in step S431 is YES, there is always a slave CPU whose total processing amount is lower than the average processing amount, so that slave CPU has the unallocated grid. By allocating the grid, it is possible to uniformly allocate the grid to each slave CPU.

Furthermore, if the determination result in step S431 is NO, that is, if there is no unallocated grid, the process ends and the process moves to the subsequent process of the process distribution unit 12 (step S320 in FIG. 8).

FIG. 10 is a flowchart showing the processing contents of the normal vector calculation execution section of the terrain processing section of the slave CPU.

As shown in FIG. 10, the normal vector calculation execution units 22 and 32 of the terrain processing units 21 and 31 of the slave CPUs 20 and 30 are in a standby state until they receive a processing command request (step S500), and the When a processing command request for arithmetic processing is received from the processing distribution unit of the terrain management unit 11, point cloud information is read from the database server 40 using the target terrain area information and terrain filter name of the processing command request (step S510). ).

Next, based on the request processing content of the processing command request, a processing method for point cloud information (in this embodiment, a normal vector calculation processing method) is read from the database server 40 (step S520), and the read processing is performed. The method is used to perform arithmetic processing on the point cloud information (step S530), and the amount of processing calculation required to perform the arithmetic processing is measured (step S540).

Subsequently, the result of the arithmetic processing and the amount of processing calculation are transmitted as a processing command response to the processing result collection unit 14 of the topography management unit 11 of the master CPU 10 (step S550), and the process is ended.

FIG. 7 is a diagram illustrating an example of a processing command response transmitted in the processing command response transmission process. The processing command response is transmitted, for example, by transmitting a data file written in JSON (JavaScript Object Notation) format, which is an object language.

As shown in FIG. 7, in the processing command response transmission process (step S550 in FIG. 10), a processing command response to be transmitted to the master CPU 10 is generated. The processing command response includes a request ID, a request sub ID, a processing result, and a processing amount.

The request ID (request ID) and the request sub ID (request_sub ID) are the same information as the processing request request (see FIG. 5) and the processing instruction request (see FIG. 6).

The processing result (serialized_result) is a serial representation of the calculation result and processing calculation amount in the normal vector calculation execution units 22 and 32 of each slave CPU 20 and 30.

FIG. 11 is a flowchart showing the processing contents of the processing result collection section of the topography management section of the master CPU.

As shown in FIG. 11, the processing result collection unit 14 of the topography management unit 11 of the master CPU 10 is in a standby state until it receives a processing command response from each slave CPU 20, 30 (step S600). When the processing command response received from each slave CPU 20, 30 is integrated (step S610), the processing calculation amount of the database server 40 is updated using the processing calculation amount included in the processing command response. (Step S620).

Subsequently, the processing distribution unit 12 determines whether processing command responses have been received from all slave CPUs 20 and 30 to which processing command requests for arithmetic processing have been sent (step S630), and if the determination result is NO, that is, If all processing command responses have not been received, steps S600 to S630 are repeated until all processing command responses are received.

If the determination result in step S630 is YES, the integrated calculation result is returned as a processing request response to the application 23 of the slave CPU 20 that sent the processing request request (step S640), and the process ends. Information about the slave CPU 20 that has issued the processing request is stored in the information storage section 13.

The effects of this embodiment configured as above will be explained.

In the conventional technology, a plurality of arithmetic units perform arithmetic processing in parallel on a task-by-task basis. However, when there is a large imbalance in the amount of calculation for each task, such as the calculation of topographical information performed by a work machine at a work site, the conventional method of performing calculations in parallel for each task cannot achieve high-speed calculation processing. It was not possible to fully implement the

On the other hand, in the present embodiment, the master arithmetic unit prepares a group of information in advance in response to a request for execution of arithmetic processing based on a group of information transmitted from one of a plurality of slave arithmetic units. In addition to dividing the small information groups into multiple small information groups based on the predetermined division rules, each of the multiple small information groups is set to execute arithmetic processing based on the small information groups. The arithmetic units each execute arithmetic processing based on the small information group set by the master arithmetic unit, and return information on the arithmetic load required to execute the arithmetic processing together with the results of the arithmetic processing to the master arithmetic unit. integrates the calculation results returned from multiple slave processing devices, and returns the integrated calculation results to the slave processing device that sent the request to perform the processing based on the information group as the processing result of the information group, and Based on the calculation load information returned from the slave calculation units, the splitting rule is changed so that the calculation loads of the slave calculation units are uniform, so that parallel processing with multiple slave CPUs It is possible to allocate calculation processing uniformly, and to perform calculations based on a large amount of information at higher speed.

<Second embodiment>
A second embodiment of the present invention will be described with reference to FIGS. 12 and 13. In the drawings, members similar to those in the first embodiment are denoted by the same reference numerals, and explanations thereof will be omitted.

In this embodiment, even among multiple slave CPUs with different performance and status, calculation processing can be uniformly allocated in parallel processing by the multiple slave CPUs, and calculations can be performed faster and based on a large amount of information. It is configured so that it can be done.

Generally, the higher the performance of the CPU, the faster the processing can be performed for the same amount of processing calculations. Furthermore, when a certain application is executed by the CPU, if another high-priority application exists, the CPU processing power is used for calculations related to that application, which delays the completion of processing of the certain application. In this embodiment, in order to take such a situation into consideration, CPU capacity (allowable amount) is calculated for each slave CPU, and distribution processing is executed based on this CPU capacity. The CPU capacity is stored in the database server 40. Since the CPU capacity may differ depending on the processing content, different CPU capacities are stored for each processing content.

FIG. 12 is a flowchart showing the processing content of the process distribution process in the process distribution unit of the topography management unit of the master CPU.

As shown in FIG. 12, as a process distribution process (step S310 in FIG. 8), the process distribution unit 12 of the topography management unit 11 of the master CPU 10 first sends information from the database server 40 to each slave CPU 20 according to the current process content. , 30 (step S400A), and calculates the average processing time by dividing the total processing calculation amount of the slave CPU by the total capacity (step S401A).

Subsequently, initialization is performed by setting the grid processed by each slave CPU 20, 30 to 0 (zero) (step S410), and setting the variable i representing each slave CPU to 0 (zero) (step S420).

Next, it is determined whether (i+1)<N (N: total number of slave CPUs), that is, whether there is a slave CPU that has not performed grid allocation processing (step S430), and the determination result is If YES, it is determined whether there is an unallocated grid (step S440). If the determination result in step S440 is NO, that is, if there is no unallocated grid to be allocated to the i-th slave CPU, the process ends, and the subsequent process of the process distribution unit 12 (see FIG. The process moves to step S320).

Further, if the determination result in step S440 is YES, that is, if there is an unallocated grid, the unallocated grid is allocated to the i-th slave CPU (step S450).

Next, it is determined whether the total processing calculation time of the grid assigned to the i-th slave CPU is greater than the average processing calculation time calculated in step S400 (step S460A), and if the determination result is NO Then, the process returns to step S440. The current processing time can be calculated by dividing the total processing amount allocated to the slave CPU by the capacity of that CPU.

Further, if the determination result in step S460A is YES, that is, if the total processing calculation time of the grid assigned to the i-th slave CPU is greater than the average processing calculation time calculated in step S400, the Then, it is determined whether the current total processing calculation time of the i-th slave CPU is closer to the average processing calculation time than the total processing calculation time before adding the latest grid (step S470A). That is, in step S470A, a comparison is made to see which of the current total processing calculation time of the i-th slave CPU and the total processing calculation time before adding the latest grid is closer to the average processing calculation time.

If the determination result in step S470A is NO, and if the total processing calculation time before adding the latest grid is closer to the average processing calculation time, the latest grid is returned to unallocated (step S480), and step Returning to the process of S430, the process moves to the allocation process for the (i+1)th slave CPU.

Further, if the determination result in step S470A is YES, that is, if the current total processing calculation time is closer to the average processing calculation time, the process returns to step S430 and allocation to the (i+1)th slave CPU is performed. Move to processing.

Furthermore, if the determination result in step S430 is NO, that is, if the allocation processing to all slave CPUs has been completed, it is then determined whether or not there is an unallocated grid (step S431).

If the determination result in step S431 is YES, that is, if there is an unallocated grid, the unallocated grid is allocated to the slave CPU whose total processing calculation time is shorter than the average processing calculation time (step S432A ), the process returns to step S431. That is, the processes of steps S431 and S432A are repeated until there are no unallocated grids. If there is an unallocated grid, as in the case where the determination result in step S431 is YES, there is always a slave CPU whose total processing calculation time is shorter than the average processing calculation time, so that slave CPU has the unallocated grid. By allocating the grid, it is possible to uniformly allocate the grid to each slave CPU.

Further, if the determination result in step S431 is NO, that is, if there is no unallocated grid, the process is ended and the process moves to the subsequent process of the processing distribution unit 12 (step S320 in FIG. 8).

FIG. 13 is a flowchart showing the processing contents of the processing result collection section of the topography management section of the master CPU.

As shown in FIG. 11, the processing result collection unit 14 of the topography management unit 11 of the master CPU 10 is in a standby state until it receives a processing command response from each slave CPU 20, 30 (step S600). Upon receiving the processing commands, the processing results included in the processing command responses received from each slave CPU 20, 30 are integrated (step S610), and the database server The processing calculation amount and capacity of 40 are updated (step S620A). Capacity is determined by multiplying the CPU time required for processing by the operating frequency of that CPU. The CPU processing time is not measured while other processes are being executed, but is simply determined as the time required for the processing. Even when calculation is performed in parallel with multiple cores, the time of each core is summed, so this method can be used even with a multi-core CPU.

Subsequently, the processing distribution unit 12 determines whether processing command responses have been received from all slave CPUs 20 and 30 to which processing command requests for arithmetic processing have been sent (step S630), and if the determination result is NO, that is, If all processing command responses have not been received, steps S600 to S630 are repeated until all processing command responses are received.

If the determination result in step S630 is YES, the integrated calculation result is returned as a processing request response to the application 23 of the slave CPU 20 that sent the processing request request (step S640), and the process ends. Information about the slave CPU 20 that has issued the processing request is stored in the information storage section 13.

Other configurations are similar to those of the first embodiment.

In this embodiment configured as described above, the same effects as in the first embodiment can be obtained.

Further, even among a plurality of slave CPUs with different performance and states, calculation processing can be uniformly assigned in parallel processing by the plurality of slave CPUs, and calculations can be performed faster and based on a large amount of information.

<Third embodiment>
A third embodiment of the present invention will be described with reference to FIGS. 14 to 16. In the drawings, members similar to those in the first embodiment are denoted by the same reference numerals, and explanations thereof will be omitted.

This embodiment shows a case where a master CPU has a database server and a plurality of slave CPUs each have a replicated database. This reduces the number of accesses to the database server and completes processing faster.

FIG. 14 is a diagram schematically showing the overall configuration of the control system according to this embodiment.

In FIG. 14, the control system includes one master CPU (master processing unit) 10B, a plurality of slave CPUs (slave processing units) 20B, 30B, and a database server 41 provided in the master CPU 10B, and each The operation of the hydraulic excavator 101 is controlled using calculation results by the master CPU 10B and a plurality of slave CPUs 20B, 30B based on information stored in replica databases 42, 43 provided in the slave CPUs 20B, 30B. Note that in this embodiment, only two of the plurality of slave CPUs are illustrated and explained.

The master CPU 10B is a functional unit that is in charge of managing processing related to the topography of the work site, such as calculation processing related to the illustrated object detection application, and is a topography management unit 11 that is realized by an application executed on the master CPU 10B. , one or more other applications 15 , and a database server 41 . The terrain management unit 11 also includes a processing distribution unit 12, an information storage unit 13, and a processing result collection, which are realized by an application executed on the master CPU 10B, as functional units in charge of processing in the terrain management unit 11. It has a section 14.

The slave CPU 20B is a functional unit that performs processing related to the topography of the work site, such as calculation processing related to the illustrated object detection application, and includes a topography processing unit 21 realized by an application executed on the slave CPU 20B, and a plurality of It is composed of other applications 23 and 24 and a replica database 42. Additionally, the terrain processing unit 21 serves as a functional unit in charge of executing calculation processing (for example, normal vector calculation processing) in the terrain processing unit 21, and executes normal vector calculation realized by an application executed on the slave CPU 20B. It has a section 22. In this embodiment, a case will be exemplified in which the application 23 of the slave CPU 20B is an object detection application. The replica database 42 stores a replica of the information of the database server 41 of the master CPU 10B.

Similarly, the slave CPU 30B is a functional unit that performs processing related to the topography of the work site, such as arithmetic processing related to the illustrated object detection application, and includes the topography processing unit 31 and the topography processing unit 31 realized by the application executed on the slave CPU 30B. , a plurality of other applications 33 and 34, and a replica database 43. Additionally, the terrain processing unit 31 serves as a functional unit in charge of executing calculation processing (for example, normal vector calculation processing) in the terrain processing unit 31, and executes normal vector calculation realized by an application executed on the slave CPU 30B. It has a section 32. The replica database 43 stores a replica of the information of the database server 41 of the master CPU 10B.

FIG. 15 is a flowchart showing the processing contents of the processing distribution section of the terrain management section of the master CPU.

As shown in FIG. 15, when the processing distribution unit 12 of the terrain management unit 11 of the master CPU 10B receives a processing request from the application 23 of the slave CPU 20B, it first receives the processing load (processing calculation amount) from the database server 41. (step S300). The processing load is given to each terrain grid. The processing load is displayed as an integer value for each grid.

Next, a terrain filter is acquired from the database server 41 (step S320). The terrain filter is information indicating which slave CPU of the plurality of slave CPUs 20B and 30B each grid is assigned to. For example, an integer value is assigned to each grid as a terrain filter, and a rule is created in which the first slave CPU is responsible for the calculation processing of a grid whose numerical value is 1, and the second slave CPU2 is responsible for the calculation processing of that grid when the numerical value is 2. Decide in advance. In this embodiment, a currently available terrain filter is acquired from among a plurality of terrain filters stored in advance in the database server 41. The database server 41 stores a predetermined number of terrain filters in order to use the plurality of terrain filters simultaneously.

Subsequently, a database partial synchronization process is performed (step S321) to synchronize the information in the replica databases 42 and 43 with the information in the database server 41.

Next, based on the obtained processing load, an area to be allocated to each slave CPU 20B, 30B is determined (step S322). In this processing distribution process, area allocation is performed in grid units.

Next, a terrain filter indicating the area allocated to each grid is created, and information in the database server 41 is updated (step S330).

Subsequently, a processing command request for arithmetic processing is transmitted to each of the slave CPUs 20B and 30B in charge of arithmetic processing for each grid (step S340), and the processing is ended.

FIG. 16 is a flowchart showing the processing contents of the database partial synchronization process.

In FIG. 16, the processing distribution unit 12 of the topography management unit 11 of the master CPU 10B first initializes the update list (step S700), and checks the unchecked grids for all grids included in the area of target topography area information. It is determined whether or not there is (step S710).

If the determination result in step S710 is YES, that is, if there is an unchecked grid, the synchronization time of that grid is acquired (step S720), and then data is updated from the sensor etc. to the database server 41. It is determined whether it is after the synchronization time with the replica databases 42, 43 (step S730). The replication databases 42 and 43 record the time when each grid last performed synchronization, and by referring to these, the synchronization time in step S720 can be obtained.

If the determination result in step S730 is NO, that is, if data has been updated from the sensor etc. to the database server 41 after the synchronization time, the grid is added to the update list (step S740), and step S710 Return to processing. If the determination result in step S730 is YES, the database server 41 and the replica databases 42, 43 are in a synchronized state, and the process returns to step S730.

Further, if the determination result in step S710 is NO, that is, if the synchronization check has been completed for all grids, the replica databases 42 and 43 of the slave CPUs 20B and 30B are synchronized for all the grids in the update list. is performed (step S711), and the process ends.

Other configurations are similar to those of the first embodiment.

In this embodiment configured as described above, the same effects as in the first embodiment can be obtained.

Furthermore, since the number of accesses to the database server is reduced, processing can be completed more quickly.

<Additional notes>
Note that the present invention is not limited to the above-described embodiments, and includes various modifications and combinations within the scope of the invention. Furthermore, the present invention is not limited to having all the configurations described in the above embodiments, but also includes configurations in which some of the configurations are deleted. Moreover, each of the above-mentioned configurations, functions, etc. may be realized in part or in whole by, for example, designing an integrated circuit. Furthermore, each of the above configurations, functions, etc. may be realized by software by a processor interpreting and executing a program for realizing each function. In addition, information such as programs, tables, files, etc. that implement each function should be stored in memory, a recording device such as a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD. may be configured.

10, 10B... Master CPU, 11... Terrain management unit, 12... Processing distribution unit, 13... Information storage unit, 14... Processing result collection unit, 15... Application, 20, 20B... Slave CPU, 21... Terrain processing unit, 22... Normal vector calculation execution unit, 23, 24... Application, 30, 30B... Slave CPU, 31... Terrain processing unit, 32... Normal vector calculation execution unit, 33, 34... Application, 40, 41... Database server, 42, 43...Replication database, 100...Topographic measurement device, 101...Hydraulic excavator, 110...Front working machine, 111...Boom, 112...Arm, 113...Bucket, 121...Boom cylinder, 122...Arm cylinder, 123...Bucket cylinder , 124... Swing motor, 125, 126... Traveling motor, 130... Vehicle body, 131... Upper rotating body, 132... Lower traveling body, 151... Cab, 171, 172... GNSS receiver, 181, 182, 183, 184... inertial measurement device

Claims (4)

A work machine that performs work at a work site and measures topographic information of the work site ,
It has one master arithmetic device, a plurality of slave arithmetic devices, and a storage device, and based on an information group consisting of a plurality of pieces of information stored in the storage device, the one master arithmetic device and the plurality of slave arithmetic devices comprising a control system that performs arithmetic processing of the topographical information of the work site using a calculation device,
The control system includes:
The master arithmetic device divides the information group based on a predetermined division rule in response to a request for execution of arithmetic processing based on the information group transmitted from one slave arithmetic device of the plurality of slave arithmetic devices. dividing into a plurality of small information groups, and setting execution of arithmetic processing based on the small information groups to any of the plurality of slave processing devices for each of the plurality of small information groups;
The plurality of slave arithmetic devices each execute arithmetic processing based on the small information group set by the master arithmetic device, and transmit information on the arithmetic load required for execution of the arithmetic processing together with the result of the arithmetic processing to the master arithmetic device. Reply to
The master arithmetic device integrates the arithmetic results returned from the plurality of slave arithmetic devices, and transmits the integrated arithmetic results to the slave arithmetic device that has sent the request to perform the arithmetic processing based on the information group. When changing the division rule so that the calculation loads of the plurality of slave calculation units are equalized based on the calculation load information returned from the plurality of slave calculation units as well as the result, one In addition to calculating the average calculation load, which is the calculation load per slave calculation unit, when the calculation processing based on the small information group is assigned to each of the plurality of slave calculation units, the total current calculation load and the A work machine characterized by comparing the total calculation load before allocating the calculation processing based on the information group to see which one is closer to the average calculation load, and changing the division rule based on the comparison result. The working machine according to claim 1,
The information group is point cloud information that indicates the surface position of an object at the work site using a plurality of coordinates in a site coordinate system set at the work site,
The control system includes:
The master arithmetic device sets a spatial area of the work site in a predetermined area in response to a request for execution of arithmetic processing based on the information group transmitted from one slave arithmetic device of the plurality of slave arithmetic devices. By dividing the point cloud information group, which is the information group, into a plurality of small information groups by dividing it into a plurality of grids, and setting any of the plurality of slave calculation devices in the plurality of grids,
The plurality of slave computing devices each execute arithmetic processing based on the small information group consisting of coordinates included in the grid set by the master computing device, and display the results of the arithmetic processing as well as the calculations required to execute the arithmetic processing. replying load information to the master arithmetic unit;
The master arithmetic device integrates the arithmetic results returned from the plurality of slave arithmetic devices, and transmits the integrated arithmetic results to the slave arithmetic device that has sent the request to perform the arithmetic processing based on the information group. Replying as a result and changing each of the areas of the plurality of grids so that the calculation loads of the plurality of slave calculation devices are uniform based on the calculation load information returned from the plurality of slave calculation devices. A working machine featuring: The working machine according to claim 1,
A working machine, wherein the plurality of slave arithmetic devices include slave arithmetic devices with different arithmetic performance. The working machine according to claim 1,
The storage device includes a plurality of storage devices respectively provided in the master arithmetic device and the plurality of slave arithmetic devices,
A work machine characterized in that the storage devices of the plurality of slave processing devices each store information copied from the information stored in the storage device of the master processing device.

Images