JP2020148122A - Control device and control method - Google Patents

Abstract

To complete processing to be required while suppressing power consumption.SOLUTION: A control device 1 is a control device 1 that controls a rotation mechanism, and comprises: a control unit 2 that causes software to process first processing of performing calculation by a neural network within a first period based on a rotation state of the rotation mechanism, and second processing of starting processing on the basis of the rotation state of the rotation mechanism and performing the processing with a higher priority than the first processing; and a hardware calculation unit 4 capable of causing hardware to process at least part of the first processing. The control unit 2 activates the hardware calculation unit 4 when a rotation speed of the rotation mechanism is equal to or larger than a predetermined threshold value.SELECTED DRAWING: Figure 2

Description

The present invention relates to a control device and a control method.

In recent years, for example, there is a control device for determining an abnormality of an internal combustion engine based on a learning model. In this type of control device, for example, the internal combustion engine outputs a determination result within a predetermined period from a predetermined rotation timing (see, for example, Patent Document 1).

JP-A-2018-178810

However, in the prior art, for example, when the rotation speed of the internal combustion engine exceeds a predetermined value, the processing may not be in time. However, it is not preferable to perform the processing by the learning model on other hardware because the power consumption increases by the amount of the hardware.

The present invention has been made in view of the above, and an object of the present invention is to provide a control device and a control method capable of completing a required process while suppressing power consumption.

In order to solve the above-mentioned problems and achieve the object, the control device according to the embodiment is a control device that controls the rotation mechanism, and is calculated by a neural network within the first period based on the rotation state of the rotation mechanism. A control unit that software processes the first process of performing the process and the second process of starting the process based on the rotation state of the rotation mechanism and performing the process with a higher priority than the first process, and the first process. A hardware calculation unit capable of processing at least a part of one processing by hardware is provided, and the control unit activates the hardware calculation unit when the rotation speed of the rotation mechanism is equal to or higher than a predetermined threshold value. ..

According to the present invention, the required processing can be completed while suppressing the power consumption.

FIG. 1 is a diagram showing an outline of a control method. FIG. 2 is a block diagram of the control device. FIG. 3 is a timing chart showing the processing executed by the control device. FIG. 4 is a flowchart showing a processing procedure executed by the control device.

Hereinafter, embodiments of the control device and control method disclosed in the present application will be described in detail with reference to the accompanying drawings. The present invention is not limited to the embodiments shown below.

First, the outline of the control device and the control method according to the embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an outline of a control method. The correction method according to the embodiment is executed by the control device 1 shown in FIG.

As shown in FIG. 1, the control device 1 according to the embodiment acquires, for example, various sensor values input from the engine 100, and estimates the internal state of the engine 100 based on the sensor values.

Further, the control device 1 includes a software calculation unit and a hardware calculation unit. The software calculation unit is software that appropriately executes, for example, the first process, the second process, and the third process at a required timing.

Here, the first process is an arithmetic process by a neural network. The first process is also referred to as an AI (Artificial Intelligent) process. The first process is, for example, an abnormality detection process, such as a knocking detection process. The second process is a process having a higher priority than the first process. The second process is a process for controlling the engine, such as fuel injection control and ignition control. Further, the third process is a process having a lower priority than the first process and the second process. Therefore, the third process is a task executed in the free time of the first process and the second process.

The hardware calculation unit is hardware different from the software calculation unit that can process at least a part of the first processing by hardware.

As shown in FIG. 1, the software calculation unit executes the second process at a predetermined crank angle period (for example, 30 CA) from the top dead center (hereinafter referred to as TDC) of the crank angle of the engine 100, and the second process is performed. The first process is executed in the free time of the process.

For example, the first process is a task required every cycle (for example, 720 CA) having a longer crank angle period than the second process. Further, the third process is, for example, a task required in a predetermined time cycle.

By the way, in general, as the rotation speed of the engine 100 increases, the processing interval of the second processing becomes shorter. In this case, it is assumed that the processing by the software calculation unit is occupied by the second processing (and the first processing) and the third processing cannot be executed.

Therefore, in the control method according to the embodiment, the hardware calculation unit is activated when the rotation speed of the engine 100 exceeds a predetermined threshold value. That is, in the control method according to the embodiment, when the rotation speed of the engine exceeds the rotation threshold value, the hard calculation unit is activated and the second processing is executed on the hard calculation unit side.

As a result, the software calculation unit does not need to execute the second process during the period when the hard calculation unit is running, and can execute the third process in the free time of the first process. On the other hand, in the control method according to the embodiment, when the rotation speed of the engine 100 is equal to or less than the rotation threshold value, the hardware calculation unit is stopped.

In other words, the hard calculation unit is stopped when the software calculation unit can execute the first processing and the third processing by the timing required in the free time of the second processing. This makes it possible to suppress the power consumption of the hardware calculation unit.

That is, in the control method according to the embodiment, when the processing by the software calculation unit cannot catch up, the hard calculation unit performs the first processing, and when the software calculation unit is sufficient, the hardware calculation unit is stopped.

Therefore, according to the control method according to the embodiment, it is possible to complete the required processing while suppressing the power consumption.

Next, a configuration example of the control device 1 according to the embodiment will be described with reference to FIG. FIG. 2 is a block diagram of the control device 1. Note that FIG. 2 also shows the sensor group 101. The sensor group 101 is a general term for sensors that detect the internal state of the engine 100.

As described above, when the control device 1 detects knocking, the sensor group 101 includes a crank angle sensor, a knock sensor, and the like. The sensor group 101 can be appropriately changed according to the abnormality detection target.

As shown in FIG. 2, the control device 1 includes a control unit 2, a storage unit 3, and a hardware calculation unit 4. The control unit 2 includes a detection unit 21, a software calculation unit 22, a capture unit 23, and a start control unit 24. The storage unit 3 stores the input information 31 and the calculation information 32.

The hardware calculation unit 4 is hardware for performing processing (first processing) by the neural network. In the present embodiment, the hardware calculation unit 4 does not need to perform a task other than the first processing, and can use the hardware specialized for the first processing, that is, the AI processing. Note that the AI process is basically a process of repeating similar operations with different input values, etc., although the amount of calculation is large, so that the amount of load reduction when the process is replaced is large, and the process is performed by hardware. Since the circuit configuration is relatively simple, it is suitable for processing by hardware as compared with other second processing and third processing.

The control unit 2 is a computer that performs software calculations, and is composed of, for example, a microcomputer, and has a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), and the like. Includes computers with input / output ports and various circuits.

The CPU of the computer functions as a detection unit 21, a software calculation unit 22, a capture unit 23, and a start control unit 24 of the control unit 2, for example, by reading and executing a program stored in the ROM.

Further, at least a part or all of the detection unit 21, the software calculation unit 22, the capture unit 23, and the start control unit 24 of the control unit 2 are hardware such as an ASIC (Application Specific Integrated Circuit) and an FPGA (Field Programmable Gate Array). It can also be configured with hardware.

Further, the storage unit 3 corresponds to, for example, a RAM, a flash ROM, or an HDD. The storage unit 3 can store input information 31, calculation information 32, and information on various programs. The control device 1 may acquire the above-mentioned program and various information via another computer or a portable recording medium connected by a wired or wireless network. The control unit 2 and the storage unit 3 may be configured by one microcomputer.

The input information 31 is information regarding the input data input from the sensor group 101. The control unit 2 can execute the above-mentioned first to third processes based on the input information 31. Further, as shown in FIG. 2, the storage unit 3 is connected to the hardware calculation unit 4, and can be accessed from the hardware calculation unit 4.

In this case, the hardware calculation unit 4 can directly acquire the input information 31 from the storage unit 3 and perform the first processing. In other words, the hardware calculation unit 4 acquires the input information 31 without going through the control unit 2. Since the amount of input data and coefficient data required at the start of AI processing is larger than the calculation result at the end, it is better to configure the hardware calculation unit 4 to directly acquire from the storage unit 3. It is better in terms of the processing efficiency of 2.

This makes it possible to suppress the processing load by the control unit 2. That is, the amount of data in the input information 31 is relatively large, and if the hardware calculation unit 4 acquires the input information 31 via the control unit 2, the control unit 2 reads the input information 31 and causes the hardware calculation unit 4 to read the input information 31. There is a risk that a relatively large amount of time will be spent on the delivery process.

The calculation information 32 is information regarding the calculation result of the above-mentioned first process. For example, when the first process is interrupted, the software calculation unit 22 stores the calculation result of the first process before the interruption as calculation information 32 in the storage unit 3.

The control unit 2 executes various processes (first process to third process) in synchronization with the crank angle of the engine 100, and when the rotation speed of the engine 100 exceeds the rotation threshold, the hard calculation unit 4 is operated. Start it.

The detection unit 21 detects the rotational state of the engine 100. Specifically, the detection unit 21 detects the rotational state of the engine 100 based on the crank angle signal input from the crank angle sensor included in the sensor group 101. Further, the detection unit 21 can also calculate the rotation speed of the engine 100 (hereinafter, simply referred to as the rotation speed) based on the crank angle signal.

The software calculation unit 22 is a calculation unit that executes each task. As described above, the software calculation unit 22 executes the first process and the second process in synchronization with the crank angle, and also executes the third process at a predetermined time cycle.

In the present embodiment, the priority of each task is 2nd process> 1st process> 3rd process. Therefore, the software calculation unit 22 performs the first process in the free time of the second process, and further performs the third process in the free time of the first process.

When the import unit 23 receives a notification from the hardware calculation unit that the calculation processing is completed by an interrupt or the like, the import unit 23 acquires the calculation result of the first processing from the hardware calculation unit 4, and the storage unit 3 calculates based on the calculation result. Information 32 is updated. That is, the import unit 23 performs an import process for importing the calculation result from the hardware calculation unit 4 and updates the calculation information 32. It should be noted that, when the hardware calculation unit 4 is activated, the import unit 23 can acquire information necessary for the hardware calculation unit 4 to perform the first processing from the storage unit 3 and pass it to the hardware calculation unit 4. .. When the hardware calculation unit 4 is configured to directly write the calculation result to the storage unit 3, the import unit 23 receives a notification from the hardware calculation unit that the calculation process is completed by an interrupt or the like. The calculation result is read from the storage unit 3 so that it can be used in the processing of the software calculation unit 22 that uses the calculation result.

The start control unit 24 is a processing unit that controls the start and stop of the hardware calculation unit 4. The start control unit 25 activates the hard calculation unit 4 when the rotation speed exceeds the rotation threshold value, and stops the hard calculation unit 4 when the rotation speed falls below the rotation threshold value.

The activation control unit 24 activates the hardware calculation unit 4 when the rotation speed exceeds the rotation threshold value. Further, when the start timing of the first process is reached while the hard calculation unit is in the activated state, the hard calculation unit 4 is notified of the start timing of the first process. As a result, the hardware calculation unit 4 reads the predetermined input information 31 from the storage unit 3 and starts the first process from the start timing.

Here, the rotation threshold value is, for example, a rotation speed at which the processing of the software calculation unit 22 cannot keep up. For example, the rotation threshold is determined in advance according to the performance (processing speed) of the control unit 2. That is, the higher the performance of the control unit 2, the more processes that can be executed per unit time, so that the rotation threshold value is set higher.

Further, the activation control unit 24 acquires a completion notification from the hardware calculation unit 4 when the hardware calculation unit 4 completes the first process. As a result, the import unit 23 acquires the calculation result from the hardware calculation unit 4 and updates the calculation information 32. The process by the capture unit 23 is an interrupt process.

That is, when the control unit 2 acquires the above completion notification, at least the third process is temporarily interrupted, the operation result is acquired from the hardware operation unit 4, and the operation information 32 is updated. Since this update process has an extremely small load, even if the second process is temporarily interrupted in addition to the third process, the other tasks by the control unit 2 are not hindered. It is possible to do it.

Next, a series of processes executed by the control device 1 according to the embodiment will be described with reference to FIG. FIG. 3 is a timing chart showing the processing executed by the control device 1. Here, the processing of the control device 1 during the period when the rotation speed exceeds the rotation threshold value will be described.

Further, FIG. 3 shows the processing timings of the first to third processing and the import processing by the control unit 2 and the first processing by the hardware calculation unit 4.

As shown in FIG. 3, the control unit 2 executes the second process in a predetermined cycle (30 CA cycle), and executes the third process in the free time of the second process. At this time, the hardware calculation unit 4 executes the first process on behalf of the control unit 2.

Then, when the hardware calculation unit 4 completes the first process, the hardware calculation unit 4 notifies the control unit 2 of the completion (time t1). As a result, the control unit 2 starts the capture process of fetching the calculation result from the hardware calculation unit 4 by the interrupt process. In the example shown in FIG. 3, the case where the control unit 2 performs the capture process in the period from the time t1 to the time t2 is shown.

As a result, the calculation information 32 of the storage unit 3 can be updated. After that, the control unit 2 notifies the hardware calculation unit 4 of the time t3, which is the start timing (TDC) of the first process. The control unit 2 can estimate the start timing (time t3) from the current rotation speed. The start timing of the first process is not limited to the TDC, and may be other rotation angle timings, or may be time-dependent timings such as every predetermined time.

Then, the hardware calculation unit 4 can execute the first process again from the start timing. In the period from time t2 to time t3, the hard calculation unit 4 may be kept on standby in a activated state, or the hard calculation unit 4 may be stopped. However, when the rotation speed exceeds the rotation threshold, the rotation speed often exceeds the rotation threshold for a period longer than one cycle (720CA from the TDC timing to the next TDC timing), and the rotation speed exceeds the rotation threshold. In order to prevent control hunting that frequently repeats the start / stop of the hard calculation unit 4 in a short time, once the hard calculation unit 4 is started, the start state is maintained for a while (for example, a predetermined cycle elapses). Until a predetermined time elapses, the rotation speed is maintained until the rotation speed drops to a rotation threshold lower than the rotation threshold at the time of activation).

Next, the processing procedure executed by the control device 1 according to the embodiment will be described with reference to FIG. The processing procedure described below is repeatedly executed by the control unit 2 in a predetermined cycle shorter than the cycle of the second processing having the highest priority.

As shown in FIG. 4, the control device 1 according to the embodiment first determines whether or not it is the second processing timing (30CA cycle), which is the timing for executing the second processing (step S101), and the second processing timing. If (step S101, Yes), the second processing call is performed (step S102).

As a result, the software calculation unit 22 performs the second processing. On the other hand, in the determination process of step S101, if it is not the second process timing (steps S101, No), the process of step S102 is omitted.

Subsequently, the control device 1 determines whether or not it is the first processing timing (720CA cycle) for executing the first processing (step S103), and if it is the first processing timing (step S103, Yes), the rotation speed is high. It is determined whether or not it is larger than the rotation threshold value (step S104).

Here, when the rotation speed exceeds the rotation threshold (step S104, Yes), the control device 1 activates the hardware calculation unit 4 (step S105), and when the rotation speed is equal to or less than the rotation threshold (step S104, No). , The hardware calculation unit 4 is stopped (step S106).

Subsequently, the control device 1 determines whether or not the hard calculation unit 4 is stopped (step S107), and when the hard calculation unit 4 is stopped (step S107, Yes), the software calculation unit 22 is contacted. In response to this, a first processing call is made (step S108).

On the other hand, when the hardware calculation unit 4 is running (steps S107, No), the control device 1 notifies the hardware calculation unit 4 of the start timing (step S109).

Subsequently, the control device 1 determines whether or not the hard calculation unit 4 is starting (step S110), and when the hard calculation unit 4 is starting (step S110, Yes), the first by the hard calculation unit 4 It is determined whether or not the process is completed (step S111).

When the first process is completed in step S111 (step S111, Yes), the control device 1 makes an update process call for the calculation information 32 (step S112). Further, in the control device 1, when the hardware calculation unit 4 is stopped in step S110 (step S110, No), or when the first process is not completed in step S111 (step S111, No), the step. The process proceeds to S113.

Subsequently, the control device 1 determines whether or not it is the third processing timing for executing the third processing (step S113), and when it is the third processing timing (step S113, Yes). A third processing call is made (step S114), and processing is terminated. Further, if the processing in step S113 is not at the third processing timing (steps S113, No), the control device 1 ends the processing without going through the processing in step S114.

As described above, the control device 1 that controls the engine 100 (an example of the rotation mechanism) according to the embodiment, and the first process that performs the calculation by the neural network within the first period based on the rotation state of the engine 100. , The control unit 2 that starts the process based on the rotation state of the engine 100 and processes the second process with a higher priority than the first process by software, and hardware at least a part of the first process. A hardware calculation unit 4 that can be processed by wear is provided, and the control unit 2 activates the hardware calculation unit 4 when the rotation speed of the engine 100 is equal to or higher than a predetermined threshold value. Therefore, according to the control device 1 according to the embodiment, the required processing can be completed while suppressing the power consumption.

By the way, in the above-described embodiment, the case where the rotation mechanism is the engine 100 has been described, but the rotation mechanism is not limited to the engine 100, and may be another rotation mechanism such as a motor.

Further, in the above-described embodiment, the case where the priority of the arithmetic processing is lower than that of the high task processing has been described, but the present invention is not limited to this. It may be the case that the priority of the arithmetic processing is higher than that of the high task processing.

Further, in the above-described embodiment, the arithmetic processing is executed by the hard arithmetic unit 4, but the present invention is not limited to this, and other tasks may be executed by the hard arithmetic unit 4 in addition to the arithmetic processing. Good. In this case, the hardware calculation unit 4 for executing the calculation process and other tasks may be provided separately.

Further effects and variations can be easily derived by those skilled in the art. For this reason, the broader aspects of the invention are not limited to the particular details and representative embodiments described and described above. Therefore, various changes are possible without departing from the spirit or scope of the overall concept of the invention as defined by the appended claims and their equivalents.

1 Control device 2 Control unit 4 Hard calculation unit 21 Detection unit 22 Soft calculation unit 23 Import unit 24 Start control unit 100 Engine (an example of rotation mechanism)

Claims (5)

A control device that controls the rotation mechanism.
The first process of performing the calculation by the neural network within the first period based on the rotation state of the rotation mechanism and the process are started based on the rotation state of the rotation mechanism, and the processing is performed with a higher priority than the first process. A control unit that processes the second process with software,
It is provided with a hardware calculation unit capable of processing at least a part of the first processing by hardware.
The control unit
A control device characterized in that the hardware calculation unit is activated when the rotation speed of the rotation mechanism is equal to or higher than a predetermined threshold value. The control unit
The control device according to claim 1, wherein the first process and the third process, which performs the process with a lower priority than the second process, are performed. The control unit
The control device according to claim 1 or 2, wherein when the hardware calculation unit receives a notification of completion of a calculation by a neural network, a capture process for capturing the result data of the calculation is performed. The control unit
The control device according to claim 1, 2, or 3, wherein the hardware calculation unit is stopped when the rotation speed of the rotation mechanism becomes less than the predetermined threshold value. It is a control method that controls the rotation mechanism.
The first process of performing the calculation by the neural network within the first period based on the rotation state of the rotation mechanism and the process are started based on the rotation state of the rotation mechanism, and the processing is performed with a higher priority than the first process. The second process includes a control process in which software processes the second process.
The control step is
A control method comprising activating a hardware calculation unit capable of processing at least a part of the first processing by hardware when the rotation speed of the rotation mechanism is equal to or higher than a predetermined threshold value.

Images